Will This Revolutionize Biology?
An interview with Michael Levin.
Michael Levin is an American developmental and synthetic biologist at Tufts University—Levin directs the Levin Lab, directs the Allen Discovery Center, directs the Tufts Center for Regenerative and Developmental Biology, and co-directs the Institute for Computationally Designed Organisms.
You hear so much overheated talk about “revolution” when it comes to science and technology, but I think that Levin’s work actually merits the term—take a look at this video to get a sense of how important Levin’s work really is:
Make sure to check out these articles:
“Acute multidrug delivery via a wearable bioreactor facilitates long-term limb regeneration and functional recovery in adult Xenopus laevis” (26 January 2022)
“Scientists Regrow Frog’s Lost Leg” (26 January 2022)
And this fascinating article:
And everyone should also make sure to check out the various audio interviews and various videos on the Levin Lab website.
See below my interview with Levin that I edited for flow, organized by topic, and added hyperlinks to.
I hope you believe in my Substack’s potential—Noam Chomsky made this comment about my Substack:
Andrew Van Wagner has rapidly turned his Substack project into a major source for those who want to understand crucial current issues in the human sciences, broadly understood, and the ways they bear on social and political life. He has engaged some of the leading and most original minds, with probing questions informed by his careful research. It is a truly impressive achievement.
Make sure to do a free signup for my Substack—I want to thrive on Substack so that I don’t have to go back to my less exciting retail job:
Remember:
it’s free
you won’t get any spam
you’ll get an email notice whenever I publish a piece
So make sure to do a free signup in order to help me out!
Exciting Projects
1) What are the most exciting projects that you’re currently working on?
There are a few—in the lab, we’ve already used computer modeling and ion channel modulation to do the following in frog in vivo:
(1) prevent and normalize cancer
(2) repair birth defects of the face, brain, gut, and heart
(3) induce limb regeneration
And we’re now trying to use similar approaches to:
(A) normalize glioblastoma in human cells in vitro
(B) induce limb regeneration in mice in vivo
We’re also working to understand Xenobots’ cognitive capacities—their preferences, memories, and behavioral responses. Xenobots are novel proto-organisms made of frog skin cells.
And we’re also trying to figure out how to program Xenobots for specific structures and functions.
In terms of theory, our most exciting direction is that we’re trying to understand morphogenesis—the construction and repair of complex anatomy—as a behavior of collective intelligence on the part of cell groups whose proto-cognitive functions and memories and goals are represented through non-neural bioelectrical circuits.
And we’re doing projects where we’re trying to take what we’ve learned about pre-neural biological intelligence and use that knowledge to build better AI:
2) What are the most exciting projects that you know of that others are working on?
One of my webpages lists a lot of the most interesting work that I know of.
Let me specifically mention a few things.
Oded Rechavi’s work on transgenerational inheritance:
Richard Watson’s work on evolution as a form of learning:
“How Can Evolution Learn?” (2015)
Jitka Čejková’s and Martin Hanczyc’s work on aqueous droplets as forms of minimal agency:
Ezequiel Di Paolo’s and Randy Beer’s work on the dynamics of basal cognition:
Ricard Solé’s synthetic biology:
There are many more things to list, but those are the ones that come to mind.
Gallistel’s Quest
1) What do you think about Randy Gallistel’s quest to find the engram?
The quest to find the engram is one of the most fundamental and significant projects in all of science.
And I’m very sympathetic to Randy’s view—my views are perhaps even weirder than his, so I find nothing taboo about the idea that learning takes place inside cells.
Randy has suggested some potential mechanisms—like RNA—for memory and computation, and I’d add the cytoskeleton as a potential medium for memory and computation:
“Cytoskeletal Signaling” (2012)
You don’t need traditional synapses for learning. We see memory in bacteria, yeast, slime molds, and other single-celled life forms—life was remembering things long before brains appeared.
And the problem isn’t finding molecular biologists—there’s no magic here that requires a special type of advanced person, and you could just recruit a postdoc with basic skills in cell biology and molecular biology.
The basic issue is instead that it’s simply a lot of work to figure out how cells store information—and learn, and compute—at the molecular level.
