BI 236 Liset de la Prida: Neurons, Ripples, and Manifolds

[00:00:04] Speaker A: It is the most intriguing phenomena that I think the brain has, which is related with the reactivation of sequences of activity that occur during the previous exploratory phase. What I can envision is that we will have a fractality again into the manifold organization in terms of different interacting hierarchies of representations which are interleaved at different levels to complement a high functional integration. In the consideration of the new problems that these approaches bring to us, they also open your mind for you to consider solutions that were not in the table at the beginning. And this is happening with these new fancy tools. Many people just applying the tools, but sometimes you need to understand what the tools is telling you and how that challenges your interpretation. And for me, that is in that creates new opportunities to address all problems or to learn better how to treat the system. So yeah, I mean, it's part of the progress. It is always like that. [00:01:34] Speaker B: This is brain inspired, Powered by the transmitter Lisette de la Prida is director of the Centro de Neurocensia Cajal at the Instituto Cajal in Madrid, Spain, where she runs the Laboratory of Neural Circuits. Today we discuss two main topics. What drew me to invite Lisette was her work on neural manifolds, which we've talked about a lot recently on this podcast. She studies how specific subtypes of neurons affect and control neuromanifolds. More on that in a second, because what drew her to study manifolds was her work on what are known as sharp wave ripples in the hippocampus. Sharp wave ripples are generally quick bursts of oscillatory activity as found in local field potential recordings. And those bursts are accompanied by little bursty sequences of action potentials fired off by sets of neurons. So those ripples have been associated with a quick replaying of some experience an organism has had with a thinking that by replaying those sequences of neural activity associated with the event, it's helping to consolidate the memory for that event in the cortex. Okay, like everything else, the story is not so simple and we talk about some of the findings that have added to the complexity of understanding what sharp wave ripples are, what they're doing, and the the varieties of them. Okay, varieties. There is key that varieties part is related to the second main thing that we discuss, which is the varieties of neuron subtypes and their roles in shaping the manifolds that have featured prominently on Brain Inspired. As a reminder, manifolds are dynamic structures along which populations of neural activity unfold over time. They're usually low dimensional relative to the dimensionality of the many neurons being recorded. And they have proved to be one effective way of making sense of how large populations of neurons coordinate their activity to do useful things for our cognition. Lisette is interested in the relation between sharp wave ripples and manifolds and in how specific subtypes of neurons affect manifolds and cognition in general. Okay. You can learn more about her and her lab through the shownotes at BrainInspired Co Podcast 236, where I also link to some of the papers that we discuss. We also briefly talk about Lisette's book, Brain Space and Time, the Neuroscience of How We Navigate Reality, Memory and the Future, which currently is only available in Spanish, although hopefully there'll be an English translation in the near future. But we have an interesting discussion about writing in one's native language for Popular Science while writing in English as a second language for scientific papers, and how to translate between those two and sort of the dynamics of that, which is interesting. Okay, here's my conversation with Lisette. Lisette, thank you for being with me here. So I didn't tell you this. I recently had Juan Gallego on. Or maybe I did tell you this. I'm not sure. And man, we talk manifolds for two hours and we're not going to do that here, but that's going to be part of the conversation. Okay, but. Oh, so the manifolds are supposed to make things easier, and here you come along and say, no, no, no, let's go look back at the. At the cell types and their functions, et cetera. So that's going to be a lot of what we talk about. But you have spent a lot of time working on hippocampal function features in hippocampus, local field potentials, sharp wave ripples, hippocampus in health and in disease. And I thought maybe we would start off very broadly. And the hippocampus itself has been implicated in so many different functions, cognitive functions. I'm wondering how your view of the hippocampus, where it is now and kind of how it's evolved over the course of your work on. [00:06:04] Speaker A: Yeah, that's pretty interesting question. So, yeah, I mean, from the very earliest consideration of hippocampus as the main circuit involved in episodic memory up to today, we have evolved a lot in consideration of what hippocampus can do initially. The idea that dominate the field is the cognitive map. Right. So we need the hippocampus to establish representations, relational information regarding things that happen somewhere. And some time, everything happens in A place and is related with events that are flowing through experience. And this is the core concept of episodic memory and the main role of hippocampus in memory. But then progressively we have learned that hippocampus is involved in the consolidation of memory, not only in retrieval the memories that, or in building memories, but also in reinforcing the storing of memories, not in the hippocampus itself, but transferring this information to the neocortex, then we still don't understand how this process goes. And we still don't understand how much we rely on the hippocampus for recovering and retrieving particular information at a given moment in time in which you need to think in events that happen, have a recollection of all this memory, and confront that information with some novel information that is around in the environment and inform your decisions based on memory. So, for instance, in the case of exploration, curiosity driven exploration, the the most simple aspect of you are moving around the world and trying to explore a city and deciding whether to make a left or right, or stop by in a store, at a coffee shop, or go to a library. How much hippocampus is involved in this map? How much of this cognitive map you retrieve. So that is more or less an evolution in myself and in the field, I guess. [00:08:16] Speaker B: Yeah, but there are lots of stories about how this functions. I mean, you just said that we still don't know how it functions in these various aspects, but. And you mentioned the transfer of memory, right, Using like, I'll say, complementary learning systems. I'm not sure if you buy into that concept, but the idea where there's this quick sort of encoding of an event, an episodic memory in hippocampus, and then over time, via sharp wave ripples, et cetera, these things get replayed and then that memory gets consolidated in the cortex over time while we sleep and while we're resting in resting periods and stuff like that. So we have lots of stories about how these things function. But do you think that we are on the cusp of understanding these things? Where are we in that? Along that journey? [00:09:07] Speaker A: Well, we are on the journey, still trying to make sense of all this information, which is absolutely vast. Yes. I mean, we have evolved from the idea that encoding and retrieval could be implemented in alternative phases of the theta oscillations, for instance, which is quite fast. Like within 150 millisecon, 200 milliseconds, you can be alternating. Right. But then when you are offline of experience, then you have these shockwave ripples as you mentioned, which is a very fast oscillation between 100200 Hertz and then there it is. The most intriguing phenomena that I think the brain has, which is related with the reactivation of sequences of activity that occur during. During the previous exploratory phase. This is presumably something that contributes to consolidation. But initially we thought the ripples were always offline, obviously when you get to sleep. But then we discover. When I said we, I don't mean my lab, I mean the community. I always talk about this. This is a community effort in trying to understand this phenomena. We have discovered that ripples also cure during exploratory pauses. There are awake ripples, ripples that occur immediately surrounding the experience. And those has different phenomenology. Then there are other ripples that occur when you get to rest at the beginning. The features and the aspects of the ripples differ as the time is passing. And then eventually you get to sleep and then we enter into the consolidation domain. So my lab has been doing a lot of work in trying to conceptualize around what is a ripple and whether all ripples are the same. We are very much in the moment of considering that we have been trying to pack all these events in a single nomenclature. But this is fundamentally wrong because they are quite diverse, the phenomenology is quite diverse. So we think ripples really represents not only one thing, but many. We still need to understand that. [00:11:29] Speaker B: Oh, that's another one of those stories that starts off real nice and simple. And then when you start digging into the details, it gets more complicated. [00:11:39] Speaker A: Yeah, absolutely. But this is. I mean, so in philosophical terms, we look for answers and we try to understand what is the truth, what is the knowledge that we can extract from nature. Right. And researchers, we typically try to conceptualize by simplifying, by naming things. So the first step is to try to order what is unknown and try to make some boxes in which we can put phenomenas and concepts and then try to elaborate our understanding based on [00:12:11] Speaker B: that, but then create an ontology. [00:12:14] Speaker A: Absolutely, absolutely. And this is essential. It's essential part of the scientific survey. But then we end up acknowledging the complexity of nature and the phenomena that we look for. And then we all need to expand our concepts and the considerations regarding the phenomena. So, yes, for the ripples in this very moment, there are some labs in the world that we are trying to target. This very big question. I mean, how many ripples there are, how we can classify them, whether we can classify them in separate categories or they just reflect and a spectrum. So all These questions, we are starting to think that of ripple not as a single phenomenon, but as a family of events with different inputs, with different structures and potentially different