Watch Episode Here
Listen to Episode Here
Show Notes
This is a ~1 hour talk on the field of developmental bioelectricity from a perspective of cognitive science and the homology between mechanisms of self-assembly of somatic and brain-based intelligence.
CHAPTERS:
(00:01) Introduction to Bioelectricity
(05:30) Cells as Collective Intelligence
(11:15) Beyond Genomic Information
(17:45) Anatomical Compiler Vision
(22:30) Bioelectric Patterns and Memory
(30:15) Organ Creation and Regeneration
(38:45) Computational Models for Regeneration
(44:00) Anthrobots: Novel Biological Constructs
(50:30) Future of Biomedical Applications
CONNECT WITH ME:
TRANSCRIPT:
And thank you for organizing this amazing meeting. Thank you for giving me the opportunity to share some ideas with you. What I'm going to do today is talk about some symmetries, what I think are profound symmetries between neuroscience and developmental biology. If you want to see any of the details, the software, the datasets, everything is at this site, and here are some personal thoughts about what I think all this stuff means.
So I'm going to frame today's talk around this question of how does it know? Because when we look at biology, and in particular, developmental biology, that's one of the first things that strikes people: how do the cells and tissues and everything else know what to do?
So I'm going to give a few main points. First of all, I'm going to claim that biomedicine and bioengineering boil down to finding out the informational structures of the material of life. That is, how does it know what to do in various circumstances, what kind of mistakes it makes and why, how do we alter decision-making towards specific set points that living material is very good at doing, and for biomedicine and then engineering in particular, how do we change these outcomes with the least effort possible?
And I'm gonna make three basic claims:
- Bioelectricity is a kind of cognitive glue that binds individual cells and other structures towards larger-scale purpose.
- Bioelectricity is also a kind of memory medium which basically stores patterns towards which cells and tissues operate, and that we can now reset those patterns.
- Towards the end, I'm gonna show you some new bioengineering, which I think is a great opportunity for bioelectricity, but it hasn't really taken off yet, but I think it will.
So the most obvious thing that anybody first asks when they see this process of embryonic development, where we all start life as a single cell and eventually we become one of these amazing things, basically, the first question arises, how do the cells know what to do? They all have the same genome. The stem cells and everything else in your hand and in your foot are the same, but why does your hand not look like your foot? How does it know when to stop? All of these things.
And so this question of how do they know what to make, I'm going to emphasize the idea that the mechanisms of knowing what to build and how are actually homologous, meaning both in terms of their mechanisms, their evolutionary mechanisms, and their algorithms, to those that underline knowing in familiar contexts, meaning behavior, brain-driven behavior. These are some of the exact same mechanisms that operate in those two spaces.
And the first clue we get to this is this interesting point that chemistry doesn't make mistakes. Chemistry just does what it does, and the laws of chemistry, we sort of roll forwards and that's all. But morphogenesis absolutely can make mistakes, and so something very interesting happens in this transition, where you go from chemistry and physics to a morphogenetic system that has goals and the ability to fail to meet them, and then to various cognitive and behavioral goals that also can be met or not.
And this kind of thing was well-appreciated by Alan Turing. He was a guy who was very interested in intelligence broadly defined in different embodiments for intelligence, machine learning and machine thinking and things like that. But he also, towards the end of his life, he wrote this amazing paper, The Chemical Bases of Morphogenesis, and we might wonder why would somebody that was interested in computation and intelligence be thinking about chemicals during morphogenesis, but I think he actually saw this profound symmetry that the story of the autopoiesis of minds is basically the same story as the autopoiesis of the body, that we have to understand how this works.
And we encounter the amazing aspects of this self-constructing material very early on in development. This is a single cell. You can see it's very active. It has lots of competencies and its own little cognitive light cone. This is a free-living being called the lacrymaria, but we're all made of cells like this, and for that reason, we are all a collective intelligence. We are all made of competent little sub-units.
