🦋 How Life Remembers: From Metamorphosis to Simulation

📖 Read: 

Memory is not metadata. It is material. It lives somewhere. And finding where it lives, what survives and what gets erased and why, may be the most important question biology has ever asked.

The chrysalis, it turns out, is not an ending. It is a test. What is truly essential enough to survive the dissolution? What, in the end, does life consider worth remembering?

Perhaps the question applies to more than butterflies.

Genius 10 Year Old’s Research Shocks Scientists Around the World
Bringing the genetically minimal cell to life on a computer in 4D
A Drosophila computational brain model reveals sensorimotor processing

and 20 other papers for context

This is Heliox: Where Evidence Meets Empathy

Independent, moderated, timely, deep, gentle, clinical, global, and community conversations about things that matter.  Breathe Easy, we go deep and lightly surface the big ideas.

Transcript

0:25

Imagine, for just a moment, that you are a caterpillar. Okay, I am picturing it. Lots of legs, very hungry. Exactly. You spend your days, you know, lazily munching on leaves in a garden, just living the dream. Sounds pretty peaceful, honestly. Right, but then biological instinct takes over, you wrap yourself up in a chrysalis, and things get incredibly weird. Yeah, that is definitely an understatement. Because your entire body literally dissolves. It turns into this chaotic biological soup. You are, for all intents and purposes, fundamentally dismantled. Right, the neural pathways that made up your tiny caterpillar brain, they just turned to liquid. Yes. And then, somehow, you rebuild yourself from that puddle of goo into a butterfly. Now, as you flutter away, here's the question for you to ponder. Do you remember the leaves you ate? Wow. I mean, did that memory just melt away with the physical structure of your brain? Exactly. And taking it one massive mind-bending step further, what if somehow your grandchildren remember those leaves, even though they were never caterpillars in that same garden? Which is just, you know, a staggering thought experiment. It forces us to ask questions. what a memory actually is at its absolute core. I mean, if it's not a fixed physical structure, if the hardware can literally be melted down to liquid and rebuilt, where does the software go? Right, because we tend to think of memory as like a file saved on a hard drive. But metamorphosis suggests the file might somehow survive the total destruction of the drive itself. Which breaks all the rules of how we think biology works. It really does. So, welcome to this deep dive. Today, our mission is to track down the physical footprint of memory. We are exploring the profound mystery of biological information transfer. Essentially, how does life remember? Exactly. And we are going to chase this question from the messy, heavily contested wetware of biology all the way into the pristine architecture of digital simulation. Because, as you will see, finding the physical hard drive of memory in living creatures is proving to be so overwhelmingly difficult that scientists are literally having to digitize life itself just to figure out how structure creates behavior. It is a wild ride. And the stack of sources we are pulling from today is fascinating. It's incredibly diverse. We are starting with a groundbreaking documentary about a 10-year-old entomologist in Japan, Joe Nagai. His backyard experiments are completely challenging our fundamental assumptions about inheritance. Just brilliant stuff. Oh, it's amazing. And from there, we will wade into the fiercely debated academic literature on transgenerational epigenetic inheritance. We'll be looking closely at critical reviews by leading geneticists like Adrian Bird and Edith Hurd. Right, because we need to look at what the scientific establishment thinks it understands and frankly, what it is desperately trying to figure out. It's a very heated debate. Very heated. And when the biology gets too opaque, we are going to pivot to two staggering computational These are the ones that really blew my mind. Same here. The first details the mapping of an entire Drosophila fruit fly brain down to the last of its 50 million synapses. 50 million? It's hard to even wrap your head around that number. And the second brings a minimal living cell to life in a four-dimensional computer simulation. We're going to look at the organizations behind these monumental efforts and ask how these models apply to the future of AI. And open source interdependent agents, embodied robotics, all of it. Yeah, it is part storytelling, part rigorous scientific exploration full of dead ends, fierce academic battles, and just the sheer delight of discovering how the machinery of life actually operates. And I really think the best place to start looking for this physical footprint of memory is with someone who approached the problem without any of the entrenched academic baggage. Yes. Let's talk about Joe. Let me introduce you to Joe Nagai. He is a 10 year old from Kobe, Japan. And from kindergarten onward, Joe exhibited the kind of observational discipline that you usually only see in a seasoned researcher. He wasn't just, you know, running around catching bugs with a net. No, not at all. He was running endurance and behavioral trials. By first grade, he was meticulously documenting the molting