WEBVTT

00:00:01.997 --> 00:00:05.517
This is Paul Verschure with the Convergent Science Network podcast.

00:00:06.097 --> 00:00:11.017
And today I'm with one of the regular visitors to our summer school here in

00:00:11.017 --> 00:00:13.757
Barcelona, Jerry Haslow.

00:00:14.677 --> 00:00:22.177
And Jerry is with us now. He's visiting Barcelona. And he talked about his recent work on the cerebellum.

00:00:22.977 --> 00:00:28.277
And um so jerry one thing maybe as an introduction but what i always found fascinating

00:00:28.277 --> 00:00:32.777
in in your case is that sort of you moved from psychology and philosophy,

00:00:33.517 --> 00:00:39.857
into not just neuroscience but a very sort of hardcore version of of neuroscience

00:00:39.857 --> 00:00:47.417
right so so how do you how do you see that transition for yourself well i wanted

00:00:47.417 --> 00:00:50.897
when i started psychology I was in a department that was mainly Freudian,

00:00:50.957 --> 00:00:56.117
and I wanted to study psychology from a more scientific point of view.

00:00:56.277 --> 00:01:01.597
And I looked at various departments and laboratories, and I found that there

00:01:01.597 --> 00:01:06.057
was a group in my hometown in Lund working on the cerebellum.

00:01:06.097 --> 00:01:08.517
They were doing things that I found extremely boring.

00:01:08.637 --> 00:01:15.497
They were looking at pathways through the spinal cord up to the cerebellum.

00:01:15.497 --> 00:01:21.017
And I found that very, very boring since I was interested in learning and memory

00:01:21.017 --> 00:01:23.237
and stuff like that, physiological psychology.

00:01:23.877 --> 00:01:26.857
But the quality of the research was extremely good.

00:01:26.937 --> 00:01:34.437
And I've come to the view that this is really, really crucial for probably most of science.

00:01:34.437 --> 00:01:41.597
The quality of the research, when things are done in a careful, precise,

00:01:42.037 --> 00:01:49.017
painstaking, systematic way, this is much more important than working in trendy

00:01:49.017 --> 00:01:51.977
subjects and following fashion and so on.

00:01:52.297 --> 00:01:56.237
So even though I found it boring, I found that the quality of the research being

00:01:56.237 --> 00:01:58.797
done in that department was just so good.

00:01:59.317 --> 00:02:04.037
I felt that these guys know what they're doing. This is what I want to learn.

00:02:04.437 --> 00:02:11.997
So I started working on the cerebellum with Carl Friedrich Eckert,

00:02:12.137 --> 00:02:18.757
Gert Andersen, and a few others just to learn the techniques and to try and

00:02:18.757 --> 00:02:20.157
emulate their way of thinking,

00:02:20.377 --> 00:02:25.417
their carefulness, and the systematic way they were doing research.

00:02:25.417 --> 00:02:30.837
And then, just by luck, it turned out that a certain form of associative memory,

00:02:31.137 --> 00:02:35.397
Pavlovian conditioning, actually occurred in the cerebellum.

00:02:35.797 --> 00:02:40.817
So, for me, that was just fantastic because it meant that I could utilize my

00:02:40.817 --> 00:02:43.657
knowledge of cerebellar physiology and...

00:02:45.833 --> 00:02:52.313
Address a task that I felt was also very interesting, namely the formation of associative memory.

00:02:52.593 --> 00:02:57.813
So that's how I ended up doing work on the cerebellum on Poblovian eye-blinking conditioning.

00:02:58.333 --> 00:03:01.113
So that's basically what you've been doing for the last 25 years.

00:03:01.213 --> 00:03:03.913
Exactly. And it's been extremely difficult.

00:03:04.093 --> 00:03:11.593
We've spent a lot of time trying to create an experimental setup that is workable.

00:03:12.273 --> 00:03:17.213
There are many constraints when you do research. You have the ethical constraints, for instance.

00:03:17.373 --> 00:03:23.033
You cannot do this on any kind of animal. You cannot do it with any kind of anesthetics and so on.

00:03:24.033 --> 00:03:27.533
But we found a way to solve these problems.

00:03:27.713 --> 00:03:35.093
But it's taken, I would say, it took more than 15 years to almost 20 years to

00:03:35.093 --> 00:03:37.193
make the preparation really workable.

00:03:38.093 --> 00:03:45.733
Now, finally, results are pouring in. It's been a very long travel to get there.

00:03:46.453 --> 00:03:51.733
So how do you map Pavlovian conditioning onto the brain, and in particular onto the cerebellum?

00:03:52.693 --> 00:03:54.133
What's the relationship there?

00:03:55.633 --> 00:04:00.153
Well, Pavlovian conditioning is a very general phenomenon. There are many response

00:04:00.153 --> 00:04:01.893
systems in which you can get it.

00:04:02.053 --> 00:04:06.453
I mean, Pavlov used autonomic conditioning. He showed how you could train dogs

00:04:06.453 --> 00:04:12.753
to salivate when they heard a sound by pairing the sound with food in the dog's mouth.

00:04:13.653 --> 00:04:20.013
We can also use the same learning paradigm to.

00:04:22.158 --> 00:04:30.418
How should I put it, to look at how certain immune responses occur,

00:04:30.818 --> 00:04:37.198
cognitive aspects of learning, how your heart rate changes when you are told

00:04:37.198 --> 00:04:38.338
something frightening.

00:04:39.278 --> 00:04:45.318
How asthmatic patients can get asthmatic attacks when they look at pictures

00:04:45.318 --> 00:04:49.238
of flowers or smell flowers and things like that.

00:04:49.938 --> 00:04:54.378
There are many, many, many applications to this, and we are only studying one

00:04:54.378 --> 00:04:57.018
of them, namely Eibling conditioning, which is a motor response.

00:04:57.558 --> 00:05:01.778
We know that this is in the cerebellum, but probably the other forms of conditioning,

00:05:01.858 --> 00:05:05.598
or many of the other forms of conditioning, involve other parts of the brain.

00:05:06.058 --> 00:05:11.218
But it's a very simple and very basic form of learning. There is the Nobel Prize

00:05:11.218 --> 00:05:17.238
winner Peter Medawar once wrote a collection of essays called The Art of the Soluble.

00:05:17.238 --> 00:05:20.498
And by that he meant that scientists should

00:05:20.498 --> 00:05:24.098
not only address the important problems they should also choose

00:05:24.098 --> 00:05:27.698
problems that are really soluble and

00:05:27.698 --> 00:05:30.538
that often means you should

00:05:30.538 --> 00:05:36.398
start with the simple problems and wait with the really huge task like explaining

00:05:36.398 --> 00:05:40.158
cognition and consciousness and stuff like that okay but now in pavlovian conditioning

00:05:40.158 --> 00:05:44.138
we actually have a very reduced setup right you just have you're talking about

00:05:44.138 --> 00:05:48.018
two kinds of stimuli yes in the end one kind of response you're going to get

00:05:48.018 --> 00:05:49.078
out of the system, right?

00:05:49.158 --> 00:05:54.718
So how do you... And these are also very simple responses. They're stereotyped responses.

00:05:54.898 --> 00:06:01.038
Exactly. So how do you map that very reduced learning system now from the behavioral perspective?

00:06:01.948 --> 00:06:05.528
Onto the neural substrate, right? So how does something like this conditioned

00:06:05.528 --> 00:06:09.408
stimulus, this initially neutral stimulus, now map onto this preparation you developed?

00:06:09.788 --> 00:06:13.288
Or how do you map your unconditioned stimulus onto that system and so on?

00:06:13.448 --> 00:06:18.008
I'm not sure what you mean by how do I map it. Well, you have a functional relationship.

00:06:18.328 --> 00:06:21.268
You know that if you present this conditioned stimulus paired with an unconditioned

00:06:21.268 --> 00:06:24.328
stimulus, you will be able to trigger conditioned response over time.

00:06:24.508 --> 00:06:28.988
So that means you have to now identify the pathways that actually transduce

00:06:28.988 --> 00:06:30.288
this information to the brain.

00:06:30.288 --> 00:06:34.808
Yeah, but we know that we can, in our case, we map it on the cerebellum and

00:06:34.808 --> 00:06:40.488
there's a lot of experimental work before us where people have showed you can

00:06:40.488 --> 00:06:42.828
get these in animals that have no forebrain.

00:06:43.068 --> 00:06:47.188
You can remove a huge part of the brain and still get conditioning.

00:06:47.188 --> 00:06:55.108
Conditioning, we have developed this into a preparation where we remove most

00:06:55.108 --> 00:06:58.888
of the brain, all the parts of the brain that maintain cognitive function and

00:06:58.888 --> 00:07:00.328
consciousness and so on are removed,

00:07:00.588 --> 00:07:03.728
and we have the cerebellum and the brainstem left intact.

00:07:03.928 --> 00:07:07.788
We can still get conditioning, and it seems to follow basically the same rules

00:07:07.788 --> 00:07:13.228
that you find in the intact animal. It takes roughly the same amount of time

00:07:13.228 --> 00:07:15.228
and training to acquire the response.

00:07:17.688 --> 00:07:22.268
They extinguish in the same way and so on. So it looks very, very similar.

00:07:22.648 --> 00:07:29.008
And then we have tried to identify precisely which cells in the cerebellum learn this.

00:07:29.168 --> 00:07:32.408
And we show that these cells, they have particular properties,

00:07:32.528 --> 00:07:34.828
they have particular input characteristics and so on.

00:07:34.828 --> 00:07:46.408
So we can, within a few tens of a microns, we can localize the cells that learn this task.

00:07:46.548 --> 00:07:50.428
And we can record from precisely those cells. And we can...

00:07:51.541 --> 00:07:56.941
The findings we have can account for almost everything you find in the behavior.

00:07:57.881 --> 00:08:03.521
So we have cellular responses, responses in the particular group of Purkinje

00:08:03.521 --> 00:08:05.101
cells in the cerebellar cortex.

00:08:05.761 --> 00:08:10.461
They have the temporal properties of the behavior. They respond in the same

00:08:10.461 --> 00:08:14.141
way when you change various conditions during training.

