WEBVTT

00:00:03.577 --> 00:00:10.497
This is the Convergent Science Network podcast. Leading researchers in the domain

00:00:10.497 --> 00:00:16.777
of neuroscience, brain theory and technology are interviewed by Paul Verschure and Tony Prescott.

00:00:25.757 --> 00:00:28.777
This is Paul Verschure with the Convergent Science Network. network.

00:00:29.417 --> 00:00:34.897
And I'm speaking here with Cyril Pennarts. And Cyril is a neurophysiologist

00:00:34.897 --> 00:00:38.877
who has been looking at the red brain for quite a while,

00:00:39.097 --> 00:00:44.037
and particularly looking at how different areas in the red brain operate and

00:00:44.037 --> 00:00:46.477
interact in the context of different tasks.

00:00:47.557 --> 00:00:51.617
So now, in your talk, you started with the notion of cognitive architecture.

00:00:52.177 --> 00:00:55.177
So why do you think that's relevant?

00:00:56.317 --> 00:01:07.517
I brought it into the talk because it's a theme you see recurring in almost every session.

00:01:09.677 --> 00:01:15.057
And it strikes me because both in robotics you see the modular systems,

00:01:15.317 --> 00:01:17.417
but also in neuroscience.

00:01:18.857 --> 00:01:23.837
And in both fields, these questions of communication and coordination pop up.

00:01:26.497 --> 00:01:30.397
But it seemed to look like you were emphasizing this notion,

00:01:30.497 --> 00:01:35.697
well, we have different modules operating in this brain, and we have to think

00:01:35.697 --> 00:01:38.357
about how these modules communicate. Is that really how you think about it?

00:01:40.435 --> 00:01:45.655
Well, what I try to emphasize is that, yeah, there's evidence for different modules.

00:01:46.675 --> 00:01:53.255
Anatomically, in the brain, physiologically, lesions will have selective effects on areas.

00:01:53.535 --> 00:01:57.075
So there are functional specialization. You could say that.

00:01:58.915 --> 00:02:04.455
Yeah, the issue of how you get a flexible communication is not really answered.

00:02:04.455 --> 00:02:13.195
And it seemed a relevant theme for robots as well or other cognitive architectures.

00:02:13.995 --> 00:02:20.135
But that means in your research, also the kind of work you were describing,

00:02:20.275 --> 00:02:21.715
and we'll discuss in a bit more detail,

00:02:22.635 --> 00:02:27.695
this notion of, let's say, well-defined modules with well-defined communication

00:02:27.695 --> 00:02:33.075
channels between them is really like a sort of a guideline in how you investigate

00:02:33.075 --> 00:02:34.495
these different areas you study.

00:02:35.275 --> 00:02:41.855
Yeah, yeah. Of course, this happens a lot by human EEG research, fMRI.

00:02:42.515 --> 00:02:48.275
But what we try to add is basically the neural coding also by way of recording spike trains. Right.

00:02:48.895 --> 00:02:56.495
And the combination of having more mesoscopic measures of EEG plus the detail spike trains,

00:02:57.655 --> 00:03:03.515
makes it more unique or more informative because now you see for instance all

00:03:03.515 --> 00:03:07.775
that whereas there is specialization in the hippocampus in Central Australia

00:03:07.775 --> 00:03:13.315
there are also commonalities but they are only revealed if you have the spike

00:03:13.315 --> 00:03:17.055
trains and can look at this remapping phenomena by.

00:03:18.775 --> 00:03:24.335
Reward predictive use for instance okay well but you sort of summarized It's

00:03:24.335 --> 00:03:27.095
not everything you're going to say in the future very rapidly.

00:03:27.435 --> 00:03:33.515
But the point is that you're saying, look, it might be nice to think about modules

00:03:33.515 --> 00:03:36.435
from a more macroscopic perspective like using EEG.

00:03:36.835 --> 00:03:41.775
But if we don't have the detailed information at a physiological or anatomical

00:03:41.775 --> 00:03:43.855
level, we actually don't really know what we're talking about.

00:03:45.395 --> 00:03:50.175
Well, you can know what you talk about if you realize the limitations of the approach.

00:03:50.635 --> 00:03:55.795
Right. Okay. But yeah, of course, with EEG, you have source localization problems.

00:03:56.935 --> 00:04:02.615
FMRI is better for spatial source, but not for time. So it's more volume averaged slow signal.

00:04:02.735 --> 00:04:07.095
So yeah, I think it's really important to have the spike signals in millisecond

00:04:07.095 --> 00:04:10.255
resolution to look at the closer synchrony.

00:04:10.395 --> 00:04:19.255
Right, exactly. But then, so your emphasis you place in your research and also

00:04:19.255 --> 00:04:20.835
in your presentation today,

00:04:21.015 --> 00:04:26.655
focus very much on areas like the rental striatum and the hippocampus. And you were...

00:04:27.730 --> 00:04:32.430
Defining or showing how these areas actually seem to follow a very specific

00:04:32.430 --> 00:04:36.690
kind of zoning of their organization. So can you say something about that?

00:04:38.170 --> 00:04:44.150
Yeah, what I showed was the zonation of the striatum in relation to the frontal cortex.

00:04:45.530 --> 00:04:50.150
Second part of the talk was more about sensory neocortex and the caudal parts.

00:04:51.650 --> 00:04:57.670
Yeah, the zonation I think is interesting because it shows how continuous the

00:04:57.670 --> 00:05:00.350
innervation pattern is of the striatum.

00:05:00.410 --> 00:05:06.550
So let's say you're lateral in the frontal cortex and you find a specific dedicated

00:05:06.550 --> 00:05:10.530
area of reception in the dorsolateral striatum for that,

00:05:11.270 --> 00:05:16.090
sensory motor, detailed associations and probably habits.

00:05:17.310 --> 00:05:19.330
And then as you follow the band

00:05:19.330 --> 00:05:22.290
more immediately, then you find areas protecting more and more ventral.

00:05:22.290 --> 00:05:25.370
So it's all very nicely topographic,

00:05:25.650 --> 00:05:31.190
but also this topography exists between the hippocampus and the striatum,

00:05:31.290 --> 00:05:39.830
ventral hippocampus, more in the ventral, the most ventral part of the ventral striatum, and so on.

00:05:42.250 --> 00:05:48.070
But it's also in agreement with the suggestion made that there's not just a

00:05:48.070 --> 00:05:51.610
strict segregation within the stratum of, let's say, an actor and a critic.

00:05:52.330 --> 00:05:57.290
Fentral stratum, according to the older scheme, as Andy Bartow proposed,

00:05:57.490 --> 00:06:00.070
it would need a critic. Doing what?

00:06:01.630 --> 00:06:09.350
Basically, forming reward predictions based on the error feedback received from the dopamine cells.

00:06:09.830 --> 00:06:15.630
And then somehow the report prediction signals would also be transmitted to

00:06:15.630 --> 00:06:17.410
the dorsal striatum acting as an actor.

00:06:18.630 --> 00:06:25.350
While there are indirect pathways to do that, we thought there's actually more

00:06:25.350 --> 00:06:29.530
homogeneity across the striatum in how the system works.

00:06:30.670 --> 00:06:35.870
Yeah, but does that mean – so we look at it, so cortex provides massive input

00:06:35.870 --> 00:06:39.890
to the striatum, loops through this whole structure of the basal ganglia,

00:06:40.090 --> 00:06:45.090
and again, via the thalamus goes back into the cortex of this massive loop, right?

00:06:45.090 --> 00:06:49.430
Then we would have similar kinds of loops running towards the hippocampus.

00:06:50.870 --> 00:06:55.130
And now we see that there are zones. That means there is some sort of topography.

00:06:55.170 --> 00:07:01.030
The specific areas are very specifically targeting certain regions in the stratum.

00:07:01.050 --> 00:07:04.070
And then the loops are basically going to follow the zoning scheme.

00:07:04.630 --> 00:07:09.370
So this is now an anatomical construct. Does it have any functional consequences?

00:07:12.510 --> 00:07:13.550
Yeah, because…,

00:07:15.055 --> 00:07:20.335
What people have done is make lesions at different sites in the stratum and

00:07:20.335 --> 00:07:22.715
test rats on different learning tasks.

00:07:22.935 --> 00:07:28.835
And then there's evidence to suggest that indeed you have most sensory motor

00:07:28.835 --> 00:07:33.715
coding of, let's say, very detailed motions like arm movements in the dorsolateral stratum.

00:07:34.535 --> 00:07:42.315
Whereas this area when lesion is also impaired in stereotyped arm movements or habits. it.

00:07:42.895 --> 00:07:49.415
But if you go more ventral, you find specific impairments of the association

00:07:49.415 --> 00:07:50.555
between action and outcome.

00:07:50.875 --> 00:07:57.115
So it's more like making the head movement for a reward. That kind of association is not properly made.

00:07:57.555 --> 00:08:02.075
And eventually you have more influences of specific cues.

00:08:03.015 --> 00:08:06.595
Lights, coffee cups, etc. And then also space.

00:08:06.955 --> 00:08:11.095
Spatial becomes important. So it seems to be a whole stream of information.

00:08:12.215 --> 00:08:17.775
The difference in its content, but the computational principles, we think, are the same.

00:08:18.315 --> 00:08:21.735
Right. For instance, if you lesioned a ventral striatum, you would lesion the critic.

00:08:22.155 --> 00:08:27.715
But in rats or other animals, the dorsal striatum can still learn despite the

00:08:27.715 --> 00:08:29.775
absence of a ventral striatum. Right. Okay.

00:08:30.115 --> 00:08:36.595
But then before we get to the actor-critic criticism, so what you're saying

00:08:36.595 --> 00:08:44.475
is, look, Look, we have these loops, these cerebral basal ganglia loops.

00:08:44.715 --> 00:08:48.575
They have specific qualities.

00:08:48.955 --> 00:08:52.835
It can be action, it can be cue, so sensation.

00:08:52.875 --> 00:08:57.455
It can be value or internal state, motivational state or space.

00:08:58.355 --> 00:09:02.155
How many of these qualities do we have, do you think?

00:09:03.835 --> 00:09:09.215
You mean qualities as dimensions? Well, yeah, or different modalities if you want.

00:09:09.755 --> 00:09:13.235
Yeah. Like we have cue, action, motivation, space.

00:09:14.315 --> 00:09:16.155
What's missing from this list? What's missing?

00:09:18.215 --> 00:09:22.115
Um, I don't think we, we miss very much.

00:09:23.695 --> 00:09:30.075
Um, sometimes you could use time as a, as a cue, uh, sort of internal time when to expect a reward.