functions. And this in our view is required to understand the mechanism, the basic mechanism, but also the function of this intriguing event. [00:13:14] Speaker B: Well, I mean I want to ask you what your guess is as to how many types of ripples there are. Right. And, and then, and also so they might be implementing different functions or associated with different functions. But is there a common underlying way to think about ripples? I mean is there some common thread that, that binds them all like it. Because there's replay pre play, there's fast, there's slow, there's. I mean I understand all the different features, but is there something we can say that sort of unifies them? [00:13:48] Speaker A: Yeah, yeah, yeah, yeah. So we have to first, the first, the first decision we need to make is we will be talking about ripple. Then we are talking about the local field potential event which has a high frequency oscillatory component. This is essential part of the definition. We need that. Right. Okay. Then we can discuss what are the frequency limits of that high frequency oscillation. And here is many comments as experts are in the field, depending on what species you are working. For instance, human typically do not get ripple too fast. Typically they get oscillations which are compatible with the definition of ripples which are between 8090 hertz. But in the rodents physiology, most researchers they try to avoid the 8090 Hertz as part of a ripple because they could be somehow mixed with the high gamma oscillation. So let's try to forget about this frontier because it's a bit unclear and we move faster than 100 Hz. This reflect the discussion. Okay, let's just leave it. Let's assume that we have two community here, the rodent and the human. And maybe they Disagree regarding the 80, 90, 100, but more or less they agree there is a high frequency oscillation, short lived companion by a shock wave, which is a kind of negative deflection. That's okay, that is a ripple. Then we start at least with something that we know this is an event. Associated to that event we have increase of firing activity from individual cells, that is the replay. But then people who study replay and detect replay by looking at the increase of firing of cells, sometimes they find replay without ripples. And sometimes we have ripples which has a kind of weird replay event, not really that we can decode. Then we can also have some ripples without a shockwave and some shall wave without ripples. Of course, I admit that this is a phenomenology and the core event, the majority of event could be a shockwave and a ripple with a replay. But you see that differences generate a lot of discussion always. [00:16:14] Speaker B: Yeah, of course, yeah, but could you say that. So we mentioned that they were, I think they were discovered during sleep. Right. But then they're also found during periods of rest and like sort of rest, even in between while you're doing things. But are they always. Is that a common feature? Is that they're always unrelated, not unrelated to, but they don't occur during the real time experience of an animal. They're always related to something that happens. Not offline, but not when they happen. They're associated. [00:16:48] Speaker A: No, not really. Actually. No, not really. Actually. If you get back to the first papers discovering these events, which were of course by Bushaki, and initially they associate ripples with consummatory behavior. You know that we get ripples when the animals are offline. I mean they are in the running but they are drinking something or during these periods they do have it. Of course, we have more ripples and more clear ripples during sleep. And then we have this two stage model that Bushaki proposes regarding the hippocampus function in which we have a phase of encoding dominated by theta and a phase of consolidation dominated by. By ripples. And that paper was fundamental paper in the field and has influenced many of us. And of course that provides, we were discussing before a concept, a box to try to categorize and organize our thinking to understand that has so much influence that for many, many years people typically focus on repurposing the consolidation and sleeping phase. But it doesn't mean that ripples are only there. Soon later there were other papers and researchers that basically started reporting a there are ripples in awake conditions. And then we started to break the barrier to discuss about things. But it doesn't mean that from the very beginning Bushaki knew that the ripple were not only during sleep. [00:18:17] Speaker B: Yeah, but could it, could it be something as simple as what we're calling ripples are a phenomenon that is constantly occurring, but it's during, during sleep, for example, maybe spontaneous, maybe activity that interferes with detecting the ripples is loud enough that it sort of is, you know, disappeared. We just can't detect them, but it's actually always ongoing. [00:18:42] Speaker A: No, no, no, ripples are not always occurring. Ripples reflect a state of the system which is, which occurs when you have a low acetylcholinergic tone, a cholinergic tone. Is very low. This doesn't happen when the animal is exploring, when you are moving around, when you are active. It doesn't happen, actually. Activating cholinergic terminals block the emergence of ripples, essentially because it shifts the cells to a different regime. They become depolarized and they start oscillating in theta following the pacemaking by the septum. And you have no ripp. So you need a state, a behavioral state in which you reduce your cholinergic tone and then ripples emerge. Now, ripples are very rare in terms of they occur at low interval between them and they are very short. They typically can last between. There is a distribution, but they could last between 30 to say the longest could be 70, 100. That really depends. Then we have like exchange of ripples one after another. And we can have double or triple ripples coming. [00:19:59] Speaker B: Milliseconds. [00:19:59] Speaker A: Yeah, milliseconds, exactly. But essentially that means that when you are in, when the animal is quiet sleeping or you have ripple face, you get ripples one every two, three seconds from time to time. You have to wake a lot. Well, not a lot, but I mean, they, they come. They come sometime in in tandems, but it's not occurring all the time. It is not. [00:20:27] Speaker B: Well, that's the ripple itself, which is how we sort of found that, that regime of activity. But I guess what I was asking is, could the function of what is associated with ripples? As you were saying, the definition of a ripple has sort of been blurred. Right. And maybe the actual function carried out by sequence of events, sometimes it's slower, sometimes it's faster and associated with the ripple. But maybe the function is like universally happening all the time or something. You know, naively. [00:20:57] Speaker A: Yeah, no, no, I think, I mean, the key point about ripples is that first it is one of the events which is more synchronous across the many brain regions. That means it has a very clear and clean signal to noise ratio. And that makes them exceptional for any analysis. Because surrounding ripples, you typically have low firing rate. That means that it's quite clean, what you can get surrounding them. Now, the firing increases could start well before the high frequency component of the ripple start. And it means we have, in my lab, we have been dealing with this concept of integration periods occurring before the ripple when you start seeing activity coming. And sometimes that activity didn't trigger the event at the population level and you don't have a clear ripple. What I tend to think is that those are the ripple which are not associated with the ripples. An increase of firing, which is not associated with ripple, essentially because that increase of firing was somehow sub threshold to the level that you need to get all the events synchronous in the population, which is a ripple itself. Then as I said, then you have a ripple lasting between 30 milliseconds to 70 millisecond. And then you have a post ripple, which is typically a post inhibition occurring with the shock wave. Because essentially what you have is you bring all cells close to the threshold, they start firing together, those which belongs to the replay assembly, and they get inhibited, they hyperpolarize and then they start firing again outside the lens. [00:22:50] Speaker B: Like a release almost. Yeah. [00:22:53] Speaker A: Surrounding a ripple you can get about 100, 200, 500 milliseconds of something that deviates from a state. And then the network goes to some state active states, which is quite specific. And that is the miracle. So why this set during this repl. Why a replay during this event? And here is where we started considering. I mean if I have these cell assemblies which are so specific, they reflect the sequences of events that happen, they replay and all these ideas, the composition of these events are various from one to another. And so the spectral features of RIP that should be varying also. And this lead us to the idea of exploring the variability of the events. But instead of trying to resort on the spectral features that in our view they were already exploited to the stimulation because we couldn't extract more information from a spectra from the spectrum of the event. Then we started considering transforming the problem into a high dimensional problem in which we conceptualize about the waveform of the different ripples and we create a representation, a high dimensional representation from where we could extract the minimal low dimensional properties that categorize the ripple. And then started from there to explore the diversity of the events. And here is where we introduced the idea that ripples reflect a continuum of events that probably encompasses like different assemblies which reflect the different memories stored in the system that we can retrieve independently during the events. And well, this is more or less how we are trying to think right now. [00:24:54] Speaker B: It's interesting that sort of integration period, it's like it. You can predict what the information will be in the ripple from that integration period, I would imagine. And then is your thought that the ripple itself is just. It needs to be that synchronous in order to communicate to other brain areas sort of. [00:25:16] Speaker A: I mean, because a ripple increases, the activity locally started typically like an increase occurring mostly in CA3, probably also in C1 integrating. And then you have the boot of activity coming, you have the oscillation. The selection reflects a composition between the action potential of individual cells, few of them very sparse, which becomes synchronous