00:08:14.141 --> 00:08:17.361
And that means that we we know within

00:08:17.361 --> 00:08:20.961
a very very very narrow area very

00:08:20.961 --> 00:08:23.781
small area where the learning takes

00:08:23.781 --> 00:08:27.041
place so how do these cells learn i mean on on

00:08:27.041 --> 00:08:29.921
these cells we have convergence pavlov already talked about looking

00:08:29.921 --> 00:08:33.201
for convergence to find the site of the n-gram or the memory trace

00:08:33.201 --> 00:08:37.481
so the condition stimulus information comes

00:08:37.481 --> 00:08:40.441
up to this perkinia cell through the parallel fibers

00:08:40.441 --> 00:08:43.241
there is a granule cells this might tell you something

00:08:43.241 --> 00:08:46.181
about the world now we have our climbing fiber coming

00:08:46.181 --> 00:08:49.281
in so it conversions on this Purkinje cell so you

00:08:49.281 --> 00:08:52.161
could argue well look that seems a fairly trivial setup to

00:08:52.161 --> 00:08:57.041
learn an association right well actually the Purkinje cells are ideally suited

00:08:57.041 --> 00:09:02.241
to make associative connections they are unique in the nervous system in the

00:09:02.241 --> 00:09:06.641
degree of convergence of inputs one Purkinje cell can receive several hundred

00:09:06.641 --> 00:09:10.921
thousand different inputs from the so-called parallel fibers.

00:09:11.281 --> 00:09:17.841
So a single Purkinje cell will be contacted by maybe a quarter of a million,

00:09:17.941 --> 00:09:19.541
maybe half a million parallel fibers.

00:09:20.561 --> 00:09:24.821
And the parallel fibers will signal everything that happens to the organism.

00:09:24.861 --> 00:09:29.221
All the sensory systems project via the parallel fibers of the Purkinje cell.

00:09:29.481 --> 00:09:34.821
Also, everything that happens in In the brain, when you think about things,

00:09:35.041 --> 00:09:38.821
when you plan ahead, when you imagine you're seeing things, all of that information

00:09:38.821 --> 00:09:43.001
is also sent via the parallel fibers to the Purkinje cells.

00:09:43.341 --> 00:09:48.021
So this is a perfect system for generating associations.

00:09:49.421 --> 00:09:55.661
And once you've seen this degree of convergence, it's not a surprise that this

00:09:55.661 --> 00:09:57.521
is where associative learning takes place.

00:09:57.521 --> 00:10:01.781
Well, but it seems a very special kind of associative learning because it's

00:10:01.781 --> 00:10:07.341
not that you just associate any idea with any other idea, to use a more Jungian interpretation,

00:10:07.841 --> 00:10:14.121
but you seem to associate in a very convergent fashion a large set of possible

00:10:14.121 --> 00:10:16.421
states onto one output state.

00:10:16.621 --> 00:10:19.121
Yeah. Right, so it's not just any form of associative learning,

00:10:19.341 --> 00:10:22.201
but it seems a very specific kind of associative learning. Yes,

00:10:22.261 --> 00:10:29.101
but you have many output channels also, small groups of Purkinje cells project to their own targets.

00:10:29.541 --> 00:10:33.161
Most of them project to muscles, but there are also Purkinje cells projecting

00:10:33.161 --> 00:10:38.621
to the prefrontal cortex and to the parts of the brain we believe are the centers

00:10:38.621 --> 00:10:40.341
of various cognitive functions.

00:10:43.008 --> 00:10:50.908
And since all these Purkinje cells projecting to small groups of muscles or

00:10:50.908 --> 00:10:53.968
areas of the cognitive parts of the forebrain,

00:10:54.088 --> 00:10:58.948
they all receive this convergent input from all the other systems.

00:10:58.948 --> 00:11:02.248
But what do you then see as the output unit?

00:11:02.388 --> 00:11:07.668
Because in the eye-blink conditioning case, which is a standard paradigm that

00:11:07.668 --> 00:11:14.128
you also have pursued vigorously, the output is a discrete movement in the end, right?

00:11:14.588 --> 00:11:20.228
So now you also say that, but other Purkinje cells would indirectly project

00:11:20.228 --> 00:11:22.228
towards the frontal cortex.

00:11:22.508 --> 00:11:26.888
Yes. So what's now this output unit of the system? Well, if you go,

00:11:27.008 --> 00:11:30.768
some of the projections from the cerebellum goes to the primary motor cortex.

00:11:30.988 --> 00:11:36.788
And there you have pyramidal cells projecting also to elementary movements or

00:11:36.788 --> 00:11:37.968
small groups of muscles.

00:11:38.248 --> 00:11:42.768
So that outward pathway is rather similar to what we're looking at in eye blink conditioning.

00:11:43.168 --> 00:11:48.908
But then as you go more forward in the forebrain, you will reach areas that

00:11:48.908 --> 00:11:54.608
control more global aspects of actions.

00:11:55.528 --> 00:12:00.548
Perhaps you shouldn't even call them movements because they are actions in a more abstract sense.

00:12:00.788 --> 00:12:06.548
So the more rostral, the more forward you come, the more complex the movements

00:12:06.548 --> 00:12:08.628
will be controlled by that area.

00:12:09.068 --> 00:12:16.568
And when you come, suppose that I said that I want to draw a triangle.

00:12:16.568 --> 00:12:19.388
Then on my primary motor cortex there

00:12:19.388 --> 00:12:22.248
will be commands for holding the pen

00:12:22.248 --> 00:12:25.208
and moving your arm

00:12:25.208 --> 00:12:29.628
in a certain direction if you go a little bit forward you will have a command

00:12:29.628 --> 00:12:35.388
signal controlling let's say draw a horizontal line and if you go even further

00:12:35.388 --> 00:12:39.668
forward than that you will have more global commands like drawing a triangle

00:12:39.668 --> 00:12:44.108
or even drawing a house or something like that And the more forward you go,

00:12:44.328 --> 00:12:48.008
the more global will the commands be, the more abstract will they be,

00:12:48.128 --> 00:12:51.568
or the more abstract will the actions be that are controlled from that area.

00:12:51.808 --> 00:12:58.828
So it looks as if the cerebellum can control not only the sort of the fine detailed

00:12:58.828 --> 00:13:03.408
parts of a movement like squeeze the pen to hold it,

00:13:03.488 --> 00:13:07.808
but also more global actions like draw a triangle or things like that.

00:13:07.808 --> 00:13:11.848
But would that not mean that, like in the eye-blink conditioning case,

00:13:12.088 --> 00:13:16.568
you were describing the output of the system as groups of muscles.

00:13:16.908 --> 00:13:22.248
But it actually is, this projection is only indirectly targeting these muscles

00:13:22.248 --> 00:13:26.248
because it hits brainstem motor nuclei.

00:13:27.068 --> 00:13:30.188
So also there wouldn't be fair to say that you control an action as opposed

00:13:30.188 --> 00:13:31.528
to groups of muscles? No.

00:13:32.160 --> 00:13:33.380
Well, you might, I suppose.

00:13:35.440 --> 00:13:41.800
And I also suppose that if you take eye blink conditioning, which is in some

00:13:41.800 --> 00:13:46.680
ways a special case, but it will activate the facial nucleus and send signals out of the eyelid.

00:13:46.780 --> 00:13:50.000
If you're looking at other forms, in a rabbit, for instance,

00:13:50.100 --> 00:13:53.580
it's retraction of the eyeball and not only the eyelid and so on.

00:13:53.580 --> 00:14:02.940
But in any case, the facial nucleus will not only respond then to change in

00:14:02.940 --> 00:14:06.440
its input signals controlling the movement of the eyelid.

00:14:06.640 --> 00:14:09.920
There must be other mechanisms controlling, let's say, the amplitude of the

00:14:09.920 --> 00:14:13.580
movement. That could also be the cerebellum in a different part of the cerebellum.

00:14:14.580 --> 00:14:19.660
And maybe that will also be controlled by signals coming from the forebrain.

00:14:19.660 --> 00:14:24.140
Like in this particular situation, it's not very good to close your eyes because

00:14:24.140 --> 00:14:32.260
you need to have a look at the dangerous environment you're being in at the

00:14:32.260 --> 00:14:33.300
moment, things like that.

00:14:34.120 --> 00:14:38.340
So, of course, in a certain sense, the facial nucleus will have to integrate

00:14:38.340 --> 00:14:42.700
or compromise between signals coming from the cerebellum but also coming from

00:14:42.700 --> 00:14:45.660
other parts of the brain and from the periphery.

00:14:47.060 --> 00:14:53.320
So now we have the whole setup, right? We have sort of our conditioned stimuli coming in.

00:14:53.380 --> 00:14:57.180
This is basically all possible states of the world or internal states of my

00:14:57.180 --> 00:14:59.660
brain that come in via the pons over these parallel fibers.

00:15:00.260 --> 00:15:04.040
I have these climbing fibers that are much more specifically targeting these

00:15:04.040 --> 00:15:09.540
Purkinje cells linked often to very specific events on the body,

00:15:09.620 --> 00:15:13.020
like unconditioned stimuli, painful stimuli, and noxious stimuli, and so on.

00:15:13.020 --> 00:15:17.180
And now we have this output pathway, which is a strong divergent pathway,

00:15:17.300 --> 00:15:22.860
or conversion pathway again, onto motor nuclei or onto frontal areas of the brain.

00:15:22.960 --> 00:15:28.220
Okay, so now you could argue, well, and indeed, because of this very almost

00:15:28.220 --> 00:15:31.980
clean, what people call crystalline design of this circuit,

00:15:32.360 --> 00:15:36.480
its operations, or some people like to call this its computations,

00:15:36.720 --> 00:15:39.060
might be fairly trivial.

00:15:39.060 --> 00:15:44.060
There have been models of this already by Marr and Albus and Ito since the 70s and early 80s.

00:15:46.340 --> 00:15:49.760
You could argue, okay, but given this anatomical arrangement and given this

00:15:49.760 --> 00:15:54.220
physiology, the computational principles appear clear because you just learned

00:15:54.220 --> 00:15:55.680
to trigger this response.

00:15:55.900 --> 00:16:00.300
You have to bridge some sort of little time interval between the onset of one stimulus and the other.

00:16:02.003 --> 00:16:05.323
What's really the problem of this system? Why is it so difficult to understand

00:16:05.323 --> 00:16:08.683
how this really works? Where are the problems coming in? I don't think it is very difficult. Okay.

00:16:09.263 --> 00:16:12.003
I think... Yeah, but still, you're performing experiments every day,

00:16:12.063 --> 00:16:13.543
so apparently they're on these old problems.

00:16:13.763 --> 00:16:17.283
I mean, we're trying to sort out many of the details, but I think that in some

00:16:17.283 --> 00:16:20.703
basic stuff, we understand pretty well now.