In fact, during embryonic development, when we look at an embryo and we say, "There is one embryo," what are we counting? What is there one of? There might be hundreds of thousands or millions of cells. What is there one of that we're counting? Well, what we're counting is alignment. We're counting commitment of all the cells to the same journey in anatomical space. What makes it an embryo as opposed to a pile of cells is that they're all committed to the same homeostatic process that's going to get them to a particular region of anatomical space. They're all going to make this thing.
And in fact, what you can do is you can make little cuts in this blastoderm, and here are some in an avian embryo that I made many years ago, and when you do this, each of these little islands, for the time that it doesn't feel the presence of the others, it self-organizes into its own embryo. And so from this excitable medium of this blastoderm, you might get one, two, zero, or up to half a dozen or more individuals.
So the question of how many individuals are here is not an obvious question. It is not set by the genetics. It is solved in real time by processes of alignment, by this, and I'm gonna mention this again and again, this notion of cognitive glue, these mechanisms that enable individual pieces to align in some kind of problem space, in this case, the anatomical space, to have a shared vision of what it is that they're going to build.
And bioelectricity is a really important, it's not the only mechanism, but it's a really important cognitive mechanism that allows wholes to form. By the way, of course, in cognitive science, we already know this is the case because electrophysiology is what makes us more than a collection of neurons. It is what allows us to have memories, preferences, goals, and so on that our individual neurons don't have, and there too, you have this exact same kind of question about how many individuals are within a certain amount of real estate because we have split brain patients and dissociative identity disorders and things like this. So this is again, very parallel.
So we need to understand how all these kinds of decisions are made. And the standard story, it's the genome. In fact, more than that, the standard story that we're told is this kind of open loop process that leans on notions of complexity and emergence. That basically there are these gene regulatory networks. They make proteins. Some of the proteins do things. They diffuse or they're sticky or they have enzymatic activity, so there's a bunch of physics that goes on in parallel. And then this magical process of emergence happens, and we know this is true. If you have lots of simple rules that you execute in parallel, often the outcome is quite complex. And so the standard story is this. This kind of open loop feed go-forward emergence of complexity.
But there's a significant distance between what's actually in the genome and the thing that we really want to understand and control. So this is a cross-section through a human torso. You can see the incredible complexity. Everything is in the right place, the right size, the right shape, next to the right neighbors. But what's actually in the genome, of course, is not anything about this. What you see in the genome is information about protein structure and some other information about when and how these proteins become expressed. Doesn't say anything directly about the size, the shape, the symmetry of the body or anything like that.
So we still have this distance that we need to bridge between what's actually the hardware specification that is in the genome and the kind of physiological software that enables cells to know what to do and when to stop. And from that, we can infer how do we convince cells to repair things that are missing or damaged. And then, of course, as I'll show you at the end of this talk more so, what can we get them to build other than the default morphology?
So one thing to realize is that the genomic information is actually really insufficient for a broad understanding of shape. So just as a simple example, axolotl larvae have little forelegs. Frog tadpoles do not. And so in our group we make something called the frogolotl, which is a kind of combination of frog and axolotl. And I could ask a simple question. We have the axolotl genome. It's been sequenced. We have the frog genome. It's been sequenced. So we have all of that. Can you tell me if a frogolotl is going to have legs or not? And the answer is, we can't. We have no idea. And we also, and if it does have legs, whether those legs will contain frog cells or whether they will be made strictly of axolotl cells, we don't know that either. Actually, to be fair, we don't, we couldn't even predict the shape of either of these things from their genome, other than comparing it to the genomes of other animals that we do know what their shape is. So our ability to predict either static shape or these kind of novel cases is actually quite poor.
And the other thing that is really not readily discernible from genomic information are the properties of dynamic robustness. So for example, this axolotl will regenerate its eyes, its jaws, its limbs, its spinal cord portions of the brain and heart. And if you amputate anywhere along this axis, the cells will grow exactly what they need and then they stop. That's the most magical thing about regeneration is that it knows when to stop. When does it stop? It stops when the correct structure has been produced. Okay? So it's a kind of... You can see already it's a kind of homeostatic process where the system can tell it's been deviated and it will do what it needs to do to get back to where it needs to go. So all of these kinds of things, the capability of these systems are not readily predictable.