in a very short period. And that gives you the negative part of the negativity of the oscillations. Then you have the inhibition, which is pacemaking at high frequency events. And then give you the positivity of the oscillation. And that is essentially what that you can see intracellularly and extracellularly in the firing. Now all this happens sparsely. You increase firing in about 30% of the population. But that is enough to generate this kind of synchronous outputs going through the C1, which project to the subiculum. The subiculum projects brain wide two structures in the frontal, in the prefrontal cortex, mamillary bodies. They project also to the ventral domain of hippocampus. They project to the nucleus accumbens. So the information flow through there, all along the cingulate cortex, up to the prefrontal cortex surrounding the hippocampal ripples. You can detect increases of activity in all these other regions. And here is where communication is being transferred. Right there is the mystery where the different type of ripples broadcast. Whether one type cuts one region more than other some cells. And here is where the other topics of the discussion comes. All cells are the same, or all of them are just neurons firing. Or there are different categories, some function and they reflect different composition and different information. So that's when things become more complex, but at the same time, for me, more interesting. [00:27:30] Speaker B: Wow, is that what I want to ask you is whether is that aspect of your research is what led you to thinking about including cell type specificity within the manifold story? But before I ask you that, I have to ask you what people, how you respond to people who suggest that oscillations are epiphenomenal and not causal. Where you are in that fight, in that story? [00:28:00] Speaker A: I am never in a fight. I try to listen every idea and then I have my own criteria. My opinions are based in my empirical experience, strong empirical experience. I am a physicist by training, but I am experimentalist. What is important for me, I don't study planetary systems that will be very physics oriented. I study biology. And for me the most precious things that biology offers to a physicist is heterogeneity and diversity that we need to understand. So for me, oscillations are part of the system. It's an intimate part of the system. The system is generates oscillation and pace activity this way. And from there that generate a Function. It is like trying to separate walking from stepping. I mean, to me, the two aspects interact each other. I mean, brain activity oscillates the heartbeat. We breathe, we walk. There are many oscillatory components in the brain and body. And the body integrates and activity flow oscillating. And during that oscillation is that where function emerges. So note epiphenomena, just part of the system and generate a function because they are part of the system. [00:29:33] Speaker B: I guess the debate that's often had is whether oscillations are causal. But in your view, there'd be sort of a certain circular causality because they're so tightly integrated. Spikes and neural activity and populations can generate oscillations, but those oscillations can also shape the neural activity. Would that be your view? [00:29:54] Speaker A: Totally, totally. So I hear that this the debate, but to me, oscillation exists. The code, the brain code. And any code that we try to extract from the activity of the cell is embedded in an oscillatory behavior. There are part of the code which doesn't require timing and oscillations, which is totally fine for me, because you also need that, but you also need the other code. And a causality in a complex system requires a different understanding because you cannot affect one component and pretend that that confirmed that that component was making a particular function. Because when you interact with a complex system, you just push the system away from their regime and they will expose what the system it is it matter to you because it coincides with your hypothesis. So lucky we are. But to me it's just an opportunity to understand how the system works. Not necessarily prove that causality in that terms. Causality requires philosophical discussion and more discussion in the field for the basic researcher point of view. [00:31:07] Speaker B: I like that. When I asked you where you were in that fight, you appealed to being a physicist, saying, I don't fight, I'm a physicist. Come on, Physicists fight. It's not. Has nothing to do. [00:31:20] Speaker A: Yes, a lot. They fight a lot. Actually. I am a we are physicist because the first thing I when I why I was studying as a. As a younger, younger student, I was surprised that some people were so certain about concepts because for me everything was in doubt. And I don't have a problem with that. I just want to make a hypothesis. And probably being wrong is not a problem, it's just advancing in knowledge. So not that much. I am not that kind of physicist who believes that I am in a position of understanding everything because I think there are so many mysteries and Fighting is like, I mean, it's more important to debate, to discuss and to try to consider opening the borders. [00:32:11] Speaker B: Right, sure. Well, I mean, a fight was maybe the wrong term for it because in, you know, in some sense it's these debates, I think that I'm not sure if they help progress more. I think they can do both. I think that they can aid in progress, but I think they can also hinder progress as people stamp their foots down. Feet down. [00:32:32] Speaker A: Yes. And in this particular question, I think it's important. Yeah. Because sometimes we, the researchers limit progress without recognizing we are doing that. Because sometimes when we review others, we try to think that, well, this is fundamentally wrong because this is a hypothesis or this is a theoretical framework. I mean, we need to let other ideas to expose. And what you need to learn is whether you technically are good or not. Whether technically there is something which is fundamentally flawed that you need to correct. But just limiting progress of competing ideas is not a good idea. It's not good. I mean, so I think that sometimes, yes, these debates, I agree that it's maybe better not to set fights, but we shouldn't fight. We should debate and let the debate to explore the options because we have no clear answers sometimes. So we need different perspectives. [00:33:37] Speaker B: Okay. Back before I derailed us with oscillations there, I was going to ask you is whether your work on the sharp wave ripples story and seeing how different cell types contribute to potentially different types of sharp wave ripples and therefore different functions, etc. Is that what led you to your interest in applying this kind of approach to manifolds? How did you get into the manifold story? [00:34:06] Speaker A: Yeah, yeah. Everything started actually as a physicist when I was essentially thinking, I was reading about these manifold ideas and then thinking what kind of data we apply this idea of manifold. So for the physicists who are, or the mathematician who are listening to us, essentially we have a matrix with the rows are the cells and the columns are the timestamps and the activity of these. So I was thinking spikes. Exactly, spikes and firing rate. Then the manifold takes the idea that you have a population vector, which is essentially you have any number of cells. So you any space of N dimensions which are the same number of dimensions of cells you have, then you follow in time how the vector is moving in this population, how these activities moving in this population. Then because you have many cells, but you have redundancy in information and the cells are connected, there is a low dimensional structure there which is reflecting, as a repetition of the firing patterns, the region of the space that you revisit from again and again in time, essentially, when you think about replay, this is what you have. What you have is a group of self firing in sequences that repeat, right? So the ripple is somehow revisiting a short part of the manifold. And I was curious about that. And this is how everything started for me, trying to conceptualize the ripples and the manifold. And I am still on that. We are still, we need, I mean any idea requires time because then you have to get the data, then you have to test the analysis. But this is essentially where we started looking. And here is where the concept of the different cell types and the ripples become somehow entangled. And again, we are on this big question trying to solve this conundrum, this difficulty in trying to identify the differences between the two. But I think they are connected. [00:36:21] Speaker B: So I mean, maybe summarize where you are right now in that. Because I'm picturing, you know, like one, let's say you're. You discover a sphere as your neural manifold and then you could have one, one kind of sharp wave rifle ripple, you know, goes along the equator of the sphere or whatever, if it's. And then in a different kind of sharp wave ripple heads to the poles or something like that. Right, but, and, or, and what we're going to talk about is actually you don't just have a sphere manifold. The manifolds are different, there are differences in the manifold based on the different cell types, etc. So where are you at right now with. In regards to manifold trajectories? [00:37:02] Speaker A: Okay, so let's try to separate the discussion. So first regarding ripples, what we discovered was that if we take the ripples and look at the waveform of the ripples, and we try to reduce the complexity of the many waveforms by using these high dimensional strategies, we end up with a cloud that represent each dot in that cloud is a ripple. And then what we found was that the ripples of different waveforms, they form a kind of continuum in the cloud. Right? And then what we discovered was that when you map onto this cloud the inputs that each ripple receive, you see order. You see that the inputs with the stronger inputs are located associated with ripples in one side of the cloud and the weaker are the other. And we look at the inputs from the CA3 and the inputs from the entorhinal cortex and the Input from the CA2 regions, which are inputs that additionally impact ripple variance in the CA1, which is the hippocampal region we work with. Then we also learned that depending on whether the ripples are from the sleep or from the awake phase, they distribute distinctly in this cloud. That allow us to conceptualize that there are a diversity of ripples and diversity of ripples reflect a diversity of inputs and a diversity of cell assemblies contributing. This is one part, let's leave it in this part of the room, then we move to