00:16:21.003 --> 00:16:25.923
We do. I think we do understand that all the sensory information is transmitted

00:16:25.923 --> 00:16:28.663
via the mossy fiber and the parallel fibers of the Purkinje cells.

00:16:28.923 --> 00:16:33.283
The climbing fibers provide a teaching signal to change the circuitry.

00:16:33.343 --> 00:16:39.783
The pre-kinder cell has the ability to output, to generate well-timed outputs that control behavior.

00:16:40.163 --> 00:16:42.663
I think this is now pretty clear.

00:16:43.123 --> 00:16:48.263
A lot of the controversy surrounding the cerebellum is that a lot of people

00:16:48.263 --> 00:16:50.943
who work, perhaps those who are working on different setups,

00:16:51.583 --> 00:16:56.623
don't accept the answers that I claim are the right ones.

00:16:57.043 --> 00:17:01.283
But there's another complication, And that is that the cerebellum,

00:17:01.283 --> 00:17:03.503
as you said, has a crystalline structure.

00:17:03.743 --> 00:17:07.423
It looks very similar throughout the cerebellar cortex, wherever you are in the cerebellum.

00:17:07.723 --> 00:17:10.263
And that has fostered the assumption

00:17:10.263 --> 00:17:14.243
that the cerebellum works in the same way in all different parts.

00:17:14.383 --> 00:17:20.483
However, it is known and has been for some time that there are biochemical differences

00:17:20.483 --> 00:17:22.643
between different zones in the cerebellum.

00:17:22.763 --> 00:17:27.063
And those biochemical differences must correspond to some sort of functional

00:17:27.063 --> 00:17:30.623
difference. Like what kind of difference would stand out?

00:17:31.343 --> 00:17:35.963
I'm not sure what they are. And this has not been investigated because almost

00:17:35.963 --> 00:17:38.843
all the in vitro work has been done in the vermis.

00:17:39.283 --> 00:17:43.883
People have taken out slices from the vermis and put them in a dish and try

00:17:43.883 --> 00:17:46.783
to explore learning properties and so on.

00:17:46.943 --> 00:17:50.503
But a large part of the cerebellum have never been subjected to that kind of

00:17:50.503 --> 00:17:54.563
analysis. And if you look at what different parts of the cerebellum do,

00:17:54.843 --> 00:17:59.423
it's not difficult to envisage that they might learn quite different things

00:17:59.423 --> 00:18:01.883
and respond in different ways.

00:18:02.683 --> 00:18:08.923
This is very speculative on my part, but if you take, for instance,

00:18:09.183 --> 00:18:14.723
a medial part of the cerebellum that controls posture and axial musculature.

00:18:15.903 --> 00:18:27.723
It's difficult to see that the cerebellum would be very useful if it only produced a kind of phasic,

00:18:27.723 --> 00:18:34.503
fairly short latency and very short lasting movements that are generated by

00:18:34.503 --> 00:18:36.923
a more lateral area that generates eye blink.

00:18:36.923 --> 00:18:42.503
So, pre-kinesio-cell controlling eye blink, you would expect it to generate

00:18:42.503 --> 00:18:47.823
a short-lasting fast movement or signal.

00:18:49.544 --> 00:18:52.644
But if you go to the part of the vermis controlling posture,

00:18:53.124 --> 00:18:58.444
you wouldn't want, you want someone to be able to stand still for a while.

00:18:58.564 --> 00:19:03.444
And it's not useful then to generate very, very short lasting fast movements.

00:19:03.704 --> 00:19:07.524
So my guess is that you would find that those parts of the cerebellum work in

00:19:07.524 --> 00:19:10.324
a slightly different way. They may have very different temporal properties.

00:19:11.624 --> 00:19:14.684
But look, you could also argue that that transformation is

00:19:14.684 --> 00:19:17.884
performed in these downstream brainstem stem nuclei that

00:19:17.884 --> 00:19:21.064
they sort of perform some sort of temporal matching

00:19:21.064 --> 00:19:23.944
of these output signals to the properties of the periphery you

00:19:23.944 --> 00:19:26.884
can but your original question was why has this and why

00:19:26.884 --> 00:19:30.044
has it been so difficult to agree on how the cerebellum works

00:19:30.044 --> 00:19:35.444
and one one guess that i have is that it has always been assumed that the cerebellum

00:19:35.444 --> 00:19:39.524
works in the same way everywhere but people working on eibling conditioning

00:19:39.524 --> 00:19:45.624
work in one part of the cerebellum those who work on in vitro on slices work

00:19:45.624 --> 00:19:47.224
in a different part of the cerebellum,

00:19:47.284 --> 00:19:52.504
and those working on the other main experimental paradigm,

00:19:52.664 --> 00:19:55.504
namely the adaptation of the vestibulo-ocular reflex,

00:19:55.844 --> 00:19:58.624
they work in a third area of the cerebellum.

00:19:58.884 --> 00:20:02.524
And maybe the properties of the cerebellum are different in all these areas.

00:20:02.704 --> 00:20:07.024
And that could also be one reason why it's been so difficult to reach agreement.

00:20:07.024 --> 00:20:10.264
But these three paradigms, how much of the cerebellum do they really cover?

00:20:11.064 --> 00:20:14.604
Well, they cover only a small part of the cerebellum. We are working in the

00:20:14.604 --> 00:20:16.964
intermediate part known as the C3 zone.

00:20:18.604 --> 00:20:22.884
The slice people work mainly on the A zone. Maybe they get some B zone.

00:20:22.984 --> 00:20:24.944
I don't know, depending on how careful they are.

00:20:25.404 --> 00:20:31.584
And the vestibular ocular reflex people are working on the flocculus,

00:20:31.644 --> 00:20:33.464
a very old part of the cerebellum.

00:20:33.524 --> 00:20:38.004
And it could be that there are significant differences between these areas.

00:20:38.124 --> 00:20:40.944
So in total, this is not more than 10%, I would guess.

00:20:41.204 --> 00:20:44.664
No, less than that. Less, okay. Okay, I was being generous.

00:20:45.884 --> 00:20:52.364
Okay. Yeah. So the problem is that if you say, okay, it might not be a uniformly

00:20:52.364 --> 00:20:59.764
operating learning system, do you have any physiological or direct anatomical evidence for that?

00:20:59.844 --> 00:21:02.304
Or is it only in terms of functional considerations?

00:21:02.884 --> 00:21:06.924
Well, it's functional considerations coupled with the fact that we know that

00:21:06.924 --> 00:21:11.584
there are biochemical differences. People working with anatomical techniques

00:21:11.584 --> 00:21:16.164
have found that there are some clear chemical differences between different

00:21:16.164 --> 00:21:17.244
parts of the cerebellum.

00:21:17.384 --> 00:21:24.684
One of the most well-known is the zebrin bands, which correspond to an enzyme

00:21:24.684 --> 00:21:30.484
called aldolase, which is important for basic metabolism of the cells.

00:21:30.844 --> 00:21:36.244
And it's very difficult to understand why should there be bands of Purkinje

00:21:36.244 --> 00:21:39.804
cells with different levels of aldolase. Mm-hmm. Um.

00:21:40.965 --> 00:21:44.125
If it doesn't correspond to some sort of functional difference between them.

00:21:44.445 --> 00:21:45.385
Yeah, that's reasonable.

00:21:45.945 --> 00:21:49.425
So you said earlier, which I liked actually, you said, okay,

00:21:49.465 --> 00:21:53.525
I know how this system works, but they don't want to listen.

00:21:53.845 --> 00:21:59.825
So how does this system then work when we take the eye-blink conditioning example, right?

00:21:59.885 --> 00:22:03.605
So here comes the sound, a bit later comes the air puff to the cornea.

00:22:03.765 --> 00:22:06.565
Well, of course. What's happening in this system now? We don't know a lot about

00:22:06.565 --> 00:22:09.645
the details here, but what we know is that Sketch it out. Yeah,

00:22:09.785 --> 00:22:14.985
the Purkinje cell controlling the eyelid is only a small fraction of the Purkinje

00:22:14.985 --> 00:22:18.145
cells, but it is a distinct group of Purkinje cells.

00:22:18.505 --> 00:22:24.545
They will receive input from tones and light and skin pressure and everything

00:22:24.545 --> 00:22:26.285
that happens to the organism all the time.

00:22:26.285 --> 00:22:33.825
But now and then, a certain stimulus like a tone will occur and be followed

00:22:33.825 --> 00:22:36.385
by some kind of stimulus to the eye,

00:22:36.525 --> 00:22:45.025
like air on the cornea or an insect irritating the skin around the eye or something like that.

00:22:45.285 --> 00:22:51.325
And that will produce a signal in a particular pathway known as the climbing

00:22:51.325 --> 00:22:53.365
fibers going up to the same pre-kinder cells.

00:22:53.565 --> 00:22:59.725
And that climbing fiber impulse will change the synapses between the parallel

00:22:59.725 --> 00:23:01.765
fibers and the Purkinje cells.

00:23:02.005 --> 00:23:07.005
And it will only change those parallel fibers that were just active at the time

00:23:07.005 --> 00:23:09.765
we're considering. And those happen to be tone.

00:23:10.305 --> 00:23:18.285
So the synapses between the tone carrying signal and the climbing fiber input,

00:23:18.365 --> 00:23:21.185
those synapses will be changed by the climbing fiber input.

00:23:21.185 --> 00:23:26.585
And that will make the Purkinje cell more likely, the next time this happened,

00:23:26.825 --> 00:23:33.005
to generate an output signal in response to the tone, before the clavicle fiber,

00:23:33.065 --> 00:23:35.805
before the corneal or eye stimulation.

00:23:36.305 --> 00:23:39.365
But how does the Purkinje cell then trigger that response?

00:23:40.345 --> 00:23:44.565
That is totally unknown. Okay. And very controversial.

00:23:44.985 --> 00:23:49.005
Okay. Yeah. How come? What's the difficulty there? Well, one difficulty is to

00:23:49.005 --> 00:23:53.165
account for the timing. if the dominant idea...

00:23:54.595 --> 00:23:59.675
In the neurobiology of learning has been for many decades, I would say,

00:23:59.775 --> 00:24:03.955
the idea that you can change the strength of a synapse.

00:24:04.715 --> 00:24:08.355
Synapses can be depressed or they can be potentiated.

00:24:08.515 --> 00:24:13.935
And the terms long-term depression and long-term potentiation are usually used

00:24:13.935 --> 00:24:16.095
to designate these two forms of learning.

