Listeners:

Top listeners:

00:00 00:00

Go to album

Paul Verschure on consciousness and distributed adaptive control CSN Podcasts
Edvard Moser on grid cells and entorhinal cortex CSN Podcasts
Giacomo Rizzolatti on mirror neurons and action understanding CSN Podcasts
Robert Axelrod on game theory and prisoner's dilemma CSN Podcasts
Adrian Owen on disorders of consciousness and vegetative state CSN Podcasts
Jonathan Whitlock on markerless motion capture and posterior parietal cortex CSN Podcasts
Luis Puelles on neuroanatomy and prosomeric model CSN Podcasts
Zoltan Molnar on subplate neurons and cortical development CSN Podcasts

Narender Ramnani on cerebellum and cortico-cerebellar loops

Episode 10 15.03.2015

PLAY EPISODE

Season 2015

0 Followers 13 Episodes

Description

What if the cerebellum is not just a motor structure but a universal learning machine wired to the entire frontal lobe? Neuroscientist Narender Ramnani explains how the anatomy of cortico-cerebellar loops forces us to rethink the cerebellum’s role , from fine-tuning movements to supporting rule learning, cognitive error processing, and the transition from deliberate to habitual behavior. Subscribe for more from the Convergent Science Network podcast series. Narender Ramnani joins Paul Verschure and Tony Prescott at the BCBT summer school to present evidence that the cerebellum communicates not only with motor cortex but with diverse regions of the prefrontal and parietal cortex through closed anatomical loops. He traces this insight to work by Peter Strick and by Schmahmann and Pandya, which revealed that the cerebellum’s connectivity is far broader than the classical motor view suggests. If the cerebellar microcircuit is computationally uniform , the same Marr-Albus learning architecture repeated across the structure , then the same transform applied to motor inputs should also apply to cognitive inputs arriving from prefrontal cortex. The discussion digs into the critical question of error signals. In classical conditioning, the inferior olive delivers a clear teaching signal. But what serves as the error signal for prefrontal-cerebellar loops? Ramnani presents anatomical evidence for at least two routes: dopaminergic projections from the VTA that send collaterals directly to cerebellar Purkinje cells, and prefrontal projections from the anterior cingulate cortex that reach the inferior olive. He also describes fMRI evidence showing that cerebellar activity during instrumental rule learning mirrors the Purkinje cell pause seen in classical conditioning , a decrease in BOLD signal consistent with reduced Purkinje cell firing during learning. Key topics include the modular independence of cerebellar loops, why cerebellar modules must communicate through neocortex rather than internally, the system-one versus system-two distinction in habit formation, how to interpret fMRI signals in the cerebellum, and the challenge of building computational models that capture these cortico-cerebellar interactions. Part of the Convergent Science Network podcast series from the BCBT Summer School.

Tagged as:
Cerebellar cerebellum Classical Conditioning cortico-cerebellar loops Inferior Olive Motor Rule Learning

About the author

CSN Podcasts

Both the triumphs of humanity and its most evil deeds have resulted from collaboration. In a time where humanity is required to aspire to the former and minimize the latter, the question arises of how collaboration arises and why it fails. Surprisingly, this phenomenon, so central to who we are, is not well understood. Hence, a collaborative effort is required to understand collaboration in its full biological, psychological, sociological, cultural, and economic complexity and to translate this understanding into operational impact. This series of podcasts is one step toward achieving these complementary goals. The Collaboration Podcast presents interviews with people who are central orchestrators of collaboration in various domains including business, government, science, art, health, sustainability, and the military. The discussions were conducted by Prof. Dr. Paul F.M.J. Verschure and members of the Program Advisory Committee of the Ernst Strungmann Forum on Collaboration (https://www.esforum.de/forums/ESF32_Collaboration.html) during 2021 and had the goal to sketch a map of opportunities, challenges, and obstacles in human collaboration. The forum took place in May 2022, and now we would like to share this series of interviews with a broader audience. The full report of the Forum will be published in 2023 by MIT Press. The podcast was produced by the Convergent Science Network (https://www.convergentsciencenetwork.org/). Context: The stability of social systems depends critically on realizing sustainable methods of “collaboration,” yet how and by which means collaboration is achieved is not clearly understood; neither are the conditions or processes that lead to its breakdown or failure. Collaboration can be understood as cooperation between agents toward mutually constructed goals. Part of the reason for our lack of understanding is that the phenomenon of collaboration is, by nature, a highly multidisciplinary problem, and effective research into its complexities has been difficult to achieve across the broad range of scientific and technical disciplines involved. The need for a fundamental understanding of collaboration, however, has become increasingly important. Not only does humankind demand answers as it attempts to address critical challenges at multiple scales (e.g., climate change, migration, enhanced automation, social and economic inequality), but ever-increasing technological and economic means of interconnecting people and societies are disrupting long-established, familiar patterns of how we interact. Radical technological changes that are ongoing have the potential to reshape collaboration in ways that are currently hard to predict or influence (e.g., by altering configurations in interaction, information creation, and modes of communication). On one hand, such changes could disrupt hitherto stable forms of collaboration by affecting critical communication channels and traditional roles, as can be observed in the rapidly changing patterns in governance, commerce, and social interaction. Conversely, technology could lead to the emergence of novel, successful forms of collaboration that deviate from traditional “hierarchical” architectures. Evidence of this can be seen in areas as diverse as highly automated manufacturing plants, the open science movement, collaborative software repositories, user-centered services, and the sharing of economy-based modes of organization. Without a fundamental understanding of the mechanisms, processes, and boundary conditions of collaboration, it is not possible to evaluate or predict which of these possible scenarios are sustainable or even plausible. The Forum “How Collaboration Arises and Why it Fails” (May 8–13, 2022, Location: Frankfurt am Main, Germany) Chairs: Andreas Roepstorff and Paul Verschure Program Advisory Committee: Jenna Bednar, Julia R. Lupp, Bhavani R. Rao , Andreas Roepstorff, Ferdinand von Siemens, and Paul Verschure