There are actually a bunch of people working on this in the basal cognition field, but neuroscientists largely ignore this work. And I think that the people doing this work in non-neural somatic cells like skin cells—and in basal free-living cells like bacteria, yeast, slime molds, and protozoa—will solve this problem and that the same mechanism will then be discovered in neurons later on.
2) What do you think about what David Poeppel told me regarding Gallistel’s project?
I agree with this comment from Poeppel:
I should actually call Gallistel and chat with him about this—my own hunch is that he should talk to people in computational biology or synthetic biology, since he could ask them: “What would be a simulated experiment or a set of steps that would go in this direction? And once the simulation points to some of these intermediate things, we’ll know what a specific experiment would be.”
That’s exactly right.
And Gallistel should also talk to the people in basal cognition who have been thinking about this problem a lot—for example, Gallistel should talk to Audrey Dussutour about her work on memory in slime molds:
There are lots of people researching memory systems that don’t involve synapses—and researching ways to make novel memory systems from biological parts—so Gallistel should talk to all of these researchers about their work.
Anti-Mainstream Views
1) Which views do you have that are even more anti-mainstream than Randy’s views?
Take a look at this:
Here are some views that I have that most scientists probably wouldn’t agree with:
cognition is a continuum that stretches all the way across the tree of life and into some very unconventional substrates
we’ll need to fundamentally rethink ethics once we recognize cognition’s extent
anthropomorphism is a prescientific notion that tacitly assumes that humans have some sort of magic that shouldn’t be bestowed—even to a small degree—on other systems
there aren’t any binary distinctions for a majority of interesting concepts, and it’s deeply erroneous to accept dichotomies like “cognitive” vs. not; “true preferences” vs. “just physics”; “robot/machine” vs. “living organism”; and so on
these dichotomies that I just mentioned aren’t long for this world, given advances in chimeric technologies and in bioengineering
morphogenesis is the behavior of a collective intelligence in anatomical morphospace
there are many somatic intelligences that operate in novel hard-for-us-to-recognize spaces like physiological space and transcriptional space and anatomical morphospace
a mature science of biology will have teleonomy—goal-directedness—as a central concept that will help us understand what Selves are
molecular medicine that focuses on the molecular level is just one piece of the puzzle that leaves a lot on the table
true regenerative medicine will only be possible when we understand that there’s no privileged level of explanation and that some things are best achieved top-down through training systems like tissues and organs, and motivating these systems, and communicating with these systems
evolution doesn’t find solutions that are adapted to specific problems and environments, but instead finds ways to encode machines—of various degrees of somatic intelligence—that solve problems in all sorts of new environments
future progress in regeneration and bioengineering will crucially depend on understanding how evolution generalizes—there are immediate payoffs whenever evolution optimizes a solution that’s adapted to a single specific environment, and yet evolution somehow (1) creates cellular hardware that can solve novel problems and (2) creates cellular hardware that can survive when the microenvironment or the external environment presents radically new circumstances
DNA isn’t the cell’s software—DNA encodes cellular hardware, and the software operates through bioelectric and biochemical and biomechanical dynamics
the software is much more powerful than DNA and is deeply reprogrammable
genomic editing and stem cell biology aren’t a universal solution to most of the things that we care about in biomedicine
genomic editing and stem cell biology will hit a ceiling that will require us to import deep ideas from computer science and cognitive science in order to really control biology
neuroscience isn’t about “neurons”—which is an unclear term—any more than computer science is about your laptop
neuroscience has deep insights that apply to cells throughout the body—evolution was using these deep principles to produce intelligent behavior in anatomical morphospace long before brains ever appeared
2) Which experiments support your own anti-mainstream views?
There are many relevant experiments from my own lab and from other labs, some of which are covered in these reviews:
“Re-membering the body” (2015)
“Bioelectric signaling” (2021)
And take a look at this video that shows a lot of the data on bioelectrical reprogramming of (1) cancer and (2) embryogenesis and (3) regeneration:
3) What explains the resistance to your own anti-mainstream views?