00:24:16.395 --> 00:24:20.755
And this has been the standard assumption in not only the cerebellum,

00:24:20.755 --> 00:24:24.215
but in all ideas about learning in the brain.

00:24:25.315 --> 00:24:32.575
The difficulty has been that the learned eye blink, the conditioned eye blink, is well-timed.

00:24:33.455 --> 00:24:38.215
It reaches, if you have a particular interval between the onset of the tone

00:24:38.215 --> 00:24:42.515
and the onset of the air puff, like 300 milliseconds or 500 milliseconds,

00:24:42.975 --> 00:24:48.355
the Purkinje cell response is adaptively timed to the interval between these

00:24:48.355 --> 00:24:50.415
two stimuli that occurred during training.

00:24:50.975 --> 00:24:56.035
And how can you account for the delay? If you use 300 milliseconds,

00:24:56.495 --> 00:25:00.095
the Purkinje cell will start firing or change its firing early,

00:25:00.275 --> 00:25:05.095
and it will reach a maximum amplitude at roughly the time when the unconditioned

00:25:05.095 --> 00:25:06.375
stimulus or the air puff comes on.

00:25:06.555 --> 00:25:13.175
If you use 500 milliseconds instead, the response will be delayed and adapted

00:25:13.175 --> 00:25:14.595
to the 500 millisecond interval.

00:25:15.295 --> 00:25:21.015
Now, how can you account for a timed response if you have only increases or

00:25:21.015 --> 00:25:22.435
decreases in stimulus strength?

00:25:25.235 --> 00:25:31.175
Just changing the strength of the response to the tone signal could explain

00:25:31.175 --> 00:25:35.495
why the Purkinje cell responds to tone, but it doesn't explain how that response

00:25:35.495 --> 00:25:38.355
can be delayed by a couple of hundred milliseconds or so.

00:25:38.555 --> 00:25:42.555
So this has been the main problem, that how do you account for the timing?

00:25:42.555 --> 00:25:46.595
And there have been many ideas about how this could be achieved.

00:25:46.675 --> 00:25:49.215
For instance, you could have delays in the input signal.

00:25:50.215 --> 00:25:55.175
This has been an idea that goes back many decades. It's called tapped delay lines.

00:25:55.635 --> 00:26:00.335
And the idea was that if you had a delay, let's say, in the mossy fibers going

00:26:00.335 --> 00:26:04.895
up to the cerebellum or delays in the granule cells and parallel fiber signal into the kidney cells,

00:26:05.735 --> 00:26:11.355
then you could have a synaptic change only between those parallel fibers Bruce.

00:26:12.915 --> 00:26:18.535
That had the right delay so that the signals would coincide with the input coming

00:26:18.535 --> 00:26:20.415
through the climbing fibers from the air puff.

00:26:20.935 --> 00:26:24.995
And that would give you an automatic way of timing because then only those parallel

00:26:24.995 --> 00:26:29.175
fibers would respond after learning. And that would give you automatic timing.

00:26:29.595 --> 00:26:36.495
We have in Lund conducted experiments suggesting that there are no such delays in the input signals.

00:26:37.255 --> 00:26:41.995
And even if we circumvent them by stimulating, Well, a few years ago,

00:26:42.035 --> 00:26:46.115
we tried to stimulate the mossy fibers, and we got well-timed responses showing

00:26:46.115 --> 00:26:52.095
that it couldn't be delays in the mossy fibers because we had delays after the mossy fibers.

00:26:52.495 --> 00:26:56.155
So then recently, we have stimulated the parallel fibers instead,

00:26:56.315 --> 00:27:02.975
and we still get nicely timed responses suggesting that it cannot be delays in the parallel fibers.

00:27:02.975 --> 00:27:08.555
It has to be something closer to the Bikindi cell, either cortical interneurons

00:27:08.555 --> 00:27:11.055
or the Bikindi cells themselves.

00:27:11.595 --> 00:27:16.995
Well, but what you're invalidating with these experiments is the notion of a

00:27:16.995 --> 00:27:21.395
tap-delay line, which would be that you have one spike basically traveling through

00:27:21.395 --> 00:27:25.495
this parallel fiber that signals, well, the interstimulus interval is 100 milliseconds.

00:27:25.855 --> 00:27:29.575
And it's on that one spike that you have to learn, right? But you cannot exclude

00:27:29.575 --> 00:27:35.415
the possibility that you have a more continuous fluctuating response on these

00:27:35.415 --> 00:27:37.535
parallel fibers that sort of represents interstem.

00:27:37.915 --> 00:27:45.375
This is also a problem for the LTD, LTP story that if you, yes,

00:27:45.475 --> 00:27:47.275
when you provide a tone signal,

00:27:47.575 --> 00:27:52.575
the mossy fibers and the parallel fibers will go on firing throughout the whole

00:27:52.575 --> 00:27:53.895
period that you have the tone.

00:27:53.895 --> 00:27:58.495
And you can get if the tone continues well after the air puff.

00:27:59.430 --> 00:28:02.290
The tone can continue several hundred milliseconds after the air puff,

00:28:02.790 --> 00:28:05.990
but that doesn't produce a continuous response.

00:28:06.090 --> 00:28:09.950
The response goes on at a certain time, but it also goes off at a certain time.

00:28:10.710 --> 00:28:14.370
Even if the condition simulates continuous of many hundreds of milliseconds,

00:28:14.670 --> 00:28:16.790
the response still goes off at the right time.

00:28:17.270 --> 00:28:21.290
But wait, I could argue that that is then accomplished by the climbing fiber signal.

00:28:21.770 --> 00:28:28.610
Yes, but the same thing happens if you give a C as alone or tone alone trial. the same thing happens.

00:28:28.750 --> 00:28:32.270
The response goes off at the time when the climbing fiber response usually comes

00:28:32.270 --> 00:28:34.090
in. So it doesn't really matter.

00:28:34.390 --> 00:28:40.070
And also you can elicit the response with a brief, if you train with several

00:28:40.070 --> 00:28:44.250
hundred millisecond tone, you can still elicit the conditioned response with

00:28:44.250 --> 00:28:47.030
just 10 milliseconds conditioned stimulus.

00:28:47.470 --> 00:28:51.970
So it looks as if it's only the initial part of the stimulus that matters from

00:28:51.970 --> 00:28:53.310
the learning point of view. Right.

00:28:53.850 --> 00:29:00.750
Okay, so this whole topic of how is the interstimulus interval represented so that you can learn it.

00:29:00.810 --> 00:29:04.390
This is one problem, right? And now you sketch some scenarios how we can think about this.

00:29:04.750 --> 00:29:11.610
But another issue might be how you transform this. Yeah, but I have to interrupt here. Yeah, please.

00:29:11.990 --> 00:29:16.690
If you have the tone, and the tone goes on for many hundreds of milliseconds.

00:29:17.410 --> 00:29:22.090
Then there will be parallel fiber signals coming throughout this whole period.

00:29:22.290 --> 00:29:25.890
And that means the first parallel fiber signals, they will

00:29:25.890 --> 00:29:29.010
be matched with say if you

00:29:29.010 --> 00:29:31.910
have an interval of 300 milliseconds they will be followed after 300

00:29:31.910 --> 00:29:34.790
milliseconds by the complex back climbing fibers

00:29:34.790 --> 00:29:40.110
in the air path but if they can continue then there will be some parallel fiber

00:29:40.110 --> 00:29:45.130
signals will be followed at a shorter interval by the climbing fiber input and

00:29:45.130 --> 00:29:49.870
some will be followed even at an even yet shorter period and some will even

00:29:49.870 --> 00:29:51.950
coincide with the unconditioned stimulus.

00:29:52.550 --> 00:30:00.310
So this also creates a problem for the idea that you just have this long-term

00:30:00.310 --> 00:30:02.590
depression of synaptic strength.

00:30:03.350 --> 00:30:08.910
Okay. So one riddle, if you want, still in the system is, okay,

00:30:08.910 --> 00:30:14.570
where does the trace reside that correlates or that corresponds to the stimulus interval, right?

00:30:14.590 --> 00:30:19.290
It's just one problem. And indeed, I think your most recent experiments shed

00:30:19.290 --> 00:30:20.790
really an interesting light on that.

00:30:21.110 --> 00:30:25.850
But another question I think that's unresolved is really how do you now transform

00:30:25.850 --> 00:30:31.250
changes in peritoneal cell firing to actually triggering this eye blink?

00:30:31.630 --> 00:30:36.750
Because it's not that this peritoneal cell is directly driving this skeletal

00:30:36.750 --> 00:30:39.770
muscle system, right? It has to go through a number of steps.

00:30:40.230 --> 00:30:42.930
Moreover, the modulation might be inhibitory. Right.

00:30:43.415 --> 00:30:47.615
So you have a sign reversal problem. So how do we map now changes in Purkinje

00:30:47.615 --> 00:30:51.655
cell responses to actually triggering that eye blink?

00:30:52.035 --> 00:30:57.915
Well, we know that Purkinje cells are inhibitory, and they inhibit the cerebellar

00:30:57.915 --> 00:31:00.315
nuclei, and they are tonically active.

00:31:00.795 --> 00:31:06.355
That means that if a Purkinje cell that is tonically active has a certain stand...

00:31:06.355 --> 00:31:08.575
At a pretty high rate, I might add. At a pretty high rate, yeah.

00:31:09.175 --> 00:31:13.655
Tens of impulses per second, and even hundreds of impulses per second.

00:31:14.655 --> 00:31:18.935
And the nuclei to which they project also have a spontaneous background activity.

00:31:20.115 --> 00:31:24.435
So in order to generate an excitatory output, the Purkinje cell has to stop its firing.

00:31:25.295 --> 00:31:31.715
And that is what it actually does. When you apply a Pavlovian conditioning protocol

00:31:31.715 --> 00:31:37.755
to the Purkinje cell, it will respond with a pause to the input signal.

00:31:37.755 --> 00:31:43.495
Signal, and that pause means that the Purkinje cell, which is inhibitory,

00:31:43.535 --> 00:31:44.955
will stop inhibiting the nuclei.

00:31:45.235 --> 00:31:49.815
That means the nuclei will be excited and will generate an excitatory signal.

00:31:49.855 --> 00:31:51.595
Who excites the nucleus?

00:31:52.615 --> 00:31:55.375
What excites the nucleus? Nothing has to excite the nucleus.

00:31:55.475 --> 00:32:00.595
This was previously considered a problem, but it's now known that the nuclei

00:32:00.595 --> 00:32:04.275
have an intrinsic mechanism for generating a background activity,

00:32:04.575 --> 00:32:08.135
so nothing really has to excite So it's a form of rebound, a rebound excitation?