Timestamp

00:00:00 - That's a cheeky question, and you're going to hold me to that,
00:00:03 - aren't you? Of course, what do you think?
00:00:04 - Why do you think we're recording this? Absolutely. This is the Convergent Science Network podcast.
00:00:11 - Leading researchers in the domain of neuroscience, brain theory,
00:00:15 - and technology are interviewed by Paul Vershoor and Tony Prescott.
00:00:20 - It's Paul Vershoor with the Convergent Science Network podcast,
00:00:24 - and I'm here with my colleague Tony Prescott. Scott.
00:00:27 - We're both in our BCBT summer school here in Barcelona 2015.
00:00:32 - And I'm here with Narender Ramnani, who was giving us actually a very different
00:00:39 - kind of view on cerebellum because,
00:00:41 - the standard view on cerebellum is really like, well, this is sort of this little
00:00:45 - appendix that hangs off your brainstem, and it might be involved in some form of motor control.
00:00:50 - But But when things really get interesting, cognition, decision-making, forget it.
00:00:56 - We go look at cortical areas, mainly prefrontal cortex.
00:01:00 - And you sort of turn that whole perspective upside down. So how did you come to that perspective?
00:01:06 - Well, that's an interesting question. So this goes back again to everything
00:01:12 - being the foundation for my views are really the anatomy of the system.
00:01:19 - And so one of the things that I became very curious about was the looped architecture
00:01:26 - of the corticocerebellar system and this idea that our knowledge of what the
00:01:33 - cerebellum is talking to is incomplete. complete.
00:01:36 - And then over the years, the gap started to become filled and some very nice
00:01:42 - work came out from Peter Strick's lab and from the work that Schmarman and Pandya
00:01:48 - did shortly before that.
00:01:51 - And this work really highlighted the idea that the cerebellum isn't simply talking to the motor to cortex,
00:02:00 - there are many other parts of the cerebellum that are communicating with really,
00:02:08 - quite a diverse range of different areas of the neocortex, such as the prefrontal cortex.
00:02:17 - And, in fact, many different parts of the frontal lobe and also the parietal
00:02:22 - cortex. So there's this diverse range of inputs.
00:02:26 - So the question is, if it's just a motor structure, what on earth is it doing
00:02:31 - just doing motor control?
00:02:34 - And so that's really where it all kicked off from.
00:02:38 - Right. You began also by talking about what is now quite a classical view of
00:02:44 - Sarah Ballin, sort of the Mar-Albus theory.
00:02:47 - Could you just summarize that for us again?
00:02:50 - And also, do you think that that's still a useful framework to think about Sarah
00:02:54 - Ballin? Oh, undoubtedly.
00:02:55 - I mean, it's a framework that hasn't changed very much, actually.
00:02:58 - I mean, people have tinkered around the edges with it, but the fundamental message
00:03:04 - of what Mara and Albus were talking about is still very much there.
00:03:09 - And the perspective that they offer is, the theoretical perspective that they
00:03:14 - offer is this idea that Purkinje cells are the computational unit,
00:03:20 - the principal computational unit,
00:03:22 - and that this is a system that's capable of plasticity for supporting learning.
00:03:28 - And that learning is achieved by adjusting the inputs into Purkinje cells.
00:03:36 - And these inputs are.
00:03:40 - Typically parallel fibers and those changes occur under the guidance of a teaching
00:03:46 - related error signal from the inferior olive now having said that there are
00:03:52 - new perspectives emerging um,
00:03:55 - They also talk about adjusting the synapses between cerebellar units and their inputs.
00:04:06 - And I think it's recognized now that this is not the only source of plasticity
00:04:11 - in the cerebellum, but you also have plasticity involving a lot of other cell
00:04:18 - types in the cerebellar cortex.
00:04:19 - But one of the basic messages out of that is that there's a sort of microcircuit.
00:04:25 - And that it's repeated across the cerebellum. Do we still hold with that?
00:04:31 - There's very little data to argue against that view. And so the microcircuit
00:04:38 - or the microcomplex idea is still something that's very much in play.
00:04:44 - And the anatomy and physiology work bears that out really,
00:04:48 - particularly in relation to the work that's been done using eye movements or
00:04:54 - using classical eye blink conditioning where people have identified very specific
00:05:00 - microzones that might contain the Purkinje cells that house that plasticity.
00:05:07 - But as you were saying, sort of classically, people have thought this microcircuit
00:05:11 - is doing something in the motor domain.
00:05:13 - Yeah. But you now want to generalize it. Absolutely. So the architecture of
00:05:18 - the cerebellum is considered to be quite uniform.
00:05:21 - Form, it's really very, very different from the architecture that you find in
00:05:24 - the cerebral cortex, where site architecture is quite diverse.
00:05:29 - So people will have come across the concept of Brodmann areas in the cerebral cortex, where the...
00:05:38 - Area 4 has a different profile of cells from its neighbors in Area 6 or wherever.
00:05:47 - The idea about the cerebellar cortex is that it's much, much more uniform than that.
00:05:54 - So the computations that are going on in any given location will be the same
00:06:01 - in one location as in any other.
00:06:04 - So that idea still holds up. Although people are starting to find some interesting
00:06:10 - biochemical differences that suggest that there may be some interesting but
00:06:17 - subtle differences between these areas.
00:06:22 - So one of them, for example, is the idea of, there's this substance called aldolase C,
00:06:33 - or zebrin, that's differentially expressed in different parts of the cerebellar cortex.
00:06:40 - And one of the interesting things about zebrin is that if you stain the cerebellum
00:06:45 - for zebrin, you get a stripy parasagittal organization.
00:06:49 - That in itself tells you something about the parasagittal functional organization of the cerebellum,
00:06:57 - but it also suggests that maybe there are some subtle differences in the way
00:07:02 - that these adjacent zones compute. Right.
00:07:08 - And how many zones do we think we have, say, in the primate?
00:07:12 - Oh, crikey. That's a difficult question to answer because….
00:07:19 - I think as you go out more laterally, they become more and more refined.
00:07:23 - And the other thing, of course, is there are slightly different ways of characterizing a zone.
00:07:30 - I mean, Zebrin is one way, but there are many, many subzones, I think.
00:07:37 - The system I know better, the Basal Ganglia, has a similar notion of a microcircuit,
00:07:42 - which is repeated across.
00:07:44 - Cross yeah but when you drill down and ask people to define
00:07:47 - the micro circuits it all gets very hazy
00:07:50 - yeah and you end up with some people say there's larger domains
00:07:53 - of which there might be a handful and
00:07:57 - some people say well there's very small micro domains i mean
00:08:00 - so is is there still a controversy in the cerebellum about
00:08:03 - the granularity of the domain i think there probably is
00:08:06 - i mean there's still this slightly old-fashioned controversy that
00:08:10 - continues to rumble on about whether you have patchy somatotopy that's determined
00:08:16 - by the parallel fiber inputs or whether there's this other organization that's
00:08:24 - determined by, you know, Zebrin.
00:08:26 - But the other interesting thing I should say about Zebrin is that it seems to
00:08:30 - correspond very closely with one of the important input systems from the inferior olive.
00:08:41 - So climbing fibers project to the cerebellar cortex almost on a one-to-one basis.
00:08:48 - So one climbing fiber will completely dominate the physiology of any given Purkinje cell.
00:08:55 - And it seems as though the parasagittal organization.
00:09:03 - That we find with Zebrin seems to correspond with the parasagittal organization
00:09:09 - as determined by the projections from the inferior olive.
00:09:12 - So there's an interesting story there.
00:09:18 - So the microcircuit is really defined by the climbing fiber.
00:09:22 - Arguably, that's what's been claimed by some authors, yes.
00:09:27 - Absolutely. So you also made the claim that there's sort of this co-evolution
00:09:31 - of cerebellum and cortex.
00:09:33 - And that in that sense, the cerebellum performs like a universal transformation,
00:09:38 - from its inputs that it receives over the pons to its outputs that go out over
00:09:42 - the deep cerebellar nuclei.
00:09:44 - But also as Maureen Albers tells us, this is all dependent, independent how this output is shaped.
00:09:49 - It's all dependent on the error signal you get from your inferior olive.
00:09:53 - So now in the case of classic conditioning, it's fairly easy to understand where
00:09:57 - that error signal comes from because it's linked into the periphery that just
00:10:00 - says, okay, you just got shocked on your orbit or you got an air puff hit on your cornea.
00:10:04 - But if we now say, well, a huge chunk of cerebellum is actually bidirectionally
00:10:09 - interfaced to prefrontal.