the other part of the room. Then we ask it, okay, so what the cells which generate the sequences that we see in the ripples are doing when the animal is exploring in there, we move into now how this population vector is organized. In that particular experiment, we resorted on a very simple memory. Well, very simple task. It's a navigation task, not memory really. And these are mice who have to run left, right, for getting reward, water reward. We have a linear track and they go back and forth, back and forth. Because they go back and forth and you have place cells and you have reward cells and you have cells which increase firing. When the animal is swishing the reward, you get like a repetition, which is directional information also. Right. And the topological aspect of this will be a ring. And this is essentially what we saw that a ring is reflected during this behavior. Now if we want to connect the ripples with the ring, what we need to do is to look for the replace sequences which will essentially revisit the ring and see how each ripple projecting to the ring by reactivating the sequences. Okay, so that is something we are still working on. Okay. We have reported the first, which is the ripple shape, which is this paper we published recently in Nature Neuroscience, which is related with the embedding the ripple embedding the structure of the ripples. The one which is the cell type manifold analysis, which was published last year in Neurom. And now we are working in the. In bridging, trying to make bridges between the two. The two ideas. [00:40:43] Speaker B: Ah, okay. And you're optimistic. Things are. Are you stuck or are you going full speed? What's your outlook right now? [00:40:54] Speaker A: You always stuck in size, right? [00:40:58] Speaker B: Always crawling, inching forward? Yeah, absolutely. [00:41:02] Speaker A: You always way. And then you have to change the way we think. But no, I mean now it's kind of amazing. I mean it is a kind of challenging. We are advancing a lot in identifying the how to set that. We are much advancing well in making the links between the two representations. But what is being a bit challenging is to properly link and align in time from a perspective of the mathematical alignment between the two representation. That is a methodological aspect that we need to develop. Sometimes there are not enough Methods, I mean, these kind of approaches rely a lot on topology, mathematics. And sometimes you end up in a moment in which you don't have methodology to address the problem. You know what you want to do, but you don't know how to effectively do that. And then you have to make a detour and develop a methodology and then get back to the problem itself. And developing a methodology requires validation. Then you need to stop, publish the methodology, get back to the biological problems. So we are in this moment, in [00:42:13] Speaker B: this very moment, just as an aside, I mean, maybe that's one consequence of the era that we're in with big data and all of these nice technological tools where we can analyze heaps and heaps of data that we couldn't before. And now that invents many new problems. And so therefore, we almost have to slow down just to invent new methodologies to address the new data that we're getting. [00:42:40] Speaker A: Yeah, no, it's true, it's true. But it is true also that in the consideration of the new problems that these approaches bring to us, they also open your mind for you to consider solutions that were not in the table at the beginning. And this is happening with these new fancy tools. Many people just applying the tools, but sometimes you need to understand what the tools is telling you and how that challenges your interpretation. And for me, that creates new opportunities to address all problems or to learn better how to treat the system. So, yeah, I mean, it's part of progress. It is always like that. Things are not static. The knowledge is never static, and knowledge never stops. We will always be addressing questions, expanding the toolbox to address the questions. And expanding the toolbox will open new questions or will let us to revisit questions that we thought were solved before with a new perspective. And all is valid. [00:43:52] Speaker B: That's beautiful. The. Okay, so manifolds fit nicely into this. Because I often. Right now I'm kind of wondering one thing, I wonder is, you know, 100 years from now, what, how we're going to view our, our current state of manifold mania, for lack of better term. Right back so back when we were recording single neurons 100 years ago, and then, you know, you had the, the single neuron doctrine. There's mother cells and you ascribe functions to single neurons. And now that looks, we see that and we think that's kind of embarrassing as a field almost, you know, and, and right. And manifolds have this promise, right? In some sense, like, oh, this is great. Oh, we have, we can now we can record thousands of neurons at a time. But look, it's actually quite simple. It's. And we have these manifolds are better because we can see these low dimensional structures and it all makes sense. And then people like you come along and are interested in incorporating aspects of single neurons, you know, small populations of neurons to, to see like, well, is it really the same manifold? Or if we, if we record from different neurons, are they different manifolds? And so we'll, we'll talk about that. But I just kind of wonder like, oh man, are we kidding ourselves with everything that seems to be manifolds right now? [00:45:22] Speaker A: Yeah. No, no, no. Okay, so let's, let's. Before addressing a specific, whether we need sales types to make things more complex or more simple. We don't know though. The first thing is, I mean, I don't think it's fair to blame ourselves for what. No, it's true. I mean, no, why we need to stop the search for knowledge just by limiting and say, oh, I mean, that is a new method. And now everything is solved. No, it is not solved. We will not, not know the limits till we go to the end. So if we want to understand what are the limits of the manifold framework, we need to exploit the framework till the very limit at the very end and then expose the limits of the technique itself. And that will open a new opportunity. That is the way I try any problem now offer a very tratable concepts in dynamical systems that we have been requiring for a long, long period in order to conceptualize how the brain operates. Because you build neural manifolds from the population activity of many cells. Traditionally you don't need to rely on the cell type definition because you are going to the manifold to forget about defining what place cell is or attuning properties of the cells are. And actually then you learn that those tuning properties become embedded into the manifold structure. They are the manifold itself. They are the manifestation of the emerging properties of the manifold at the single cell level. Right. Now, why do we need cell types if we already have the manifold framework to deal with the emerging properties? [00:47:13] Speaker B: Essentially, that's part of the reason we like manifolds is it gets rid of the need for cell types. [00:47:17] Speaker A: So what are you doing? What are you telling me about why you want to bring cell types to this problem? And here is where I came and said these are not planetary systems, guys. These are cells, biological cells. And cell types are diverse and they represent the compositional primitives of the system. I mean, the different cell types, neuronal types are determined genetically by developmental programs. They wire in specific ways and More importantly, they are the entrant point to control the system genetically. We will not be able to control the system to control a place field, specifically in a genetic approach, because we don't have a definition, a genetic definition of place field, of a place cell. It doesn't exist. We will be able to control glutamatergic cell types, or we will be able to control a subset of interneurons or a subset of glutamatergic cell types. The fact that they are diverse and that reflect the developmental program and the building blocks of the brain, for me, makes them important to get the definitions. This is essentially why we bring the cell types into the manifold domain also, because when we were recording the manifold and building manifold, we quickly learned that they were lacking information. If we don't have the appropriate cell types represented in the manifold reconstruction, what [00:48:57] Speaker B: information were they lacking? [00:48:58] Speaker A: For instance, we discovered that the transformation between allocentric and egocentric representation is encoded into the different manifold structure of deep and superficial CA1 cell of hippocampus. This is a simple example. Okay. When we had the animals running in a linear track, and then we rotate the linear track in the room, only the manifold from the deep cells, the deep C1 cells, were informed about the transformation. The manifold from superficial cells was absolutely agnostic to the transformation because it remained anchored to the allocentric information of the room. [00:49:45] Speaker B: Okay, so let's just back up, because you're describing the results of one of the papers, you know, that we want to discuss here. So, I mean, You've referenced the CA1 part of the hippocampus multiple times, and maybe we can just to start more generally, and let's just talk about the results of this. And I jumped the gun a little bit there. But. Okay, so we're talking about a region of the hippocampus which is the CA1 projects is the output of the hippocampus, right? Yeah. [00:50:16] Speaker A: Well, no, C1 projects to the subiculum and maybe to the entorhinal cortex. But then it's. [00:50:21] Speaker B: Entorhinal cortex. [00:50:22] Speaker A: Yeah, it is. The real output of hippocampus is a subiculum. Yeah. [00:50:28] Speaker B: In either case, CA1 has structure within it and layers within it. And the hippocampus itself has like, kind of a unique layered structure. That's really well known. But we're talking specifically of one area of the hippocampus here and maybe talk through why we think that there are different types of cells in. In the CA1 that this would even be something to look at. [00:50:51] Speaker A: Yeah, yeah. There are different type of cell in CA1 and also in the CA2, CA3, the other regions of hippocampus. Right. In terms of glutamatergic cells, which are excitatory cells, there are different composition across the layer, the sub layer within the hippocampus, deep and superficial. They are generated at different embryogenic period during the development and they wire differently, project a bit differently, and are integrated differently with local circuit gabergic