The resistance probably comes from the same factors that produce resistance to any significant change to the intellectual status quo—it’s uncomfortable to have to rethink things that we thought we knew, things seem easier when you think in sharp categories, and it’s difficult to integrate ideas for the first time across very different disciplines.
And in biology, there’s an old idea that everything has to be explained at the level of chemistry.
But I’m not complaining—I’m actually pretty happy with the traction that my ideas and data have gotten over the years, and the normal process of science includes initial resistance to new work.
Problems and Mysteries and Challenges
1) What are the biggest problems and mysteries and challenges in your field that you want to solve?
We want to crack the bioelectric code: How do bioelectric patterns within tissue encode solutions to morphogenetic problems? We already know the molecular mechanisms through which voltage levels control gene expression in single cells, but we want to answer the higher-level question of how multicellular patterns encode concepts like “Make an eye here”—this higher-level problem is the somatic equivalent of cognitive neuroscience.
And we want to solve the problem of scaling Selves: How do large-scale emergent goals arise when you join collections of competent subunits in specific ways?
And we also want to solve the problem of how evolution gives rise to problem-solving machines and pivots these strategies toward successful behavior in different problem spaces that these machines have never encountered before.
2) With unlimited money, which experiments would you do to move the needle on these problems and mysteries and challenges?
We need to improve our technology if we want to be able to read and write bioelectric states in tissues—my lab has developed some tools, but my lab’s tools are just the tip of the iceberg.
We need much better technology for getting physiological data in vivo, and much better ways to assemble—in the way that neuroscientists now do for brain and behavior—comprehensive datasets on different organisms’ bioelectrical states when organisms are in various states like developmental states and regeneration and cancer and aging.
We also want to move from existing models that evolved naturally to synthetic organisms like our Xenobots that were created with chimeric and bioengineering technologies. We want to develop 100s of unique organism types that are largely free from evolutionary lineages’ baggage and from evolutionary lineages’ frozen accidents—these unique organism types will be novel in morphological and in behavioral and in other ways.
We won’t be able to explain these unique organism types’ novel traits with the idea that evolutionary pressure acted over eons and eons to select for fitness for specific environments. So a large-scale research program that develops these synthetic organisms will push us to develop theories of emergent morphology and emergent behavior and emergent goals—these theories won’t lean on evolution and will give us a much better understanding of life’s plasticity and life’s intelligence; novelty’s origins; the evolutionary process’s intelligence; how the genome relates to form and function; and so on.
An Electrical Language
1) In your TED2020 conversation “The electrical blueprints that orchestrate life”, you describe an electrical language via which cells communicate with one another—what exactly sends out these electrical signals, through what medium do these electrical signals travel, and what exactly receives these electrical signals?
Take almost any neuroscience paper about electricity—replace “millisecond” with “minute” or “hour”, replace “neuron” with “cell”, and everything usually holds. The mechanisms are the same—the only difference is the space in which the mechanisms solve problems.
Changes in cells’ resting potentials carry the signals—ion channels produce these changes and gap junctions propagate these changes.
These changes move through all cells and tissues. And then other cells and tissues receive the signals. So it’s like what you see in the brain—bioelectric states change and these changes propagate through tissue.
Evolution uses this process of electrical communication as a kind of computation in order to regulate morphogenetic change.
Neurons process electrical signals in order to control muscle movement in 3-D space, but all cells process electrical signals in a network in order to control cell behavior in order to move the organism’s shape through anatomical morphospace—the state space of all possible anatomical configurations.
2) How exactly is information encoded into these electrical signals and how exactly is information decoded from these electrical signals?
We’d be in a far better place than we are if we knew all the answers—we have a lot of work to do.
The key is to recognize that “How exactly” doesn’t just mean “at the molecular level”. It also means the large-scale question—which is exactly parallel to neuroscience’s neural decoding problem—of how a particular bioelectric pattern signifies “eye” vs. “hand” or how a particular bioelectric pattern defines the number of heads in an organism:
“Re-membering the body” (2015)
We’ve been able to decode anatomical specification in specific cases—like head number, head shape, and head layout—regarding the frog embryo and the planarian:
“Bioelectric signaling” (2021)
So we’re starting to build up the bioelectrical dictionary for anatomy, but there’s a lot of work ahead.