Not only that, not only can these systems get back to where they need to be after injury, after some kind of external perturbation that damages a body part, but there's some really incredible problem-solving competencies within this material. So this is one of my favorites. If you make polyploid newts, meaning extra copies of the genetic material, what happens is that these kidney tubules, which usually have about eight to 10 cells that work together to leave this lumen, the cells get bigger to accommodate the new genetic material. The newt stays the same size, so it uses fewer but larger cells to do exactly the same thing until the cells get truly gigantic, and then it'll use just one cell bent around itself, okay? And give you the same structure.
Now, the ability to use different molecular mechanisms, so here's cell-to-cell communication, here's cytoskeletal bending. The ability to use different affordances in your toolkit to solve a problem you haven't seen before is basically a standard definition of intelligence. It's what's measured on all the IQ tests. What we have here is the ability of a newt when it comes into the world, it doesn't know in advance... Never mind the environment has uncertainties, but its own parts are unreliable. You don't know how many copies of your genome you're gonna have. You don't know how big or how many cells you're going to have. And you have to do the job using different molecular mechanisms, right? So different affordances from your genome in ways to solve the problem.
So again, all of this is completely not obvious from anything that we're going to get from the kind of typical molecular profiling. In fact, even below the single cell level, just the molecular pathways themselves, never mind the cells, never mind the brain, but just the molecular pathways by themselves have six different kinds of learning capacity, including Pavlovian conditioning. And so this is something else that we're doing in our group is trying to take advantage of some of these properties for things like drug conditioning and so on, the fact that the molecular pathways inside of cells can learn as well.
So all of this forms this kind of amazing multi-scale competency architecture where the material has various capabilities at different levels and all of the different levels have agendas and abilities to solve problems in different spaces, in gene expression space, in physiological space, in anatomical space.
So ultimately, what we would like to do is this. We would like to have something I call an anatomical compiler. The idea is that someday you should be able to sit in front of a computer and draw the plant, animal, organ, biobot, whatever it is, in any shape and configuration that you want, okay? And the system should be able to then, if we properly understood how the material of life works, the system would compile this into a set of stimuli that could then be given to cells to get them to build exactly this, in this case this three-dimensional flatworm, three-headed flatworm.
The key here is that this is not something like a 3D printer, which is going to basically build as if it were Legos, build the whole structure. This is actually a communications device. It's a translator from the goals of the engineer to those of the cellular collective, and if we had something like this, all of this would go away, right? We could... If we knew how to communicate novel goals to groups of cells, all of this would become a non-issue.
We're actually very far away from this. We don't have anything remotely like this, and you might wonder why, because molecular biology, genetics, biochemistry have been going very strong for many decades now. Why don't we have something like this? And partially, it's because we've been really focused on this idea of the reliability of development. So this happens all the time and it happens correctly almost all of the time, so you have these eggs and they give rise to a very specific thing, and we think we tend to think that, "Okay, well, this is what the genome encodes. This is what the genome is capable of doing."
And we're still missing the deep lessons of both neuroscience and computer science where a reprogrammable material is far more than its hardware specification, because when you have something like this, and I purposely start off with a plant example and we'll move into animals, what these cells are actually capable of is building things like this. This is called a gall. It's made of the actual cells of the plant. You would have no idea that the cells that reliably build this nice, flat, green structure, very stereotypical, we think that's what the genome does, but we have no idea that they're actually capable of building something like this, until a non-human bioengineer comes along, in this case, a wasp, that is able to prompt these cells with cues to build these incredible structures.
And so who knows what else they're capable of? Probably the latent space is extremely large. But what we do know is that more advanced engineers produce more elaborate constructions. So bacteria make these kind of featureless blobs. Fungi do something similar. Nematodes are starting to do a little better. There's some kinda non-trivial structure happening here. Mites do the same. By the time you get to insects, you get these incredible constructs. So clearly the space of possibilities is far wider than the genomic default, than the thing that we see all the time, and what we would like to do, I mean, presumably it took that wasp millions of years to get to be able to do this. We would like to do it a lot faster.