00:32:08.595 --> 00:32:12.255
You don't need that either. Okay. If the Purkinje cell stops firing,

00:32:12.475 --> 00:32:17.955
stops inhibiting the deep nuclei, they will immediately increase their firing rate.

00:32:18.195 --> 00:32:22.295
So that's not a problem. There may be problems, and this, of course,

00:32:22.315 --> 00:32:24.515
has to be addressed in a future physiology presentation.

00:32:26.662 --> 00:32:31.462
How is this transformation from the inhibitory perikinetic cell signal to the

00:32:31.462 --> 00:32:34.162
nuclei, to the red nucleus, to the facial nucleus?

00:32:35.322 --> 00:32:39.862
How is this transformation? What does it look like? Is it linear or is it something else?

00:32:40.222 --> 00:32:43.762
And this, of course, we don't yet know. And this has to be found out.

00:32:45.022 --> 00:32:49.062
And there are, of course, interesting aspects because the system also has to

00:32:49.062 --> 00:32:49.942
regulate the amplitude.

00:32:50.182 --> 00:32:54.722
The blink has to have, the eyelid movement has to have the right force.

00:32:54.922 --> 00:32:59.322
Right. If the eye is a little bit dry, for instance, it will need a larger force and so on.

00:32:59.642 --> 00:33:05.062
So that has to be regulated also. And that could well be downstream from the Purkinje cells.

00:33:05.602 --> 00:33:08.942
It could be in the nuclei, cerebellar nuclei.

00:33:09.242 --> 00:33:12.722
It could even be learned in the cerebellar nuclei. We don't know.

00:33:12.862 --> 00:33:18.002
Okay. But now there's another puzzling component of this story,

00:33:18.082 --> 00:33:20.202
which is that these deep nuclei.

00:33:20.202 --> 00:33:24.302
So now we have an idea of how we transform the pons and the Purkinje cell firing

00:33:24.302 --> 00:33:27.882
into triggering an action of the eyelid.

00:33:28.902 --> 00:33:35.002
But these deep nuclear cells also receive a collateral input from the pons, right?

00:33:35.142 --> 00:33:37.682
Well, no. So does it play any role of significance?

00:33:38.402 --> 00:33:42.882
I don't think so. Actually this has been a highly controversial subject.

00:33:43.202 --> 00:33:47.762
The nuclei do receive some mossy fabric collaterals from some sources.

00:33:50.356 --> 00:33:56.776
It's highly questionable if they do so from the pontine nuclei and from the

00:33:56.776 --> 00:33:58.916
mossy fibers arising in the pons.

00:33:59.256 --> 00:34:03.216
There have been some claims that they do. Other people have claimed they do not.

00:34:03.536 --> 00:34:09.876
And in any case, the number of such collaterals is likely to be very,

00:34:09.996 --> 00:34:14.276
very small compared to the convergence of input you have in the cerebellar cortex

00:34:14.276 --> 00:34:18.596
with a quarter of a million parallel fibers to a single Purkinje cell.

00:34:18.876 --> 00:34:22.976
So how many mossy fibers would collateral to the deep nucleus you think?

00:34:23.096 --> 00:34:26.616
I don't know. I mean some people have questioned that there are any at all. Oh really? Okay.

00:34:26.736 --> 00:34:33.316
So my guess is it wouldn't be near, I mean it would totally different order

00:34:33.316 --> 00:34:35.436
magnitude from what you have recorded.

00:34:35.636 --> 00:34:40.196
So we ignore this for now. It wouldn't be nearly enough to explain associative learning.

00:34:40.456 --> 00:34:49.676
But it might be enough to explain explain, for instance, a consistent modulation of reflex strength.

00:34:49.976 --> 00:34:52.336
Exactly right, yeah. So if you had mossy fibrous collateral,

00:34:52.636 --> 00:34:55.556
perhaps not from the pons, but from somewhere else, regulating,

00:34:55.596 --> 00:35:01.076
let's say, the background firing rate of the deep nuclei, that could mean that

00:35:01.076 --> 00:35:03.456
the eyelid response, the eye blink amplitude.

00:35:04.256 --> 00:35:06.276
Could be stronger or weaker.

00:35:06.556 --> 00:35:09.596
And that kind of learning could well... And this is a relevant issue, right?

00:35:09.636 --> 00:35:12.976
Because what you're conditioning is the amplitude time course of the response.

00:35:12.976 --> 00:35:17.916
And the Purkinje cells tell us how we can get the timing, but not this amplitude modulation.

00:35:18.336 --> 00:35:23.296
Precisely. And in other areas, for instance, in the vestibular ocular reflex,

00:35:23.576 --> 00:35:29.276
the Purkinje cell, or sorry, the cerebellum does modulate the amplitude of reflex.

00:35:29.696 --> 00:35:36.416
This has also been a view, if you go back decades, this has been the view of

00:35:36.416 --> 00:35:41.516
what the cerebellum does with all our spinal reflexes, that they modulate the reflex.

00:35:42.974 --> 00:35:45.714
Gain right so that the strength of

00:35:45.714 --> 00:35:48.894
the reflex it always has to be adapted when the

00:35:48.894 --> 00:35:52.534
body grows your muscle strength changes the

00:35:52.534 --> 00:35:59.394
mechanical properties changes all the time so you cannot these things cannot

00:35:59.394 --> 00:36:04.334
be inborn they have to be subject to modulation right as we grow and as we train

00:36:04.334 --> 00:36:07.514
as our muscles become stronger and then as we grow older as the muscles become

00:36:07.514 --> 00:36:09.754
weaker speaker, they will always have to adapt this.

00:36:09.954 --> 00:36:14.494
So you need some kind of adaptation of reflex gain throughout your life.

00:36:14.514 --> 00:36:17.634
Or even you could argue as soon as we start to use tools, for instance,

00:36:17.894 --> 00:36:24.674
if you talk about posture, posture control with tools, then you have to adapt

00:36:24.674 --> 00:36:26.014
very rapidly to that. Oh yes.

00:36:26.494 --> 00:36:29.894
But now, so okay, so let's say we're not going to worry too much about these

00:36:29.894 --> 00:36:37.014
mossy fiber collaterals, but another aspect of this downward pathway towards the action.

00:36:37.854 --> 00:36:42.174
Is that it also has a collateral from the deep nucleus back to the inferior

00:36:42.174 --> 00:36:44.274
olive, which seems puzzling, right?

00:36:44.334 --> 00:36:48.974
And you actually have been investigating this specific aspect of the whole circuit quite a bit.

00:36:49.234 --> 00:36:52.294
Because over the inferior olive, we have, if you want, our error signal coming

00:36:52.294 --> 00:36:54.954
in that tells you, okay, you just got a shock or you just got an air puff,

00:36:55.194 --> 00:36:56.474
something bad happened to you.

00:36:57.294 --> 00:37:01.234
But now in itself, these inferior olive neurons receive an inhibitory input

00:37:01.234 --> 00:37:04.074
from the deep nucleus, which we are triggering when we're saying,

00:37:04.174 --> 00:37:05.414
okay, close your eyelid now.

00:37:05.914 --> 00:37:11.674
So what's the role? How do you see the role of that connection, of that link?

00:37:11.714 --> 00:37:17.114
Well, we think that this is a negative feedback pathway for controlling learning,

00:37:17.294 --> 00:37:18.754
the amplitude of the learning.

00:37:19.641 --> 00:37:24.221
The amplitude of learning? The amplitude of the learned response. Okay.

00:37:25.181 --> 00:37:30.061
This was very puzzling initially. We discovered this by chance that this pathway

00:37:30.061 --> 00:37:32.361
was inhibitory. It was always believed to be excitatory.

00:37:33.501 --> 00:37:39.181
And when it's inhibitory, it's very difficult because Bikini cells are also inhibitory.

00:37:39.841 --> 00:37:42.941
And the climbing fibers are excitatory.

00:37:42.941 --> 00:37:49.701
The net effect would be a positive feedback system which is unbiological and

00:37:49.701 --> 00:37:53.881
and it can never be stable and there are all sorts of problems with that but

00:37:53.881 --> 00:37:59.381
if and this so this looked as a problem when it was thought that the effect

00:37:59.381 --> 00:38:02.261
of the climbing fibers was the excitatory.

00:38:03.241 --> 00:38:08.281
Providing excitatory input to the pre-kindle cells but if the pre-kindle if

00:38:08.281 --> 00:38:16.021
the climbing fiber induces a weakening of the parallel fiber to Purkinje cell

00:38:16.021 --> 00:38:20.321
synapses or if they produce post-responses in the Purkinje cell,

00:38:20.481 --> 00:38:23.501
then it becomes a negative feedback pathway.

00:38:23.821 --> 00:38:26.741
And that makes sense because it might control learning.

00:38:26.981 --> 00:38:29.961
So the idea that

00:38:29.961 --> 00:38:33.361
we have tried to pursue has been that when

00:38:33.361 --> 00:38:36.901
the cerebellum the

00:38:36.901 --> 00:38:39.861
cerebellum learns to produce an excitatory output

00:38:39.861 --> 00:38:43.261
signal generating a learned

00:38:43.261 --> 00:38:48.681
movement like an eye blink it will at the same time send a signal to the inferior

00:38:48.681 --> 00:38:54.681
olive that is inhibitory and that weakens or turns off the learning the teaching

00:38:54.681 --> 00:38:58.281
signal coming from the olive so it means that when the learning has reached

00:38:58.281 --> 00:39:01.961
a certain level you turn off the learning machinery So,

00:39:02.241 --> 00:39:04.561
you would prevent a kind of overlearning.

00:39:05.141 --> 00:39:09.981
And that in itself might not seem to be so important, but it does have important consequences.

00:39:10.341 --> 00:39:18.961
And one of them is that if the subject learns to respond to a certain stimulus like a light.

00:39:20.640 --> 00:39:23.400
And then, or say, sorry, like a tone.

00:39:23.740 --> 00:39:28.140
And then you add. So suppose you learn to blink to a tone. So each time you

00:39:28.140 --> 00:39:30.420
hear the tone, you blink nicely, a conditioned blink.

00:39:30.920 --> 00:39:34.280
Then you add a light on every trial.

00:39:36.000 --> 00:39:39.540
And pair the light with the tone and the air puff.

00:39:40.180 --> 00:39:45.220
If you then test learning to the light afterwards, you will find that somebody hasn't learned.