interneurons, which are extremely heterogeneous. The work of Peter Somogi has defined more than 25 type of GABA G cells within the CA1 region and also in CE3 and the other region. So there is heterogeneity. They connect differently, they receive differently from the inputs, integrate differently and project differently. And that is an essential part of the system. [00:51:52] Speaker B: Is this the case where you know, what you want is a nice clean story? Right. Like there are 12 different types of neurons and they have properties that are orthogonal to each other, but like everything else that we've been discussing, they're not orthogonal, they have like overlapping. Right. So it projects like 30% more to the subiculum. Not like these cells project to the subiculum, these do not. Right, so it is a gradient, as you say. [00:52:16] Speaker A: No, absolutely. But this is how biological function emerge from gradients distribution. Yeah, absolutely. But the key question is that we don't know how these gradients map into this gradient, maps into the other domains. And this is what we look for. Connectivity, participation in different functions and modulation by different aspect. I mean neuromodulatory action for iter, which is essential in the function. The expression of receptors for different neuromodulators is distributed in these cell types. Some will be more responsive to serotonin, the other acetylcholine or. I mean, and you will activate different subsets of cells depending on that. So we need this information. I mean, it's part of the system, how the system is built. The very same system that we are trying just to conceptualize emergently, but ignoring the composition. That's the point. But if you think, I mean, we are doing this effort in two opposite direction. I mean, with cell type atlases and all the work by the Allen Institute and all this categorization of cell types is making granularity to the point that we go to the single cell level at that point, then at the other extreme we are functionally trying to compose everything in an emergent population concept, which is manifold. And then we ignore that. That's the challenge for me. To try to get information from these two extremes of investigation of the system and try to interact, eventually being able to extract more, more information. [00:54:12] Speaker B: So what is your take on the ever finer grain classification of cell types? Right, right. Does it depend on your question? Does it depend on. [00:54:23] Speaker A: Yeah, but for me it's operationally. I mean, we need to define this in operational terms. What is the question? If you are looking for the connectome, you need that information. If you are looking for communications between regions, that is essential. We need to know who is broadcasting what where. We need to know that. So we need that information. If we are placing electrodes in two or three regions in the brain, we need to know how they are expected to be connected. That information is required if we want to go to the single cell level. If we are talking about engram cells, we are talking about engrams. What proportion of place cells are Engram cells? And from there why are they engram and the others are not? And then what give a cell to become relevant for a memory is one single cell. For memory is one single cell essential to retrieve a short memory item. So we place the bar where we need, depending on the question. [00:55:29] Speaker B: Okay, so in this case the bar is placed in the fact that there are superficial CA1 pyramidal cells and D CA1 pyramidal cells. So what's the story there? What's the difference there? What did you guys do and what'd you find? [00:55:43] Speaker A: Yeah, so, well, actually this started like in 2011. That was the first paper and it was Mushaggi's paper lab noticing differences of pyramidal cells across the C1 layer. And in that moment the angle was it looks like we have more place cells in the deep deep versus the superficial. Then after that initial discovery, many labs became interested in this angle. What is this deep and superficial? And then we Learned that deep C1 cells, they are early born cells. Along development, they receive more input from parvalbumin, gabaergic cells, superficial cell, they get they are born later in the development and they receive more input from other type of Caborg cells, CCK cells. They also integrate differently. Input from the CA3 and input from the entorhinal cortex. They distribute it distinctly whether they are more proximal or more distal into the CA3 to CA1, CA1 to subiculum axis. Then we start learning that they are more reflecting place feel associated with somatosensory landmarks in the deep layers. And those are the superficial. They are more oriented allocentrically. In my lab we started also noticing differences in their trend to neurodegeneration and how they contribute to the temporal load sclerosis and how they participate differently into the, in the sclerotic hippocampus and in seizure type of activity. So they started to emerge as something. This is an angle that we should try to look at. [00:57:42] Speaker B: And this is, you said originally 2011, the suggestion was there just seems to be more place cells in one of the layers than the other. But they're essentially both contributing to the same cognitive map. [00:57:59] Speaker A: They essentially contribute more or less into the same cognitive map. One of them shift more during REM sleep to different theta phases. That was deep. And then essentially was more or less proposed that they could participate differently in different assemblies. But as I said, that was the beginning of the phase. Right now, I mean there are some labs, many labs looking at that and we are now learning a lot about the hippocampus cells just by considering this dimension, the deep and superficial. It doesn't mean that everything is deep and superficial. It's just a convenient dimension treatable in experimental terms because genetically they can be defined. We can use specific CRE lines which are a technology that allow us to constrain the expression of some proteins like channelrhodopsin for optogenetics or gcamp for calcium imaging in deep or in superficial layers of the C1 and then eventually allow us to address questions. So right now they are are main part of the research program in leaden labs, including mine, because of that. [00:59:23] Speaker B: Okay, so before we go on, I just want to emphasize. So earlier I said, oh, manifolds are supposed to be this great thing because then we don't need to worry about the distracting details, all the varieties at the lower level, at the cellular level, for example. But what I want to emphasize is that one of the hardest problems in complex systems like biological systems is connecting between levels. And what your work is promising to do is connecting in some way from those lower level details to this, I don't know, meso level, mid level manifold kind of descriptions. And of course the hope is to connect at all levels and then you have this multi level sort of understanding. So I just want to emphasize that we should be excited about this. Not just like, oh no, we have to deal with these things. [01:00:17] Speaker A: Thank you. No, no, no. There are many people excited about that, including us, of course, of course. No, no, I appreciate this coming a lot because I absolutely agree we need to connect across hierarchies and this is still unsolved in the field. And yes, this is a way to try to address this question also for me, again, it is important to bridge. Remember, we start discussing, we start discussing how, I mean, how many ripples are, how different they are, how the ensembles are. I mean, how we will end up understanding the diversity of memories just by pulling all of them together. I mean, my God, we need to keep granularity and emergency at the same time. Time. [01:01:07] Speaker B: Right, right. [01:01:08] Speaker A: Treat the beast. Because these are like two heads of a beast. Treat the two heads of the beast, the granularity and the emergency together. [01:01:18] Speaker B: Yeah, yeah, yeah, I agree. Well, I mean, that's one thought I had about your work and I promise we'll get back to. Because I do want you to describe what you found. But it runs the risk, right. When you start considering different cell types and their effects on manifold and then you think, o, well, then we have this combinatorial explosion problem where instead of this nice one ring structure manifold, now we have about 40,000 sub ring types, manifold structures, and is that 40,000 manifolds? And oh no, we're already in. You know, there's too many things going on and the simplified story of manifolds is ruined. So that's the worry. [01:02:01] Speaker A: Well, but if you want to decode, if you want to decode properly, I mean, if you want to decode a general principle, then with a low dimensional representation, that's enough. But if you want to decode more, we use, listen, we use the manifold to decode what the animal was doing. I mean, to decode and in principle to predict what the animal is going to do. Right. And that opens another question because what we saw was that when we use the manifold to the coat, we saw that the animal sometimes were where the manifold predict. Okay, but in some circumstances we saw that the animal were not where the manifold was. The animal was stuck, for instance, in the somatosensory cue. While the animal was away from the somatosensory cue, does it reflect in the fact that the animal was thinking in the somatosensory cue even if it was not in the somatosensory queue? And then what else is the animal thinking if they are thinking too much that we can decode for a general emergent manifold, or do we need to go into the low levels to decode more properly at a more refined level? That's open question decoding. Using this information to learn how the brain is representing what we are thinking, we will need to go from the emergent level to the bottom level for sure. [01:03:31] Speaker B: Okay. Okay. So tell us a little bit more about what you found in this study. Connecting the cell types and manifolds. [01:03:38] Speaker A: So what. Essentially what we did was. So the experiment was quite simple. So we take two type of animals. We express, essentially we express GCAM in deep or in superficial cells using two different three lines. And we also build cross animals to have a mouse, which allow us to express differently in the two layers at the same time. So we use either a green calcium indicator, such as GCAN alone, or we combined it with a red shifted calcium indicator so that we can target the deep and the superficial cell in the very same animal at the same time. Right. So what we found was that essentially the geometrical properties and the topology, the simple topological