Bioelectricity, Collective Intelligence & Regeneration | Prof. Michael Levin

We don't start by asking why is there cancer. We ask why isn't it cancer all the time? Because the whole point of understanding cancer is as a disease of multicellularity. So normally every amoeba divides as fast as it possibly can. It goes wherever it wants. It doesn't coordinate with anybody else. Why does that ever change? It's amazing that our cells get together and all work on the same project. Normal amoebas don't work on the same project. They have tiny little individual goals. In an embryo, the reason you say it's one embryo as opposed to a million cells is because they all have commitment to the same goal state. So anytime you have that kind of a system, there's going to be a failure mode. The failure mode is what happens when one or more of those cells can't remember what the goal is. And the way that can happen is that it can disconnect physiologically from the other cells. So what I'm going to argue is that cancer is basically a dissociative identity disorder. It's like multiple personality, what they used to call it, because having all the cells connected in a physiological circuit gives them all a high-level unification where there's this embryo that exists over and above all the individual cells. But as soon as a cell disconnects from that mind meld, it is basically back to being an amoeba. But the cognitive light cone is defined as the scale of the biggest goal that a system can pursue. So not how far you can see, not how far you can affect things, but what is the size of the biggest goal you can pursue. So individual cells have tiny little cognitive light cones. When they merge together in embryogenesis, they have much bigger cognitive light cones, like build a limb, build an organ. When a cell disconnects electrically from the collective, that cognitive light cone shrinks. And as far as they're concerned, they're just amoebas. The rest of the body is external environment to them. They go where life is good. They migrate. They divide as much as they can, back to amoeba. Tiny little goals. We've shown that you can go in the opposite direction and actually prevent or normalize tumors in an amphibian model. Human oncogenes. We don't kill the cells with chemotherapy. We don't fix the genetic defect. We simply reconnect the cells electrically to the environment and they join the collective and keep working on nice skin, muscle, whatever they were making before. So a normalization strategy. We've done this in the amphibian model. We are now working on human spheroids. So colon cancer and also glioblastoma and some other things like that. The idea is to remind them of their goals. Their goals were always to make a nice healthy organism. Our goal is to remind them of them.

This is nice.

Yeah. I'm a professor at Tufts University. I also have an appointment at Harvard's Wyss Institute, and I run the Allen Discovery Center at Tufts. My background initially was computer science, and then a PhD in developmental biology and genetics. And my group here at Tufts works on various questions of intelligence in different embodiments, for example cells, tissues, things like that, and also various synthetic and hybrid kinds of forms. And the implications of the work become hopefully applications in regenerative medicine. So birth defects, cancer, injury, aging, those kinds of things.

Yeah. Well, first I think let's take a couple steps back because in order to understand why we're interested in bioelectricity, we need to understand a much more basic aspect of cognitive science, which is this issue of how is it that you know things that your individual cells don't know. So you are a collection, you and I are a collection of cells, neurons and other things, and you have goals, preferences, various competencies that your individual cells don't. How does that work? So what we're really talking about here is collective intelligence. We are talking about how groups do things that individuals don't know how to do. How do they do them in other spaces? For example, individual cells may have homeostatic goals in physiological space, in metabolic space. So they have tiny little goals. They're interested in, you know, tiny little sets of states that are about the size of a single cell. But a group of cells might have very grandiose goals. For example, it might be trying to build a limb or an organ or a whole body. Why do I call it goals? Because it meets the basic criterion for goal-directedness from cybernetics. If you have a salamander arm and you cut it off in the middle, the cells will very quickly grow the exact same arm and then they stop. Okay? So it's very clear if you deviate them from their set point, they will do their best to get there, like your thermostat does. They have this ability to keep a large-scale goal state and try to meet it. If a human embryo gets cut into quarters, you don't get parts of embryos, you get monozygotic quadruplets and twins and things like that. So we need to understand how the collective intelligence of cells works. That's first of all. And neuroscience is basically, I think, not the science of neurons. It's the science of cognitive glue. It's the science of how is it that electrophysiology and the mechanisms of electrical signaling allow you as a large-scale cognitive being to live in a collection of neurons. So once we understand that, we can ask the question, okay, when did this amazing trick happen evolutionarily? Where did the brain get this amazing ability to run this kind of a cognitive software on the cellular hardware that's given by the genome? Well, it turns out that it's incredibly ancient. This did not come about for the first time when neurons and muscles evolved. It's extremely ancient. So all of the things you learned about in neuroscience, including the ion channels, the electrical synapses, the neurotransmitters, the active inference, you know, all the stuff that you learn in neuroscience actually is very old. It was here since the time of bacterial biofilms and probably before that. And so this is why you need to understand bioelectricity. Bioelectricity is a very important kind of cognitive glue. It's probably not the only one, but it is an extremely important kind. We already know how it operates in the brain, at least some of the important aspects of it, and it turns out that that same cognitive glue determines what kinds of structures get built during embryogenesis, what kinds of structures get repaired during regeneration, and what goes wrong during cancer and aging. All of these things are disorders of collective intelligence. They are primarily not hardware problems. They are primarily disorders of collective intelligence. So that's why bioelectricity is important. So what is bioelectricity? Well, every cell in your body has ion channels. And so these are little proteins on the cell surface that allow certain charged molecules, potassium, sodium, chloride, to go in and out depending on the physiological gradients. And when that happens, there is a voltage potential across the membrane. And at that point that cell may communicate that voltage potential to its neighbors if it has electrical synapses, also known as gap junctions. So what you have is a network of cells that can acquire particular electrical states, and then those states can propagate through the network in complicated ways that we try to model as computations, for example. And those computations, in the case of neuroscience, they carry the cognitive content of your mind. So neuroscientists who try to do neural decoding and read your mind by looking at brain states believe that your mental states are encoded in this electrophysiology. So the exact same thing happens with the rest of the body. So if you want to understand where the shape of different organs is dictated, why you have two eyes and not seven, why the symmetry of your body is a particular way, a huge part of the answer is precisely in the bioelectric information that is processed by the groups of cells that orchestrate morphogenesis. And that's what we study when we study developmental bioelectricity.