00:39:46.040 --> 00:39:48.980
Now, that makes biological sense, because why should it

00:39:48.980 --> 00:39:51.820
learn to respond to the light when you already have

00:39:51.820 --> 00:39:54.780
the tone and the tone enables you to avoid

00:39:54.780 --> 00:39:57.560
the air puff without paying any

00:39:57.560 --> 00:40:00.980
attention to the light so it would be just a waste of circuitry

00:40:00.980 --> 00:40:03.800
a waste of energy and so on to also learn to respond

00:40:03.800 --> 00:40:06.720
to the light but this system this

00:40:06.720 --> 00:40:09.980
mechanism will do precisely that because whenever

00:40:09.980 --> 00:40:13.220
you have the tone and the air puff the

00:40:13.220 --> 00:40:18.000
the subject generates the condition blink an output signal you turn off the

00:40:18.000 --> 00:40:24.840
learning mechanism so when the light comes on the subject doesn't learn anymore

00:40:24.840 --> 00:40:28.120
it has already learned to respond to the tone so it doesn't need the learning

00:40:28.120 --> 00:40:31.640
machinery anymore and it will prevent the animal from forming.

00:40:32.880 --> 00:40:37.640
Associations that are of no use but they're like redundant yeah exactly they're

00:40:37.640 --> 00:40:41.940
redundant right but then information but the interesting implication of that

00:40:41.940 --> 00:40:43.140
is that this learning machine.

00:40:43.940 --> 00:40:47.140
Has set itself up to never be a perfect learner

00:40:47.140 --> 00:40:50.020
because as soon as you started and block this

00:40:50.020 --> 00:40:54.540
teaching signal of the the climbing fiber signal you are creating conditions

00:40:54.540 --> 00:41:00.120
for extinction yes right so in some sense the learning plateau is now defined

00:41:00.120 --> 00:41:05.320
by by let's say the correct prediction then setting in motion extinction that

00:41:05.320 --> 00:41:08.160
forces you again then to learn because you start to make mistakes, right?

00:41:08.660 --> 00:41:16.840
Well, my view is if you have a thermostat system, the system will,

00:41:16.920 --> 00:41:21.020
when the temperature increases, it will turn off the heating.

00:41:21.140 --> 00:41:25.540
The heating, temperature go down, heating will be turned on,

00:41:25.640 --> 00:41:28.220
and you will have an oscillation around a set point.

00:41:28.840 --> 00:41:32.560
And I guess the same thing happens here. You have a set point,

00:41:32.580 --> 00:41:35.820
and you reach some kind of equilibrium,

00:41:35.820 --> 00:41:38.820
and you get a little bit of extinction and that

00:41:38.820 --> 00:41:41.740
suffices to turn on the learning the teaching machinery

00:41:41.740 --> 00:41:44.680
and you will learn and you will

00:41:44.680 --> 00:41:47.740
oscillate around a certain certain level

00:41:47.740 --> 00:41:52.820
okay but it also means it's always testing the hypothesis is predicting right

00:41:52.820 --> 00:41:58.880
yeah by just this this if you want non-specific extinction that it imposes upon

00:41:58.880 --> 00:42:04.320
itself yeah okay but now another Another aspect of this circuit is that in turn

00:42:04.320 --> 00:42:06.920
the inferior olive also projects back to the deep nucleus.

00:42:08.659 --> 00:42:14.599
Right. Is that of any significance in this learning perspective? I just don't know.

00:42:14.799 --> 00:42:19.059
My guess is that you can have some other form of learning in the nuclei and

00:42:19.059 --> 00:42:25.839
maybe those are also elicited by these collaterals.

00:42:26.099 --> 00:42:33.439
I don't know. Another possibility is that when the teaching segment to brachyndosal

00:42:33.439 --> 00:42:38.859
will cause the brachyndosal to fire and provide an inhibitory signal down to the nucleus.

00:42:39.219 --> 00:42:44.539
And maybe that inhibitory signal interferes with the normal movement.

00:42:44.899 --> 00:42:50.919
So if you have a signal coming via the collateral, it would help to cancel that

00:42:50.919 --> 00:42:53.499
meaningless inhibitory signal coming from the Purkinje cell.

00:42:53.599 --> 00:42:55.299
I don't know. This is pure speculation.

00:42:56.059 --> 00:42:59.319
But on the other hand, what is intriguing about this is that the inferior olive

00:42:59.319 --> 00:43:03.059
has its own spontaneous activity, right?

00:43:03.159 --> 00:43:05.759
Which is about one hertz to 10 hertz, depends who who

00:43:05.759 --> 00:43:08.879
you talk to um but now so

00:43:08.879 --> 00:43:12.159
it's not the case that that you only have climbing fiber activity when

00:43:12.159 --> 00:43:16.319
you actually have this unconditioned stimulus it's actually interspersed with

00:43:16.319 --> 00:43:20.819
a spontaneous level of yes and and it looks as if the climbing fabric input

00:43:20.819 --> 00:43:26.759
not only teaches the perkin cell to respond to particular specific stimulus

00:43:26.759 --> 00:43:30.879
it also regulates the background firing rate of the Purkinje cell.

00:43:31.219 --> 00:43:36.259
So normally the olive will fire with around one hertz or so.

00:43:36.339 --> 00:43:41.619
If you increase that to just over two hertz, you will silence the background

00:43:41.619 --> 00:43:44.279
firing of the Purkinje cell. It will totally silent.

00:43:44.799 --> 00:43:51.419
And if you reduce the background firing rate of the olive down to half a hertz or zero,

00:43:51.659 --> 00:44:00.159
after half a minute or so, the Purkinje cell will start firing with very increased rate.

00:44:02.819 --> 00:44:07.379
But when that happens, when the Purkinje cell background firing goes up,

00:44:07.499 --> 00:44:10.859
it will suppress the nucleus.

00:44:12.098 --> 00:44:17.678
Which will reduce the inhibitory signal to the olive, which will increase the

00:44:17.678 --> 00:44:21.738
background firing rate of the olive, and that will depress the Purkinje cell

00:44:21.738 --> 00:44:22.578
background firing rate.

00:44:22.698 --> 00:44:26.078
So you have a negative feedback system, not only for controlling learning,

00:44:26.378 --> 00:44:30.178
but also have a negative feedback system for controlling the background firing

00:44:30.178 --> 00:44:31.258
rate of the Purkinje cell.

00:44:31.378 --> 00:44:36.638
But now, in some sense, you could argue that maybe, if you think about eyeballing

00:44:36.638 --> 00:44:41.758
conditioning, it's actually a very tiny perturbation on the intrinsic dynamics of this system.

00:44:42.098 --> 00:44:46.098
So, maybe this learning system is just trying to keep, let's say the Purkinje

00:44:46.098 --> 00:44:51.398
cells are just trying to keep the inferior olive at some optimal level of activity, right?

00:44:51.438 --> 00:44:57.338
And tries to regulate it up and down by regulating the deep nucleus essentially, right?

00:44:57.398 --> 00:45:00.598
So, it's really homostatic negative feedback. Yeah, this makes it very difficult

00:45:00.598 --> 00:45:04.498
to theorize about it because it's not clear to me what is regulating what.

00:45:04.738 --> 00:45:08.438
Okay. Is it the Purkinje cells that are regulating the olive or is it the olive

00:45:08.438 --> 00:45:09.798
that are regulating the Purkinje cells?

00:45:09.798 --> 00:45:13.898
Well, it would basically mean that if you have, let's say, this closed-loop

00:45:13.898 --> 00:45:19.678
homostatic system where the Purkinje cells are regulating the inhibition of a deep nucleus,

00:45:19.938 --> 00:45:24.918
so you in turn regulate the inhibition onto the inferior olive,

00:45:25.038 --> 00:45:28.498
so you can keep it at a fixed rate, somewhere between 1 to 10 hertz.

00:45:29.078 --> 00:45:32.518
Now, if I have a climbing fiber of an unconditioned stimulus coming in,

00:45:32.618 --> 00:45:35.698
I start to speed up my inferior olive, right? So that would say,

00:45:35.758 --> 00:45:39.678
okay, now I have to sort of start to increase my inhibition to an inferior level

00:45:39.678 --> 00:45:42.618
to bring it back to this baseline that I want to keep it at, you see.

00:45:42.898 --> 00:45:47.938
So this might be, in that context, you could also account for cluster conditioning.

00:45:48.138 --> 00:45:51.438
But then what the learning system is still trying to do is keep this inferior

00:45:51.438 --> 00:45:55.418
olive neurons at the right level of firing and it doesn't act.

00:45:55.518 --> 00:45:59.918
And that you get an eye blink associated with it is something you get for free.

00:46:00.038 --> 00:46:04.598
That is how this system is wired up. So the emphasis of the learning system

00:46:04.598 --> 00:46:08.998
is on the controlling the fear olive as opposed to controlling the periphery. Is that reasonable?

00:46:10.504 --> 00:46:15.404
Well, it's certainly consistent with what we know, and I just don't know.

00:46:15.544 --> 00:46:19.964
I find this a very difficult subject to speculate on. Okay. It has puzzled me

00:46:19.964 --> 00:46:22.464
for a long time and confused me. Right.

00:46:23.144 --> 00:46:27.244
But we have an idea now about what this circuit does, right, how this could operate.

00:46:27.664 --> 00:46:32.984
And what is interesting about this, and now we also know that all the data we

00:46:32.984 --> 00:46:37.064
have is actually only describing a tiny fraction of that system, right? Right.

00:46:37.484 --> 00:46:40.904
So, what's the rest of the cerebellum actually doing?

00:46:43.044 --> 00:46:48.564
Well, a large part of the cerebellum will control movements that are in a certain way similar to eyelid.

00:46:48.724 --> 00:46:55.224
It will control various muscles like leg withdrawal or finger movements and things like that.

00:46:55.964 --> 00:47:00.504
Some parts of the cerebellum will control the forebrain.

00:47:00.644 --> 00:47:05.164
You have a huge part of the cerebellum projecting to the forebrain.

00:47:05.164 --> 00:47:09.804
And what it does there we don't know but presumably at least to some extent

00:47:09.804 --> 00:47:12.404
what it does is it changes the.

00:47:15.564 --> 00:47:20.184
Excitability of pyramidal cells controlling movements so that for instance if

00:47:20.184 --> 00:47:24.184
you make an error when you try to reach something,

00:47:24.824 --> 00:47:29.884
the cerebellum can learn to correct your movements and exciting a little bit

00:47:29.884 --> 00:47:34.204
more those muscles that would tend to move your arm in a certain direction and

00:47:34.204 --> 00:47:37.684
so on So that is one thing that's happening.