00:10:11 - What's the error signal in that case? Well, that's another really fascinating question.
00:10:15 - I think one of the problems that we're confronted with is that there's so little
00:10:20 - data, so little physiological data that can answer that question.
00:10:25 - And it kind of highlights how much work there is left to do.
00:10:29 - But some of those questions could be answered in terms of the anatomy.
00:10:33 - We know, for example, that there are parts of the dopamine system that encode prediction error.
00:10:41 - We also know about the fact that elements of the dopamine system are sending
00:10:49 - their outputs to the cerebellum,
00:10:52 - and so there's a route there for this information to reach it.
00:10:57 - For example, the VTA not only sends its error signals off to quite large areas
00:11:06 - of the prefrontal cortex.
00:11:07 - But interestingly, there are collaterals that branch off from those projections
00:11:13 - that end up forming connections with Purkinje cells. And these are dopamine connections.
00:11:21 - So, it's as if the cerebellum is receiving a copy of the error signals that
00:11:29 - are reaching the neocortex.
00:11:31 - Okay. So, you're saying maybe for interaction with frontal areas,
00:11:36 - if your olive becomes less important to convey error and will be more derived
00:11:41 - from a VTA-dependent dopamine signal, which might signal something like a reward prediction error.
00:11:46 - Does that would be the idea? Well, that's one thing that one could claim.
00:11:51 - I think we shouldn't underplay the importance of the olive.
00:11:54 - There is recent data that has demonstrated there
00:11:59 - are in rats prefrontal inputs to parts of the olive that then project on to
00:12:06 - areas like cruce one and cruce two that receive input from the prefrontal cortex from the other system.
00:12:15 - So there's a lot of different anatomical routes through which that could happen.
00:12:20 - And one of these areas in the prefrontal is the anterior cingulate.
00:12:24 - And we know that the anterior cingulate does play a major role in the monitoring
00:12:30 - of errors that are cognitive in origin, you know, errors in your own cognitive
00:12:35 - processing, for example.
00:12:37 - And those areas do seem to send in rats and output to the olive.
00:12:42 - And then the olive then sends those projections back to... For you,
00:12:47 - which area would that happen?
00:12:48 - Would that then play out in red? What areas are we talking about?
00:12:52 - Which neocortical areas? So the anterior cingulate cortex.
00:12:56 - I don't remember off the top of my head which particular anterior ACC areas,
00:13:01 - but Tom Rickrock has basically characterized that system.
00:13:06 - And if we can just run over the thalamus to the inferior olive, Or is this a collateral?
00:13:11 - No, this would go directly from the neocortex to the olive.
00:13:19 - And from there on into cerebellum are direct inputs.
00:13:25 - And these are topologically mapped. That means these anterior cingulate projections
00:13:32 - that end up in the inferior olive target climbing fibers.
00:13:36 - Yes. The target regions of the cerebellum are recurrently coupled to that same bit of neocortex.
00:13:42 - That's right, because the study was conducted using transsynaptic traces.
00:13:49 - And because these traces have the ability to jump synapses, it's possible to
00:13:54 - do that point-to-point connectivity.
00:13:56 - Right. So that's cool, right? Because it means you're heuristic to say,
00:13:59 - okay, in case we don't understand, let's follow the anatomy.
00:14:02 - Yeah. It's then again paying off. At least the idea of having the finger olive
00:14:06 - as your error signal generator is still intact also for these interactions. Absolutely.
00:14:12 - And I think the other thing it does is to open up the possibility of physiological
00:14:19 - studies to try and investigate these circuits at a functional level.
00:14:24 - Right but now to finish up the inferior olive okay I just want to say that,
00:14:30 - the other thing which however I felt
00:14:33 - I didn't see back in your in your diagram in a too prominent role is a negative
00:14:38 - feedback between the deep nucleus and the inferior olive I mean it might sound
00:14:42 - like a detail but if it turns out that these Purkinje cells are actually actively
00:14:47 - controlling the errors that they receive over the inferior olive so um.
00:14:53 - Is this kind of error feedback control that occurs within the cerebellum for
00:14:58 - you computationally less relevant
00:15:01 - at this stage or just something that you ignore for other reasons?
00:15:04 - I think it's an extremely important facet of the circuit.
00:15:09 - And in fact, I think if I'm not mistaken, there was an inhibitory output coming
00:15:15 - out in those diagrams that I showed projecting back to the olive.
00:15:18 - One of the things that this seems to be involved in is the damping down of error-related activities.
00:15:31 - So, for example, in classical eibling conditioning, you'll be aware,
00:15:36 - perhaps more than I would, that this pathway is important for decreasing,
00:15:45 - in some ways, the salience of the unconditioned stimulus. this.
00:15:49 - So, you know, Jerry Hessler's work has shown that the,
00:15:55 - and Chris Yeo's work has also shown that the amplitude of the US is decreased
00:16:01 - in the presence of a conditioned response.
00:16:07 - So if there's a copy, there's an inference copy, if you like,
00:16:10 - of the CR that leaves the cerebellum and on its way,
00:16:15 - it basically goes down back to the olive and reduces the salience of the US.
00:16:23 - There are some interesting parallels that one could apply that to in the cognitive domain.
00:16:31 - If there's a system like the cerebellum that's specialized for feed-forward
00:16:37 - behavior that that doesn't really benefit from feedback,
00:16:41 - and indeed could be impaired by any sort of ongoing feedback if it causes an
00:16:46 - interruption to the processing.
00:16:48 - You do want a system that dampens down the effects of any ongoing error feedback coming into the system.
00:17:04 - I wanted to clarify about your diagram. So you have the inferior olive feeding
00:17:08 - back to three areas of cerebellum, and then you have VTA feeding into another area. Is that right?
00:17:16 - Yeah. So is VTA, is dopamine signaling something similar then to what the inferior
00:17:25 - olive is signaling in these other areas?
00:17:27 - Well, the diagram that you're talking about is simplified, And there are a few
00:17:33 - speculative things about it.
00:17:34 - But in essence, there will be some things that are similar.
00:17:43 - But there will probably be some things that are different too.
00:17:47 - So in answer to your question, what would be similar?
00:17:53 - One of the things that it could signal are the higher level cognitive errors
00:18:00 - that we spoke about earlier.
00:18:02 - Um the the what could be different um again you know this is speculative because
00:18:10 - nobody's ever sat down and recorded from both of them to see how they differ um.
00:18:18 - But the dopamine system signals reward related error and so that's why you know
00:18:25 - i've included it in that uh in that pathway and that diagram yeah i mean so
00:18:29 - you have these overall domains of cortex,
00:18:32 - which I guess they're defined in part by where they get their input from in cortex,
00:18:37 - but then also they're defined in part by where they get their feedback from.
00:18:43 - Yes, although I'm not ruling out the possibility that the olive could also be
00:18:48 - feeding into the high-level circuits that I was mentioning. I'm not ruling that out.
00:18:55 - It's just that we need to see some sort of concrete evidence for their existence
00:19:02 - before we could do that. Sure.
00:19:06 - So then, you presented us with a really fascinating diagram,
00:19:12 - which I liked a lot, and there's a lot to discuss about it.
00:19:16 - But, of course, there's also data behind it. And I think what's really impressive
00:19:22 - is that you've been able actually to replicate a lot of the phenomena that we
00:19:27 - know about cerebellar learning in humans.
00:19:29 - So could you, to what extent is, let's say, at least in your mind,
00:19:35 - is the response of the human cerebellum actually identical or similar to what
00:19:41 - you might expect in a rodent or a macaque,
00:19:45 - under classical conditioning conditions or VOR?
00:19:49 - Yeah, so one of the things that I alluded to in my talk was this idea.
00:19:57 - Of a conditioned stimulus being processed similarly, regardless of whether it's
00:20:06 - involved in the process of classical conditioning or whether it's involved in
00:20:10 - the process of instrumental learning.
00:20:11 - And so one of the things that I found very interesting was to basically draw
00:20:16 - from the classical conditioning literature because things are so very well known in that area.
00:20:21 - And one of the things that I mentioned was the idea that when you record from
00:20:32 - the eye blink microzones in the cerebellar cortex,
00:20:35 - you find a decrease in the activity of the Purkinje cell in response to a conditioned
00:20:41 - stimulus during classical conditioning.
00:20:45 - So the issue is that if If there is such a thing as a universal cerebellar transform,