00:09:30.675 --> 00:09:35.575
Uh, but usually time is accompanied by distinct signals, uh, going along with it.

00:09:36.575 --> 00:09:41.775
But we also, uh, know now that, yeah, the hippocampus when time is relevant

00:09:41.775 --> 00:09:46.655
for a task can encode, uh, moments of time or little episode moments.

00:09:46.655 --> 00:09:50.915
So, that could also be a mechanism to transmit timing information.

00:09:52.035 --> 00:09:55.215
Okay. So, if anything is missing, it would be time.

00:09:55.615 --> 00:10:00.195
And then it's not obvious that time would be looping through these structures in the same way.

00:10:02.455 --> 00:10:05.515
Right. Not as it is in the hippocampus, yeah.

00:10:05.635 --> 00:10:10.775
The striatum or basal ganglia are probably important for the perception of time

00:10:10.775 --> 00:10:14.995
or the estimation of time because drugs that typically work on stratum like

00:10:14.995 --> 00:10:19.455
amphetamine also change the perception of time, cannabinoids, et cetera.

00:10:19.995 --> 00:10:26.135
And that might have more to do with these ramping responses that you see in the firing rate.

00:10:26.255 --> 00:10:31.275
So when the animal's in a situation where he expects some cue to appear or a

00:10:31.275 --> 00:10:35.735
reward to come, you will already see cells increasing the firing rate slowly

00:10:35.735 --> 00:10:37.475
until it finally happens.

00:10:37.775 --> 00:10:40.975
And you believe that's an intrinsic property of these cells or of this structure?

00:10:40.975 --> 00:10:45.295
Of the loop, yes. Yeah, but this goes with the cortex.

00:10:47.015 --> 00:10:48.535
But where's the clock?

00:10:53.455 --> 00:10:58.455
Well, it's possible that there are pacemaker cells somewhere, but it's not so likely.

00:10:59.055 --> 00:11:03.835
The dopamine cells have some pacemaker properties, but then the firing of those

00:11:03.835 --> 00:11:07.275
cells is also variable and they burst and reset. set.

00:11:07.295 --> 00:11:11.595
So it's more likely to be an emergent property of the whole circuit, I think.

00:11:11.695 --> 00:11:14.795
So it's not a cerebellum or so? Um...

00:11:16.154 --> 00:11:20.714
Surveillance is supposedly important for timing, but also in the very fine time domain.

00:11:22.294 --> 00:11:26.374
Stratum might be more for longer stretches of time, like seconds.

00:11:27.174 --> 00:11:31.794
That's the beauty of this scheme that you didn't consider.

00:11:32.414 --> 00:11:38.534
We have to find out why. But in some sense, the stratum gives you a beautiful event-based system,

00:11:39.154 --> 00:11:43.994
with these different modalities or qualities that it processes while it can

00:11:43.994 --> 00:11:49.054
then, by using the cerebellum, which is still about 60% of your brain,

00:11:49.734 --> 00:11:51.134
get high-resolution time signals.

00:11:51.414 --> 00:11:55.734
Because the cerebellum's timing precision stops at about one second. Right.

00:11:56.154 --> 00:11:59.234
So you might have a dual loop, right? That you have sort of an event stream

00:11:59.234 --> 00:12:06.034
across basal ganglia, hippocampus, cortex, and then a real-time high-precision

00:12:06.034 --> 00:12:08.154
stream, an interval stream over cerebellum.

00:12:08.374 --> 00:12:11.914
Exactly, yeah. Would you buy that as an explanation of time?

00:12:13.454 --> 00:12:17.234
As a mechanism for timing, yes. Yeah, timing of movement.

00:12:18.114 --> 00:12:22.234
No, but also timing of these ramping responses, right? For expected rewards, for instance.

00:12:22.534 --> 00:12:25.774
Yeah, yeah. I don't know how that would work in the cerebellum,

00:12:25.774 --> 00:12:33.234
but if you, let's say, you train animals or humans on a sequence of movements like finger tapping,

00:12:33.394 --> 00:12:37.194
but now you change one key of the instrument and you make it hard to press,

00:12:37.214 --> 00:12:42.514
then the cerebellar cells would notice that and you have to adjust the strength

00:12:42.514 --> 00:12:44.714
of your finger tap on that. Right.

00:12:45.014 --> 00:12:47.034
And similarly with timing issues.

00:12:47.494 --> 00:12:52.514
So it sounds quite likely that Sir Benhamin there is for the fine timing.

00:12:52.774 --> 00:12:58.314
But this sort of little discourse or detour, if you want,

00:12:59.114 --> 00:13:07.394
was interesting to see whether notion of cue, action, motivation and space would

00:13:07.394 --> 00:13:10.394
be the four main domains or whatever, we missed something fundamental.

00:13:10.554 --> 00:13:15.914
And now at least we invented a story, the two of us, that would in some sense

00:13:15.914 --> 00:13:18.434
suggest that we could leave time out.

00:13:19.440 --> 00:13:25.460
You don't need to include time in those qualities. Yeah, at least not by a separate structure.

00:13:25.620 --> 00:13:31.540
Exactly. But I do wonder how the cerebellar timing is coupled to the corticostratal timing.

00:13:32.140 --> 00:13:34.200
Somewhere the thalamus is in between.

00:13:35.180 --> 00:13:39.640
Sure, there are dense projections between these structures, right? Yeah, yeah.

00:13:40.540 --> 00:13:44.260
Okay, this is outside a little bit of what you were presenting today.

00:13:44.380 --> 00:13:46.620
I don't think you ever measured from the cerebellum, did you?

00:13:47.700 --> 00:13:51.620
No, no, that's right. So it's a very global idea.

00:13:51.820 --> 00:13:56.380
It's never too late to start, you see. But then, okay, so now we have to consider the zones.

00:13:57.520 --> 00:13:59.080
That's good. We have the four qualities.

00:14:00.700 --> 00:14:04.640
And now the key point you made on the basis of that is, look,

00:14:04.820 --> 00:14:08.840
these loops with these varying qualities also carry, if you want,

00:14:08.860 --> 00:14:10.720
motivation in one of these loops.

00:14:10.720 --> 00:14:17.040
So why separate action from motivation or why separate value now from action

00:14:17.040 --> 00:14:19.960
as you might do in an actor-critic system, right?

00:14:20.020 --> 00:14:26.380
It's more a distributed solution where if you want value and action are both

00:14:26.380 --> 00:14:30.840
processed on equal terms in some sense. Is that how you think about it?

00:14:32.358 --> 00:14:40.458
Yeah, so, yeah, you always see, no matter what task the rat does or some other

00:14:40.458 --> 00:14:43.918
animal, you always see this desolation of sequence.

00:14:44.358 --> 00:14:47.038
That's also in the orbital frontal and medial prefrontal.

00:14:47.538 --> 00:14:51.618
So every little element of task is basically coded.

00:14:53.678 --> 00:14:58.138
But then what to do with that? Well, we regard that sequence as a scaffold to

00:14:58.138 --> 00:15:01.878
which you can associate things like reward value.

00:15:02.798 --> 00:15:07.138
Which would mean that, in a sense, at the straighter level, you attach a weight

00:15:07.138 --> 00:15:11.318
to a particular action as being important because it's reward predictive or

00:15:11.318 --> 00:15:13.418
predictive of an hour outcome.

00:15:16.478 --> 00:15:22.998
The only exception could be that habits are not strictly dependent on reward.

00:15:23.238 --> 00:15:27.198
You can delete reward and your habits will still keep going.

00:15:28.958 --> 00:15:34.578
But in a sense, if you take the notion of outcome to be more abstract and not per se reward related,

00:15:34.838 --> 00:15:39.858
you could also say, well, an action itself, the completion of an appropriate

00:15:39.858 --> 00:15:44.978
action where you touch an object can also be regarded as an outcome.

00:15:47.218 --> 00:15:54.018
Even for a learning infant, being able to reach some object could be regarded as an accomplishment.

00:15:55.298 --> 00:15:58.278
But the one thing I don't understand fully is that you're saying,

00:15:58.378 --> 00:16:05.578
well, you could attach a reward to single events. Like you have this tessellation of the response.

00:16:05.838 --> 00:16:11.178
Like you're engaged in a task. Here I am. I'm the rat. I'm pushing all these levers and whatever.

00:16:12.658 --> 00:16:17.498
Now, in my brain, this sort of corticostradial system and my hippocampus,

00:16:17.498 --> 00:16:21.418
I'm now decomposing this task and all its small elements.

00:16:22.478 --> 00:16:26.438
Relevant, irrelevant. relevant doesn't matter and now you're saying i'm now

00:16:26.438 --> 00:16:32.458
tagging if you want some of these elements with um with value with with reward

00:16:32.458 --> 00:16:34.238
predictions is this really what you have in mind,

00:16:35.617 --> 00:16:37.697
Yeah, yeah, basically, yes.

00:16:40.157 --> 00:16:45.677
Where, of course, yeah, if we train animals, the animals are trained in such

00:16:45.677 --> 00:16:51.737
a way that they will only do or perform these tasks if finally some reward is coming.

00:16:51.937 --> 00:16:57.337
Sure. So in a sense, we don't test for real spontaneous behavior or behavior

00:16:57.337 --> 00:16:58.697
where there's no reward at all.

00:16:59.117 --> 00:17:04.837
In the end, somewhere there will be a reward. so in that sense the sequence

00:17:04.837 --> 00:17:08.217
that you see might always have to do with the final outcome.

00:17:08.577 --> 00:17:15.577
Right. Now tell me where's the site where this n-gram so this memory trace of,

00:17:16.217 --> 00:17:19.377
action reward is established and maintained?

00:17:21.737 --> 00:17:25.117
We don't know but probably the,

00:17:26.037 --> 00:17:31.477
corticosteroidal synapse itself could be a good site for that at least when

00:17:31.477 --> 00:17:39.757
it comes to associating, for instance, place to reward or cue to reward or actions to reward.

00:17:41.817 --> 00:17:47.637
Because then we have only to suppose that the hippocampal cells or the subicular

00:17:47.637 --> 00:17:52.757
cells, C1, subiculum, project to the striatum. They do.

00:17:53.337 --> 00:17:55.577
We know that there is plasticity in that connection.

00:17:56.217 --> 00:17:58.117
The fibers are glutamatergic.

00:18:00.037 --> 00:18:05.937
So in a fairly straightforward, Hebbian way, given the effect of reward feedback,

00:18:06.117 --> 00:18:09.237
there can be a plasticity going on.

00:18:10.737 --> 00:18:18.477
And this lines up with the behavioral evidence based on lesions and disconnection.