patterns of caterpillars. he was timing the survival rates of butterflies stuck inside their chrysalises. Which is just incredible dedication for a first grader. Right. And that observational rigor led to a specific anomaly that caught his attention. He noticed that the swallowtail butterflies he raised from eggs inside his house would flutter toward him when they emerged. Okay, so they seemed friendly. Yeah, but whereas wild swallowtails they would just scatter immediately. Oh, interesting. So they treated him differently. Exactly. And his 10-year-old brain formulated this highly sophisticated hypothesis. He thought, maybe the memory of me, or at least a positive association with my presence, actually survived the biological soup of the chrysalis. And to understand why that is such a bold hypothesis, you have to consider the prevailing scientific consensus for decades. Oh, completely. Entomologists largely believed that complete metamorphosis, what they call hollow metabolist development, effectively resets the insect's central nervous system. Right. The slate is wiped clean. Exactly. The larval brain is almost entirely broken down through programmed cell death, which is called apoptosis, and a brand new adult brain is constructed from these isolated clusters of cells called neuroblasts. It's like taking a computer motherboard, melting it down into liquid metal, building a completely new machine from that puddle, and then expecting your old Word documents to still be sitting there on the desktop. That is the perfect analogy. How does the software survive the melting of the hardware? It shouldn't be possible, but we have this 10-year-old suspecting that memories survive apoptosis. So he digs into the literature. As 10-year-olds do. Right. And he finds a study by an entomologist named Martha Weiss at Georgetown University. Now she'd actually managed to prove that moths could retain associative memories from their caterpillar stage. Okay, so there is precedent. Yes. And Joe decides he wants to adapt her highly complex behavioral study for his swallowtail butterflies. So he writes her a letter. A four-page, handwritten letter mailed all the way from Japan to the United States, detailing his years of observational data. That is just so charming. And she actually writes back, right? She does. Weiss receives this letter, and very graciously... suggests he might want to try a simpler project. Which is fair. I mean, it's a lot for a kid. It is. Replicating her protocol requires strict environmental controls, behavioral conditioning, rigorous statistical analysis. It's university-level science. But I love his response to her because he basically writes back saying, thanks, but I've already built a lab in my bedroom. I've adapted your protocols. And actually, I have some critiques regarding your methodology. Which is just amazing. A 10 year old critiquing a Georgetown professor. And he specifically challenged her use of ethyl acetate as the scent the caterpillars were supposed to smell during the conditioning phase. Why was that a problem? Well, his critique was incredibly astute. You see, ethyl acetate is frequently used as a killing agent in entomology jars. Oh, wow. So it's literally toxic to them. Exactly. It triggers pain and survival pathways in the insect's nervous system. It smells like nail polish remover. And Joe recognized that immediately. He reasoned that using a toxic smell completely skews the behavioral assay. Right, because if you want to test pure associative memory, you need a neutral stimulus. If you use a poison, you aren't testing memory. You are testing a biological baseline aversion. You are measuring their instinct to not die rather than their ability to remember a conditioned response. Exactly. It's brilliant experimental design. So he swapped the ethyl acetate for lavender oil. Because swallowtail caterpillars have no natural evolutionary relationship with lavender. Right. It's a totally neutral scent to them. And this shift in the experimental design, born entirely out of Joe's empathy and intuition, isolated the memory variable far better than the original university-level protocol. It is just stunning. So to build that associative memory, he needed to pair the lavender scent with a negative stimulus. He used an electric shock. Yeah, and standard practice in many labs is to just administer a blanket voltage that guarantees a response. Blast them with electricity so you know they feel it. But Joe didn't want to do that. He wanted to find the absolute minimum threshold of discomfort. His shock device had 15 intensity levels. And he didn't just test it on the bugs? No. No. He applied the electrode pads to his own arm to feel the voltage himself. That is true dedication to the scientific method. He was conducting a meticulous titration protocol. He started at level one and worked his way up until he observed a clear physiological response from the caterpillars. And what was that response? Well, at level four, the caterpillars deployed their osmoterium. Ah, okay. Let's break down the biology of that for the listener. The osmaterium is this fleshy, forked organ hidden inside the caterpillar's thorax. It looks a bit like a snake's tongue. Yeah, exactly. And when a predator like a wasp or an ant attacks, lymphatic pressure forces the organ to literally pop out and it releases a foul-smelling chemical cocktail of terpenes. Or