00:20:52 - where a similar form of information processing is applied to, let's say, inputs...
00:21:02 - Going into a part of the cerebellar cortex that's receiving from prefrontal
00:21:06 - cortex, if you take a higher level form of learning, for example,
00:21:11 - like instrumental learning of rules, the question would be, do you see something similar?
00:21:17 - Do you see a decrease in the activity of a patch of cerebellar cortex when a
00:21:28 - CS is presented at the same time?
00:21:30 - And the work that I mentioned seems to bear this out. There seems to be a decrease in activity.
00:21:35 - Which is important, right? Because on the one hand, you replicate what we would
00:21:40 - expect from physiology, which is we obtain a pause in pericardial cell firing,
00:21:46 - which means less oxygen consumption and reduced blood flow.
00:21:50 - But it also means the pause is not dependent on another, let's say,
00:21:55 - population of interneurons And so now inhibits the perky cells because then
00:21:59 - your blood flow pattern might not have changed so much.
00:22:01 - Would you agree with that? Absolutely. And I think one of the things we have
00:22:04 - to be very careful about when we interpret imaging is, you know,
00:22:09 - what does an increase mean?
00:22:11 - You know, because this could mean an increase in inhibitory activity which could
00:22:16 - be manifest in a different way if you're using electrophysiology.
00:22:19 - One of the things that was very interesting about the work that I've just mentioned
00:22:26 - that uses electrophysiology is the use of gabazine.
00:22:31 - And so gabazine inhibits the action of neurotransmitter, inhibitory neurotransmitter,
00:22:38 - inhibitory inputs onto Purkinje cells. else.
00:22:43 - But the CS-related decrease that you observe is still there,
00:22:49 - despite the fact that the gabazine has acted on the system.
00:22:54 - And what that suggests is that the input of the,
00:22:59 - that's driving this inhibition are probably the excitatory inputs coming in via parallel fibers.
00:23:07 - And the value of understanding that for imaging is that if you have a decrease
00:23:16 - in activity during imaging, it's probably the case that there are,
00:23:22 - you know, there's a similar explanation involved that doesn't involve the ramping
00:23:28 - up of activity by increased metabolic demand, for example, from inhibitory endurance. Yes.
00:23:34 - Is there a sort of subtleties about how you interpret the fMRI data with cerebellum?
00:23:40 - Because when you have learning, you actually have decreased activity.
00:23:44 - Is that right, generally? So as you see learning proceed, you don't necessarily
00:23:51 - expect once the system is tuned up Do you see a lot of activity there?
00:23:55 - I think sometimes it's very difficult to interpret this sort of data.
00:23:59 - We know what we can say is that it's consistent with the electrophysiology.
00:24:04 - But there are studies that show learning-related increases in cerebellum too.
00:24:10 - And of course, we will never know why because we don't know what cell types
00:24:15 - are generating that information.
00:24:19 - We basically don't know what is driving that activity.
00:24:23 - So, for example, in a study of learning where you see an increase,
00:24:31 - that could be that you're,
00:24:36 - you're ramping up activity in the cerebellum for other reasons.
00:24:38 - It may not be that you see this sort of Mar and Albus type change that I spoke of before,
00:24:47 - or this Albus type decrease that we were comparing with.
00:24:55 - It's possible that you're getting inputs in that are ramping up the signal from
00:25:00 - other areas of cortex for other reasons.
00:25:03 - Um learning related activity
00:25:06 - is not always going to be manifested as
00:25:09 - a decrease in the cerebellum so what's the
00:25:12 - strategy then what do you have that's a hard one um imaging is is is a great
00:25:20 - tool to use for many reasons i mean not least because you can image the entire
00:25:25 - brain in one go and plus the fact um uh.
00:25:33 - You could look at anatomically very specific areas and so on.
00:25:39 - There are lots of benefits to imaging, including the fact that you could do
00:25:43 - it in humans, you can examine tasks in humans that you wouldn't otherwise be able to do in animals.
00:25:49 - But the key thing to remember is that fMRI is otherwise a very imperfect tool.
00:25:57 - Every tool has its drawbacks. And with fMRI,
00:26:01 - you're not going to be able to tease apart things like which cell type is contributing
00:26:06 - to my signal and the indirect nature of it,
00:26:10 - the fact that you're depending on a secondary measure of electrical activity
00:26:16 - rather than the electrical activity itself.
00:26:19 - So there are lots of things that get in the way of the interpretation,
00:26:23 - but it's still an excellent tool to use. Right.
00:26:26 - But now, so the paradigm you've used on humans, which is really interesting,
00:26:33 - sort of moves away a bit from the traditional division we have between operant
00:26:38 - learning going back to Thorndike and what's called classical conditioning going
00:26:42 - back to Pavlov, where in the first case, we are learning about the effect of
00:26:46 - our own actions on the world.
00:26:47 - And in the second case, the world exposes itself to us and we have to pick up
00:26:52 - whatever correlations are in that data.
00:26:54 - So you have tried to find a paradigm where you really merge instrumental aspects
00:27:00 - of learning with classical aspects of it. So how do you exactly balance that?
00:27:04 - How do you make sure that you can really still call it instrumental?
00:27:08 - Well, we can call it instrumental because of the nature of the contingencies in the trial design.
00:27:16 - So with instrumental learning, well, let me start with classical conditioning,
00:27:21 - where classical conditioning, you have a conditioned stimulus,
00:27:25 - and that will always be followed, regardless of what the subject does,
00:27:31 - by an unconditioned stimulus, which will cause an unconditioned response.
00:27:35 - And so there's no contingency between the
00:27:40 - way that the subject behaves and the
00:27:43 - outcome the outcome will always be delivered so that's that's fine you know
00:27:48 - that's that's the classical that that's the classical scenario the pump the
00:27:53 - instrumental scenario of course is very different we make sure that the outcomes
00:27:59 - are dependent on the subjects behavior.
00:28:02 - So there's a very clear contingency there. So the trial design really takes care of that, I think.
00:28:10 - But now, in the operant case, you would also believe that, let's say,
00:28:16 - it's something like there's a larger time interval we're talking about.
00:28:21 - And also the feedback the subject gets has a different kind of quality.
00:28:25 - So in this case, your human subject, I guess, is not being shocked anywhere.
00:28:29 - So we're not talking about an aversive stimulus that's being applied.
00:28:33 - So do you think that is a qualitatively change for the cerebellar learning system
00:28:37 - or actually it turns out to be still operating in an identical way, in your mind at least?
00:28:44 - Well, that's an interesting question. I think that the best way to answer that
00:28:49 - is just to make clear that the inputs that come into the cerebellum come in through,
00:28:59 - you know, so it's basically in the trials that I'm using, in the methods that
00:29:05 - I'm using, I use a visual feedback.
00:29:07 - There are reports, lots of reports showing that the cerebellum has access to
00:29:15 - that information, at least at a sensory level.
00:29:20 - So whether, you know,
00:29:24 - in classical learning theory will tell
00:29:27 - you that the more intense
00:29:30 - a stimulus the faster you learn and that sort of thing those
00:29:34 - same kinds of manipulations can be applied to instrumental learning and so you
00:29:41 - can you can have a partial reinforcement schedule and so on the kinds of signal
00:29:46 - that we see coming from from the cerebellum seem to mimic the amplitude.
00:29:57 - Well, the rate of learning, I would say. And so they all seem to follow these
00:30:01 - sort of Rescorla-Wagner type rules in that sense.
00:30:06 - But now, when a task becomes instrumental, we also have to start to think about
00:30:12 - how do we represent a rule that's hidden in this task, right?
00:30:16 - And how do we then pick up the relationship between that rule and my own actions?
00:30:20 - So in your mind it is that rule representation that would be the frontal neocortical
00:30:27 - contribution and the action generation would then be the cerebellar contribution
00:30:31 - that's how you think about it yeah so the,
00:30:35 - not really I mean I tend to think so what you're suggesting is that the rule representation is is,
00:30:46 - neocortical and that the the action-related representation is coming from the cerebellum.
00:30:53 - I don't follow that path at all
00:30:55 - because if you, again, look at the wiring diagram, it doesn't support it.
00:31:01 - I find it surprising because I thought we just had agreed that all the wires
00:31:06 - are there to link these two systems together.
00:31:11 - They are. But if you look carefully at the system,
00:31:16 - the cerebellar modules are all separate