00:18:18.477 --> 00:18:22.157
So if you lesion the striatum in that loop where you have the place representation

00:18:22.157 --> 00:18:25.457
in the striatum or place processing….

00:18:26.795 --> 00:18:30.595
Do you see this system compromised? Or you see this behavior compromised?

00:18:30.735 --> 00:18:35.435
Right, yeah. So if you lesion, in this case, the part of the ventral stratum,

00:18:35.475 --> 00:18:39.815
which receives most of the hippocampal input, then you lose your place reward

00:18:39.815 --> 00:18:42.455
association, or at least it's not behaviorally expressed.

00:18:43.415 --> 00:18:48.855
And if you lesion the hippocampus on one side and this ventral stratum part

00:18:48.855 --> 00:18:50.395
on the other side, it's also lost.

00:18:51.075 --> 00:18:57.335
Lost um this could mean that uh place reward association is association is still

00:18:57.335 --> 00:19:03.235
else stored somewhere elsewhere but at least the behavioral expression is lost so there's no uh,

00:19:04.235 --> 00:19:10.255
translation into invigorated actions um but for me it's more natural to think

00:19:10.255 --> 00:19:15.975
that reward prediction is motivation it's right as soon as you have a reward

00:19:15.975 --> 00:19:17.795
prediction that's enough to drive behavior.

00:19:18.175 --> 00:19:20.415
Yeah, because the point of what we're really talking about is that the animal

00:19:20.415 --> 00:19:23.395
is learning about places, right? It learns about locations in space.

00:19:24.335 --> 00:19:29.075
And it's also associating, let's say, these reward predictions to locations

00:19:29.075 --> 00:19:33.495
in space or to some cues in the environment. But it can be either of the two.

00:19:34.015 --> 00:19:39.515
Yeah, or both. Yeah. But does it also mean that if you interfere with plasticity

00:19:39.515 --> 00:19:45.675
specifically in this place system of the ventral striatum, that you then also

00:19:45.675 --> 00:19:47.835
don't see any acquired place preference?

00:19:49.475 --> 00:19:54.615
Yeah, you can interfere with NMDA blockers, dopamine antagonists also,

00:19:54.815 --> 00:20:00.995
and you will, it's not been shown that place preference per se is impaired, but other,

00:20:02.039 --> 00:20:05.939
learning processes are impaired, like the Pavlovian auto-shaping,

00:20:05.979 --> 00:20:07.799
where you associate cues with reward.

00:20:08.079 --> 00:20:13.959
Okay. But I would not consider the ventral stratum as a place system because

00:20:13.959 --> 00:20:17.639
the neural firing patterns are not that spatially specific.

00:20:18.779 --> 00:20:25.619
They're more specific for what you do at a certain place. So we're here.

00:20:26.019 --> 00:20:29.239
If you want to get coffee, you know you have to get out of the room.

00:20:29.239 --> 00:20:34.479
So when you get up, there will be striatal cells firing, but that's relating to your reaction.

00:20:35.319 --> 00:20:38.739
If you would get the coffee by going in the other direction,

00:20:38.819 --> 00:20:41.799
they would also fire, or it would be other cells firing. Yes.

00:20:42.119 --> 00:20:46.259
So these were these experiments you have been doing in this Y maze,

00:20:46.419 --> 00:20:49.399
right, where the animal could find reward at different sites.

00:20:50.559 --> 00:20:55.879
Reward locations were indicated by a cue light, and then the animal should just

00:20:55.879 --> 00:20:59.319
wait for this cue light to appear, and then it could go out and find a reward or not.

00:20:59.679 --> 00:21:03.899
And in those experiments, you were particularly looking at the information exchange,

00:21:04.039 --> 00:21:08.079
if you want, between the hippocampus and the ventral striatum with the idea,

00:21:08.179 --> 00:21:10.979
okay, hippocampus we know is dedicated,

00:21:11.419 --> 00:21:15.139
among other things, to learning about location, space.

00:21:15.439 --> 00:21:20.379
We will have these place fields with a very specific firing responses in certain positions in space.

00:21:20.779 --> 00:21:25.639
And the question is, is this information directly entering into this ventral

00:21:25.639 --> 00:21:28.819
striatal system or not? Was that really what you looked at?

00:21:31.979 --> 00:21:37.299
What I presented was not directly about the communication in terms of whether

00:21:37.299 --> 00:21:38.719
that's involving oscillations or so.

00:21:39.139 --> 00:21:42.099
Well, but look at the level of correlation between these structures.

00:21:42.259 --> 00:21:42.719
Yeah, yeah, yeah, certainly.

00:21:42.999 --> 00:21:46.979
Yeah. Well, to buttress it, sorry, you would have to present,

00:21:47.119 --> 00:21:48.659
let's say, cross correlations.

00:21:49.499 --> 00:21:54.059
Previously, we looked at these replay sequences during sleep where you do see

00:21:54.059 --> 00:22:00.199
actually hippocampal play cells firing first. and then the reward action cells in the striatum.

00:22:02.199 --> 00:22:06.559
So, yeah, you have to do these tighter correlations in the spike balance.

00:22:06.599 --> 00:22:10.399
No, but what I'm after is that I thought that what you were after was to say,

00:22:10.439 --> 00:22:15.159
okay, the ventral striatum has a response that's maybe action-dominated,

00:22:15.319 --> 00:22:18.339
but it has a spatial component, right?

00:22:18.379 --> 00:22:24.879
It's not devoid of space. and whether this space, the specificity for place

00:22:24.879 --> 00:22:30.439
in the ventral striatum was in some way dependent on the hippocampus or not.

00:22:32.479 --> 00:22:37.999
Yeah, but that's really a minority of cells that have this spatial specificity. Okay.

00:22:38.059 --> 00:22:40.879
It could arise on the one hand by very sparse firing.

00:22:41.059 --> 00:22:44.759
So it's kind of an undersampling problem where some of the spikes happen to

00:22:44.759 --> 00:22:49.479
end up in more in one of the chambers than the other. um,

00:22:51.102 --> 00:22:56.722
Yeah, the other possibility is really that you have some kind of heritage of the place cells.

00:22:57.662 --> 00:23:02.402
But at least in a setting where you eliminate the local cues and you make the

00:23:02.402 --> 00:23:04.042
behavior dependent on path integration.

00:23:06.462 --> 00:23:12.702
We think it's more that the ventral shade of cells take the hippocampal input

00:23:12.702 --> 00:23:16.882
and translate it into a current reward prediction and base directions on it.

00:23:17.502 --> 00:23:24.162
If that action means initiate an approach to a goal site, then that will generalize

00:23:24.162 --> 00:23:27.302
across all the chambers in which to make that action.

00:23:27.402 --> 00:23:29.642
So it's not really spatially specific.

00:23:30.062 --> 00:23:34.702
Okay. So it's really more policy dependent or action specific.

00:23:34.862 --> 00:23:38.102
It says, look, when I see this light to the right, I turn left.

00:23:38.282 --> 00:23:40.622
That's a great thing to do and I will do it wherever I am.

00:23:41.422 --> 00:23:45.682
Right, exactly. Yeah. Okay. Yeah. But now the other thing that you observed,

00:23:45.922 --> 00:23:50.502
which I found very curious, is that there was a modulation of the size of this

00:23:50.502 --> 00:23:53.862
place field in the hippocampus by reward itself.

00:23:55.582 --> 00:24:03.382
Yeah, so what we saw is that these nine reward sites tend to have occupancy of micro place fields,

00:24:03.562 --> 00:24:09.282
whereas you don't see these micro place fields in the bigger non-rewarded compartments

00:24:09.282 --> 00:24:11.662
of the maze. How do you explain that?

00:24:13.362 --> 00:24:16.082
Uh, well, on the one hand, they're extremely relevant sites.

00:24:16.382 --> 00:24:20.762
So it's really important for the rat to know where to stick your nose in precisely.

00:24:21.382 --> 00:24:26.142
So a finer spatial scaling could be very useful. Uh, and it's also a couple

00:24:26.142 --> 00:24:29.182
of more specific small scale behaviors.

00:24:29.542 --> 00:24:31.502
Uh, for instance, uh.

00:24:33.490 --> 00:24:38.150
Licking while you approach the site, the rat would have to stop.

00:24:38.230 --> 00:24:42.130
That's already a deceleration motion in a small stretch of space.

00:24:43.270 --> 00:24:46.530
Then the licking behavior, then waiting for a certain while before the reward

00:24:46.530 --> 00:24:47.370
comes, and then licking.

00:24:47.490 --> 00:24:53.650
So it's actually a lot happening, which all probably has to be coded somewhere.

00:24:53.930 --> 00:24:58.870
So you're saying, if we would replot place field size versus something like

00:24:58.870 --> 00:25:05.750
behavioral complexity, it should give us a fairly uniform form curve. Yeah, right. Okay.

00:25:06.790 --> 00:25:08.670
Complexity on the x-axis and

00:25:08.670 --> 00:25:14.370
the inverse of size on the y-axis give you a straight line. Okay. Yeah.

00:25:15.690 --> 00:25:21.510
Is that a known feature of these hippocampal place cells? No, not really. No, no.

00:25:22.970 --> 00:25:28.650
So, yeah, there have been studies that showed a greater density of place fields

00:25:28.650 --> 00:25:33.750
around important sites, like the hidden platform in the Morris Water Maze,

00:25:34.630 --> 00:25:36.630
attract more or less more place fields.

00:25:36.790 --> 00:25:40.530
But it could be that this involves the same phenomenon.

00:25:40.630 --> 00:25:46.630
So if you have a broad coverage of the space by big place fields and coverage

00:25:46.630 --> 00:25:51.990
by small place fields, if the sites are really relevant, then you end up with

00:25:51.990 --> 00:25:54.650
a higher density of place fields. Right.

00:25:55.070 --> 00:26:00.610
But now the other thing that was interesting is that there also seems to be

00:26:00.610 --> 00:26:05.990
a correlation between the peak firing rate in the place field and the size of the place field.

00:26:06.610 --> 00:26:17.150
Right. So apparently the larger size of place fields also gave you higher peak

00:26:17.150 --> 00:26:19.290
frequency than the small size place fields.

00:26:19.350 --> 00:26:22.670
I mean, are these kinds of correlations intuitive to you?

00:26:23.230 --> 00:26:26.990
Or does this make sense? We haven't systematically looked at it.

00:26:27.330 --> 00:26:32.710
It could be the case. We'd have to go through the entire set of cells to see if there's a relation.

00:26:34.410 --> 00:26:39.090
It does make a little bit of sense in the sense that the way you define a place

00:26:39.090 --> 00:26:47.510
field or when you plot a bright yellow spot is somewhat correlated to the absolute firing rate.