So in my group, in collaboration with Josh Bongard, we are building this kind of automated robot scientist platform that is basically making hypotheses about the laws of morphogenesis, actually providing stimuli to real cells in parallel, observing the morphogenetic outcomes, revising its hypotheses and going back. I mean, you'll recognize this as the typical cycle that we all do in science, ideally, this will go a lot faster, so that we can actually start to have some control and understanding of this process.
So where we are now is that the community is very good at this kind of information, figuring out the hardware, which proteins and RNAs interact with each other and so on, but what we would really like to do is this. And one kind of positive control that we can look at is what happened in computer science and information technology. This is what programming looked like in the 1940s and '50s. You had to physically interact with the hardware. She's sitting there rewiring the machine to get it to do something different. And this is what most of today's biomedicine and molecular medicine especially is all about. Genomic editing, pathway rewiring, protein engineering. It's all about the finer and finer control of the hardware of life.
What we really need to understand is things like this, and so this was discovered in our group by Danny and Laura Vandenberg, and this amazing thing that yes, normally you have this stereotypical change of these tadpoles become frogs and they have to rearrange their face and move their various craniofacial organs. But this is not a hardwired process, because if you scramble all of them and you make these so-called Picasso frogs where the eye's on the back of the head, the mouth is off to the side, everything is scrambled, you still get largely normal frogs, because all of these things will actually adapt to their novel starting conditions. They will move in novel paths. Sometimes they go too far and they actually have to double back a little bit. But they still get their goals met.
And so what you see is that the genetics is not giving you a piece of hardware that does the same thing every time. What it actually gives you is an error minimization scheme and a system that can recognize unexpected states and to take corrective action. And so that leads to a very obvious problem which is how does it know what the right pattern is if it's going to do the, if it's going to become a frog from this state or any of the other things I showed you? How does it know what the correct pattern, final goal is?
So what we do in our group is augment this kind of conventional open loop with this homeostatic component where actually deviation from this state, and this is something that typically open loop systems like cellular automata and so on don't do this. When you try to deviate the system from whatever that outcome was, it actually works really hard, using all the mechanisms at its disposal. So of course, the genetics, but also the physics. Works really hard to try to get back there, and this is not just injury, but it's also mutations. It's also teratogens. It's also many, many different things.
So what we would like to know then is how does this work? First of all, this is, once you add this loop, you, for the first time, this is where you, for the first time, the notion of mistakes enters the picture. Because until you have a homeostatic loop towards a specific outcome that is willing to expend energy to get back to where it was, there's no reasonable definition of what a mistake is, right? Otherwise, it just sort of rolls forward and whatever happens, happens. But now you see that any situation that pulls it away from this state, and that might be a situation that's fixable or one that's not fixable, is now you can define this idea that how hard is the system willing to try to get back, how much stress is the system under given that you've deviated it from its goal state and so on?
And so, what we need to understand now is, how does the system know what the set point is? Any homeostatic process has to have a set point with respect to which it's measuring error. And so now we need to understand, what does it actually mean to store a target state in cells, and what does it mean to try to reduce the delta from where you are now to where you are? How could cells possibly execute something like this?
Well, we have a non-controversial example of that. And that, of course, is brain-based behavior. So in the brain, what we have is a bunch of hardware which is constituted by cells in a network that communicate electrically. And on top of this hardware runs some very interesting physiology, which allows that network to do context-sensitive behaviors to store memories, and do problem solving of various degrees of complexity from simple habituation and sensitization all the way up to anticipation, planning, and many other things. And so what we know is that this system is using the remarkable properties of electrical networks and information processing in those networks to move your body through three-dimensional space. I mean, it does many other things now that we're humans and we play chess and do things like that. But fundamentally, it was, it evolved to move you through three-dimensional space.