00:47:38.324 --> 00:47:40.764
You have all the responses...

00:47:42.229 --> 00:47:45.869
Of the this your system of balance that the

00:47:45.869 --> 00:47:48.689
cerebellum controls it could operate in

00:47:48.689 --> 00:47:51.549
a way the similar type link conditioning but my guess is that there

00:47:51.549 --> 00:47:57.789
are significant differences but now a few years ago you also proposed that these

00:47:57.789 --> 00:48:02.289
kinds of learning systems could actually underlie our ability to simulate right

00:48:02.289 --> 00:48:07.589
to support internal simulation or because now your examples you seem to emphasize

00:48:07.589 --> 00:48:11.009
very strongly the motor aspect of this.

00:48:11.209 --> 00:48:15.929
Yes, but my view of internal simulation is that it's very similar to actual

00:48:15.929 --> 00:48:21.249
motions, just as you turn off the output signals from the primary motor cortex.

00:48:22.109 --> 00:48:26.229
And for all we know, it could well be the cerebellum that is doing it.

00:48:26.269 --> 00:48:30.149
It could be the cerebellum that turns off or prevents movement from coming out.

00:48:30.269 --> 00:48:34.209
We normally think of the cerebellum as producing signals to excite the motor

00:48:34.209 --> 00:48:38.309
neurons in the motor cortex. But it could just as well be the opposite.

00:48:38.509 --> 00:48:44.549
They could very well be there in order to inhibit or stop movements that could be dangerous to you.

00:48:44.789 --> 00:48:49.869
So just as the cerebellum can produce a movement of the eyelid in anticipation

00:48:49.869 --> 00:48:54.389
of something that might be harmful,

00:48:54.689 --> 00:48:57.869
like an air puff or an insect coming into your eyes or something,

00:48:57.869 --> 00:49:06.249
It could very well also project to stop or prevent movements that might be harmful.

00:49:06.449 --> 00:49:10.649
For instance, you are walking, you are approaching something that is dangerous,

00:49:10.829 --> 00:49:14.949
you should stop walking before you fall off a cliff or something.

00:49:15.429 --> 00:49:22.469
Maybe the cerebellum is producing a signal that will enable you to stop the

00:49:22.469 --> 00:49:25.489
ongoing movement before something dangerous happens.

00:49:25.489 --> 00:49:28.309
Happens and the cerebellum is very good at it because it's

00:49:28.309 --> 00:49:31.229
so well timed and it's very fast compared to these signals

00:49:31.229 --> 00:49:35.549
coming from let's say the visual cortex to the fore to the forebrain it's very

00:49:35.549 --> 00:49:44.589
very slow and you need for quick adaptation anticipating dangerous events you

00:49:44.589 --> 00:49:50.649
need a fast system and the cerebellum is good at that but now do you believe that this um that

00:49:50.729 --> 00:49:56.049
also plays a role in, let's say, conscious states or conscious processing.

00:49:56.789 --> 00:50:01.589
Do you see this kind of simulation that you describe occurring in the cerebellum

00:50:01.589 --> 00:50:06.969
or being supported by cerebellar processing also feeding back into conscious

00:50:06.969 --> 00:50:10.829
states or do you really see those as completely distinct subsystems?

00:50:10.829 --> 00:50:15.709
I don't think the cerebellum in itself could be conscious, But I do think that

00:50:15.709 --> 00:50:22.569
the cerebellum is an integral part of what is going on in the cerebral cortex.

00:50:23.069 --> 00:50:29.569
And I think that consciousness has to do, well, consciousness is a term that

00:50:29.569 --> 00:50:30.769
covers too many different things.

00:50:30.869 --> 00:50:35.689
But in my view, the most critical part of consciousness is the fact that we

00:50:35.689 --> 00:50:41.209
have an inner reality that we can perform movements and see things and hear things.

00:50:43.021 --> 00:50:47.901
Internally in the brain without connection to the external world and that is

00:50:47.901 --> 00:50:52.681
what i mean by simulation so the cerebellum would play the same role in simulated

00:50:52.681 --> 00:50:58.961
movements as it does in overt movements and it would generate anticipated movements

00:50:58.961 --> 00:51:01.981
even if they are not and it would stop.

00:51:04.961 --> 00:51:09.421
Continuation of a movement that just incipient and you just started start doing

00:51:09.421 --> 00:51:12.041
the cerebellum could prevent that from occurring.

00:51:12.721 --> 00:51:18.301
And by doing that, it will always modulate the internal simulation.

00:51:18.421 --> 00:51:22.461
It will also, in that sense, modulate what's going on in your inner reality

00:51:22.461 --> 00:51:24.481
and, if you want, in your consciousness.

00:51:24.641 --> 00:51:31.721
Okay, but then it's more, let's say, an assistive function that then presents

00:51:31.721 --> 00:51:33.741
this information to the conscious scene.

00:51:33.781 --> 00:51:40.201
It's not really directly, let's say, it's not directly carrying that conscious scene. No, it's not.

00:51:40.321 --> 00:51:44.481
But it has a particular kind of importance because one of the important aspects

00:51:44.481 --> 00:51:48.321
of consciousness is your ability to anticipate what is going to happen.

00:51:48.901 --> 00:51:55.101
And you are aware of something might be dangerous. If I do this, then that will happen.

00:51:55.221 --> 00:52:00.141
This is something you're always aware of and it's a very important part of the

00:52:00.141 --> 00:52:01.521
function of consciousness, I think.

00:52:01.521 --> 00:52:07.101
And the cerebellum is very good at anticipation, and therefore I think that

00:52:07.101 --> 00:52:12.421
the input from the cerebellum to the frontal cortex is very important for consciousness.

00:52:12.961 --> 00:52:17.901
But I could argue that also unconscious functions have predictive components

00:52:17.901 --> 00:52:19.801
to them. Oh, absolutely, absolutely. Absolutely.

00:52:19.921 --> 00:52:25.081
And in fact, when you, I don't know if you've ever experienced being subjected

00:52:25.081 --> 00:52:27.161
to eye blink conditioning, but

00:52:27.161 --> 00:52:33.121
when you, if you do, you will notice how surprised you are when you blink.

00:52:33.461 --> 00:52:38.301
Okay. It's not, it's not as if you're aware that, that, oh.

00:52:39.799 --> 00:52:42.739
I hear the tone soon there will be an air puff i'd better

00:52:42.739 --> 00:52:45.839
blink so that i i avoid the air puff it's actually

00:52:45.839 --> 00:52:50.019
when you notice that the movement you're really surprised and i think the same

00:52:50.019 --> 00:52:55.379
thing happens when you learn let's say to play the piano or any musical instrument

00:52:55.379 --> 00:53:01.879
that that you you train and then suddenly you reach a point where your fingers

00:53:01.879 --> 00:53:03.999
move by themselves you're not really.

00:53:04.839 --> 00:53:09.239
You don't feel that you are moving the fingers you feel as if they are doing it themselves.

00:53:09.859 --> 00:53:13.539
Because of an unconscious process and what has happened is that the cerebellum

00:53:13.539 --> 00:53:17.259
has taken over into doing right exactly but so another so now that we solve

00:53:17.259 --> 00:53:18.399
the puzzle of consciousness,

00:53:18.899 --> 00:53:21.719
let's let's try another one which which i find

00:53:21.719 --> 00:53:25.819
very very intriguing with respect to the cerebellum which is the one of causality

00:53:25.819 --> 00:53:29.699
because which actually you also studied prior to entering into neuroscience

00:53:29.699 --> 00:53:33.299
yes i worked in philosophy for a few years before yes and and so what's interesting

00:53:33.299 --> 00:53:37.519
there is in the amount of conditioning we have this debate between contingency

00:53:37.519 --> 00:53:40.459
and contiguity being dominant in conditioning.

00:53:40.599 --> 00:53:44.299
That means either it has a specific order of events that matters,

00:53:44.379 --> 00:53:48.319
or is it just that events co-occur in space and time, right?

00:53:48.419 --> 00:53:52.659
So now in some sense, you could argue that the cerebellum is actually wired

00:53:52.659 --> 00:53:57.459
up to be sensitive to a very specific order of events, right?

00:53:57.539 --> 00:54:00.279
So that would be the contiguity view of conditioning.

00:54:00.679 --> 00:54:05.879
But I could also now make an argument. You could argue that from an evolutionary perspective,

00:54:06.079 --> 00:54:11.099
what is wired into this learning system is that causal relationships do exist

00:54:11.099 --> 00:54:16.719
out there in the world because I have wired myself up to pick them up within

00:54:16.719 --> 00:54:19.439
a timeframe of up to one second, right?

00:54:19.559 --> 00:54:24.739
So do you feel that this wiring of the cerebellum from a sort of evolutionary

00:54:24.739 --> 00:54:29.579
epistemological perspective tells us that causality is actually out there in

00:54:29.579 --> 00:54:32.339
the real world and that this is the way of the brain to deal with it?

00:54:37.584 --> 00:54:42.064
Well, you are now touching some extremely difficult questions.

00:54:42.784 --> 00:54:49.884
I mean, we are really entering deep waters here.

00:54:51.384 --> 00:54:58.964
And I can't comment on everything you said here, but let me first say that in

00:54:58.964 --> 00:55:06.124
any behavior, the order of different components of your behavior is very important.

00:55:06.124 --> 00:55:11.324
I mean, in language it's very obvious that the order of words is extremely important

00:55:11.324 --> 00:55:14.044
for the comprehensiveness of the message.

00:55:14.784 --> 00:55:18.964
Order can be achieved by a temporal code.

00:55:19.184 --> 00:55:26.604
If every component of your speech has a specific time, then order will be secondary.

00:55:26.604 --> 00:55:30.224
It will impose an order on it because if

00:55:30.224 --> 00:55:34.584
if the first phonemes

00:55:34.584 --> 00:55:37.404
are are generated early and the

00:55:37.404 --> 00:55:40.064
later phonemes are generated late then the order of the

00:55:40.064 --> 00:55:44.264
phonemes become sort of an unavoidable consequence but

00:55:44.264 --> 00:55:47.924
of course it's also possible that

00:55:47.924 --> 00:55:50.704
the cerebellum and the cerebellum might be

00:55:50.704 --> 00:55:53.644
important for that but it's also possible that the cerebellum imposes an

00:55:53.644 --> 00:55:56.584
order you can you can imagine for instance either you

00:55:56.584 --> 00:56:02.104
have let's say that you play a piano and you have you're moving a certain a

00:56:02.104 --> 00:56:07.964
certain finger in a certain number of times on a key and each key press has

00:56:07.964 --> 00:56:11.064
a specific time on it so you can

00:56:11.064 --> 00:56:15.384
imagine the purkinje cell sending out controlling that muscle sending out.