3) What exactly do these electrical signals instruct cells to do and how exactly do cells turn these electrically transmitted instructions into actions?
We now know at some level of detail how bioelectric networks’ outputs are transduced into:
second messenger cascades like calcium signaling
changes in neurotransmitter distribution in neurons and beyond
changes in morphogen distribution
changes in gene expression
We know that morphogenesis results when bioelectrical guidance tells cell groups to divide, differentiate, turn genes on and off, and migrate. The issue is that we need to better understand the large-scale anatomical decisions that result from all of this activity—we need to better understand the computations and the information processing.
Life’s Complexity
1) To what extent do we understand how each cell in the body knows what cell it’s supposed to be? How does a cell know it’s supposed to be a tooth cell or an eye cell or a brain cell?
Traditional molecular development biology has made huge strides on this, so there’s a lot of information on this at the individual cellular level.
But the deepest problems aren’t to do with what you’re asking—stem cell biology has proceeded well and has made tissue types, but the large-scale organization is missing, so the much more interesting thing is how large-scale decisions are made.
A teratoma has skin and hair and teeth and muscle and so on, so it has many of the right cell types, but it lacks the normal 3-D arrangement.
And a salamander limb can be cut off and then the cells will rapidly build a new limb and then stop rebuilding—how do those cells know when to stop rebuilding and how does the system determine that a “correct” salamander arm has been completed?
And a tadpole rearranges its face to become a frog face, but when we make tadpoles with scrambled faces where everything is in the wrong position then you still end up with a pretty normal frog face because the organs move around in novel paths until they get to the right place and then they stop—how do the organs know how to do that, and how did evolution create such a clever error-minimizing system that can handle novelty instead of the easier hardwired system that always moves everything in the right direction the right amount?
And there’s a phenomenon where a deer’s antler rack will be damaged in a certain spot and then after the antler rack has been shed and regrown an abnormality will appear in the spot where the damage happened—what possible model could you construct that would explain (1) how cells at the deer’s scalp could remember where in the antler rack’s 3-D structure damage had occurred and (2) how these cells at the deer’s scalp could store that information for months and (3) how these cells at the deer’s scalp could then use that information to guide bone-branching decisions months later?
These examples of plasticity and anatomical decision-making are more profound, more difficult, and—ultimately—more important than the cell-level questions of how cells know what cell type to become. And these more profound questions can’t be answered with the same conceptual tools or lab tools.
2) Is there any interesting literature on how a human face turns out to look the way that it looks? I guess that that’s not all in the genome—I’m not sure how different identical twins’ faces are at birth and how much identical twins’ faces diverge throughout life.
My own lab doesn’t work on this, but other labs do—here’s a review that addresses some of the various extra-genomic factors like physical forces and uterine position that affect embryos:
3) How do the brain cells know when to “turn left and stop growing in the previous direction”? Look at what David Poeppel told me about this:
There’s a remarkable genetic instruction set that builds our brains—language riles people up, but it’s a very simple and important fact that the brain has six layers, and so as the brain forms, the cells have to grow and grow and grow, and then there’s a genetic command that says “You’ve reached your target!” and then the cells turn left and stop growing in the previous direction. How’s that supposed to work?
The question here is more general: How do organs and tissues know to do anything at all?
We know that it’s not just “in the genes”—genomes encode protein-level hardware, but they don’t encode an actual blueprint of the resulting anatomy or functionality. Frog cells that have a perfectly good frog genome in them can make Xenobots and can maybe make other types of organisms too. And you can revise a planarian’s bioelectric pattern memory so that the planarian—even though it has a normal genome—will make a line of permanently two-headed planaria.
So “How does the thing know to do X?” is the number one question in many domains, not just in the brain domain.
But brain-wise, we’ve recently discovered the following in the frog model:
there’s a bioelectric prepattern that guides overall brain size and overall brain shape
this prepattern extends throughout the body—belly cells determine through their electrical activity the brain’s size and shape
So it’s not just local environment that matters—cells communicate in order to decide what to do.