This is a major question that neuroscientists are trying to answer too. So you have specific memories. How are your specific memories stored in the electrical activity of your brain? So this is exactly the same kind of thing. And we produced, in 2000, we produced the first molecular tools to read and write electrical memories out of non-neural tissues. So neuroscientists try to do it from the brain, we do it from the rest of the body. And so when you read these electrical patterns, one way to do it is with a voltage-sensitive fluorescent dye. So these are fluorescent molecules. You flood your embryo or, you know, whatever organ you're looking at, and what you see, with all kinds of controls that you need to do, you see an image of what the voltage distribution is like, and then from there you need to try to decode it. Some of those images look exactly like the thing they encode. For example, the electric face. There's a pattern we call the electric face because it basically tells you what the shape of the face is going to be in vertebrate embryos. The electric face pattern. It looks like a face. You can just look at it and you can see, well, here are the eyes. Here's where the mouth is going to be. Here's the, you know, the side tissues. You can see exactly, it looks like a face. Other patterns are much more indirect. So there has to be a lot of decoding to figure out what does this actually mean. So that whole task is called cracking the bioelectric code, and we are trying to do that. We only have an understanding of very little of it at the moment, but enough to make some really interesting progress that we can talk about. Some of the things we've done by rewriting, I mean, that's the other interesting part of this, is that you can read the patterns so you can know ahead of time what pattern the tissue is trying to build, but you can also rewrite the patterns. And that's the mark of a goal-directed system. When you change the goals, the system acts towards the new goals. You don't have to rewire it. You don't have to change the hardware. You just change the goals. It will do whatever. So these bioelectrical patterns, they store the set point towards which the cells are going to try to minimize error. So it's just like in your thermostat, there's an electrical memory location that stores the set point that you're trying to get to, and then everything else is relative to that set point. Same thing here. You look at the tissues and you can see, ah, I see a pattern, and this pattern says one head, one tail, or it says here's what the face looks like, or it says here's where the eye is going to go, or here's where the boundaries of a tumor are going to be, whatever. And then that's what the cells interpret as.

Well, so there's a couple of different ways. One way is that two adjacent cells can have an electrical synapse that connects them. So a gap junction is basically a pore that has, there's six proteins, let's say connexin proteins on either side, and they dock like this, and there's a direct physiological connection, aqueous. So it's water-soluble, small molecules can go from one end to the other. So charge can sort of flow through. What's interesting about gap junctions is that they are themselves, they are highly controllable. So they can open and close. They can execute pretty complex functions of being open or closed depending on the voltage on either side. So they're not just empty sort of pores. They're controllable computational elements. And that network, that allows cells to build up a network where potentially quite complicated patterns are spreading through. There are also other ways that they can communicate electrically. For example, there are ion pumps that actively emit, or in fact, in the case of transporters, can take in various ions. This can be calcium, this can be protons, this can be potassium, whatever. And cells can sense that, right? And so they can communicate either directly, cytoplasm to cytoplasm, or they can communicate electrically through the extracellular surface.