00:31:20 - from each other they don't talk to each other the only
00:31:22 - way the only scope for us
00:31:25 - for the cerebellar signal
00:31:28 - to influence the
00:31:31 - neocortical system or rather the
00:31:34 - only way in which these modules can communicate with each other is through their
00:31:40 - links via the neocortical areas that do wire up to each other so you've got
00:31:46 - prefrontal premotor primary motor cortex um all talking to each other a lot um.
00:31:54 - The cerebellum modules that are independently wired up to these areas are operating
00:32:00 - independently of each other.
00:32:03 - There is no wiring within the cerebellum that can allow them to talk to each other.
00:32:08 - So you're saying they have to operate through the neocortex?
00:32:13 - Yeah, basically that's right. Right, yeah. But in theory, they could also, let's say,
00:32:19 - via brainstem motor nuclei like the red nucleus, the peduncle pontine nucleus,
00:32:24 - you might loop back to the pons and then generate as an inference copy an input
00:32:29 - to a cerebellar, a next cerebellar circuit and loop it straight through the cerebellum.
00:32:34 - Would that be anatomically defendable?
00:32:36 - Don't run that one past me again. I missed that. So an alternative is that I
00:32:41 - send my motor command from the deep nucleus of the cerebellum down to the red
00:32:46 - nucleus and or to any other brainstem motor nucleus, peduncle pontine nucleus,
00:32:52 - from which I get an afferent copy back into my pontine nucleus,
00:32:56 - input station to the cerebellum.
00:32:58 - And now I can loop it over a next cerebellar circuit.
00:33:01 - So I think in theory and atomically, you could devise a scheme that they can
00:33:07 - loop multiple cerebellar circuits without ever talking to the neocortex. Would you buy that?
00:33:13 - Uh, I would probably find that, um...
00:33:20 - I think if if the pathways existed then i would you know i think that that may
00:33:24 - be supportable but what is the evidence uh that those those did the anatomy
00:33:30 - exist that could support those,
00:33:33 - uh that could support that uh well red
00:33:37 - nucleus for sure projects back to inferior olive for instance yes it
00:33:40 - does but but i think i think the the
00:33:44 - olive is is a very modular structure and
00:33:48 - so the red nucleus is not going to project to
00:33:50 - every part of the olive um similarly does
00:33:55 - the red nucleus you know to what extent do different elements of the red nucleus
00:33:59 - communicate with you know eventually with with areas like prefrontal and premotor
00:34:05 - cortex um well i think i can play it another way to to see if i since you like
00:34:11 - the wires i could also say okay we got an reference copy,
00:34:14 - execute the motion, comes back to somatosensory cortex,
00:34:18 - and from there straight into your pons.
00:34:20 - Right? So still, I would be able to loop multiple cerebellar circuits without
00:34:26 - ever talking to my motor cortex.
00:34:30 - I think the cerebellum is a very, very modular structure, not just intrinsically,
00:34:36 - but also the way that it is wired up with lots of other areas.
00:34:43 - I think all of the evidence to date suggests that these loops don't communicate with each other and,
00:34:51 - that's a it's a generalization in some ways but I think you know you we have
00:34:55 - to we have to go with that because that's the evidence that we have available.
00:34:59 - But is it important to your model at least how you think about the circuit functionally
00:35:03 - is this an important axiom of your of your model like it that if it would be
00:35:08 - violated the whole model collapses well that's an interesting question um.
00:35:14 - There's no computational instantiation of my model yet.
00:35:20 - So there hasn't been an opportunity to sit down and write this into a piece of software and test it.
00:35:25 - It makes it much easier to speculate about it. That's a good thing. Yes, yes.
00:35:30 - I agree. Your model sort of claims the system one, system two distinction.
00:35:36 - Yeah. And the system one, which is where you say the cerebellum has the lead
00:35:42 - role, is a sort of habit learning system.
00:35:44 - But what you've described about the anatomy would seem to suggest that the habits
00:35:50 - have to be enacted via the cortex, even if they're learned in the Sarabala.
00:35:54 - Yeah, because I think the apparatus for delivering the habit-based behavior
00:36:00 - is housed in the neocortex.
00:36:02 - Well, it could also be housed in the brainstem, you know, elsewhere.
00:36:05 - There's a lot of machinery there for controlling the motor system,
00:36:09 - where you would not need motor cortical input.
00:36:12 - That's right. And you could imagine that with the cerebellum,
00:36:16 - you could wire up all sorts of interesting responses in the brainstem without
00:36:21 - having to bother with motor cortex.
00:36:24 - I think that's a possibility. But the neocortical architecture that supports very complex behaviors.
00:36:37 - Is better placed, I think, to execute those kinds of behaviors than the circuits in the brainstem.
00:36:44 - So in that case, system one would be about wiring up circuits in motor cortex
00:36:51 - to perform the habit behaviors.
00:36:54 - So is it more of a... I'm still not sure how this can work if the loops don't interact.
00:36:59 - I guess you've got your loops in motor cortex. But what's my prefrontal cortex
00:37:03 - loop? How is that able to help me learn a motor habit?
00:37:08 - Or am I learning something from the habit? No, you're learning these modules
00:37:12 - are learning independently.
00:37:14 - So what's habit learning in the context of prefrontal cortex?
00:37:18 - Okay, so what habit learning in the context of prefrontal cortex is,
00:37:21 - for example, the execution of a rule.
00:37:25 - The execution of, for example, the kind of experiments that I talked about earlier
00:37:29 - where you're arbitrarily pairing.
00:37:32 - Pairing, you're pairing a stimulus and a response in a completely arbitrary manner.
00:37:43 - Now, you can think of that as having motor control demands.
00:37:47 - You can think of that as having sensory demands, but you can also think of that
00:37:50 - as having cognitive demands.
00:37:52 - It's the cognitive element of that that is being basically.
00:38:00 - Being run by the prefrontal cerebellar loop. Yeah, but,
00:38:07 - To me, this doesn't make any sense, really. I'm sorry to say,
00:38:11 - but look, the point is, so let's start with Schiffer and Schneider, right?
00:38:15 - So we talk about controlled versus automatic processing.
00:38:18 - Yeah. And it also means the controlled deliberate system figures out the rules.
00:38:22 - It figures out how to map sensory states, action states, given goals.
00:38:27 - With that, you train your habits that now can be executed by your automatic
00:38:30 - system. And why do you want to do this?
00:38:32 - Because you want to be fast. You want to optimize your response latency.
00:38:34 - Or you want to free up resource and frontal contact.
00:38:37 - Yes, absolutely. Well, I would think a real fitness requirement is latency more
00:38:44 - than working memory space, but we can debate that, okay?
00:38:48 - So here I am, now I have my automatic process.
00:38:51 - What am I automizing? Time, okay, latency.
00:38:54 - So now, also in the perspective of Kahneman, you would have system one, system two.
00:38:59 - System two is the slow system, it's a controlled system, deliberate.
00:39:02 - And the system one is a fast automatic system. Great. Okay.
00:39:07 - But now you're proposing that this fast system that puts in this huge amount
00:39:11 - of metabolic resource into getting fast.
00:39:15 - And the Purkinje cells and the Cibrella circuit is fast.
00:39:18 - Right. And it really controls latency very precisely by disinhibiting the deep nuclear cells.
00:39:24 - So you rely on the rebound, which is controlled in latency to trigger an output.
00:39:29 - But that's not what you're building on. You say, no, no. no,
00:39:31 - these guys are really, maybe very precise.
00:39:34 - But again, that's the thing that goes back to motor cortex, which sits there,
00:39:37 - a piece of cortex, it's integrating, it's sloppy, it's jittery, whatever.
00:39:41 - And then that piece of cortex is going to issue this final motor command that
00:39:45 - cruises down my corticospinal tract to control the skeletal muscle system.
00:39:52 - But I'm adding latency now. Why would I do that?
00:39:58 - Because you have to parcellate up different elements of the task.
00:40:03 - So, I mean, what you're suggesting, what you're suggesting is that somehow this
00:40:08 - very high level thing that you've learned finds its way into the motor system,
00:40:14 - without going through the motor cortex.
00:40:18 - But there's no route. I mean, how would it do it?
00:40:21 - I mean, if you think about the connections of cruise one and cruise two,
00:40:25 - which are the parts of the cerebellum that are supposed to be learning these
00:40:28 - things, they have no output to the motor system.
00:40:32 - So there is no other route. But maybe they're doing something else and controlling action.
00:40:38 - Maybe they're part of the slow system and it helps the slow system to figure
00:40:43 - out, let's say, interval information, timing between events.
