Download MCB105 Motor Learning Lecture by Bence Olveczky 2015 Apr 8

MCB105 Motor Learning Lecture by Bence Olveczky
2015 Apr 8 annotated lecture notes
Motor exploration (trial and error) – evaluate performance – reinforce good motor
programs
Improved performance
Reduced variability
Tennis serve –beginner (high variability) vs professional (no variability)
How does the brain underlie motor skill learning and execution?
What models can be used for motor learning?
Primates
Songbirds
Songbirds – the experts
Rodents – the generalist
Zebrafinch
Beginner to expert
Sensory stage – listens to his father. Stereotyped syllables in stereotyped
sequence. Zebrafinches only have one motif
Motor stage (age 35days or so)
Starts to sing – at first just “babbling”, but improves over time. Just before sexual
maturity (90 days), song is perfected.
Syrinx (similar to larynx)
Syringeal muscles to change configuration of voice box
Air pressure through syrinx through control of respiratory muscles
Airsac pressure waves (due to respiratory muscles) at first all over the place
At 90 days, super stereotyped
Motor pathway: HVC (similar to supplementary motor cortex) – RA (robust
nucleus of archeopallium – equivalent to motor cortex) – Brainstem (innervates
muscles involved in singing – syringeal and respiratory).
Anterior forebrain pathway:
Record from HVC while bird is singing
Record from 8 neurons during HVC. Action potentials of different neurons occur
at a specifically time point.
What kind of circuit can produce this kind of activity pattern
A kind of “delay line”
Put the neurons in a chain (one neuron excites the next and so on) – synfire
chain
Chain propogates as a function of time.
Record from neurons in RA – each neuron drives one muscle/muscle
group. RA represents muscle (motor activity).
How do you test that HVC represents time?
Try to slow down signal propagation through network.
Different temperatures – postsynaptic response starts to be delayed
Temperature also affects signal propagation in Central Pattern Generator (in
insects)
Wing beat pattern is also a function of temperature
Also cricket song at different temperatures affected because central pattern
generator is affected. So temperature can affect synaptic transmission.
Birds maintain temperature at 42 deg
Can cool their brains with a peltier cooler (cool down HVC without cooling the
rest of the brain).
What does slowing down of HVC do?
Stretch the song uniformly (syllables become longer, and space between
syllables)
Back to RA’s representation – muscle activity
Each neuron fires multiple times.
How to connect HVC neurons to RA neurons to get specific RA activity
pattern?
This is the task the brain has to solve.
How to connect the correct HVC neurons (timekeeper) to specific RA neurons.
This is the learning they have to do. They have to find the right connections by
trial and error.
HVC temporal representations don’t need to be learned – the “clock/timekeeping”
signal is stereotyped during learning.
However, RA activity during learning has much more variable activity, especially
at the beginning.
Early in learning –RA neurons receive many inputs from HVC, but not strong.
Variable, no high frequency bursts
Later in learning – select synapses that are good, prune the ones that are not.
This leads to specific connectivity and specific firing patterns that lead to specific
muscle activation.
How to study HVC to RA connectivity
E.g. retrograde tracing (but won’t tell you about strength of connections)
Experiment:
More importantly, can stimulate HVC neuron fibers (axons) and patch onto single
RA neurons.
Increasing stimulus intensity recruits incrementally more HVC neurons.
Adults have higher synaptic strength
But lower number of synapses than juveniles.
How to explore HVC to RA connections in order to decide which is the best
connections?
LMAN induces variability in the song (LMAN is an experimenter). LMAN injects
noise into RA neurons. It is the output of the learning (AF) pathway.
If it promotes exploration, then silencing LMAN should get rid of that.
Experiment: Remove LMAN
No more experimentation/exploration.
LMAN is the source of motor exploration.
Record from LMAN neurons in young bird – they are highly variable when
aligned to a song.
Record from RA neurons in young bird – inactivate LMAN at the same time
(using microdialysis probes, inject lidocaine/GABA agonist to inhibit).
RA firing pattern becomes stereotyped.
At a particular time, there’s a specific connection pattern of firing between HVC
and RA (revealed when you silence LMAN)
Turn LMAN on – either a burst or nothing for one connection. Then either
strengthen or weaken one particular synapse. Pruning occurs.
Now, when you shut off LMAN again, you see a different stereotyped pattern,
because specific synapses have already been pruned during the few hours of
pruning/experimenting.
Variability is reduced after learning.
LMAN inputs to RA don’t change over development
But HVC – RA drive decreases
FI curve (firing rate as a function of input drive to RA –HVC+LMAN). Sigmoid.
During learning, medium strength connections, within the linear regime. LMAN
input can change RA firing very well.
After learning, either input to RA neuron is strong, or saturated (different regime
on the curve). LMAN firing no longer has consequences on HVC-RA
neurons.
System becomes robust to sources of noise.
How about a mammalian model?
Can train rodents to perform stereotyped motor sequences.
Specific sequence the rat has to learn:
Press 1, delay 700ms, press 2, reward. Operant conditioning.
Reward a range of delays around 700ms. Later on, reduce the range.
Over many trials – he converges on around 700ms. About a month of training.
Task is unconstrained – each animal does it slightly differently, in order to keep
time. Sometimes weird behaviors get rewarded – e.g. sticking out tongue
Credit assignment is difficult, so sometimes unnecessary behaviors are rewarded
and reinforced as well.
Their behavior starts variable, and then becomes precise.
What is an important part of the brain for this behavior?
Cortex?
Basal ganglia?
Cerebellum?
Lesion motor cortex (all of it), in animals that have learned motor sequences.
Their performance is just as good.
Lesion motor cortex before they start learning. They cannot learn the correct
behavior.
Motor cortex is not required for executing learned motor skills, but required for
learning.
Motor cortex can reprogram basic subcortical circuits (a tutor of subcortical
circuits, although probably also a controller).