a definitive binary indicator of acute stress. stress exactly by identifying that level 4 voltage trigger the osmaterium Joe knew he had established a negative stimulus without causing unnecessary tissue damage right so he sets up the experiment he has a control group that received no training and an experimental group that was shocked while smelling lavender and then they grow up yes both groups entered the chrysalis underwent apoptosis total melting of the brain, and emerged as adult butterflies. And this is where the Y maze comes in. Exactly. He built the Y maze, literally a tube that splits into two paths. Both paths contained sugar water, but one path included a cotton pad soaked in lavender oil. Okay, the moment of truth. The control group went in and split evenly down the middle, 50-50. They showed no baseline preference or aversion to the lavender. Which proves the lavender really is neutral. Right, but the trained butterflies, the ones that were shocked as caterpillars, they showed a 70% avoidance rate for the lavender path. That is incredible. The statistical significance of that 70% avoidance rate is what made Joe Nagai the first person in the world to prove that Swallowtail butterflies retain associative memories through the complete morphological destruction of metamorphosis. A 10-year-old kid in his bedroom lab did that. That alone is a stunning achievement. But the investigation into the physicality of memory doesn't stop with the butterfly's own lifespan. No, it goes much deeper. Think about your own family history for a second. Have you ever wondered if, you know, an irrational fear or a specific anxiety you hold was somehow inherited from your grandparents? It's something a lot of people wonder about. Well, Joe had a similar thought. He noticed that he, his mother, and his grandmother all suffered from hay fever, a shared physical sensitivity. And that sparked a massive question for him. Right. He wondered, if traits are passed down in humans, could my butterflies pass down their newly learned fear of lavender to their offspring? And this hypothesis crosses the threshold into one of the most hotly debated frontiers in modern biology, transgenerational epigenetic inheritance, or TEI. TEI. Can an acquired environmentally induced trait be passed down through the germline to future generations? So what does Joe do? Joe takes the trained butterflies, the ones that were shocked as caterpillars. and he breeds them. He raises the F1 generation. The children. Yes. And crucially, this new generation never experiences an electric shock. They are entirely naive to the training. They've never even smelled lavender before. Exactly. He puts them in the Y maze and they avoid the lavender path at almost the exact same 70% ratio as their parents. Wow. And he didn't stop there. No, he breeds the F1 generation to create the F2 generation, the grandchildren of the original trained caterpillars. Wow. Yeah. The aversion holds. It presented particularly strongly in the male offspring. Joe presented these findings at the 2024 International Congress of Entomology, and it completely disrupted the room. I can imagine. To understand why scientists were so shocked, we need to define the layers of biological memory here. Because when we talk about inheritance, we are usually talking about the DNA sequence itself, right? The actual genes. Right. The A-T-C-G letters of the code. Exactly. But Joe's caterpillars didn't magically evolve a new gene for lavender fear in a single generation. The DNA sequence didn't change at all. So what did change? We are talking about epigenetics. These are the chemical modifications that sit on top of the DNA and dictate how it is read. And we really have to separate this into three distinct categories to understand what's happening. Okay, let's break them down. First, you have mitotic memory. This is simply cell-to-cell inheritance within a single organism. So like when liver cell divides. Exactly. The daughter cell remembers to be a liver cell because epigenetic tags like methyl groups physically block the cellular transcription machinery from reading the genes that would turn it into a skin cell or a bone cell. Makes sense. What's the second layer? Then you have transcriptional priming. This is when an environmental stressor like a severe drought or a famine alters those epigenetic tags during an organism's lifetime. It changes how its genes are expressed to help it survive in that new environment. Okay, so a responsive change. So if DNA is the alphabet, epigenetics is the punctuation and the highlighting. That is a brilliant way to put it. So what we're asking is, if a parent highlights a chapter in a book because of a traumatic event, is the child born with the book already highlighted? Yes, and that is the third category. True meiotic transgenerational memory. This is the Holy Grail. This is when that altered epigenetic state, the memory of the stressor, is packaged into the reproductive cells, the sperm or the egg, and passed down to the F1, F2, and F3 generations. But I am stuck on the mechanics of that packaging process. I mean, how does a neurological fear of lavender in the brain literally get written into the sperm or egg of a butterfly? The brain and the gonads are completely different physiological systems. And you have hit on the core mechanical problem of TEI. That is the million-dollar question. To