00:40:48 - And it is not involved at all in generating actions. For that,
00:40:51 - you train up other parts of the cerebellum. Yes. Well, that's basically what I'm suggesting.
00:40:56 - So you have modular parts of the cerebellum.
00:41:02 - So you have some parts that are specialized for dealing with prefrontal problems,
00:41:07 - some parts dealing with the primary motor cortex.
00:41:13 - And that it's basically speeding up elements of the information processing that
00:41:20 - would otherwise go on in each of these neocortical areas.
00:41:24 - But you need the neocortical areas to actually deliver the response.
00:41:28 - There is no output from CRUS-1 and CRUS-2 that can access this corticospinal
00:41:35 - tract Except through going back to places like prefrontal cortex and then,
00:41:42 - you know, sending the signal down through premotor cortex and primary motor cortex.
00:41:47 - You also exclude that there's any kind of divergence in, let's say,
00:41:52 - the deep cerebellar nuclei that might allow them to indirectly target brainstem motor nuclei.
00:42:00 - Well, that's an interesting question. So.
00:42:07 - The perspective that Peter Strick has offered us is that these are highly modular systems.
00:42:14 - Now, the missing information in all of this anatomical work that we've talked
00:42:20 - about before is the absence of any point-to-point connectivity that's been shown.
00:42:27 - So it's never been shown, for example, that there's
00:42:32 - a particular neuron in the primary motor cortex
00:42:35 - that sends its outputs to
00:42:38 - specific Purkinje cells and that
00:42:41 - those specific Purkinje cells send outputs back to
00:42:45 - thalamus and back to that particular neuron in the primary motor cortex so that
00:42:49 - point-to-point connectivity has been missing from the single cell level at the
00:42:54 - single cell but it does reach that volume of cells yeah so so that the broad areas are you know Um,
00:43:02 - so it's been shown, for example, that if you put antragrade and retrograde tracer
00:43:07 - into M1, you'll end up with, um,
00:43:11 - label in the same kinds of, the same cerebellar lobules.
00:43:16 - Um, and, uh, and that suggests quite broadly, but it suggests that there are
00:43:23 - these independent loops and that they don't communicate with each other.
00:43:26 - However, I think to be conclusive about it and to be sensible about developing
00:43:34 - computational models, you have to demonstrate that this point-to-point connectivity exists.
00:43:39 - Well, I'm not sure that's a fair constraint, really, because certainly,
00:43:43 - so what is the shortest distance at which people have found these recurrent
00:43:49 - projections to terminate?
00:43:50 - I mean, we're talking about, what, hundreds of microns? This will be the scale.
00:43:56 - So, as soon as you hit cortex, you have a highly excitable substrate with dense
00:44:03 - local coupling, with lateral coupling.
00:44:05 - So, as long as you target, let's say, an area that is, let's say,
00:44:09 - 100 micron away, you can be quite sure that activity will percolate down to that neuron.
00:44:16 - Well, that's right, except that the things that we were talking about earlier
00:44:21 - were about ways in which inputs from prefrontal cortex can start to influence
00:44:28 - the motor output and so on.
00:44:32 - These are in different lobules altogether. Sure, but I wasn't,
00:44:36 - indeed, we weren't finished with that, and I have a solution.
00:44:39 - Because maybe the thing to consider is that even though there's a universal
00:44:43 - cerebellum transformation,
00:44:45 - the cerebellar areas that are interfaced to prefrontal don't care about action
00:44:50 - generation in terms of the motor plant itself,
00:44:54 - but they assist prefrontal areas to figure out, let's say, interval information
00:44:58 - about task relevant or decision variables.
00:45:01 - Because that's the unique contribution they can give you. They can tell you
00:45:05 - something about interval timing.
00:45:07 - And this is as much necessary by acquiring the habit as it is for executing
00:45:12 - the habit. So how about that?
00:45:15 - That's plausible, I guess.
00:45:19 - So not all is lost for my canoe right now.
00:45:24 - That's like a compromise to me. Oh, really? I'm already, okay, watering it down. Okay.
00:45:31 - So, okay, good.
00:45:35 - So, now, okay, so we have different interpretations of how, let's say,
00:45:40 - the dense interaction of cerebellum and prefrontal.
00:45:44 - Would you say, I mean, how much of the hardware, the cerebellar hardware is
00:45:49 - committed to interacting with these premotor areas, you would say?
00:45:53 - The premotor areas? Well, or prefrontal,
00:45:55 - just decision-making areas in the brain, let's say. Right, okay.
00:45:58 - Okay. Estimates have varied, but a very, very significant chunk of it.
00:46:04 - I mean, I think Peter Strick, in one of his papers, if memory serves me correctly,
00:46:09 - said there was something like,
00:46:12 - 40% of the territory of the cerebellum in primates was accounted for by the motor system.
00:46:21 - Then including the decision-making parts of what we now call motor system.
00:46:26 - No, no. So the issue is what is the rest of it doing?
00:46:31 - In theory, very significant chunks of it are going to be devoted to communication
00:46:38 - with prefrontal cortex.
00:46:40 - And it sort of makes sense that that's the case, because prefrontal cortexes
00:46:46 - in the human brain has expanded enormously during the course of human evolution.
00:46:53 - And if you take on board the fact that connected systems tend to evolve together
00:47:00 - because of the selection pressures being applied on systems as a whole,
00:47:06 - then yes, that sort of makes sense.
00:47:09 - It makes sense. But I find that such a strange heuristic because I might also
00:47:14 - imagine I'm an engineer, right?
00:47:15 - We're the God engineer. We're building brains. We have these modules.
00:47:18 - As you say, right? It has a very modular structure, the cerebellum.
00:47:22 - So I think also Tony mentioned it in the talk.
00:47:24 - You could imagine I engineer some animal who has to do lots of really fast movements
00:47:29 - in a complicated dynamical world.
00:47:31 - I give it a huge cerebellum and it doesn't need to solve any problems.
00:47:35 - It doesn't solve crossword puzzles and Wisconsin card sorting.
00:47:39 - Sorting is also not available in the jungle.
00:47:41 - So it just reduces its neocortex. So why do these things by necessity have to co-evolve?
00:47:48 - Well.
00:47:50 - By necessity? I mean… That's what you seem to say, no?
00:47:53 - I'm suggesting that if selection pressures apply to systems that operate functionally
00:48:02 - as a unit, then you're going to find an expansion on the whole.
00:48:08 - You know, you're going to find an expansion of the parts of the cerebellum that
00:48:13 - are communicating with the prefrontal cortex.
00:48:15 - And that's indeed what we've found. And it's expansion of the prokenia cell
00:48:19 - level, the granule cell, all of them together?
00:48:22 - Those are empirical questions. We have no idea.
00:48:25 - But one thing you did say was that there's a kind of dorsal ventral,
00:48:30 - I think, distinction in the amount that PFC talks to cerebellum,
00:48:36 - so there's much more connectivity with the dorsal area.
00:48:38 - Well, there's a dorsal ventral split in the dentate.
00:48:43 - So that the dentate nucleus is is
00:48:48 - you can you can split it up in terms of
00:48:51 - the dorsal part which is wired up
00:48:53 - with the motor system and the ventral part which is wired up with a prefrontal
00:48:58 - system right and um it's been shown that previous studies have shown when you
00:49:04 - compare across primates that there has been a selective expansion of the ventral
00:49:10 - part of the dentate nucleus nucleus.
00:49:12 - And we're now starting to explore those sorts of issues.
00:49:16 - I think I was referring to, I was hearing in your talk, you were saying that
00:49:20 - there were parts of prefrontal cortex which have a strong projection and other
00:49:23 - parts which have a weak projection. Yes, exactly.
00:49:25 - So one of the interesting things that was found by Jeremy Schmarman and Deepak
00:49:30 - Pandya was that the dorsal parts of prefrontal cortex,
00:49:36 - so areas that are dorsal to sulcus his prince upon us.
00:49:41 - Have a greater tendency to project to cerebellum than ventral parts.
00:49:45 - Right. And so you could speculate about why that might be.
00:49:49 - And, you know, that might tell us something about the kinds of information that
00:49:55 - the cerebellum wants to process, as it were.
00:50:02 - But now the other, so let's say, okay, let's say we agree, okay?
00:50:07 - So they all evolve together.
00:50:10 - Very pleased to hear that. Then we can talk about basal ganglia hippocampus
00:50:13 - and spirochaliculitis if they
00:50:14 - also co-evolved the same way and they might have done that or not, okay?
00:50:18 - But, so now we have a control system which definitely will depend on neocortex,
00:50:25 - frontal areas, and basal ganglia.
00:50:27 - I don't know, maybe you want to leave that out, I don't know.
00:50:29 - But now I have to compile my habit into my cerebrally circuit,
00:50:35 - right? So, how do I first segment a habit?