00:26:47.510 --> 00:26:52.070
So if a cell sort of has a dynamic range from zero to 20 hertz,

00:26:52.230 --> 00:26:58.210
and the zero is kept for most part of the maze except in one chamber,

00:26:58.490 --> 00:27:07.970
then it's more likely that the cell passes the threshold for the place field

00:27:07.970 --> 00:27:12.210
more easily, so to say, utilizing the entire dynamic range. Right.

00:27:12.750 --> 00:27:19.330
So now the other thing that you mentioned is that you have this notion that

00:27:19.330 --> 00:27:22.170
there's something like a state transition occurring in these neurons,

00:27:22.270 --> 00:27:24.710
both in hippocampus and ventral striatum.

00:27:25.030 --> 00:27:26.910
So what does it really mean?

00:27:29.055 --> 00:27:34.175
I think the interesting aspect about it is that it's a global population measure.

00:27:34.335 --> 00:27:38.755
So there's no single cell that easily dominates such state transitions.

00:27:39.635 --> 00:27:44.875
And yet they are coherent because you see them happening in a lot of cells.

00:27:44.955 --> 00:27:51.175
At least there are marked firing rate changes at the point of state transition or close to it.

00:27:52.535 --> 00:27:57.415
So it seems to be a population phenomenon that both works in one structure and the other.

00:27:58.255 --> 00:28:01.135
And then the most interesting is that those are correlated also.

00:28:02.515 --> 00:28:05.835
Yeah, but I don't really understand the phenomenon. I mean, so here you have

00:28:05.835 --> 00:28:08.575
your recording of large numbers of cells.

00:28:08.655 --> 00:28:11.395
And how big is your pool of neurons you're doing this analysis on?

00:28:11.895 --> 00:28:16.815
This Y-Mains data set is about 600 neurons. Okay, so that's plenty of neurons, right?

00:28:17.175 --> 00:28:24.075
And now you are sorting these neurons on a specific metric, right?

00:28:24.175 --> 00:28:27.715
And that then gives you this notion of phase transition. but what's this metric

00:28:27.715 --> 00:28:29.455
on which you sort your neurons?

00:28:30.595 --> 00:28:35.435
Yeah, so basically if you would have 10 cells, you make a 10-dimensional space

00:28:35.435 --> 00:28:43.035
and then you plot your spatial bin firing rates into that space for all the 10 neurons.

00:28:43.255 --> 00:28:50.695
You cluster it basically based on K-means. It can be likened to maximizing the

00:28:50.695 --> 00:28:52.395
Euclidean distance between the clusters.

00:28:52.635 --> 00:28:56.015
Okay. So you seek for the best partitioning

00:28:56.015 --> 00:29:00.035
plane lane okay um and yeah

00:29:00.035 --> 00:29:04.455
so that's what you do but um remarkably it's

00:29:04.455 --> 00:29:08.815
not some kind of arbitrary cut because the cells are visibly sensitive

00:29:08.815 --> 00:29:14.255
to it not all of them but uh i would say a majority of the cells that do react

00:29:14.255 --> 00:29:21.795
in advance of the state switch or have a reaction afterwards um but you know

00:29:21.795 --> 00:29:28.335
i We would have predicted here that an event like the Q flipping on, well,

00:29:28.495 --> 00:29:33.775
might have some effect on the hippocampus, but would be more strongly felt at

00:29:33.775 --> 00:29:37.895
the ventral straddle level because our psychological colleagues would tell us,

00:29:37.935 --> 00:29:41.735
well, it's the amygdala that does the transmission of this motivational Q.

00:29:43.255 --> 00:29:46.635
So it's a bit of a surprise to see that the impact on the hippocampus is that

00:29:46.635 --> 00:29:50.135
big, whereas it does not disrupt the finer spatial code.

00:29:51.895 --> 00:29:55.935
Because it's expressed as a remapping effect. So, but it also means in your

00:29:55.935 --> 00:29:58.275
physiology of this, in the population recording,

00:29:59.095 --> 00:30:04.475
you would also see an instantaneous transient across all the cells you're measuring

00:30:04.475 --> 00:30:07.215
from in some sense, or a big subset of them.

00:30:08.255 --> 00:30:13.275
Yeah, but if you just look at the plane recordings as they're going on and the

00:30:13.275 --> 00:30:18.575
rat is doing its task, it might not be that obvious, because it's very hard

00:30:18.575 --> 00:30:21.615
to listen to all these tens of neurons at the same time. Right, exactly.

00:30:21.895 --> 00:30:27.015
So that's why it's handy to have this algorithm to do it for you. Mm-hmm. Um.

00:30:28.282 --> 00:30:33.662
And yeah, I should also emphasize that it's not only the cue lights that do it.

00:30:33.702 --> 00:30:36.402
So they are sort of a strong trigger for a state switch.

00:30:36.962 --> 00:30:42.982
But you also see significant enhancements of the switches when the rat enters a chamber.

00:30:43.422 --> 00:30:49.022
So when he sort of knows I'm going now into a direction where I'm close to the reward. Right.

00:30:49.462 --> 00:30:55.842
But now, are these neurons really doing something qualitatively different before

00:30:55.842 --> 00:30:59.082
and after this transition, this phase transition?

00:31:00.102 --> 00:31:06.202
They change their average frequency or they shut off completely or they turn on?

00:31:06.882 --> 00:31:11.922
Yeah, sometimes they switch up completely. So there are some cells that in one

00:31:11.922 --> 00:31:13.922
state have a place field and in the other state not.

00:31:14.602 --> 00:31:18.042
But usually it's a bit more subtle. So you would, for instance,

00:31:18.082 --> 00:31:24.882
have a gain modulation of a factor 2 or so or 3, and sometimes a shift in the place field.

00:31:26.462 --> 00:31:28.862
And also at the ventral straddle level.

00:31:32.082 --> 00:31:35.942
So the upshot of this is that we shouldn't

00:31:36.202 --> 00:31:42.102
Only look at single-cell detailed coding, but also have the global population

00:31:42.102 --> 00:31:47.762
picture, which says, okay, there's another type of coding change going on. Right, exactly.

00:31:48.262 --> 00:31:51.182
But then, what's your functional interpretation of this?

00:31:55.682 --> 00:31:59.502
The functional interpretation would be to say, well, there's,

00:31:59.502 --> 00:32:05.262
again, this scaffold, fault you might say of basal coding in this hippocampus

00:32:05.262 --> 00:32:09.022
in the spatial task that is a spatial layout,

00:32:10.282 --> 00:32:17.422
but then again you can attach or associate events on top of that which do things to your basal code,

00:32:18.602 --> 00:32:26.242
same thing for a commons yeah but if you think about it it makes sense I think because,

00:32:28.582 --> 00:32:34.162
for an episodic memory it's very useful to have a sort of basal spatio-temporal

00:32:34.162 --> 00:32:39.782
framework onto which you can tag important events that need to be remembered.

00:32:42.962 --> 00:32:47.662
So in other words, it's kind of handy to have a scaffold to build your memory on.

00:32:49.282 --> 00:32:55.002
In the artificial way that Romans had at memory art, they imagined themselves

00:32:55.002 --> 00:33:01.842
walking into a house and storing things in caches in the wall or behind doors as a way to remember.

00:33:02.162 --> 00:33:06.962
And maybe that could be a metaphor for how this scaffolding mechanism works.

00:33:07.402 --> 00:33:10.302
Okay, and then you see this, what you call a phase transition,

00:33:11.142 --> 00:33:19.122
as a signature of this kind of, let's say, attaching specific cues into that scaffold. Right.

00:33:20.342 --> 00:33:25.502
Right, yeah. Or is the scaffold itself? The scaffold in the hippocampus here

00:33:25.502 --> 00:33:30.862
would be the basal spatial coding, the rate maps and place fields of all the cells.

00:33:32.002 --> 00:33:36.182
Yeah, in this case, you do have a very important event because it totally drives

00:33:36.182 --> 00:33:43.642
the animal's behavior and becomes not known as an association to one little place field,

00:33:43.702 --> 00:33:47.762
but rather while this cue appears in the whole space,

00:33:47.862 --> 00:33:50.062
the animal can clearly see it.

00:33:50.062 --> 00:33:55.342
So it becomes more of an overriding event that impinges on the global coding.

00:33:55.542 --> 00:33:59.142
And that would be modulated by something like a reward prediction.

00:34:00.022 --> 00:34:06.182
Because the cue light comes on. Yeah, yeah, yeah. We don't know how that would happen.

00:34:06.462 --> 00:34:10.502
It could be driven by the visual system, which initially takes in the visual information.

00:34:12.082 --> 00:34:18.542
But at some higher level of visual processing says, this is a reward predicting cue. Right.

00:34:20.062 --> 00:34:23.282
It could also happen at the prefrontal level or at many places.

00:34:24.162 --> 00:34:28.302
So now you observe this, what you call phase transition. I'm not sure phase

00:34:28.302 --> 00:34:31.502
transition is a good label really, but okay, let's keep it for now.

00:34:32.462 --> 00:34:38.102
You see it both in ventral striatum and hippocampus, two areas that are densely coupled.

00:34:38.622 --> 00:34:42.262
And you also looked at the cross-correlation of these events.

00:34:42.522 --> 00:34:44.082
So what did that tell you?

00:34:45.682 --> 00:34:48.822
The interesting thing there to see

00:34:48.822 --> 00:34:52.062
is that the state transitions although

00:34:52.062 --> 00:34:54.962
they are computed only locally for

00:34:54.962 --> 00:34:58.942
each structure are still correlated with each other and

00:34:58.942 --> 00:35:05.522
it could be at least partly externally driven because the queue event occurs

00:35:05.522 --> 00:35:11.642
at one moment and could trigger state transitions in both structures but apart

00:35:11.642 --> 00:35:15.322
from the queue events there are also other are lots of more spontaneous transitions,

00:35:16.842 --> 00:35:19.582
which do appear still to be highly correlated.

00:35:21.142 --> 00:35:25.582
So, it doesn't mean per se that the hippocampus switches first and then predicts

00:35:25.582 --> 00:35:27.622
its altered spike patterns due to the accumbens.

00:35:27.782 --> 00:35:34.242
It could also be indicating that these state transitions are a more global phenomenon

00:35:34.242 --> 00:35:38.542
and also involving other cortical areas like the prefrontal cortex.

00:35:38.902 --> 00:35:42.322
But how about, so in terms of interpretations that I say, well,

00:35:42.362 --> 00:35:46.462
look, You know, the animal sits here in this Y maze. There's nothing better to do.

00:35:46.902 --> 00:35:53.062
A cue comes on and I just have a nonspecific attentional effect orienting response.