And so we see all kinds of creatures doing all kinds of clever things, such as these crows that pick up cigarette butts and get a reward when they drop them off. And you ask, how does it know what to do? And so the commitment of neuroscience is that if we were to scan and read out this electrophysiology, we would be able to eventually decode it and extract the cognitive algorithms. We would know what is the animal thinking about, what are the memories it has, what are the goals, the preferences we would... Like, all of that cognitive stuff is encoded and implemented in the electrophysiology of the brain. And that's what these things think about.
Now, it turns out that this amazing trick is, of course, not just about brains. This is evolutionarily very ancient. It evolved around the time of bacteria and then bacterial biofilms. And if you ask what those networks think about, I think the answer is they think about the movement of your body configuration in anatomical morphospace. Basically, I think what evolution did was pivot some ancient tricks that these electrical networks were doing to navigate anatomical space, and it basically sped them up quite a bit and then pivoted them into control of motion in three-dimensional space.
And you can ask the same kind of questions. When Farinella grafted these tails to the flank, to the side of a salamander, they slowly remodeled, right? These tails slowly remodeled into limbs. And in fact, the tail tip of here, which, you know, these cells are in the correct local environment, they're sitting, they're tail tip cells sitting at the end of a tail. But they become fingers. This whole thing starts to remodel to better match the large scale pattern of what the body plan is supposed to be like.
And you can ask exactly the same question. How does it know what to do? It's basically a very parallel kind of system to what's done in neuroscience. And so we're trying to do exactly the same thing, to do this kind of neural decoding, except not in neurons. And to ask, how are the decisions, the memories, the set points and so on, how are these things encoded in the electrophysiology of somatic tissues? Okay. So same kinda question using many of the same tools.
And so here are some tools that were developed. So these are voltage sensitive reporter dyes. And so here's a movie that Danny Adams made years ago of a frog embryo during the time when all the cells are figuring out who's gonna be left, who's gonna be right, who's anterior, who's posterior. You could see all the conversations that these cells are having with each other. Here are some explanted cells. Also some explanted amphibian cells and culture making some decisions about who's going to stay within this group and who's going to leave. And you can see there's some important bio-electrical signals that Patrick McMillan in my group is analyzing. So we have the ability to use both dyes and genetically encoded reporters to read the electrical information in vivo.
We do a lot of computer simulation all the way from the molecular networks that give you those ion channels up through large scale network properties of things like pattern completion, such as people study in connectionist computer science. We try to understand how neural, how artificial neural networks encode the pattern memories and how they can be repaired and so on.
And so here are some examples of what these patterns look like in vivo. This, again, is this famous Electric Face movie that Danny made where you can see in a time lapse of this frog embryo putting its face together. Here's one snapshot from that video where you could see basically a subtle pre-pattern that tells you here's where the animal's right eye is gonna be, here's the mouth, the structures out to the sides of and so on. And not only is the bioelectrics critical for the cells in this structure to know what they're going to make. So it's a glue that coordinates the actions of individual cells towards a large scale structure within one embryo. These kind of dynamics also work across scales. So for example, here's an injury wave that propagates between individuals.
So when you... You know, when I poke this one, all of these find out about it, right? And so you can see here that it kind of, in effect... So this is Angela Tung's work who showed that groups of embryos are actually much better at resisting teratogens than singletons precisely because they communicate as a larger scale, a second-tier collective intelligence. And again, you can see in these explanted cells that Patrick made, you can see the kind of two layers. You can see the slow bioelectrics that are happening and then also here are the neurons, right? So we have the ability to observe all of this on different scales of time, on different scales of space.
But then better than just being in fact, probably more important than just being able to observe these patterns is the ability to modify them. You have to be able to do functional experiments and to insert information into the networks to know what that information is actually doing and to be able to control outcomes. So in our group, we do not use applied fields or electrodes. No electromagnetics, no waves, no frequencies. What we do is manipulate the interface that cells are normally using to hack each other. That is, all the things I'm gonna show you is not they're not happening because we're so smart. They're happening because we're simply hijacking a system that already exists by which these cells try to tell each other what to do and synthesize into a larger collective.