00:50:41 - And then secondly, how do I get those segments into my cerebellum that I can
00:50:45 - call them up on future occasions?
00:50:49 - That's a really interesting question and a tough one to answer. We need answers.
00:50:53 - Yeah, absolutely. Absolutely. So, the issue of how do you segment a habit,
00:50:58 - I think that's a tough one. Yeah.
00:51:02 - I think all you can say is that if there are associations being formed,
00:51:09 - if there are CSs being processed in areas of the prefrontal cortex.
00:51:18 - Then there'll be a spike train corresponding to that going down to the cerebellum by the pons.
00:51:23 - Um how you can sort of break that very high level uh representation down and
00:51:33 - and talk about it in terms of what will the spike train look like and how how
00:51:37 - will that be represented in uh uh uh in.
00:51:43 - Terms of purkinje cell activity or whatever i think i think it's it's not possible
00:51:47 - to answer that it's a tough one do you think it will exploit the segmentation
00:51:52 - do you think it will exploit already the structuring of action primitives or
00:51:57 - behavioral primitives at the brainstem level?
00:51:59 - Because as a cerebellum, if you take the eye blink conditioning analog,
00:52:02 - I am talking to my red nucleus, I'm triggering discrete behavioral primitives.
00:52:09 - I don't think so. And the reason for that is, let's think, for example,
00:52:14 - about the kinds of information that gets processed in the hierarchically organized neocortical system.
00:52:24 - So if you take, for example, cells in the prefrontal cortex,
00:52:30 - and you try to figure out what they're most interested in, the work of Earl
00:52:36 - Miller shows that they're interested in rules.
00:52:40 - They're not interested in the fundamentals of motor control.
00:52:45 - You won't find motor unit activity being represented in the cells of the prefrontal. Um...
00:52:55 - Equally, if you look at cells in the motor cortex, you won't find them being interested,
00:53:03 - and having their response properties tuned to particular rules,
00:53:08 - as you would find in the cells of prefrontal cortex.
00:53:15 - So I don't think that you're going to be, I don't think it's a useful exercise
00:53:21 - really to try and look in places like the brainstem.
00:53:26 - For things that are, you know, rule-related, as it were.
00:53:30 - If we think about the simple rule, which is like later on you might want to
00:53:36 - escape from the studio and go to the lounge, something like this, right?
00:53:41 - And if we do sufficient or enough interviews with you, then you would be having
00:53:45 - this habit of running to the lounge and breaking all the furniture,
00:53:49 - okay? So that's a habit now.
00:53:51 - But this habit will have primitive elements, which might also be getting up
00:53:55 - out of your chair, controlling posture, going to the door, you know, moving through a door.
00:54:00 - So they're very simple elements in the end that get stringed together.
00:54:03 - So what is special about the habit is its macroscopic structure that even brings
00:54:09 - you from here to the lounge where you smash up all the furniture.
00:54:12 - So the point is that with these primitive elements, we can easily account for
00:54:18 - at a brainstem level, you know, like just to get up, maintain your posture, locomote, and do that.
00:54:23 - Can you? I'm not so sure. Okay. So, give me an example.
00:54:28 - I mean, the kinds of representations that you have at the very lowest end of
00:54:34 - the motor system would be things like the control of individual muscle units and so on.
00:54:42 - No, look, if you go to the midline structures in your brainstem,
00:54:45 - where you have the central gray, for instance, we have whole behavioral patterns
00:54:48 - that are species-specific encoded in a local way. There'd be sort of hardwired things down there.
00:54:54 - Sure. There might be gaze, gaze controls, completely predefined particular formation.
00:54:58 - Yeah. So there'll be sort of aggressive aggression and all of that sort of.
00:55:04 - Sure, but they're behaviors, right? They're behavioral programs that relate
00:55:08 - to certain motivational systems.
00:55:09 - Yeah. So you don't see them as part of then the habit.
00:55:14 - You really see the habit as targeting, let's say, higher level behavioral elements
00:55:21 - that you might have learned?
00:55:23 - I think it's possible. It's perfectly possible for those lower-level behaviors.
00:55:32 - So if you think about Ma's idea, I'm building everything on Ma,
00:55:37 - as it were, and what he claimed was,
00:55:40 - that these higher-level representations can trigger the pre-learned representations in cerebellum.
00:55:51 - And cause them to execute a particular kind of behavior.
00:55:56 - There's no reason, I think, why...
00:56:00 - Maybe the brainstem could participate in that.
00:56:04 - So many years ago, the work of Leapin and Supple showed that there were cerebellar inputs.
00:56:14 - If you lesion the cerebellar vermis, you get sham rage.
00:56:19 - It's possible that that's happening because of the impact that cerebellum has
00:56:26 - on those lower-level brainstem centers.
00:56:29 - Centers right okay so you're saying it's
00:56:33 - not necessarily in the in the focus of your idea about
00:56:36 - how habits are learned or expressed but you
00:56:39 - might you cannot exclude it no that's right right okay
00:56:42 - absolutely i mean there's lots of evidence showing that there's interactions
00:56:45 - right exactly brainstem centers and the cerebellum of course so the other thing
00:56:50 - that you mention or that also becomes apparent as you said earlier and sometimes
00:56:54 - you're saying look all this the
00:56:56 - The modeling of the cerebellum is like a footnote to Marin Albus. Yeah.
00:57:00 - Right? So you really think that there has been not that much progress in our
00:57:05 - theoretical understanding or computational understanding of the cerebellum beyond
00:57:09 - what Marin Albus have identified? I think there's certainly been some progress, of course.
00:57:15 - You can't deny that we've come on leaps and bounds since 1969.
00:57:21 - What I would say is that the fundamental ideas about cerebellar plasticity and the way that they work.
00:57:31 - Have not changed all that much. The idea that the cerebellum has as its most
00:57:38 - fundamental unit the Purkinje cell and that the inputs into the Purkinje cell,
00:57:45 - the parallel fiber and the climbing fiber inputs being the principal inputs
00:57:52 - into the Purkinje cell, is still a dominant idea.
00:57:55 - Dear um that's not to say
00:57:58 - of course that we shouldn't be
00:58:00 - exploring other forms of plasticity in the cerebellar cortex
00:58:03 - there are lots of other cell types um and
00:58:08 - it's it's very clear that that you know there's been there's experimental evidence
00:58:12 - to support to support their involvement as well so um so here we are right so
00:58:20 - so you have this really very if you want ambitious and also So,
00:58:25 - an advanced scheme where you
00:58:28 - see how prefrontal cortex works together with cerebellum to realize both,
00:58:33 - let's say, controlled processing and automated processing, right?
00:58:37 - So, in that sense, I guess you also have had to sort of fight your battles and,
00:58:41 - you know, accumulate your scars to get that point across.
00:58:47 - So given your experience now in trying to understand the brain at the system
00:58:52 - level, what is Narendra's law that we should follow to understand the brain at this level?
00:59:01 - There's no law as such. Of course. Come on. I think we just have to be faithful to the data.
00:59:09 - I think the starting point, if the starting point is the anatomy,
00:59:13 - then I think you're on very sound footing. Um...
00:59:18 - It needs to be based on that sort of evidence. And the other thing is to build
00:59:24 - from what we have learned in the motor system.
00:59:28 - The motor system is what we understand best. So if we can take a set of principles
00:59:33 - that we can apply in the motor system,
00:59:37 - and find a way of extending those into the cognitive domain,
00:59:40 - I think that would be a good way forward.
00:59:44 - So look, Tony likes traveling, and then he always has me pay it,
00:59:50 - so I don't like to spend money.
00:59:52 - So we only have trips within England in the coming five years.
00:59:57 - So five years from now, he's going to come to your lab to check whether a prediction
01:00:05 - you're going to make today was actually confirmed or rejected.
01:00:08 - So what's the most important hypothesis that you want to see tested,
01:00:13 - validated in this five-year framework perspective?
01:00:18 - That's a cheeky question. And you're going to hold me to that,
01:00:21 - aren't you? Of course. What do you think? Why do you think we're recording this? Absolutely.
01:00:27 - There are so many interesting questions. I suppose
01:00:32 - one of them would be to
01:00:36 - go on testing the idea that a conditioned stimulus is processed in the same
01:00:46 - way in conditions of instrumental learning and classical conditioning.
01:00:54 - We've often talked about the possibility that there is this universal cerebellar
01:00:59 - transform, and it's an idea that people have been on about for a long time and for a good reason.