00:35:53.162 --> 00:35:54.762
Will something change in the world?

00:35:55.062 --> 00:35:59.942
So it's not specific in any way to the task, to reward prediction,

00:36:00.202 --> 00:36:01.942
just something changed in the world.

00:36:01.982 --> 00:36:06.222
I have a nonspecific attentional effect and that's this highly synchronized

00:36:06.222 --> 00:36:08.162
phase transition that you observe.

00:36:09.792 --> 00:36:13.012
Uh yeah um let's

00:36:13.012 --> 00:36:16.372
see well we um we don't

00:36:16.372 --> 00:36:19.312
have some other uh secondary kind of

00:36:19.312 --> 00:36:27.892
cue that would signal um address still has to do something else um and we don't

00:36:27.892 --> 00:36:33.052
have a way to probe whether this is selective attention so yeah it's possible

00:36:33.052 --> 00:36:38.652
but yeah if i would talk here about a motivational cue that subsumes attention.

00:36:38.972 --> 00:36:42.892
It's sort of a change in the animal's state, motivational attention.

00:36:43.012 --> 00:36:48.232
One thing, so you would predict if you would switch a cue light that has never,

00:36:48.292 --> 00:36:54.852
ever been coupled to reward, so it's a neutral cue light, you should not see this phase transition.

00:36:56.832 --> 00:37:01.432
It would be very hard to have it totally neutral because a novel stimulus is

00:37:01.432 --> 00:37:05.152
also interesting or either scary or interesting to explore. It's not equally,

00:37:05.392 --> 00:37:08.232
it should not be equally leading to reward predictions.

00:37:08.692 --> 00:37:12.532
Yeah, yeah. So that means especially the phase transitions in the ventral striatum

00:37:12.532 --> 00:37:17.232
should then be sort of not there because there's no sense of reward.

00:37:17.392 --> 00:37:19.692
Yeah, this would be an interesting experiment. You could say,

00:37:19.692 --> 00:37:24.612
well, I'm taking a second cue maybe of a different light or a sound cue that is loud enough.

00:37:25.690 --> 00:37:29.950
And at one point it's null, but then you keep on repeating it with the same

00:37:29.950 --> 00:37:33.070
loudness and the animal learns to ignore it because it's irrelevant.

00:37:33.410 --> 00:37:34.510
And then see what happens.

00:37:35.090 --> 00:37:40.690
That will be the control test. Yeah, but I would predict that if there's this

00:37:40.690 --> 00:37:45.590
learned irrelevance about it, that the state transitions become weaker or less

00:37:45.590 --> 00:37:47.010
frequent. Right, yeah, sure.

00:37:47.510 --> 00:37:52.050
Okay, so now we have this idea of the transition.

00:37:52.050 --> 00:37:58.530
But the other thing that made me worry about an alternative interpretation is

00:37:58.530 --> 00:38:05.450
that the latency that you saw between the straight transitions in hippocampus

00:38:05.450 --> 00:38:07.550
and ventral stratum appeared very short.

00:38:07.630 --> 00:38:12.590
They seemed really practically synchronous in this change of their overall dynamics.

00:38:12.590 --> 00:38:18.610
And then I could argue, well, look, that's the perfect signature of a nonspecific

00:38:18.610 --> 00:38:22.810
attentional global signal that sort of engages all these systems in parallel

00:38:22.810 --> 00:38:25.290
with zero latency difference between them.

00:38:26.130 --> 00:38:28.450
So have you worried about that?

00:38:30.890 --> 00:38:38.450
Yeah, we did try to define our bins on an even finer scale.

00:38:38.450 --> 00:38:45.390
But we also found that this estimation of local firing rates works best in,

00:38:45.430 --> 00:38:47.430
let's say, a resolution of 100 milliseconds.

00:38:47.850 --> 00:38:52.170
If you go below it, it becomes a little bit messier and more noisy.

00:38:53.610 --> 00:38:59.490
So we can actually only say that these joint state transitions occur with a

00:38:59.490 --> 00:39:02.190
resolution of around 100 milliseconds or a bit less.

00:39:02.970 --> 00:39:08.290
And that's not enough to say whether the hippocampus would really switch first and then the accumbens.

00:39:08.450 --> 00:39:12.890
But actually, you could check this in the data you have, right?

00:39:14.090 --> 00:39:17.010
Yeah, we could do more detailed analysis, for instance.

00:39:19.610 --> 00:39:23.510
Using more single spikes or... That's right. ...counting the number of spikes

00:39:23.510 --> 00:39:25.770
per theta cycle or so, as the rat moves along.

00:39:26.090 --> 00:39:34.210
Yeah, because, I mean, the expected latency, if the loop, as you described it,

00:39:34.290 --> 00:39:40.910
is C1 subiculum in hippocampus, to your ventral striatum, right?

00:39:40.990 --> 00:39:48.190
And then you have a latency, a transduction latency in that projection of about 25 milliseconds.

00:39:48.790 --> 00:39:54.310
So that means you would expect a very specific patterning of this phase transition

00:39:54.310 --> 00:39:58.050
if this is the projection that's actually engaged. Right.

00:39:59.062 --> 00:40:07.202
Of course, at the level of the stratum, you will have easily converging activity

00:40:07.202 --> 00:40:10.022
from the amygdala, prefrontal cortex, and thalamus.

00:40:11.582 --> 00:40:18.122
So in a way, if there is stratal firing, the correlation to hippocampal activity might be partial.

00:40:18.922 --> 00:40:21.762
But you can do this, basically.

00:40:22.182 --> 00:40:28.122
But in some simple-minded view, you could say, well, if the source is in hippocampus,

00:40:28.122 --> 00:40:29.562
then this is the latency you should see.

00:40:30.942 --> 00:40:37.822
Yeah. Yeah. We could certainly try to resolve that at, let's say,

00:40:37.842 --> 00:40:41.442
the time scale of, well, the theta cycle is actually around 100 milliseconds.

00:40:41.842 --> 00:40:45.402
It could be maybe happening also in the gamma cycles somewhere.

00:40:45.602 --> 00:40:50.082
Sure. 20 milliseconds or so. Maybe that works. But would such a post hoc control

00:40:50.082 --> 00:40:55.062
now still be worth your while or you think this is really doesn't matter anymore this

00:40:55.302 --> 00:41:00.962
story is done now you move on um well

00:41:00.962 --> 00:41:06.942
the analysis as we did it now especially jayden jackson was already quite extensive

00:41:06.942 --> 00:41:13.282
but more trying to tease out whether um these tech transitions really correlate

00:41:13.282 --> 00:41:17.502
to a motivational change or motivational attentional change of the the animal.

00:41:18.502 --> 00:41:22.722
So the additional analysis he did were more directed at finding out whether

00:41:22.722 --> 00:41:25.702
for instance the chamber entry makes a difference in the state switches.

00:41:27.242 --> 00:41:30.662
It's also interesting to see if the animal, I didn't show the data,

00:41:30.742 --> 00:41:32.762
but if the animal approaches the reward site,

00:41:33.838 --> 00:41:40.238
Then actually the state transitions, the rate of switching decreases in the accumbens.

00:41:40.978 --> 00:41:45.178
That might be because the network is converging to a stable state of,

00:41:45.198 --> 00:41:47.158
let's say, solid reward prediction.

00:41:47.438 --> 00:41:52.578
You're there, you made it, and now you can stop. Right, exactly. Yeah.

00:41:52.978 --> 00:41:58.978
Whereas in similar non-rewarded behaviors, where there's an inter-trial interval,

00:41:59.318 --> 00:42:03.358
no queue, the L1 makes the same approach, you see a higher switch rate.

00:42:03.358 --> 00:42:10.778
As if it also reflects uncertainty or, let's say, ambiguity in the system.

00:42:10.838 --> 00:42:13.058
It keeps on flipping back and forth. Right, okay.

00:42:13.378 --> 00:42:19.198
So the next part, so now we have a bit of an idea how this mental stratum hippocampal

00:42:19.198 --> 00:42:22.838
system might be combining, let's say,

00:42:22.958 --> 00:42:29.218
value reward information with information about space and Q, okay?

00:42:29.218 --> 00:42:34.218
And in some sense, indeed, what you did in those experiments and look at in

00:42:34.218 --> 00:42:38.738
too much detail was really how could these components of the nervous system,

00:42:38.838 --> 00:42:41.718
these modules of the nervous system, really exchange information.

00:42:42.518 --> 00:42:48.878
So that was sort of the next part of your presentation where you actually emphasized

00:42:48.878 --> 00:42:53.398
very much this notion of neural oscillations and spike coherence.

00:42:53.418 --> 00:42:56.558
So remember, it's a dynamics-oriented perspective on communication.

00:42:57.138 --> 00:43:02.738
Yeah. So what are the main considerations to sort of actually look in that direction

00:43:02.738 --> 00:43:07.098
at this sort of intermodule communication and not just at, let's say, rate coding?

00:43:08.578 --> 00:43:15.138
Okay, yeah. When we purely compare rate codes of one area to the next.

00:43:16.758 --> 00:43:21.818
It's a little bit hard to say that there's actually an influence directly in

00:43:21.818 --> 00:43:25.118
the way of a phase relationship or a cross-correlation.

00:43:25.118 --> 00:43:31.258
It could be done at spike level, but usually within an area,

00:43:31.358 --> 00:43:35.138
let's say a pyramidal cell and an end-neuron can have a very tight cross-correlation.

00:43:35.198 --> 00:43:41.378
But between areas, it becomes easily sloppy or not so well-defined, the delays and so on.

00:43:42.798 --> 00:43:50.938
So I do think that the oscillations are an interesting way of looking at the

00:43:50.938 --> 00:43:52.378
communication mechanisms.

00:43:52.378 --> 00:43:58.598
Although, of course, yeah, like I illustrated for gamma, it's certainly not

00:43:58.598 --> 00:44:00.818
guaranteed that oscillations per se are important.

00:44:01.618 --> 00:44:07.258
The gammas precisely show that probably they have a local function in the network

00:44:07.258 --> 00:44:12.658
and are not for this long-range hippocampal to sensory vortex communication. Right.

00:44:12.898 --> 00:44:18.458
But you emphasize also this relationship between local field potential and EPSPs

00:44:18.458 --> 00:44:24.758
or possible impact on plasticity through spike time dependent learning.

00:44:25.098 --> 00:44:27.438
So what other attractive features

00:44:27.438 --> 00:44:31.978
do you see in this sort of this synchronization view on communication?