00:56:16.636 --> 00:56:20.416
Temporal signals to to press the key now

00:56:20.416 --> 00:56:25.956
and now and now and now but you could also imagine you had a chain of responses

00:56:25.956 --> 00:56:30.736
so the first response would cause the second response which causes the third

00:56:30.736 --> 00:56:37.256
response right and that is a very different kind of phenomenon and one of the

00:56:37.256 --> 00:56:39.716
The difference is that in this case,

00:56:39.776 --> 00:56:45.636
you could scale up and down the speed of the movement.

00:56:46.736 --> 00:56:51.056
So an interesting thing is that you can talk a little bit slower,

00:56:51.376 --> 00:56:56.616
but the order of the various components of your speech will be the same.

00:56:57.596 --> 00:57:04.236
This is difficult to understand if you have timed signals coming all the time,

00:57:04.396 --> 00:57:08.816
determining the exact time of each component.

00:57:10.916 --> 00:57:15.956
It's much easier to understand it if you have a sequence of cause-effect relationship

00:57:15.956 --> 00:57:18.876
as if one conditioned response triggers

00:57:18.876 --> 00:57:21.916
a second conditioned response that triggers a third one and so on.

00:57:21.916 --> 00:57:26.496
But what I was after was also to look at just the anatomy of this system, right?

00:57:26.556 --> 00:57:32.696
Because it's very peculiar that you have these widely divergent parallel fibers

00:57:32.696 --> 00:57:40.756
conveying states of the world and of the brain itself converging with these

00:57:40.756 --> 00:57:43.676
highly specific climbing fibers and really tell you something,

00:57:43.836 --> 00:57:45.336
let's say, intrinsically meaningful,

00:57:45.716 --> 00:57:47.976
right? Okay, now this was painful, right?

00:57:48.236 --> 00:57:51.676
And it is already structured in a way to deal with.

00:57:51.916 --> 00:57:58.056
Time, right? That something happens before something else, but they can co-occur, right?

00:57:58.136 --> 00:58:03.236
So isn't that one of the necessary conditions of a causal relationship? Yes, of course.

00:58:03.916 --> 00:58:08.756
So this is why I was wondering whether we could not argue that really the cerebellum

00:58:08.756 --> 00:58:11.736
is really biased towards, let's say, picking up.

00:58:12.360 --> 00:58:15.520
Causal relations but in a very permissive

00:58:15.520 --> 00:58:19.240
fashion right so it can also pick up illusory i wouldn't

00:58:19.240 --> 00:58:22.080
necessarily say causal because i because i think

00:58:22.080 --> 00:58:25.900
this would happen whether the whether the relationship is causal or pseudo correlation

00:58:25.900 --> 00:58:30.760
from the point of view of the cerebellum it doesn't really matter what matters

00:58:30.760 --> 00:58:36.100
is that the air puff tends to follow the tone even if the tone is not even if

00:58:36.100 --> 00:58:41.840
the air puff is not caused by the tone so but that's a force in the lab right Yeah,

00:58:42.080 --> 00:58:48.840
but from the biological point of view, if you look at it in sort of infunctional terms,

00:58:49.100 --> 00:58:52.060
the important thing is that you have a predictive relationship,

00:58:52.360 --> 00:58:56.420
which could be a pseudo-correlation that doesn't have to be causal.

00:58:56.860 --> 00:59:01.540
So my guess is that the cerebellum would learn a non-causal correlation just

00:59:01.540 --> 00:59:06.240
as well as a causal one, and it doesn't really matter.

00:59:06.240 --> 00:59:09.140
Okay, the only thing I could then say in my defense is that,

00:59:09.160 --> 00:59:13.320
of course, given that the system always learns to extinction,

00:59:13.680 --> 00:59:19.920
right, it does with time zoom in on truly causal relations because these will

00:59:19.920 --> 00:59:22.920
be the only one that are persistent in an interaction with an environment.

00:59:22.920 --> 00:59:23.880
That's true, that's true.

00:59:23.980 --> 00:59:27.860
And this is what is, yeah, and this is what often happens with pseudo-correlations.

00:59:27.900 --> 00:59:28.840
They break down eventually.

00:59:29.140 --> 00:59:32.040
Right, exactly. Because they are, because they're not causal,

00:59:32.080 --> 00:59:33.980
it's possible to interfere with them in a different way.

00:59:33.980 --> 00:59:37.820
That's exactly right yeah okay great so now that we also solved the causality

00:59:37.820 --> 00:59:43.160
problem that's wonderful Jerry we're making great progress here so let's finish up with.

00:59:43.880 --> 00:59:48.080
My last two questions so look you're in the business for a long time 25 years

00:59:48.080 --> 00:59:52.740
working your way through the cerebellum and I think we made amazing progress

00:59:52.740 --> 00:59:54.000
understanding cerebellum.

00:59:55.620 --> 01:00:01.200
Also if you have farmed out to issues in consciousness so based on your experience

01:00:01.200 --> 01:00:06.160
and your knowledge what would be jerry's law in our study of the brain,

01:00:08.960 --> 01:00:16.120
pre-kinder cells learn temporal relationships not only changing strength of synapses okay,

01:00:17.040 --> 01:00:24.020
okay that's my punchline all right and then if i'm going to visit you five years

01:00:24.020 --> 01:00:28.140
from now at lund and i'm going to say okay jerry you made the prediction five

01:00:28.140 --> 01:00:31.940
years ago and i want to see what happened with it what's the one prediction

01:00:31.940 --> 01:00:34.280
you would like to sort of commit yourself to today.

01:00:38.251 --> 01:00:43.651
That is very difficult to say because I can see before me a lot of some very,

01:00:43.731 --> 01:00:46.471
very interesting avenues of research.

01:00:46.891 --> 01:00:51.791
You have to give me one prediction. I don't know what they will generate,

01:00:52.171 --> 01:00:57.071
but I suppose I can tell you what I think we will have learned.

01:00:57.531 --> 01:01:03.291
We will have learned something about the molecular underpinnings of learning

01:01:03.291 --> 01:01:04.371
a temperate relationship.

01:01:04.371 --> 01:01:09.611
Relationship so i think we will have gone much deeper into receptor physiology

01:01:09.611 --> 01:01:14.811
and second messenger systems and so on in the pre-kinder cells but there are

01:01:14.811 --> 01:01:18.211
also a number of very difficult and,

01:01:19.031 --> 01:01:24.951
at the moment unsolved problems about how the cerebellum interacts with the

01:01:24.951 --> 01:01:31.651
cerebral cortex and if you take one very nice example prism adaptation if you

01:01:31.651 --> 01:01:35.431
wear prisms that displace the outside world laterally.

01:01:36.031 --> 01:01:42.051
Then if you try to point to a certain point at a certain object, there will be an error.

01:01:42.231 --> 01:01:45.731
You will point to the side of that object.

01:01:46.711 --> 01:01:52.031
And pretty soon, if you train to do pointing, wearing these goggles with displace

01:01:52.031 --> 01:01:57.291
the outside world, you can fairly quickly adapt to it and you can point in the right direction.

01:01:57.411 --> 01:02:00.511
The same thing happens if you have prisms that turn the world upside down.

01:02:00.631 --> 01:02:02.631
Adapt to that also pretty fast.

01:02:03.711 --> 01:02:09.431
Now, this is believed to be done by the cerebellum. And given what we now believe

01:02:09.431 --> 01:02:13.251
about learning in the cerebellum, it means that pre-kindi cells controlling

01:02:13.251 --> 01:02:18.511
the muscles that point your arm and finger,

01:02:19.271 --> 01:02:25.211
those pre-kindi cells will send some kind of corrective signals up to the motor cortex.

01:02:25.751 --> 01:02:32.291
Now, how do these pre-kindi cells learn to do that? they would need climbing

01:02:32.291 --> 01:02:38.051
fiber signals to change their responses and where do the climbing fiber signals come from.

01:02:39.260 --> 01:02:44.740
What triggers the learning is the mismatch between the intended movement and the actual movement.

01:02:45.040 --> 01:02:49.980
You're pointing not where you intended. There's nothing wrong in pointing in a certain direction.

01:02:50.220 --> 01:02:53.320
It's wrong only when you intend to point at something else.

01:02:54.280 --> 01:03:00.020
So there must be some way for the brain to detect a mismatch between what you

01:03:00.020 --> 01:03:01.340
intend and what you actually do.

01:03:01.340 --> 01:03:07.760
And that mismatch has to generate, activate inferior olives so that you get

01:03:07.760 --> 01:03:09.280
a climbing fiber signal to the cerebellum.

01:03:09.560 --> 01:03:13.180
This is extremely intriguing and we just don't know.

01:03:13.580 --> 01:03:19.700
We don't even know how the intended movement is encoded. We don't know really what that means.

01:03:20.320 --> 01:03:23.660
And how do we detect the mismatch? Where does that occur?

01:03:24.340 --> 01:03:28.200
And it seems to me that this will, if you can answer those questions,

01:03:28.400 --> 01:03:31.460
you would have gone a very long way to understanding how the cerebellum interacts

01:03:31.460 --> 01:03:34.320
with the forebrain because as you know

01:03:34.320 --> 01:03:37.760
the cerebellum takes part in all your movement and probably

01:03:37.760 --> 01:03:41.200
also in your cognitive function when you just simulate movement but but

01:03:41.200 --> 01:03:44.280
it it takes part in all of it it fine-tunes the

01:03:44.280 --> 01:03:50.000
amplitude the precision the timing and so on but it in order to learn it it

01:03:50.000 --> 01:03:55.480
will have to to receive climbing fiber input not only when you do something

01:03:55.480 --> 01:03:59.900
painful but also when you do something that you didn't intend to do and how

01:03:59.900 --> 01:04:01.500
is that signal being generated.

01:04:01.660 --> 01:04:07.760
And this is something that I believe we will know perhaps not in five years, but in 10 years.

01:04:07.980 --> 01:04:10.780
That's great. Well, Jerry Haslow, thank you very much for this conversation.

01:04:11.060 --> 01:04:11.920
Thank you. It's very nice.