01:01:07 - Reason, I think this is a good way of doing that.
01:01:10 - I think it's a good way of pinning down that problem and saying,
01:01:16 - well, look, you've got a conditioned stimulus.
01:01:19 - Is it being processed in the same way, regardless of whether it's higher level or lower level?
01:01:26 - Are the areas that are activated when you process that instrumental CS,
01:01:32 - how are they talking to the cerebellar cortex?
01:01:37 - Or indeed, are they speaking, sorry, to the prefrontal cortex?
01:01:42 - Which parts of the prefrontal cortex are they talking to?
01:01:46 - We still don't have a handle on...
01:01:51 - Exactly what areas of prefrontal cortex are saying to the cerebellum.
01:01:58 - And so I think it would be good to try and work out methods for sorting out
01:02:03 - communication between the two. Great.
01:02:07 - Rolando Romani, thank you very much for this conversation. Thank you. Thank you.
01:02:13 - That was fun. You gave me certainly a good grilling there. The CSN podcast was
01:02:19 - produced by the Convergent Science Network of Biometrics and Biohybrid Systems.
01:02:24 - A project funded by the European 7th Research Framework Program.
01:02:29 - For interviews, recorded lectures or upcoming conferences in the field of biometrics
01:02:37 - and biohybrid systems, go to csnnetwork.eu.
01:02:43 - And thank you for listening.
01:02:45 - I think this is the idea of the podcast that we
01:02:49 - can go a bit behind what you would put on
01:02:51 - the slides right and and also i think it's important
01:02:55 - for people to listen to this to understand look there's uncertainty what we
01:02:57 - do yeah right there's absolutely we make assumptions and to get
01:03:00 - that a little bit more to the foreground is really i
01:03:03 - think what we want to to achieve here and if
01:03:06 - you want to also show a bit more the more
01:03:09 - human level decision making that we have to do yeah
01:03:12 - we do these things that's often missing from talks isn't it yeah
01:03:15 - exactly but no it's really good fun i enjoyed that
01:03:18 - terrific i enjoyed it actually we were
01:03:21 - really um i'm also very much trying to figure out this whole idea of how do
01:03:26 - we compile habits you know so uh indeed how can the cerebellum contribute to
01:03:30 - that this is quite relevant for what we are doing so that's why i please yeah
01:03:35 - it's a question of getting is trying to trying to understand what is a motor habit and what is
01:03:40 - a cognitive habit, really. Right, exactly.
01:03:43 - And that's something that, yeah, I think I could have answered that a little
01:03:48 - more effectively. But, yeah, it's a tough one.
01:03:51 - But you have to admit, right, it seems really odd to claim it loops back over M1.
01:03:56 - Yeah. You're screwed. You add lots of uncontrollable latency to this whole thing.
01:04:03 - Yeah. Yeah, a big chunk of the timing is taken up with the decision-related aspects,
01:04:15 - and that's what you speed up with that prefrontal component.
01:04:20 - Yeah, but maybe we give too much importance to M1. I think M1 is pretty stupid.
01:04:26 - But there's no way out of the system without going through M1.
01:04:30 - Yeah, but I see M1 really just… How can you reach the corticospinal tract without
01:04:34 - getting M1? Exactly, but you just dump your motor program in M1.
01:04:38 - You give it a very short time window to play it down into the corticospinal
01:04:42 - tract. Yeah. You close it up again. That's it. Finish.
01:04:45 - It's just the output station, you know. Yes, but how do you get it in?
01:04:50 - Via your premotor network.
01:04:55 - Uh, premotor network being through red nucleus and so on. No, wait, wait, wait.
01:04:59 - In my model, my cerebellum does not need to talk to M1. It doesn't give a damn about M1.
01:05:04 - It just plays it out downstream immediately because it wants to be quick.
01:05:08 - Yeah, yeah. I think also from an evolutionary perspective, that's the one thing
01:05:12 - that matters. This is what makes you survive.
01:05:14 - Be quick. Yes, yes. That's why you want to do this at all. Yes,
01:05:17 - I understand. Right? Yeah, yeah.
01:05:18 - So for that, this is why you just want to play it straight down.
01:05:21 - I suppose the crux of the issue is how do you get information out of...
01:05:30 - CRUISE 1 and CRUISE 2, and into your premotor network.
01:05:36 - And I don't think there is a route except through, I mean, what would it be?
01:05:41 - Well, look, CRUISE 1 and CRUISE 2, they are talking to deep nucleus.
01:05:45 - You might have divergence. Yeah. We cannot exclude that. This is the thing with
01:05:49 - the jet. Actually, right.
01:05:51 - Okay. I would bet, certainly since it's a brainstem, damn sure you have divergence. Yes. Right?
01:05:57 - Then you throw away modularity, right?
01:06:02 - Yeah, but I wasn't committed to modularity. That was your problem. Right, yes.
01:06:09 - But we still have conserved it at the cerebellar level.
01:06:13 - But outside of the cerebellum, I just allow things to diverge or to be linked together.
01:06:18 - You still, for me, just the key is, why don't you allow lateral interaction cerebellum?
01:06:24 - You should get latency. you just you say if I go through that loop I want to
01:06:28 - get out in a predictable way can be no screwing around this is why it's important to figure out,
01:06:37 - the intrinsic connectivities within the cerebellar nuclei then absolutely and
01:06:43 - the data is not there are you sure in primates it's not there,
01:06:48 - I should ask Bjorn Merker he might know he's a good friend great expert on brainstem them.
01:06:54 - We should look at red nucleus, for instance.
01:06:58 - I remember discussing this with him a few weeks ago, but not so specifically for the red nucleus.
01:07:05 - He didn't reject the idea completely, but I didn't ask for the references. No.
01:07:10 - Yeah, a few years ago, I think this issue did bother me, and I looked and I looked.
01:07:18 - I couldn't find the anatomical evidence. And the simplest route out was through M1.
01:07:27 - Possibly because we know about the connections. Sure. And we don't know about
01:07:32 - the connections, the intrinsic connectivity within the deep-coated-tel-all-out-convections.
01:07:37 - I think I did well because I have two options now.
01:07:41 - I have plan B. Plan B is the prefrontal-oriented cerebellar regions don't care
01:07:51 - about linking to the motor output.
01:07:53 - They just care about setting the right interval information for even the amplitude
01:07:57 - time course of, let's say, a memory response prefrontal cortex. Yes, yes.
01:08:01 - They don't give a damn about motor. No.
01:08:04 - I think that's more likely, actually, because the language is different.
01:08:10 - The language that the prefrontal cortex speaks, when I say language,
01:08:14 - I mean the neural activity and what it's time-locked to and so on.
01:08:23 - It doesn't give a damn about motor
01:08:27 - it just doesn't exactly right and
01:08:31 - so maybe it isn't bothered
01:08:34 - I would buy that because then what it could
01:08:37 - look like is that you just have this prefrontal cerebellar interaction cerebellum
01:08:40 - helps keep time in that system you get your deliberate action plans you play
01:08:46 - out over M1 goes down over the cortical spinal tract but you have your collaterals
01:08:51 - into a range of brainstem areas.
01:08:54 - You have your afferents copied back to the pons. You have a lot of source information
01:08:57 - for the cerebellum now to suck in these regularities.
01:09:01 - And then from then on, I would speculate, right? It would just play this out automatically.
01:09:09 - Exploiting recurrent projections into the pons that just allow to loop multiple circuits.
01:09:14 - The function of this will make sense to me. Then I really have optimized my
01:09:17 - latency. This is what I want.
01:09:19 - Yeah. I think there's a lot of merit to that actually yeah I agree just the
01:09:23 - anatomy is missing but that's a detail.
01:09:28 - A forgettable detail exactly but I'm sure I'm sure given that people haven't
01:09:34 - looked we it's a reasonable hypothesis to go to check so I should convince some some red runner to,
01:09:43 - do some traces yeah or even even somebody who works on monkeys yeah even better
01:09:48 - or humans why can't we do this Isn't there a way to do this?
01:09:52 - Can't you look at brainstem areas?
01:09:55 - Paul, I'm going to have to go and move to a hotel. All right.
01:09:58 - So then we see you at 8.30 at the Silicon for dinner? Yeah. Possibly?

CSN Podcasts

Narender Ramnani on cerebellum and cortico-cerebellar loops

Season 2015

Description

CSN Podcasts

Episodes

Matthew Diamond on whisker system and decision-making

Stephen Noctor on cortical development and precursor cells

Ranulfo Romo on decision-making and somatosensory cortex

Mark Blumberg on rem sleep and twitching

Kate Jeffery on spatial cognition and grid cells

Timestamp

Be the first to leave a comment

Leave a commentCancel

Legal

Follow us on

Narender Ramnani on cerebellum and cortico-cerebellar loops

Description

Episodes

Timestamp

Be the first to leave a comment

Leave a commentCancel

Tags

Legal

Follow us on

Login to enjoy full advantages

Please login or subscribe to continue.

Go Premium!

Enjoy the full advantage of the premium access.

Stop following

Cancel subscription