Yeah. So first of all, just the name. So what we're talking about is the process by which all of us start life as one cell. So we start life as a fertilized egg, and then eventually you end up a tree, a snake, a giraffe, a human, whatever you're going to be, or, you know, or a xenobot or something else. And so that's the process. Now those same processes, you know, start to manifest when, for example, there's an injury. So if there's an injury, you need to rebuild. Some creatures are very good. So deer will regenerate their antlers. Axolotls will regenerate most of their body organs. Planarian flatworms will regenerate every cell in their body. So this is the phenomenon we're talking about. Now I should point out that the question you just asked, how do they know what to build? Every developmental biology student asks this when the first time you see something. First of all, you see that development is reliable, that every time it works the same way to make something complex. But not only that, that's just the start. The more interesting thing is that it's highly context-sensitive and it solves problems. Meaning that if you deviate the path in a new way, you know, and developmental biologists are extremely clever in figuring out new ways to mess with the process, you will see the cells trying very hard to get back to the goal that they originally had. So that gives you a very natural question: how do they know? At that point, pretty much always in the developmental biology class, especially in a molecular developmental biology class, the instructor will say, "That's a terrible question. Don't ask how they know things. Nothing here knows anything. We're going to explain all of this in terms of a formal model of chemistry. The formal model of chemistry has the following feature: nothing knows anything. Nothing has any goals. Everything is simply following certain mechanical rules. Then we have a complexity theory which tells you that if you do that enough in sort of parallel, you have enough small units following mechanical rules, eventually something very complex and interesting might happen. That's emergence. So complexity theory and emergence, and so this is the commitment of modern approaches to morphogenesis, is that we are going to use this formalism where nothing actually knows anything, and it's all kind of an outcome of these open-loop processes that just sort of go, and whatever comes out the other end comes out the other end. I want to point out, because it's important, that this idea that we are going to do everything in terms, we're going to explain all the important features of this biology in terms of a formalism in which nothing knows anything and nothing has any goals. This is not a result. This is not something that we learned was critical. This is an axiom. It's an assumption. It's an assumption going in. We're just assuming that that's how we're going to do it. And originally that was a useful assumption because what you didn't want is a bunch of explanations where you just say, "Well, it just kind of likes to have five fingers. That's why it has five fingers." So you don't want those. Those are not useful explanations. So of course people said, "Okay, well, that doesn't help. Let's see how far we get by assuming that nothing actually, you know, has any features that come from cognitive or behavioral science." So if you say it wants, it tries, it remembers, it calculates, it compares, it decides, any of those things, those things come from the behavioral science literature. And we're going to assume that none of those things are relevant. At one point, that was okay. But my important emphasis, I think, here is that now that since the 1940s we've had the science of cybernetics, and we now know that it's not the choice between being completely dumb like rocks and being smart like humans. There's a huge spectrum in between. And the idea that you're going to do all of biology and everything that biomedicine needs at the lowest level of the cybernetic hierarchy, at the level of description where nothing knows anything and nothing has goals, this is a terrible assumption. And so my point is that this is not a result. It is not a requirement. It is an axiom. And I think we need to drop it because in the past it was considered magical. If something has goals, then it was like a magic human. At this point we understand, no, you don't need magic. Even your thermostat has little simple little goals. And now the task is clear: how many of the tools of behavioral science are applicable to different aspects of morphogenesis? Sometimes considerable, sometimes not much, sometimes it's just, you know, sort of the lowest level. But this becomes an empirical question. This is not a philosophical question. We can't make assumptions about these things. You have to do experiments. So this is what we've been doing for the last, I don't know, 25 years, is checking to see what types of cognitive frameworks are useful in developmental and regenerative biology. And so this is it. Now we've learned that in certain cases you have a system that has goals. We can now visualize the goals. We can see the encoding of the goals using the bioelectric imaging. We now see the system read the goal, try to implement the goal, and we can see it having various degrees of intelligence. And I mean that quite literally, not because it's complex, not because it's reliable, because it solves problems. Because if you try to interfere, if you try to intervene between the system and its goal, it will often do very clever things to try to resolve the problem. That is the definition of intelligence. So this is my poem about morphogenesis.