00:44:35.322 --> 00:44:43.162
Well, one advantage of synchronization is that if you have strong synchronization,

00:44:44.322 --> 00:44:51.542
a network that does that is in a better position to affect a target area or

00:44:51.542 --> 00:44:54.282
a common cell, for instance, where the cells converge upon.

00:44:56.142 --> 00:45:00.022
If the synchronization at least happens in the gamma range, you're talking about

00:45:00.022 --> 00:45:03.922
spike timing differences of in the order of 10 milliseconds or so,

00:45:04.022 --> 00:45:06.582
because otherwise you're covering the whole gamma cycle.

00:45:07.302 --> 00:45:13.622
And then you get into a range of one spike eliciting an EPSP with at least with

00:45:13.622 --> 00:45:16.822
the tail should overlap with the next EPSP 10 milliseconds later.

00:45:17.802 --> 00:45:24.162
So that's an interesting range for EPSP starting to summate and generating spikes. spike.

00:45:24.202 --> 00:45:30.422
So at the same time, if one cell generates an EPSP and the next one an IPSP,

00:45:30.562 --> 00:45:32.742
you only get very short lasting excitations.

00:45:34.022 --> 00:45:39.762
And then, yeah, this time scale of gammas correlates quite well with the time

00:45:39.762 --> 00:45:42.982
range where you would see spike timing depend plasticity.

00:45:44.482 --> 00:45:52.042
There's some nice work by Laurent where he indeed shows that spike timing regulated

00:45:52.042 --> 00:45:56.862
in the gamma range more or less does indeed alter synaptic responses.

00:45:57.862 --> 00:46:03.982
Right. It is a realistic scenario. But then, so here we have,

00:46:04.102 --> 00:46:07.402
let's say, a communication channel between two areas in the brain.

00:46:07.622 --> 00:46:14.022
It's organized along some temporal dynamics, some temporal code.

00:46:15.122 --> 00:46:19.482
What kind of code do you really have in mind? I mean, how complex would this code be?

00:46:20.242 --> 00:46:24.922
Is it really just like I have like a carrier wave and that, let's say,

00:46:24.942 --> 00:46:29.082
enslaves all my target neurons to oscillate in a certain frequency and then

00:46:29.082 --> 00:46:34.202
I can sort of more efficiently inject EPSPs or create EPSPs there.

00:46:34.242 --> 00:46:37.782
How complex is this temporal code then in your mind?

00:46:38.722 --> 00:46:44.322
Yeah, well, in my mind, there's also the debate of what this could do.

00:46:45.582 --> 00:46:52.662
On the one hand, gamma oscillation or other oscillation could be useful to bring

00:46:52.662 --> 00:46:54.742
the notion of iterations into the system.

00:46:54.802 --> 00:47:00.202
So we say we make a processing step, all the local neurons interact with each

00:47:00.202 --> 00:47:05.142
other and recompute their firing rate at the end of the cycle.

00:47:05.862 --> 00:47:11.482
Then there's a stop, maybe also to allow the system to communicate with other

00:47:11.482 --> 00:47:15.782
areas and get feedback, which allows the next iteration to happen.

00:47:16.802 --> 00:47:22.542
So this could be a functional notion of, why there is also an inhibition between...

00:47:25.700 --> 00:47:30.100
Another thing is to partially the information to make ordered sequences.

00:47:30.240 --> 00:47:37.680
You want some discretization in the system of place all ordering or sensory

00:47:37.680 --> 00:47:41.620
information ordering, not sort of happening

00:47:41.620 --> 00:47:45.080
in a jambalaya with every cell overlapping with every other cell.

00:47:46.540 --> 00:47:49.620
So ordering and sequencing could be a real function.

00:47:49.980 --> 00:47:55.200
Okay. And then what I alluded to was also the notion of phase coding.

00:47:55.200 --> 00:48:01.260
So that there is, besides global rates, additional information in when the spikes are fired.

00:48:02.620 --> 00:48:06.480
Of course, with the prime example of theta phase precession in the hippocampus,

00:48:06.520 --> 00:48:12.760
where you can really decode quite accurately the position of the animal from phasing of the spikes.

00:48:15.700 --> 00:48:23.180
But theoretically, I also couple that to employing different modes of coding, actually,

00:48:24.160 --> 00:48:29.960
because whereas you might need your rate code for feature coding representing,

00:48:30.220 --> 00:48:35.600
I have a cell here, it's a simple cell for orientation and right now that orientation

00:48:35.600 --> 00:48:38.300
is very appropriate to code so we drive with the firing rate.

00:48:38.780 --> 00:48:43.520
The other thing could be to shift that firing actually and then you create an

00:48:43.520 --> 00:48:49.940
additional phase code where that simple cell relates to other cells so it causally

00:48:49.940 --> 00:48:53.360
influences the phasing of other cells and then predict backwards.

00:48:54.280 --> 00:48:57.200
Okay, but actually, so these are the possible scenarios, right?

00:48:57.800 --> 00:49:01.800
But you actually went in there and you measured from quite a number of areas. Okay.

00:49:03.591 --> 00:49:08.511
So what was the real setup you built up there? Which areas did you measure from to test these ideas?

00:49:08.811 --> 00:49:11.951
And what was the task, the animal had to perform?

00:49:12.731 --> 00:49:20.991
Yeah, so the task was to train rats on this discrimination task with the visual

00:49:20.991 --> 00:49:23.011
stimuli being a discriminandum,

00:49:23.911 --> 00:49:29.231
or at least the positioning of CS plus versus CS minus stimulus would be the

00:49:29.231 --> 00:49:33.131
thing to be discriminated by the rats, determining their left or right choices.

00:49:33.791 --> 00:49:40.111
With the addition of tactile cues, also tickling the whisker or barrel cortex,

00:49:40.331 --> 00:49:45.011
because these are sandpaper cues where the rat would pass by and gain information

00:49:45.011 --> 00:49:47.011
about future amounts of record.

00:49:48.531 --> 00:49:53.131
So that was the idea. And by recording from both the visual cortex and barrel

00:49:53.131 --> 00:49:57.171
cortex, we can get an idea of how they interact.

00:49:57.171 --> 00:50:03.871
So, for instance, does the appearance of the visual stimulus affect also barrel

00:50:03.871 --> 00:50:05.811
activity, whisking activity?

00:50:08.031 --> 00:50:14.551
That would also be in line with a prediction from a hypothesis on multimodal

00:50:14.551 --> 00:50:19.471
integration, which I proposed a couple of years ago, and which also relates

00:50:19.471 --> 00:50:22.471
to consciousness or how modalities are actually coded. it.

00:50:23.451 --> 00:50:27.151
But then the additional areas are the perirhinal and hippocampus to look at this,

00:50:27.891 --> 00:50:32.671
let's say, potentially forward propagation of sensory information into the MTL

00:50:32.671 --> 00:50:39.351
hippocampus memory system with the additional hypothesis that at some point

00:50:39.351 --> 00:50:44.031
when the hippocampus start replaying the sequence might come back out and reach

00:50:44.031 --> 00:50:47.171
back to the neocortex again in reverse order.

00:50:47.991 --> 00:50:52.071
Okay. But how does this relate to consciousness? Oh, this last part does not.

00:50:52.251 --> 00:50:58.251
Oh, no. Okay. Well, indirectly, perhaps, because if we would accept that,

00:50:58.331 --> 00:51:02.651
let's say, notions of recognition are also part of your conscious experience,

00:51:02.811 --> 00:51:09.151
then things like periorhinal feedback to the neocortex could be very relevant. Mm-hmm.

00:51:09.671 --> 00:51:14.271
Okay. But I would also maintain that if you lose the hippocampus,

00:51:14.291 --> 00:51:14.991
you're still conscious.

00:51:15.331 --> 00:51:21.011
Right. Exactly right. Yeah. So now, the point is that now we have,

00:51:21.171 --> 00:51:23.711
you looked at four areas, right?

00:51:23.771 --> 00:51:28.511
We have the smetocentric cortex, CA1 in the hippocampus, you have primary visual

00:51:28.511 --> 00:51:31.991
cortex, you have perirhinal cortex, okay?

00:51:32.051 --> 00:51:37.331
And they're also cleanly organized in an anterior-posterior axis, right?

00:51:37.391 --> 00:51:43.071
So I guess you also did it on purpose so you can actually measure from them in a reliable way.

00:51:44.311 --> 00:51:50.791
So now then you developed a new measure that helps you to sort of look at these

00:51:50.791 --> 00:51:54.651
phase relationships between these different areas which you called,

00:51:55.231 --> 00:52:02.911
WPLI the weighted phase locking index which looked very interesting and then what did you find?

00:52:04.150 --> 00:52:07.770
Okay, well, yeah, so I confined the story today to gamma rhythms,

00:52:08.950 --> 00:52:14.670
which are these high-frequency, roughly 40 to 80 hertz oscillations.

00:52:16.510 --> 00:52:22.170
The first point of contention is to what extent the somatosensory cortex generates the gammas.

00:52:22.390 --> 00:52:24.670
The visual cortex is less contentious.

00:52:25.510 --> 00:52:30.070
We find that there are clear gammas. They're enhanced during active behavior.

00:52:30.870 --> 00:52:34.110
This might correspond to active or passive whisking.

00:52:36.010 --> 00:52:42.430
And in addition, the gammas are quite local. So at least the coherence of different

00:52:42.430 --> 00:52:49.110
field potentials is high within the local area of the somatosensory cortex,

00:52:49.430 --> 00:52:56.810
but not, let's say, somatosensory to visual coherence is almost nonexistent, very low.

00:52:56.810 --> 00:53:01.670
And the same for perirhinal somatosensory to hippocampal.

00:53:02.290 --> 00:53:07.170
Whereas if gamma would be really a central mechanism for communication between

00:53:07.170 --> 00:53:10.290
all brain areas sort of in a very global brain-wide fashion,

00:53:11.050 --> 00:53:12.130
this would not be expected.

00:53:13.430 --> 00:53:15.610
Were you surprised by that outcome? Did that surprise you?

00:53:17.790 --> 00:53:22.650
Not really, no. No. Not really because, yeah, on the one hand,

00:53:22.790 --> 00:53:26.930
there have been previous findings is an inter-areal gamma coherence in the visual system.

00:53:29.270 --> 00:53:34.090
But yeah, not all of those studies corrected for potential volume conduction problems.

00:53:34.410 --> 00:53:38.130
A lot of studies did not have the spikes in there to show that the gamma is

00:53:38.130 --> 00:53:41.050
really local, local cells are entrained to it.

00:53:42.070 --> 00:53:48.210
And so there are all kinds of ways to buy out of this idea of global gamma synchronization.

00:53:48.710 --> 00:53:52.410
Our findings do not contradict sort of short-range.