properties of the deep and the superficial manifolds were slightly different. For instance, when the animal was running back and forth in the linear track, we get a ring representation, right? Which is expected because you have directionality and you have place cells information. Right. But then when we look at the ring, we saw that the ring was more ringy. How to say that in English? A ring was topologically more compatible with a ring if you change the ratios around each point. [01:05:04] Speaker B: Is this the eccentricity differences that you're talking about, so that it was more oblong? [01:05:08] Speaker A: Yeah, well, eccentricity is more a geometrical property topologically. Whether a ring persists longer, this is topology, this is persistent homology. Okay, so we found that the ring persisted longer topologically, in topological terms in the deep than in the superficial. Right. Then we also look at the geometrical property, as you mentioned, eccentricity of the ring, and we found it was more rounded in the deep than in the superficial. For instance, we saw some simple differences between maybe you can say, okay, well, not a big deal. I mean, two rings, one more ring than the other ring, but at the end of the day, more or less similar. Well, the beauty. [01:05:55] Speaker B: Sorry, can I ask you just a small technical question because I'm learning more about topology and persistent homology. Is the interpretation of a longer persistent homology. How do I interpret that? Is it because the density of the interactions between cells somehow it is. [01:06:17] Speaker A: The interpretation is that when you look in the vicinity of your point and you look for the behavior in the vicinity of a point, the topological structures that you are looking for persist longer if you look farther from your point. If they persist longer than others, they are more persistent. Like we, we have indication of a ring in a longer, in a longer radius around the point for the deep, numb for the superficial. [01:06:49] Speaker B: Meaning it's more stable. [01:06:50] Speaker A: Right, More stable topologies. Exactly. More stable, meaning that is more that the spatial scale in topological terms. Is longer, is. Is larger, it's less granular, it's more stable at the level of space, for instance. [01:07:11] Speaker B: Yeah, yeah, okay, okay, Sorry. Sorry to interrupt. I just. [01:07:14] Speaker A: No, no, no, no. At all that's relevant. And yes, then the geometrical properties were also different between them. And then the beauty was not this simple topology or geometry only because at the end it was just classifying a ring or characterizing a ring. The beauty was when we try a transformation, we try to generate a conflict between the local cues that were in the linear track and the cues that typically you have in a room. You get an orientation in the room. Now, if I get to your room now and I rotate your table, you still are able to orient yourself, but your brain will tell you what happened here. I mean, you have rotated my table. Right. And we rotated the linear track in the room. And then we looked for how the two layers, the two sub layers, were reflecting that transformation, which is an essential simple transformation for orientation. Okay. And what we found was that the superficial cells, the ring was invariant, totally invariant. It keep anchored totally, which is essential in a navigation system. You need to anchor to. Right. And then the deep was. [01:08:38] Speaker B: This is. Sorry, so this is. That means it's like. So the inference here is that it's more anchored to the global properties of the room. Where the door is, where the window is, et cetera. [01:08:48] Speaker A: Exactly. Which is. Yeah, exactly. We were more or less expecting that this happened based on the knowledge that we about the place cells of the deep. Okay. But it was not so clear that we will be so clean. But it was super clean. Right. And then in the deep and the deep, because we have these somatosensory cues, then we thought, okay, let's see what is going on. And what we saw was that the deep ring was rotated. It rotates. There was a transformation, a geometric transformation of the manifold associated with the conflict between local and global queues, which was reflecting a change in the war that occur in the war. And this is what you need essentially for a representational system. When you have an agent that has to record, when you have an agent trying to get oriented, you need this agent to. To be representing in real time what is occurring in the environment. So if there is a transformation, you need to keep you oriented and anchored to what is fixed, but also reflecting what is not. And so what our data show and our experiments show is that you have these two transformations there. It's a very simple task. We are now elaborating more with other tasks, including memories, different type of memories. And different rules for alternation, for instance. And we also see these differences in deep and superficial informing the emergent representation. [01:10:30] Speaker B: Was this memory still along a track, like a navigation kind of task involved? What I was going to ask is how to think about. We started off by talking about how you think about hippocampus. And many things have been found in hippocampus. Hippocampus, for example, cognitive maps in abstract feature space. Right. Not just navigation, not just memory. And so I was thinking like, how would these properties of superficial and deep layers, local and global, map onto how we think about cognitive operations in a relational abstract space like that? [01:11:09] Speaker A: Absolutely. No, no, no. This, this, this is an experiment we want to do. Also we are discussing in the lab up different tasks that could exploit the different phenomenology. As you said, we are now trying, more interesting in trying to move the problem to the idea of representation concepts representation of the trait between generalization and specificity. The traits of how much? Because we feel that the two sub layers will have roles on that too. Because again, any representational system requires a gradient of generalizability so that you can generalize what is common between the different contexts, but at the same time some separability. So the idea of pattern completion, pattern separation, which typically is discussed in the dentatigyros and in CA3, we also think it has some. Some function into the CA1. Of course, we can move to CA3, but we try just to keep the problem treatable in experimental terms. Implantation of the green lengths and obtaining data, you require a learning curve to get the. And therefore you're ready to implement this aspect there. So also CA1 is projecting to the subiculum. The subiculum is a structure which is extremely heterogeneous, is much more heterogeneous than the CA1 and the CA3. And there is very little that we understand about the subiculum. And the subiculum, as I said, is the one that project neurons in the subiculum. They project brain wide. And therefore it is in a position, in a very strong position to influence behavior structure only. Also, we have learned that the C1 projection to the subiculum is structured depending on the deep and the superficial and how they innervate. So this problem of generalization separability, we are also thinking on that. [01:13:28] Speaker B: Isn't that where the cortex is supposed to come into the story? Also, I mentioned complementary learning systems earlier. Isn't the cortex supposed to be the generalizer and the hippocampus supposed to be the separator? [01:13:41] Speaker A: Yeah, well, again, this is where you put, you bring the level, right? I mean, if you go within the hippocampus, if you go to CA3, proximal CA3, closer to the dentarius is more pattern separation. And distal C3, which is farther from the dentarius, is more pattern complexion. Of course, in there, if you, you open the focus and then you go from the cortex and hippocampus, then you can try to give this generalization. I mean, this is like a fractal organization, right? If you open, you see the structure, you see generalization, separability. But then if you go and bring your focus within generalization, then you look at within generalization, you can also have a substructure. And if you go to the substructure, you have more substructure structure. It is also a property of a complex system and it is how the brain works. [01:14:39] Speaker B: Oh, I'm so glad that you said that, because I think that's the perfect way to look at it. And the fractality, people will love the scale free people will love this, this view of, of the brain as a whole. But so in any case, like one of the nice things, one of the nice take homes from your work that we've just been talking about is the different cell types affect. They don't destroy the manifold, they don't radically change the manifold. The manifold itself is pretty robust, but it shows that the manifold can be bent. It could have different topological properties, not radically different. So in some ways they're really working together and they're kind of like control nodes in terms of shaping the manifold. [01:15:22] Speaker A: Totally. They have to. Otherwise we will lose the coordination of the information. I mean, if you have a transformation, you need a fixed structure. Somehow you need something to generalize and something to separate, which is coordinated. So the key is the coordination of the representation. And this is another thing which is important because the two cell types, they. Well, in the case of hippocampus, it's a pretty interesting system because. Because we have learned over this year, working in this deep and superficial organization of the circuit, that the two layers are somehow parallel. They are not connected between each other. They are designed to be orthogonal. Right. Another issue will be a brain region in which the two layers will be connected directly. Their separability will be more challenging and probably the function will be different. Okay, but in this particular case we have this advantage or the principle, the organizational principle is this parallel processing in the layers which are reflected this way. [01:16:37] Speaker B: Does this make you think of going back to the phrenology Days we want to name different areas of the brain, and often they get names, like, based on their function at the time that people thought that they. They were doing. So should we, should we consider these two sub. Sub layers, different brain areas? How should we think about, you know, how do you parcelate that? [01:17:01] Speaker A: Okay, categorization again? [01:17:04] Speaker B: Well, we have to, we're human, we have to categorize, right? [01:17:07] Speaker A: Of course, of course. No, no, no, of course. I mean, the anatomy