Yeah. Well, that's not a silly question at all. That's a very good question that many people are working on this question. There are sort of three basic theories about this. One is that it's a problem of physics. In other words, you accumulate errors. There's entropic decay, and over time you just can't keep up. The error correction machinery just can't keep up. I mean, entropy tells us that over time everything is just going to get worse and worse. And so, okay, so one set of theories is about it's a physics problem, it accumulates errors. There's another set of theories which says, no, it's a biology problem, meaning that evolution wants you dead at some point for some reason. There's some evolutionary benefit to having the adults go away once they've reproduced. Those are the two kinds of models. And in those cases, one of the things that we've observed is that over time the bioelectric patterns change. In fact, they degrade and they get fuzzier. And so one of the things that happens is if there is an encoded pattern that tells the cells what a correct healthy adult organism looks like, if that pattern degrades over time, for example, either as an accumulation of errors or because evolution decided not to maintain the system past reproductive age, then of course you're going to have an aging phenomenon. And that suggests that by artificially maintaining or reinforcing the bioelectric patterns, we should be able to extend longevity. And this is definitely something that we're working on. There is, however, a third potential. So this is a hypothesis on our part, which is that, yeah, there could be a physics problem, there could be a biology problem, but there also could be a cognitive problem. And the cognitive problem goes like this. I'll introduce it with a strange kind of visualization. Imagine the sort of Christian version of heaven. Okay? So imagine that you, your pet snake, and your dog end up in heaven. So in heaven, there is no entropy. There are no bacteria. Nothing is ruining your cells. Your cells are going to be fine. You're there for an infinity of time. So think about what the three of you are like over time in this system. The snake, probably fine for a trillion years doing snaky things every day, right? Probably no problem. The dog, I don't know, but you could make the argument that if it's a nice environment with, you know, a farm full of rabbits or whatever, the dog will be fine for a very long period of time. The human, though, think about it. You're there for the first 10,000 years, you sort of find things to do, maybe longer, but a couple of billion years in, are you still sane at that point? And if the answer is not, why? It's not an organic problem. There's nothing physically wrong with your brain. There is no damage. There is no injury. There's no disease. It's a different class of problem. It's a cognitive problem. It doesn't come from evolution or physics. It comes from dynamics of cognition or computation. And so we could ask, so now we know there is such a thing as problems that are not organic disease, they come from a psychological domain. So now we can ask about this in terms of aging, in terms of organismal aging. If I'm right, as I've painted morphogenesis as a system that tries to achieve anatomical goals, so it traverses the space of possible anatomical states, it works to reach the adult configuration, and then it maintains that over time, it just sort of keeps trying to reduce error. If I'm right about that and it's a cognitive system, then it should also have this kind of failure mode. The failure mode is what does a cognitive system do after it has completed its goal? It's a goal-seeking system, it's completed its goal, what happens next? Now, your thermostat is simple enough that if the hardware holds out, it will just keep doing the loop again, you know, for an unlimited amount of time. But I think what's happening is that in biology, the cellular intelligence is much, much greater, and it has metacognitive loops on top of that basic homeostasis that are disruptive to holding on to a single goal for too long. And the reason I say that is for two reasons. One is we did some simulations of cells minimizing error to achieve morphogenesis. So you see the quality gets better and better. They build the embryo. Then they sort of hold on to it. The error is constant for some amount of time. That's adulthood. And all of a sudden they kind of self-destruct. And we were looking into this and like, why are they all of a sudden self-destructing? And we're still investigating it because in our model we don't have that kind of damage. We don't have a special pressure to die, an evolutionary pressure to die. There's some other reason why it's happening, and we're still investigating it. But my suspicion is that it's because you have a fairly sophisticated goal-seeking system that has met its goal. We call it, for that reason, we call it the boredom theory of aging. You know, and we all know the phenomenon when somebody retires, if they don't have anything to do, you know what... So there's something else that's interesting that we tested, a prediction of this model. Because in the model, what happens is the cells originally, to make a good body, the cells have to be in coordination. They all have to agree on what they're doing. In fact, the whole embryo is basically just an alignment of goals, just a bunch of cells that are all in agreement that this is what we're making. So we went and tested this with this question of what happens during old age due to this alignment, and we found out something quite interesting. And this is the work of Leo Pio-Lopez in my group. And this we call this phylostratigraphic dissociation. What we found is that if you look at, so there's a technique called phylostratigraphy, which looks at the genes you express and it tries to estimate the evolutionary age of those genes. Okay, how old on the evolutionary tree, how old are the genes that you're expressing? So what we found is that in a young tissue, and this isn't every tissue in the body, but some percentage of tissues, in a young tissue, in a young person, all of the cells in that tissue are in agreement about where you are on the evolutionary tree of life. In other words, all the cells are expressing the genes that are very recent, you know, sort of modern Homo sapiens kinds of genes. When you look at old tissue, you start to see a much greater divergence. Cells start to express genes that are ancient and very, very different. So your individual cells no longer agree with each other on where on the tree of life you should be. And they start expressing ancient genes, you know, from our evolutionary past. And so you will see exactly this during old age. There's this dissociation where I think what's happening is that the goal state of upkeeping an organism is no longer very compelling to the cells, and they all start to regress. They go, they sort of get pulled backwards on the evolutionary tree because it's just not as compelling for them to work towards this. It's a kind of a reversal of multicellularity, basically.