And so what we basically do is we can control the connectivity of the cells via targeting gap junctions or the actual voltage states of the individual cells. And we do this with optogenetics or with drugs that open and close these different channels. And so that corresponds to synaptic or intrinsic plasticity in the case of neuroscience.
So here are some things that we're doing. So in work with David Kaplan's group, we showed years ago that one thing you can do on a single cell level is control differentiation. So you could tell individual cells to be more stem-like or be differentiated depending on the voltage that you can control. But more importantly than individual cell state, because I think really the big impact of bioelectricity is going to be not at the single cell level, but actually in its collective forming properties, are the idea that when you connect cells, actually anything, but in this case, cells, into a larger network, one of the things you're doing is scaling up their cognitive light cone.
And what I mean by the cognitive light cone is simply the size of the biggest goals that they can pursue. So if we just sort of collapse space onto one axis and time onto the other, well, you could see that individual cells have little tiny goals. These are things like maintaining pH, maintaining metabolic state. They're all basically... You know, they have a little bit of anticipation potential, a little bit of memory going back, but basically, everything is in single cell organisms is concerned with maintaining the conditions at the level of a single cell.
But groups of cells such as tissues, organs, and whole embryos, they can have actually very large scale goals. They can start building things like this. So during development and during evolution, what actually happens is the, by joining together, the size of the goals that they're pursuing gets radically increased. So while individual cells only care about what's going on inside their own borders, these cells work really hard on this grandiose goal of maintaining a large structure. No individual cell knows what a finger is or how many fingers it's supposed to have, but the collective absolutely knows in a very functional sense, meaning that you can... If you know that they know this, then you will predict that once you introduce damage, it will actually build the right kind of stuff, and that that's when it will stop. Okay? So this system has a functional ability to get back to a goal state that is much, much larger than the tiny sort of scalars that individual cells pursue.
But of course, this kind of thing has a failure mode, and that failure mode is cancer. So one of the things that happens when cells electrically disconnect from the collective is that they can no longer pursue these large grandiose set points. They basically become... They basically roll back to their ancient unicellular lifestyle. And so here's a video of some glioblastoma cells that Juanita in our group studies. And if there... You know, there have been many, many papers looking at how these things are basically traveling back in their evolutionary history to occupy themselves with very small kinds of set points and basically treating the rest of the body as external environment. What's happened is the border between self and world has shrunk.
Now, this failure of this multicellularity does not require DNA damage. It doesn't require mutations. It can be triggered purely bioelectrically because these networks that keep cells harnessed towards specific goals are in large part bioelectric. And so this is some work by Maria and Doug from our group where we basically showed that if you interfere with the communication, the electrical communication between melanocytes, these little pigment cells, and another population of... A rare population of cells we call the instructor cells because they're the ones who keep these guys under control mostly. If you interfere with that, then these little melanocytes go crazy and they acquire this hyper-invasive morphology. They start to drop down and invade into the neural tube and the brain, and here in the blood vessels. You can see, I mean, this has a lot of the anatomical and molecular markers of metastatic melanoma. There is no primary tumor here. All of the melanocytes go crazy and do this, and there also is no genetic damage. There are no carcinogens here. There are no oncogenes. But they will turn on a bunch of markers that are associated with the metastatic melanoma, and that's all done just by disrupting the coordination, the electrical coordination between cells.
Better than inducing this kind of transformed behavior, you can actually suppress it, and this is something that Brooke Chernig showed in our group where you can... First of all, if you do inject human oncogene, so nasty things like K-RAL and dominant negative P53 and so on, you will get tumors. But if you also co-inject a channel that forces the cells into the appropriate bioelectrical state and keeps them coupled to their neighbors, then you won't get a tumor in a good chunk of the cases. You basically... This is the same animal, and so the oncoprotein is blazingly expressed. You can see it all over the place, right? It's marked with this fluorescence, but there's no tumor. Because it isn't the genetics that drives, it's the physiology. And once these cells are connected to their neighbors, they're basically gonna keep working on the skin and muscle and everything else that they were doing.