00:53:52.530 --> 00:53:56.450
Right, exactly. I think this is the key thing that you observed, right?

00:53:56.490 --> 00:54:03.990
That in all these areas you measured from, you found strong local coherence in gamma.

00:54:04.510 --> 00:54:08.270
So that means the neurons you're measuring from are all happily firing together

00:54:08.270 --> 00:54:11.970
in a gamma range. Right. At some phase relationship to each other. Yeah.

00:54:12.230 --> 00:54:17.070
But you do not find a similar coherence between these areas, okay? Right, yeah.

00:54:18.090 --> 00:54:22.850
There's a positive control. The areas do have gamma, but not with each other.

00:54:22.850 --> 00:54:25.790
Yeah, exactly. So this raises a number of interesting issues.

00:54:26.210 --> 00:54:30.990
So how do you then look upon this? Let's first look at the areas individually.

00:54:31.670 --> 00:54:39.590
So if you compare, let's say, V1 with somatosensory or perirhinal or hippocampus, CA1...

00:54:41.176 --> 00:54:45.516
Is that dynamics in the gamma range really very different between these areas?

00:54:46.916 --> 00:54:50.276
Between the somatosensory and visual cortex, not very much.

00:54:50.396 --> 00:54:54.976
No, no. They both have similar gamma range. They show good phase locking.

00:54:56.116 --> 00:54:59.456
In the hippocampus and perirhinal, the theta becomes very strong.

00:54:59.896 --> 00:55:07.516
In the slipstream of theta, you also see beta, which is roughly double the frequency, so 16, 20 hertz.

00:55:08.976 --> 00:55:11.896
So there are clearly different things going on

00:55:11.896 --> 00:55:18.996
in hippocampus you see more a chopping of the gamma because of this theta rhythm

00:55:18.996 --> 00:55:23.896
so yes you see a few spikes in gamma and then the system shuts down for a little

00:55:23.896 --> 00:55:29.676
bit and then it comes back again and that choppiness you would not see in V1 or perirhinal.

00:55:32.596 --> 00:55:37.496
Perirhinal also has a quite strong theta rhythm together with the hippocampus,

00:55:38.436 --> 00:55:44.376
probably the gammas that are locked to the theta cycle so they are not going on all throughout,

00:55:47.156 --> 00:55:50.196
the visual cortex tends to have strong gamma

00:55:50.196 --> 00:55:53.216
during the visual stimulation but also during

00:55:53.216 --> 00:55:56.776
the movement because actually the scene of the rat is totally or

00:55:56.776 --> 00:56:00.756
always changing so that means with the measurements

00:56:00.756 --> 00:56:04.696
you did we have these two cortical areas that really show their own kind of

00:56:04.696 --> 00:56:08.976
gamma dynamics and then we have some more hippocampal related areas where we're

00:56:08.976 --> 00:56:12.716
sort of theta starts to dominate the dynamics much more so we have two kinds

00:56:12.716 --> 00:56:17.996
of subsystems yeah so how do you then explain this,

00:56:18.736 --> 00:56:23.016
gamma dynamics in cortex so how is this generated yeah,

00:56:25.215 --> 00:56:29.775
Right. We know from the visual cortex that stimuli can drive the gamma.

00:56:31.995 --> 00:56:37.795
And so we presume, but cannot directly prove in this case, that also whisking

00:56:37.795 --> 00:56:41.875
movements or other somatosensory stimuli would drive the gamma.

00:56:41.975 --> 00:56:47.395
In addition, the gamma could be enhanced, for instance, by attentional processes

00:56:47.395 --> 00:56:52.355
or prefrontal feedback to the area, because that's also been shown in monkey studies.

00:56:53.675 --> 00:56:57.635
And yeah the factors

00:56:57.635 --> 00:57:01.555
driving the strong gamma coherence locally I don't

00:57:01.555 --> 00:57:05.115
think can be precisely disentangled here because during the active movement

00:57:05.115 --> 00:57:09.575
phase there's lots of things going on visual input reward expectation but wait

00:57:09.575 --> 00:57:14.715
I think you can say something about it now because you you have distinguished

00:57:14.715 --> 00:57:20.955
the different neural types involved in this and you could distinguish the

00:57:21.035 --> 00:57:24.715
inhibitory interneurons from your excitatory pyramidal cells.

00:57:25.075 --> 00:57:28.815
And you also found very specific phase relationships between them.

00:57:29.035 --> 00:57:32.275
Oh, yeah, in terms of the local circuits, we can make statements.

00:57:32.515 --> 00:57:35.135
Yeah, so there were two interneuron classes.

00:57:36.555 --> 00:57:41.015
These are the fast sparkers. That's the way you identify them in extracellular recordings.

00:57:41.955 --> 00:57:45.755
The broad sparkers are more like pyramidal cells, maybe some stellate cells.

00:57:47.175 --> 00:57:52.235
And there the special finding is that there are two classes of interneurons.

00:57:52.315 --> 00:57:58.915
One fires early in the gamma cycle, one late, whereas the pyramidal cells fire in between.

00:57:59.175 --> 00:58:06.235
So there's one class of interneurons that fire early and they're firing to the

00:58:06.235 --> 00:58:12.255
gamma or the entrainment to the gamma can be easily explained by pre-firing of the pyramidal cells.

00:58:13.152 --> 00:58:18.332
So that more points to an interneural network gamma mechanism, the ING mechanism.

00:58:19.212 --> 00:58:25.412
Whereas the late cells could more be involved in recurrent inhibition driven by the pyramidal cells.

00:58:25.912 --> 00:58:31.292
But would you see these early inhibitory cells, as I say, as a separate network

00:58:31.292 --> 00:58:34.392
of more like master controllers of the gamma?

00:58:38.012 --> 00:58:41.152
Well, they're certainly the earliest cells to fire in the gamma cycle.

00:58:42.072 --> 00:58:47.272
So what we think happens is that there's local excitatory input to these interneurons

00:58:47.272 --> 00:58:52.892
that excites them, but not from the same local population of pyramidal cells.

00:58:54.472 --> 00:58:58.312
They do inhibit each other, but in the rebound of this inhibition,

00:58:58.412 --> 00:59:00.532
they can also become excited.

00:59:00.732 --> 00:59:04.892
So there's rebound excitation or once the shunting inhibition is lost.

00:59:05.172 --> 00:59:09.632
And some of these interneuron classes have gap junctions. So if one spikes,

00:59:09.692 --> 00:59:15.672
it can trigger or stimulate firing in the gap junction coupled cell non-synaptically.

00:59:15.892 --> 00:59:21.632
So that could explain why there is a class of early firing cells. Right, exactly.

00:59:21.992 --> 00:59:26.772
But so they are exclusively coupled with gap junctions that would allow very

00:59:26.772 --> 00:59:30.792
rapid transduction with their fellow inhibitory early cells.

00:59:31.112 --> 00:59:37.772
Right, yeah. Okay. Yeah. Yeah, so we don't have an identification of what those

00:59:37.772 --> 00:59:43.352
cells are, whether those VIP interneurons or somatostatin or basket or chandelier cells,

00:59:43.532 --> 00:59:49.372
but there seem to be various classes that could conform to this ING scheme of

00:59:49.372 --> 00:59:50.912
interneuron-driven gamma. Exactly.

00:59:51.632 --> 00:59:56.812
So that might mean you have, let's say, a very tightly coupled network of interneurons

00:59:56.812 --> 01:00:04.832
that are sort of locally initiating then than a pyramidal interneuron-driven gamma oscillation.

01:00:05.852 --> 01:00:10.732
Yeah, yeah, yeah, yeah. It could be that the pyramidal cells start firing because

01:00:10.732 --> 01:00:14.792
they come out of that inhibition or because they receive additional excitation

01:00:14.792 --> 01:00:19.492
from other areas, but then most likely also excite each other locally.

01:00:19.792 --> 01:00:25.072
That would suggest that you also should see a spatial parcellation or fragmentation

01:00:25.072 --> 01:00:29.312
of this gamma oscillation in the red cortex. Is that true?

01:00:31.152 --> 01:00:37.152
Yeah, what people find generally is more gamma superficially in the superficial layers. Yeah. Right.

01:00:37.612 --> 01:00:44.172
Here, I have to say, we did record also deep, and there you also see some gamma.

01:00:44.612 --> 01:00:49.472
But these were not laminar probes. So we don't have an exact identification of the depth.

01:00:49.652 --> 01:00:53.572
So now we have learned a lot about local gamma, which is really cool.

01:00:53.692 --> 01:00:56.972
Okay. Okay. Thank you. But it turns out it has nothing to do with your original

01:00:56.972 --> 01:00:59.512
question, because you wanted to know about inter-aridal communication,

01:01:00.072 --> 01:01:02.732
which you believe will be in gamma, and it's not.

01:01:04.392 --> 01:01:07.972
At least a negative say, well, okay, it's not gamma. It should be something

01:01:07.972 --> 01:01:12.312
else. That's my question. So what is it? Yeah. Um...

01:01:13.810 --> 01:01:19.110
I think there are two possibilities. We're now looking at data and beta ranges for communication.

01:01:19.950 --> 01:01:23.870
Those seem to be working, especially for the hippocampus pyramidal system.

01:01:25.030 --> 01:01:30.890
But sometimes during some behavioral phases there, we see beta coherence with the sensory cortices.

01:01:31.330 --> 01:01:34.230
So maybe we're looking in too high a frequency range.

01:01:35.110 --> 01:01:42.910
It could also be the case that the really long range interesting stuff is in

01:01:42.910 --> 01:01:44.230
desynchronized assemblies.

01:01:47.650 --> 01:01:52.770
So, yeah, because, you know, conscious processing goes on in a largely desynchronized

01:01:52.770 --> 01:01:58.590
EEG state, it's not a given that it should happen in an oscillation mode.

01:01:58.590 --> 01:02:00.650
No, this is an interesting consequence, right?

01:02:00.730 --> 01:02:06.750
Because maybe by looking for synchronized states, they're maybe not as ordered

01:02:06.750 --> 01:02:12.470
as a rate code, but they are fairly ordered. and maybe that's still not the way to think about it.

01:02:12.510 --> 01:02:16.010
So if you really have to make a bet, do you think you're going to find any kind

01:02:16.010 --> 01:02:20.930
of informational coupling at lower frequency ranges? Do you think that's really plausible?

01:02:23.390 --> 01:02:26.750
Yeah, in terms of phasing, that could well be.

01:02:28.170 --> 01:02:31.070
In a way, we see that in Theta phase precession.

01:02:32.330 --> 01:02:39.070
Despite the low frequency of the Theta rhythm, there is distinct information coding in the phase.