understand the leading theory on how this works, we have to look at another organism, the microscopic nematode worm C. elegans. Okay, moving from butterflies to worms. Right. Researchers have demonstrated that if a worm ingests a toxic pathogen, like Pseudomonas aeruginosa, it learns to avoid it. And incredibly, it passes that avoidance behavior down for four generations. Four generations from one exposure. Yes. And the mechanism they discovered here relies on small RNAs. Wait, hold on. If the worm eats the pathogen, the sensory recognition happens in its gut and its rudimentary nervous system, right? Correct. So how does an RNA molecule travel from the gut, survive the worm's internal environment, and penetrate the reproductive organs without being degraded? Cells have natural nucleases that destroy rogue RNA. It is a remarkable piece of biological logistics. The worm utilizes something called systemic RNA interference. There are specific transport proteins, notably the SID1 transmembrane channels, that actively ferry these small double-stranded RNAs across cell boundaries. So it's like a dedicated highway system just for this information. Exactly. In the case of the pathogen avoidance, a specific small RNA called P11 is generated in the nervous system or the gut. It travels through those SID1 channels straight into the germline. And once it's in the reproductive cells, what does it do? It triggers an amplification loop. An amplification loop. So it doesn't just passively sit there waiting to be inherited. No, it actively recruits an enzyme called RNA-dependent RNA polymerase. Right. This enzyme uses the p11 RNA as a template to manufacture thousands of secondary short interfering RNAs, or CERNANS. That's making copies of the warning signal. Thousands of them. And these secondary CERNANS then bind to the worm's DNA and recruit proteins that deposit suppressive histone modifications. So the environmental experience in the gut is translated into a mobile RNA signal, transported to the gonads, amplified, and permanently alters the epigenetic landscape of the offspring. That is mind-blowing. So in a worm, or presumably a butterfly, we have a mapped physical highway for this memory transfer? We do. But this is where the biological wetware gets incredibly messy. because when we shift the focus to humans and mammals in general, we hit a massive, seemingly insurmountable biological roadblock. Ah, yes. The bottleneck problem. Bottleneck. Tell us about the bottleneck. This is the major battleground for geneticists right now. In mammals, There are two distinct global waves of epigenetic erasure. Erasure, like wiping a hard drive. Exactly like wiping a hard drive. The first wave occurs when the primordial germ cells are forming in the developing fetus. Almost all of the epigenetic marks, specifically the DNA methylation that silences genes, are actively stripped away. And just to be completely clear on the mechanics of methylation for the listener, a methyl group is a tiny hydrocarbon molecule. It physically attaches to the cytosine base of the DNA. Right. It acts like a molecular speed bump. Exactly. Physically preventing the RNA polymerase from latching on and reading that gene. Yes. And during germ cell formation, enzymes actively remove those speed bumps. The slate is wiped nearly clean. That's wave one. Then a second wave of erasure happens immediately after fertilization. When the sperm and egg fuse, the resulting zygote undergoes another massive remodeling of its chromatin structure. So a double wipe. Yes. This double erasure is nature's mechanism to ensure the early embryo is cotipotent. Cotipotent, meaning it has the potential to differentiate into any cell type in the body, completely unconstrained by the epigenetic history of the parent somatic cells. Right. It needs to be a blank slate so it can become a liver cell or a brain cell or a skin cell. The biology actively fights to destroy the physical footprint of the parents memories. Which brings us to a severe contradiction. I mean, here's where it gets really interesting. If the slate is wiped completely clean twice, how do we explain all those viral news articles about humans inheriting the trauma of their ancestors? It's a paradox. And to explore it, let's look at the famous Dias and Ressler experiment at Emory University. This is the study everyone cites. They exposed male mice to acetophenone, an odor that resembles cherry blossoms, while simultaneously administering mild foot shocks. Okay, so very similar to Joe's caterpillars, just with cherry blossoms and mice, they learned to fear the scent. Exactly. They then bred these trained males with naive females. The resulting F1 offspring and eventually the F2 grandchildren exhibited an intense fear response to the smell of acetophenone. Even though they had never met their fathers and had never been shocked themselves. But wait, if the epigenetic marks were erased during fertilization because of the bottleneck, how did the fear survive? Did they find a physical mechanism? They found compelling physical evidence. They observed targeted hypomethylation, meaning a distinct lack of those silencing methyl speed bumps, specifically on the OLF1-5-1 gene in the father's sperm. And what does that gene do? It codes for the exact olfactory receptor tuned to acetopheno. Furthermore, the offspring developed physically enlarged olfactory