4) One person told me that we are “still a good way off from understanding complex gene regulation” and that you can see our ignorance in Figure 4 of this 2017 paper—the person gave this explanation:
This is one of the best-studied pathways in all of plant biology, and they are targeting the expression of a gene that we understand the expression pattern of deeply. The expression of CLAVATA3 is very tightly regulated in a plant meristem (it controls differentiation in stem cells), and it is conserved all the way back to the most basal species. That being said, we have no idea what is going on in this figure. They throw out some theories and identify some obvious patterns, but really it remains a big mystery. And let me emphasize this is 2000 bp of a well-studied organism in one of the most studied pathways in plant biology.
Gene regulation is important, but it’s not the part of the puzzle that we most care about—the issue isn’t the mechanism that turns specific genes on and off, but instead the issue is how the collective knows which genes to turn on and off in order to solve familiar problems and new problems.
5) What’s exciting and interesting about each of the papers below?
“Re-membering the body” (2015)
“Life, death, and self” (2020)
The first paper explains the idea that all cells—not just neurons—form decision-making tissues that have memories in anatomical space and goals in anatomical space.
The paper shows that you can import from neuroscience various concepts and tools in order to solve problems in regenerative medicine, and the paper discusses how evolution discovered bioelectricity as a convenient way to compute and implement memory long before neurons showed up.
I propose in this paper that every level of behavioral science—not just the synaptic machinery and the neurotransmitters, but all the way up to psychiatry—is relevant for regenerative medicine and relevant for understanding where bodies come from.
The second paper reviews remarkable data showing that memories can survive drastic neurological remodeling—memories can survive the transition from caterpillar to butterfly.
And memories can even regenerate after you amputate trained planaria’s whole heads.
It’s exciting to see this memory stability because:
there’s a design challenge—we have no hardware in our technology that will somehow continue to store memory even throughout the process of totally disassembling and reassembling the storage medium
more broadly, this means that we understand very little about how memory is actually stored—it clearly can’t all happen in the fine-tuning of specific synapses because some memories will remain even after the whole thing has been taken apart and removed and rebuilt
In the third paper, I try to abstract from our familiar categories of intact “natural” organisms—I show how deep our knowledge gaps are, how plastic biology is, and how the obvious default outcomes that we usually see aren’t the whole story.
The paper discusses several corner cases in biology that exhibit weird biology and fluid boundaries. For example, the paper discusses the process of individual cells assembling into familiar or novel organisms after the “individual” disbands—you can easily imagine a scenario where an organism like a fish dies and then many of its cells transition to an amoeba state where they live on by themselves, and later these same cells might reassemble into yet another living construct like a Xenobot.
6) What in an organism’s genome determines whether that organism will be the size of a mouse or the size of a blue whale?
It’s tricky to ask whether something—for example, a genome—“determines” something else.
Does the structure of silicon and germanium in your calculator’s transistors “determine” what the programmable calculator is going to do? Yes, but that’s only part of the story.
The genome specifies the microscale hardware structure of organisms’ cells, but once you’ve made a specific machine then the genome can harness really interesting laws of physics and computation in order to implement that machine’s behavior—we really need to understand those laws and how different kinds of hardware can manifest those laws.
7) What in an organism’s genome determines what shape the organism will be?
Once again, it’s tricky to ask about “in the genome”.
Two-headed planaria make two-headed planaria when cut—you couldn’t have guessed their head number from the genome because they have a normal genome.
Through bioelectric changes, you can induce metastatic melanoma in tadpoles that have no oncogenic mutations and you can prevent cancer in tadpoles that have human oncogenes.
And Xenobots have a normal frog genome.
So in all of those cases, the genomic sequence doesn’t tell you what shapes will be formed.
And more broadly, imagine a “frogolotl” embryo made of 50% frog cells and 50% axolotl cells—a frog larva won’t have legs and an axolotl larva will have legs, but will the frogolotl larva have legs? And if it’s all about the genome then why can’t you look at the genome in order to answer the question?
And what if we mix cells from planarian species that have different head shapes—what head shape will the cells make?