And so we did all this in frog. We are now moving, as Juanita showed you yesterday, we're now moving into humans, well, into mammalian cells and human cell spheroids. So this is some work that she had looking at ion channel drugs as electroceuticals to manipulate the phenotype in glioblastoma cells in vitro. And we're now moving into 3D culture as she showed you yesterday.
Okay, so we've talked about the importance of bioelectricity in maintaining multicellularity and maintaining a collective commitment to morphogenesis instead of unicellular kinds of lifestyles. And so I wanna shift now to the role of bioelectrics as a pattern memory, and for that, I wanna talk about this organism. So this is a planarian. This is a remarkable organism. Not only are they incredibly regenerative, so you can cut them into pieces, the record is something like 270, I believe. And each piece will restore the full worm. They are also extremely cancer-resistant and they don't age. Okay? They're... the asexual forms are basically immortal, and it's very strange actually, and it took us years to understand why this is happening, why the animal with the best regenerative capacity, the most cancer resistance, the least susceptibility to aging is actually the one with the extremely noisy genome, right? Because of their asexual reproduction, they have they're mixoploid, you know? Their cells have different numbers of chromosomes. It's very noisy. So why is that? Well, that's for a whole other talk. We could do a whole other talk on that.
But this is an amazing model system because every piece here has this kinda holographic memory of what the collective looks like and it can rebuild this. And so what we found was that individual pieces or whole animals have this electrical gradient that tells them one head, one tail. Okay? And so if you amputate this animal, you get one head, one tail. Remarkably, what you can do is you can manipulate that gradient using ion channel drugs and you can move it to this state that's as basically two heads. If you amputate this animal, you will get a two-headed worm. This is not AI or Photoshop. These are real animals.
Now, something very important to note is that this voltage map is not a map of this two-headed animal. This voltage map is a map of this perfectly normal-looking, so anatomically normal and molecularly normal, meaning that if you look at the markers, anterior markers only on one end, not on the other end, animal. And only when you cut this animal do you realize that it had a different idea of what a normal planarian is going to look like. In other words, the body of a normal, of an anatomically and molecularly normal planarian can store at least two, probably a lot more, but we've nailed down two, at least two representations of what to do if it gets injured at a future time. That's basically a counterfactual memory. Neuroscientists will recognize this as kind of a very early primitive form of the sort of mental time travel that brainy systems can do when they can recall or anticipate things that are not happening right now.
So this is a pattern that will sit there. It's a latent memory. Doesn't do anything until the animal gets injured and when it does, that's when it becomes relevant because that is the ground truth of what the cells consult when deciding what to build at each end. I keep calling it a memory because if you actually cut these two-headed worms, no more manipulations of any kind, just cutting them in plain water, they will, in perpetuity, continue to make two-headed worms. That's it. The that pattern is permanently changed. We've not touched the genome. There's not been any genomic editing. If you were to genomically sequence these animals, you would be none the wiser that they have a radically different body plan, architecture, behavior. Everything is different. You would not know any of that.
And so there is really important information that is stored for very long periods of time, possibly permanently, as far as we can tell it's permanent, although we do know how to set it backwards, how to set it back to normal now, that is kept in this electrophysiological layer. It is not genetics. The genetics does not tell you whether the planarian's gonna have one head or two. Gives you hardware that settles into a pattern by default that encodes one head, but it's rewritable like any good memory should be. And again, this should not be surprising to anybody in neuroscience. That is what the purpose of the nervous system is, is to be flexible and to store patterns that were not genetically set in stone at the beginning.
Okay. So given that, given that we have the ability of cells to interpret these bioelectrical patterns, because the bioelectrical pattern is saying, "Build a head here," but the other cells have to interpret it and do all of the molecular, the hard work of turning on the various genes and differentiating cells and so on, into eyes and brains and so on. Given that cells have the ability to interpret it, can we modify what they're doing?