01:02:39.390 --> 01:02:43.850
Well, but it modulates it still on top of a gamma code, right?

01:02:43.930 --> 01:02:46.430
Otherwise, there's nothing there. Yeah, yeah.

01:02:47.870 --> 01:02:52.210
The local gamma codes in visual cortex, for instance, also have some phase coding

01:02:52.210 --> 01:02:58.250
in the sense that there is stimulus information in the phase. It has been shown also.

01:02:59.270 --> 01:03:05.870
But it might also be that most of the information transmission is effective

01:03:05.870 --> 01:03:10.490
in a desynchronized mode. You still have synchronous spiking assemblies, but they're sparse.

01:03:10.990 --> 01:03:15.750
They're not having a particular special relationship to the mass synaptic potentials

01:03:15.750 --> 01:03:18.650
because there are just too few of them. You wouldn't pick them out.

01:03:19.470 --> 01:03:23.970
And yeah, by their regular external projections, they reach their targets.

01:03:26.430 --> 01:03:31.470
And that's also still a viable scenario. Right, but a bit of a messy one.

01:03:33.442 --> 01:03:38.222
Yeah, and you would like to have, you know, gain control over that.

01:03:39.002 --> 01:03:46.522
But in your picture on this, would you still believe that at least these pathways

01:03:46.522 --> 01:03:47.702
are highly coordinated?

01:03:48.162 --> 01:03:53.742
That means, let's say it runs all over the thalamus, so at least there's only one hub doing this.

01:03:53.962 --> 01:03:56.822
Well, on the other hand, what we already talked about, the ventral striatum

01:03:56.822 --> 01:04:00.242
or hippocampus, you see they have convergent input from many different areas,

01:04:00.442 --> 01:04:07.942
right? So would you still think about some anatomical ordering of this very divergent face code?

01:04:08.162 --> 01:04:14.362
Or do you also see that as fairly open, like many different anatomical channels

01:04:14.362 --> 01:04:18.822
providing these kinds of codes to all parts of the brain?

01:04:20.662 --> 01:04:24.802
Yeah, that's a difficult one. Yeah, so like in the visual system,

01:04:24.942 --> 01:04:33.442
you do recognize mappings or sometimes retinotopic, sometimes craniotopic, but lots of mappings.

01:04:33.602 --> 01:04:40.862
And I think it's actually a key question how or whether neurons on similar locations

01:04:40.862 --> 01:04:46.002
of maps in the same framework communicate better with each other.

01:04:48.362 --> 01:04:53.622
So that you would kind of use the spatial location of a feature as also a binding

01:04:53.622 --> 01:04:57.142
queue to make it belong to some other queue at the same location.

01:04:58.417 --> 01:05:02.817
If we don't have a correspondence between spatial mappings or retinotopic mappings,

01:05:03.017 --> 01:05:07.937
it seems an intrinsic problem of how you piece things together in space.

01:05:07.937 --> 01:05:10.717
No, but it's interesting, right? Because you're looking for order in some sense.

01:05:10.797 --> 01:05:13.657
And you also started with this modular view. Like we have modules.

01:05:14.297 --> 01:05:15.957
Modules have their local operations.

01:05:16.717 --> 01:05:19.857
They're encapsulated informationally. They have to be, although it's just a module.

01:05:20.217 --> 01:05:23.457
And then in a very ordered way, they exchange information with each other.

01:05:23.457 --> 01:05:27.277
Like hippocampus might send place information to the ventral striatum,

01:05:27.357 --> 01:05:30.117
and the ventral striatum does something about the reward prediction.

01:05:31.277 --> 01:05:35.237
And then, I mean, you follow a very logical process there. And then you said

01:05:35.237 --> 01:05:39.317
like, okay, let's see how then these structures really exchange this information.

01:05:39.917 --> 01:05:41.677
And then actually you don't find anything.

01:05:42.837 --> 01:05:47.957
So is this whole modular scheme you originally proposed maybe already looking

01:05:47.957 --> 01:05:52.097
in the wrong direction? Is the brain really that cleanly modular as you would

01:05:52.097 --> 01:05:57.337
like to have it being an experimentalist who actually has to know where to go, right?

01:05:58.517 --> 01:06:03.877
Yeah. Well, the position I tried to defend was that it's modular and also not modular.

01:06:04.017 --> 01:06:09.757
So there are also modulations like this state switching on top of everything

01:06:09.757 --> 01:06:15.317
that indicate that there might be some more global modulation going on.

01:06:15.817 --> 01:06:20.677
Where that comes from, we don't know yet. It could be thalamus or cortical or

01:06:20.677 --> 01:06:22.377
maybe prefrontal driven.

01:06:22.557 --> 01:06:29.597
But yeah, there are certain events in the system that can happen and switch

01:06:29.597 --> 01:06:31.337
multiple structures at the same time.

01:06:32.257 --> 01:06:34.437
What would such an event be?

01:06:37.157 --> 01:06:43.457
Well, in the case of the hippocampus and ventral striatum, where the cue,

01:06:43.617 --> 01:06:45.037
the light that switches on,

01:06:46.137 --> 01:06:51.397
would signal to the rat like okay you can go now there's a reward to be obtained,

01:06:52.177 --> 01:06:54.397
and you could see the way that,

01:06:55.437 --> 01:06:59.337
Large parts of the brain and the body have to do with this reward acquisition.

01:06:59.857 --> 01:07:03.877
So you better get going and change your system.

01:07:04.017 --> 01:07:10.377
It could be that you do need a system that recognizes the relevance of the cue.

01:07:10.517 --> 01:07:14.997
This could be a mycolyte prefrontal, maybe earlier on, perirhinal maybe.

01:07:16.857 --> 01:07:21.357
And for instance, the prefrontal is in a pretty good position to modulate these systems.

01:07:21.937 --> 01:07:27.157
Projected directly into striatum, but also indirectly through the parahippocampus

01:07:27.157 --> 01:07:28.817
and perirhinal into the hippocampus.

01:07:29.737 --> 01:07:35.877
So at least you have a pretty solid top-down mechanism there to do it,

01:07:35.937 --> 01:07:38.337
but speculation weather. Right.

01:07:38.837 --> 01:07:41.337
But still you're stable to say, well, it's modular and not modular.

01:07:41.537 --> 01:07:43.757
You understand it sounds somewhat paradoxical.

01:07:44.797 --> 01:07:50.597
Yeah, of course. Yeah, yeah, yeah. On the one hand, you want to reconcile the evidence for...

01:07:51.357 --> 01:07:54.617
Local specialization because there's lesion evidence there's neural

01:07:54.617 --> 01:07:57.777
coding evidence on the other hand you say

01:07:57.777 --> 01:08:01.637
well hey but yeah there are also these overriding events that have common effects

01:08:01.637 --> 01:08:08.097
in both places right so then um so surely you have been marching through this

01:08:08.097 --> 01:08:13.217
road of brain for quite a while now and gained an incredible amount of knowledge

01:08:13.217 --> 01:08:16.277
about that system at the system level.

01:08:18.117 --> 01:08:21.557
So then what should be a surrealist law in our study of the brain?

01:08:23.777 --> 01:08:24.697
That's a good one.

01:08:27.637 --> 01:08:29.677
A general law? Yeah.

01:08:35.597 --> 01:08:42.197
I think that anything meaningful that happens in the brain is a network phenomenon. That would be my law.

01:08:42.957 --> 01:08:47.497
Okay, cool. That, you know, single cells that fire don't mean anything.

01:08:48.477 --> 01:08:49.797
They don't signify.

01:08:50.837 --> 01:08:54.337
You're not conscious of it. It's just, you know. Noise.

01:08:55.557 --> 01:09:00.597
At least we, yeah, we have to look at it from at a higher level, I think. Very good.

01:09:00.817 --> 01:09:04.737
And then, so five years from now, I'm going to come up to Amsterdam and I'm

01:09:04.737 --> 01:09:08.637
going to confront you with a hypothesis you're going to declare to me today.

01:09:10.637 --> 01:09:16.497
So what's the key prediction that you see in front of your mind's eye right

01:09:16.497 --> 01:09:20.717
now that you know you're going to have confirmed five years from now?

01:09:21.931 --> 01:09:25.111
But that would be a very low-level, easy prediction.

01:09:27.691 --> 01:09:35.271
It just turns out another way. A prediction that could be confirmed in five years from now.

01:09:37.291 --> 01:09:42.011
Something ambitious and impressive. Yeah, not something small,

01:09:42.291 --> 01:09:44.071
petty thing like gamma rhythm.

01:09:45.491 --> 01:09:46.991
You can do better than that.

01:09:50.191 --> 01:09:55.931
Well, what we're working on a lot is multimodal integration these days,

01:09:56.031 --> 01:09:59.831
but also perception actually in rats and mice.

01:10:00.931 --> 01:10:06.271
So we do have these four area recordings also going on in, let's say,

01:10:06.371 --> 01:10:10.731
lower and higher visual areas, including singlet and parietal.

01:10:10.731 --> 01:10:19.851
And there I would predict that visual perception, as we also link it to consciousness,

01:10:22.191 --> 01:10:30.671
involves the discrete and repeated iterative interactions between lower and

01:10:30.671 --> 01:10:34.431
higher areas but not exclusively in a top-down fashion.

01:10:35.451 --> 01:10:41.411
Look, you're being rather demanding on my short-term memory here.

01:10:42.071 --> 01:10:42.991
So what's the prediction?

01:10:45.491 --> 01:10:55.071
Well, basically that visual perception also is a network phenomenon that not

01:10:55.071 --> 01:10:59.871
only depends on high to low level feedback,

01:11:00.011 --> 01:11:01.591
in this case for the visual system,

01:11:02.731 --> 01:11:11.211
but should be more viewed as an ongoing, short-lasting reverberation also involving

01:11:11.211 --> 01:11:12.691
the higher systems. Very, very good.

01:11:12.831 --> 01:11:15.411
So, Cyril Pennard, thank you very much for this conversation.

01:11:15.811 --> 01:11:17.251
Thank you, Paul, as well.

01:11:18.320 --> 01:11:24.240
Music.

01:11:23.731 --> 01:11:29.411
The CSN podcast was produced by the Convergent Science Network of Biometrics

01:11:29.411 --> 01:11:35.851
and Biohybrid Systems, a project funded by the European Sevens Research Framework Program.

01:11:37.311 --> 01:11:42.671
For more interviews, recorded lectures, or upcoming conferences in the field

01:11:42.671 --> 01:11:48.911
of biometrics and biohybrid systems, go to csnnetwork.eu.

01:11:49.200 --> 01:11:57.840
Music.