So it’s not about the genome per se—it’s about how collectives of cells make decisions about what they will build. And so it’s useful—but not sufficient—to know about the genomically specified hardware.
8) What does your work say about how the brain works? Does the brain’s computational capacity depend mostly on how the brain is wired or mostly on individual cells’ information-processing capacities?
I haven’t studied that question with respect to the brain specifically.
But I think it’s both.
Half the magic is in the cells’ competency—understanding minds requires us to pay attention to the individual cells’ goals and how interactions with neighboring cells shape the goals.
And half the magic is in the laws of physics and computation that are exploited when you join cells in specific ways—emergent Selves appear when you scale up tiny cell-level goals into bigger ones.
I think it’s not so much about how the brain is wired. I suspect that you could implement specific cognitive capacities in many—but not just any—different kinds of wiring.
9) Are there any parallels between what this 2018 article says about trees and what your own work says about cells?
The main parallel is in the idea that biology is intelligent at many scales and in many different kinds of media—intelligence isn’t just about brains, and many biological systems solve problems at different scales, and there’s competition and cooperation among all of the different overlapping Selves that comprise any biological system:
“Cognitive all the way down” (2020)
“Scale-Free Biology” (2020)
“Minimal physicalism as a scale-free substrate for cognition and consciousness” (2021)
Medicine
1) What work do you know of that’s most medically interesting?
I’d have to say the work on cancer normalization from Mary Hendrix, Thea Tlsty, Mina Bissell, and others—this paper shows that you can correct cancerous cells’ behavior and that cancerous cells aren’t irrevocably broken:
We’ve had data for a long time showing that specific environments—embryonic and regenerative—can normalize cancer cells.
And my own lab has done some normalization with bioelectric cues:
But the existing work is a drop in the bucket—we need to make a concerted effort to understand how signaling from normal cell networks can shift these defecting cancer cells’ goals away from the amoeba-scale goals that these cells have reverted to and back up to the goals of building tissues and organs:
2) What did you think about the quote below from the 2015 documentary Cancer? The following quote from the documentary gives a sense of how powerful cancer is and also gives a sense of how we might use something equally powerful to defeat cancer:
No matter how powerful or precise, however, combination targeted therapy still must confront cancer’s complexity and its relentless drive to adapt and survive.
“To me the best way to think about cancer is that it is literally evolution in a bottle. It’s like taking the enormity of the power of three billion years—unimaginable timeframes that have created the unimaginable diversity of life. Imagine capturing all of those forces within a single cell and putting that inside someone’s body. That is to me the metaphor of cancer—it is all of the history of life [playing] out at a billion times the speed of evolution.”
If cancer exploits the power of evolution to survive, perhaps only a commensurate weapon—equally adaptable and also perfected over millions of years—can overcome it. That weapon—many scientists believe—is the human immune system.
Cancer is multicellularity’s inevitable price.
There are processes that connect individual cells into networks that have larger-scale goals like making organs, but sometimes these processes break down and cells revert to the ancient unicellular past’s small and simple goals of (1) reproducing and (2) migrating out to wherever life is good. Cancerous cells treat the rest of the body as “outside environment”—their cognitive boundary has collapsed to the one-cell scale.
The immune system is a great way to fight cancer.
And another way to fight cancer is to reconnect cancerous cells with the electrochemical network that previously harnessed them to perform useful functions within the body collective.
3) Why do biologists need to figure out ways to induce the human body to regenerate limbs and hunt down cancer cells and so on—shouldn’t humans have naturally acquired the ability to do all of these things?
There’s no “should” in evolution—evolution doesn’t find perfect solutions, doesn’t optimize function, and doesn’t optimize for happiness.
Evolution just produces things that survive—in whatever capacity possible.
Can humans survive without very desirable regenerative capacities? Apparently yes, so here we are.
The notion “Don’t mess with nature” makes no sense—nature results from a randomly driven hill climbing process and doesn’t necessarily optimize for any of the things that we value. So we can—and in medicine, we ethically must—do better than what nature has given us.
Great article that raises some fascinating questions.