And so this is one of our earliest applications when we used ion channel RNA. And so this is Sherry Au and ViHoff Pai's work where they introduced potassium channels into the early embryo in different regions and were able to tell the cells to build an eye. Okay? And so these eyes have all the right lens, right? And the optic nerve. They have all these things. So note a couple of important features here. First of all, this tells you that the bioelectric signals are instructive. They don't just screw up development or produce errors. They actually communicate organ level information. The other thing is that it's highly modular. In other words, we didn't tell the cells how to build an eye. We didn't talk to the stem cells. We didn't control gene expression. We have no idea how to do any of that. What we found is a very high level subroutine call that communicates the idea that this is where an eye should be. And like any good cognitive system, you can take a very simple stimulus and do all of the downstream processing, right? So hierarchical control to do all of the things needed to implement what it wants to do.
We also found something very interesting, that if you section these eyes, you can find that only a few of the cells... so here's a lens sitting out in the tail somewhere, in a tadpole, only a few of these cells were directly injected by us. What they did was instruct their
I mean, now we're getting into some smart technology that we're implanting into people, but mostly drugs are pretty dumb. They just have a target, and ideally they bind it, and that's that.
But you can imagine something like this made from your own cells, so you don't need immunosuppression. It has a billion-plus years of shared history with you, so these cells know what cancer is. They know what inflammation is, what stress is. You could imagine a million applications of using things like this to do helpful things in the body that we actually have no idea how to micromanage.
And so I suspect—and I didn't even get to talk about this; we have a lot of things going on around cell training and tissue training and so on—my suspicion is that future medicine is going to look more like a kind of somatic psychiatry than it is like chemistry, because we're going to have to really pay careful attention to what it is that cells and tissues, what inputs they received, what memories they form, what set points they are pursuing, what they are capable of pursuing, what their problem-solving competencies are. A huge chunk of that is mediated by bioelectricity.
And so I think that we have this amazing capability now in this field in particular to address this. I'm just going to say—sorry, losing my voice—that now there are some great tools coming online, in particular some AI tools and things that we're doing around creating systems to actually talk, using natural language, to gene regulatory networks, to talk to cells and tissues, to talk to organs, and using AI as a translator module to the different layers of information in the body that normally are linked by bioelectricity at different scales.
And so I'll stop here and just summarize the main points. I think that definitive regenerative medicine is going to require improved methods for communication and collaboration with the agential material of life, to recognize that we're not dealing with passive matter anymore, that we can really take advantage of some of the competencies. And for that reason, I think the relevance of bioelectricity is not just as another piece of the mechanics; it's actually a component of cognitive science and computer science that can let us reach new outcomes that we couldn't do otherwise.
I think that bioelectricity is an amazingly powerful interface to two classes of things: First, rewritable goal states, and second, the borders of the decision-making agents. It's the cognitive glue that makes individual cells into larger scale computational units. And in addition to writing these goal states, I think we can also use bioelectrics to learn a lot from the cells themselves as to how they're solving different problems. We have many examples that I haven't shown you today about how cells themselves react to and solve a bunch of problems that we actually have no idea how to do, so hopefully we can learn from them by using this interface.
And all of the things that I've shown you about the Anthrobots and the Xenobots and some others—we have some other things like that coming soon—all of those things are poised to take advantage and to benefit from everything that's being developed in this field and all the amazing advances in bioelectricity, because they allow us to now start to probe not just the stability and the robustness of life towards standard target morphologies, but actually towards the plasticity to completely new things that have never existed before, and to use that as a sandbox for complete control over growth and form in that anatomical compiler that I mentioned at the very beginning.
So okay, I'm going to stop here. I want to thank the amazing people who did all of the work that I showed you today, so here are all the post-docs, grad students, and staff scientists who did all of the things that I told you about. We have incredible technical support. Many valuable collaborators work with us on all of these things. We've had different kinds of funding over the years. Three disclosures that I have to do: morpheceuticals, AL, and Fauna Systems, and of course the most important thing to thank is the model systems, because they do really all the heavy lifting, and they're the ones who teach us all about this stuff. So, thank you so much.