is providing the limits of some categorization then. I mean, are they different brain regions? No, they are not. Anatomically, they are. Are in the very same region, actually. I mean, the distance between one soma to another are within the distance between the somas at the different regions. So in this case, there is no doubt that they are just different. And actually there is a gradient in the expression of the different genes and the transcriptional landscape. [01:17:45] Speaker B: Shocking. [01:17:46] Speaker A: Exactly, exactly. But then you define clear borders between CA1 and other regions. With CA2, for instance, there are clear markers. I mean, if you look for proteins or you look for gene expression in CA2 and CA1, you see clear differences at that. And there is a clear transition there that indicates that this is another region cat to. And then when you go to the subiculum, you clearly identify anatomically transcriptionally immunost. Using immunostaining, you see differences, clear differences. [01:18:21] Speaker B: So we don't, we don't need to rename CA1 into, okay, CA1A and B or whatever. No, no, but we will refer to. [01:18:30] Speaker A: Yeah. Another thing is what, what is the organization of the dorsal hippocampus from the denti gyrus CA3, CA2 and CA1. Whether. I mean, if you. We don't see them in a space, in the anatomical space, and we close our eyes and we think on them in the connection space as a network. Maybe deep are Farther from the CA3 than superficial, you mean. And then if you kind of unfold the network, maybe it will become closer to CA3, the superficial and then the deep will become closer to the CA2. And then if you draw that, ignoring the anatomy, it will give you like a different landscape within the same region. But anatomically they are within. They are intermixed in the. Along the CA1, along the Cornus ammonis. [01:19:38] Speaker B: All right, so we've talked about a little bit about your work connecting manifolds to single cells, types of functional identifiable cells and genetically identifiable cells. And we also discussed, you know, well, this is the grand scheme after I said, or this is the grand project is to connect levels after I said, let me correct myself and make sure that I say that this is a great thing because it connects levels. So do you think about to bring you back up to the manifold and going to maybe a larger. Higher hierarchies are bad, but connecting manifolds to higher cognitive functions. Right. What do you think the promise is of. So you're getting some grasp on this, connecting the lower level to the manifold. What do you think? [01:20:29] Speaker A: So my vision, which is a vision and could be wrong or right. Right. But this is, this is my, my role as a researcher. What I can envision is that we will have a fractality again into the manifold organization in terms of different interacting hierarchies of representations which are interleaved at different levels to complement a high functional integration. [01:20:59] Speaker B: Okay, wait, so you're talking manifolds interacting with each other. [01:21:02] Speaker A: Manifold, manifold and yeah, yeah, because the super. The space. I mean, so here is the concept of super spaces, right? The idea is that you have a large number of neurons which essentially can represent the whole aspect that the brain region will. Will be representing reality. And then within that you have like subset of manifolds or subset of representation that coexist and interact between each other. But these are also part of other regions they communicate with. And there is a subset of them which communicate with other region. And that subset give you a super space of communication. And that communication superspace will be broadcasting information from one region to another. Meaning that if you have a representation, say which is rotating or transforming in one. Let's think about the C1 subiculum, C1 deep and superficial rotation, no rotation. In this idea, you will have this information being transferred to the next hierarchy through this communication sub space. And then the whole system will contain all representations. So it is like everything everywhere, all at once. But sub specializes in the different super representations that interact each other to inform a global representation. We can read the code at the global level, while at the individual levels you are also transforming the information. This is how I see this problem. For that we need to record massively from many, many regions and many. And have the animal running task. We have different levels of information to be encrypted in different systems so that we can represent this whole aspect. And for me that will provide more complete models for informing under coding the information in agents, in artificial agents, or in the human. [01:23:16] Speaker B: Oh yeah. So I was going to say in that vision and you just alluded to doing this in rodents. But of course I wanted to take us directly to subjective experience and just the cognitive capacities of humans that differentiate Them from, let's say, rodents. Right. So in that vision, right or wrong, what would differentiate at that level, like humans, for example, from rodents? [01:23:44] Speaker A: We are now in the speculative phase of the conversation. Right, okay, fantastic. Well, we have no idea, but I guess that the complexity of the human mind and the brain representations, the parallelization that we can regarding different contents aspect that we can keep also attention is a fundamental angle to understand how you read and how you prioritize information. So I think all these aspects will be emerging as we progress with the manifold idea and with the, I mean the manifold is the population level representation. It's just as we progress, as we progress, we will be learning more about that. And I guess that we need these interspecies comparisons for sure. Of course we tend to be anthropocentric in the question that we ask and in the interpretation. But I think that we can learn a lot from other species just because they do different things with their brain. They represent the world differently. And then we can learn about that. I mean, for instance, bats and mice, they are both rodents, one fly, the other run within the forest and they have different configurations, they both have hippocampus. And we have learned lot on differences between bats and mice. Just looking at this, this will open new perspectives. So speculatively, yes, I mean, we need to check. [01:25:19] Speaker B: Yeah, well, what is it like to be a bat? That's an illusion too. Okay, I want to make sure you got the illusion, but I guess that's. So you were talking about like super spaces with the manifolds. And part of my question was wondering, are we going to find a, a difference in kind or is it just going to be more, you know, a difference in quantity. Right. When, when we go across species, for example, I mean, I can see like different, like the bats world with echolocation and flying, there will be different manifolds. Right. Interacting relative to the mouse navigating on terrestrial land, for example, and whiskering and stuff. So but then, but across like species and we can be anthropocentric or not. I mean, I guess ultimately I want to ask you about subjective experience and whether you think that this is going to be able to connect all the way up and we'll have a nice explanation for our subjective experience, which is. [01:26:19] Speaker A: No, that I don't know, Paul. I mean, I haven't thought on that carefully. And I do think that we need more human data for that, that we need to control experiments more specifically just to try to make inferences on that core principle of cognition which are so human in a way, because we cannot study that in animals. But just speaking about bats and mice, I mean, I think that for instance, animal that migrates like bats or like birds, they migrate longer or that require different representational capacities because they anchor local and more global information. So probably they have specialization within the navigational system of the brain, like in the hippocampus to cope with that. And probably the differences that we see are reflecting that it's not the same to project your map into a very narrow vision. In the forest where you move on, everything is close to you. Somatosensory information is so informative of everything that happened that for a bat that is flying around, relying in the orientation with the light, distance, seeing the world from distance and orienting themselves regarding other aspect that we are not integrating in the case of the mice. So I think that these two representations scale differently, so different regarding the inputs regarding the internal map that they have to build to get oriented, that I think that we will learn a lot about that if we understand that they are two species. I mean, that they are different in the way they have to handle information to survive and to live. Okay, yeah. [01:28:13] Speaker B: I mean, just off the cuff, couldn't they just exploit different regions of their scale free fractal representations? Right. One would just. It doesn't take once you get up into, you know, a couple orders of magnitude, one more order of magnitude, all of a sudden you're at planetary distance. [01:28:28] Speaker A: Absolutely. [01:28:29] Speaker B: So it's not that hard to navigate the globe. [01:28:32] Speaker A: Totally. Absolutely. But seeing yourself, imagine yourself when you get out and walk in the city and you are orienting yourself in the city using very close landmark, like next to corner the metro station, etc. And now try to make the exercise of internally make your brain look at, look out and look at the sky. And now try to imagine how this block where you are walking looks like from above and trying to orient yourself there, that will expose you how just changing the scale requires for your brain the exercise to anchor different. What I predict is that your representations are moving from one scale to another in this fractality representation. And if you have to navigate the city like a superman flowing through, now you need to rely on that not in the coffee shop and the corner and maybe the tree in front of the coffee shop. You need to think about the blocks and the landmarks in the city that makes a different demand to your navigational system. [01:29:46] Speaker B: Cerebro espacio y tiempo la nora ciencia de como navagamos por la real la memoria. Or El futuro. I'm so sorry, I just butchered that so much. But what is it? Brain space and time. The neuroscience of navigating the real memory and the future. [01:30:10] Speaker A: Absolutely. Thank you. Thank you. [01:30:12] Speaker B: You do it for me. I'm so sorry. [01:30:14] Speaker A: It's fantastic. No, no. Yes. So it is. It is the book I have written in Spanish. Yes, yes. Yeah. [01:30:21] Speaker B: So this is. Yeah, so it's only in Spanish and I had not realized that you had written this book. But so tell me a little bit about. About the book. Do you mention fractality? I assume you don't mention subjective experience. [01:30:34] Speaker A: No. Is an attempt to bring together space, time and memory into a single framework. And to explain that in principle to the general public. To try to get simple concepts about that in Spanish and to communicate. And then it's a kind of reflection not only about what we know about the representational system of the brain. Place cells, grid cells, head direction cells, cells. But also how that builds a representation during ripples and sequences. And how that representation is intimately related with memories. With memories. And how these memories fragile. And how. Because the way we build this representation and we build memory, they are somehow how personal and not accurate reflection of the real world. And the book discuss about all these problems also with some philosophical touch in terms of what we are, how we understand the world, how certain we can be this aspect. So that's the idea basically. [01:31:56] Speaker B: And it's geared toward public. It's like science communication type book. [01:32:00] Speaker A: Pop. [01:32:00] Speaker B: Pop science book. I have not been able to. It's in Spanish and I know that it's. You're working on a translation, potentially, is that right? [01:32:09] Speaker A: Yes, I'm trying to. Yeah, yeah, yeah. I mean we. I wrote the book in Spanish, which by the way is my mother language. And when I was writing the book, I realized how important for a brain is to. Is to activate in. In the language. Your language is a reflection of your work, is your construction. And for me I am getting used to write scientific papers in English. But as a person I am more complex than a scientific paper. I also read philosophy, I also think about literature and other aspects. And writing this in Spanish was an opportunity to me just to see how I can merge the different ways world in writing something that probably if I have to write in English, I will not have the same freedom because my vocabulary is more. Is built for science, not built for literature, but my Spanish vocabulary. My brain is built for life. And therefore I can do literature. I can do literature and write more literally. And this is the book so the book was written in Spanish and now has been published by. And it's quite successful. I mean, we saw the first edition and we are now with the second edition and I think it's well appreciated in Spanish. And we are looking for some English publisher to try to discuss whether we can we have a translation to English and also in French and in Italian. It can be done. So. So we are open to that. [01:33:53] Speaker B: What did you learn with those two different cell, two different versions of yourself coming together? [01:34:00] Speaker A: Well, they communicate more each other. That was really an experience. When I start trying to explain in Spanish what a placel is and what a ripple is. And then I started to interact more with my. My culture, my understanding of the self, with many, many, many writers. Borges, for instance, Jose Luis Borges. Jose Luis Borges, for me, is one of the main writers and probably someone who knows so much about representations, about what is true and what is not about time. He has a lot of ideas regarding time. [01:34:53] Speaker B: I've had so many people tell me that they love Borges and it's all there already. He already understood it all, is what people say over and over. [01:35:02] Speaker A: Yeah, Borges was an incredible writer and an incredible mind. And the way he extracted, he built the metaphors for. I mean, we communicate each other through metaphors. Metaphors are everywhere. We also make metaphors for science. When we said manifold, neura manifolds, when we said consolidation. And we build a metaphor, what a consolidation is, and then we try to fit in that so that the. The role of metaphors is somehow more appreciated in literature, but in science it is there. And Borges, he has built metaphors of many aspects and he has reflections regarding time, space, the fragility of the representation, what really means to dream, what really means to think about some ideas. So, which resonates to me too much with the aspect of neuroscience. And having the opportunity to write in Spanish for me was also the opportunity to link with Borges, because I read Borges in Spanish and I inspirational in many aspects. [01:36:22] Speaker B: I mean, you're just saying that the book is, you know, it's meant to be understood by the public. And people write books with different aims in mind and for different reasons and to communicate to. With different people. But I, you know, and I often wonder what the right time is to write a book, because often people, via writing their books, learn something new because of what they're grappling with or because they're putting together concepts that they had not tried to put together before. Did writing the book change your mind about anything? Or did you learn anything via writing? Or is it more bringing together the ideas? [01:36:59] Speaker A: Well, in two sense, writing the book was inspirational for me. The first was that by. For a first time, telling or talking about place cells and all the scientific knowledge that I had, typically in English, in Spanish, allowed me to access to a different understanding. And that for me was really shocking because. Because I was sure that I understand the concepts very well by studying them in English. But then I realized how deep they went into my consideration when I started to talk about them in Spanish in many. And this made me think that sometimes we, the researcher who are not English speakers, we do a huge work in trying to. To communicate our. Our signs in other language, which is totally fine because we have to do that. But thinking in your language make a lot of intuitions that sometime thinking a different language does not. So this what you mean, you. [01:38:11] Speaker B: The. The huge work is in communicating to your native language speakers or the other way around, or translating yours to internally. [01:38:23] Speaker A: What I mean is that when I was trying to communicate and to elaborate the concepts in Spanish, I realized that I gained some intuitions in Spanish that I didn't have when I was using this concept in English. [01:38:43] Speaker B: Yeah, I understand that. I thought you were alluding to the fact that that might unlock it for reader. For the readers in Spanish. Right. It might unlock these concepts that they might not get it in the English version because you have the expertise in the English version and you do that massive work in translating it. [01:38:59] Speaker A: Yeah. And also for me as a person, if I were to be able to do my science in Spanish and communicate my son in Spanish and think my signs on Spanish, probably I will gain intuitions more easily than doing this in English because my brain is not bilingual. Well, my brain is Spanish. And I just. I mean, talking with you is great and I can expose a lot of ideas with you, but I am more and more fluent in Spanish and I can find better words for more nuances in Spanish that I can do in English. Right. And I think that limits the representation that I can handle within my brain at the moment that I am thinking about. About that. That's one thing. Okay. The other thing that I learned by writing the book in Spanish was that. And the reason for writing the book also is that I myself have. Over my life, I have learned a lot about life and about how to understand my role in life, how I can understand how we as well as a society behaves by being a researcher, a neuroscientist. I have learned a lot about that, for instance, by knowing how fragile memory is. I think I have learned how fragile we are when we claim memories as an argument, for instance, by understanding how context dependent we are are. I have learned a lot and this is something that I learned from neuroscience. I have learned a lot how context dependent we are when we position ourselves as a human person in front another. And I do think that this message is important for the public that we are not perfect because we are not designed perfect. Our brain is not perfect by design. Therefore we need to understand our imperfection and try to behave better to others by understanding that. So I try also to convey this information in the book and also I discuss about this when I get interviews regarding the book and we talk about this in Spanish, of course. And I think this is a message that research I can also help that people understand themselves in this fragility and this imperfection which is part of our system. [01:41:43] Speaker B: That's great. So I'm looking forward to an English version if and when it comes out. I mean so, and I'll share this and I hope, I hope Spanish readers do read this. I mean, fortunately for you, you have been given a demonstration of near perfection with my Spanish reading reading that I demonstrated earlier. And unfortunately for me, even though I'm monolingual, it still doesn't help my science writing or my literature writing. So I don't know what I'm going to do. But I appreciate that there are folks like you doing the work that you do. So I think that's a great place to leave it and continued. You don't need the luck, but continued luck in your work and I appreciate you having the conversation with me and I think it's really cool, cool that you are connecting the cellular level properties with these manifold properties because like, like you say, I. Well, I wonder many years from now how we're going to think of manifolds and you're doing that work to elucidate all that. So, so thanks for being here. [01:42:45] Speaker A: Thank you. Thank you very much, Paul, for this time. I enjoyed very much the conversation and I really, really appreciate this, this moment in which we can share ideas about the work and life. Thank you. [01:43:06] Speaker B: Brain Inspired is powered by the Transmitter, an online publication that aims to deliver useful information, insights and tools to build bridges across neuroscience and advance research. Visit thetransmitter.org to explore the latest neuroscience news and perspectives written by journalists and scientists. If you value Brain Inspired, support it through Patreon. To access full length episodes, join our Discord community and even influence who I invite to the podcast. Go to BrainInspired Co to learn more. The music you hear is a little slow, jazzy blues performed by my friend Kyle Donovan. Thank you for your support. See you next time, [01:43:48] Speaker A: Sam.