bulbs in their brains, specifically dedicated to processing that exact scent. I mean, that seems like an airtight case for mammalian epigenetic inheritance. You have the environmental stressor, you have the molecular change in the sperm, and you have the neuroanatomical change in the offspring. Why is this fiercely debated? Because the replication of these results across other labs has been highly inconsistent. And more importantly, the specific molecular bridge across the bottleneck, how exactly that one gene escapes the double erasure, remains totally undiscovered. Ah, so we see the start and the end, but the middle is missing. Exactly. And this is where critical perspectives from heavyweights like Adrian Bird and Edith Hurd come into play. They review this field and they argue that we are conflating rare anomalies with systemic biological rules. For instance, we often hear the agouti mouse study cited as the ultimate proof of mammalian TEI, right? Oh, constantly. This is the experiment where a mother's diet rich in methyl donors like folic acid permanently changed the coat color and the obesity risk of her offspring. Right. But Bird and Herd point out that the agouti locus is a highly unusual genetic architecture. It involves an intrasisternal A particle. Which is a retrotranspose and essentially a piece of ancient viral DNA that inserted itself upstream of the agouti gene. Exactly. And this specific transposable element happens to be somewhat resistant to the global erasure process. It's a quirk. Because when researchers attempt to map this dietary inheritance across the rest of the standard genome, the effect largely vanishes. So the Ooty mouse is a fascinating anomaly of a viral insert, not a universal paradigm for how memory escapes the bottleneck. That is the argument, yes. And the debate becomes infinitely more sensitive and the methodology far more problematic when we move from mice to humans. It really does. Because we are looking at studies involving the descendants of the Holocaust or the survivors of the Dutch hunger winter and the Swedish famines. We obviously approach these histories with immense empathy for the human suffering involved. But from an impartial, strict scientific methodology standpoint, what are the critics finding when they look at the data from these human studies? They are finding profound methodological vulnerabilities. Let's examine the prominent Holocaust survivor study led by Rachel Yehuda. This study suggested that trauma altered the methylation of the FKBP5 gene, which is a regulator of the stress hormone cortisol, in both survivors and their children. Okay, so biological proof of inherited trauma? That was the claim. But the primary critique from geneticists is the sample size. The initial highly publicized findings were based on a cohort of only 32 individuals. Wow. 32. I mean, a sample size of 32 is virtually anecdotal when you are trying to establish a universal rule of human inheritance, especially given the sheer immense scale of the human genome. Compounding that small sample size is the statistical issue of multiple testing. In genome-wide methylation studies, you are analyzing hundreds of thousands of potential methylation sites. Right. If you run enough statistical tests on a tiny sample, mathematical probability dictates that you will eventually find a correlation purely by random chance. Unless you apply incredibly stringent statistical corrections, you risk identifying noise as a signal. And beyond the math, there is the massive confounding variable of simply being raised by a human being. How do you separate true molecular inheritance in the DNA from cultural, behavioral, or even microbiome transmission? The Leningrad siege studies illustrate this perfectly. Researchers study the descendants of those who survived the brutal starvation of the siege, looking for inherited metabolic changes. And what did they find? They found that the offspring exhibited distinct altered eating patterns. Specifically, a significantly higher consumption of red meat and calorie-dense foods compared to control populations. But is that a metabolic shift encoded in the mesylation of their DNA, or is that simply the psychological result of being raised by parents who survived a famine? Exactly. A family that endured starvation is going to inherently teach its children to consume high-calorie foods whenever they are available. The anxiety around scarcity is passed down through conversation, through behavior, through parenting, not necessarily through the germline. Exactly. It's the nurture versus nature debate all over again. And Adrian Bird raises a brilliant evolutionary paradox regarding this entire field of human transgenerational trauma. Oh, the evolutionary argument. Let's unpack this. In Darwinian terms. The C. elegans worm inheriting an RNA-based immunity to a toxic pathogen makes perfect sense. It is a clear adaptive advantage. Right. It helps the grandchildren survive. But why would a complex mammal evolve a mechanism to inherit a susceptibility to disease, trauma, or obesity? Inheriting your grandfather's PTSD does not make you more fit for survival. It makes you maladaptive. It is carrying baggage that only hinders your survival in a new environment. If the tiger is gone, you shouldn't still be jumping at shadows just because your grandfather did. Exactly. And we are discovering that mammalian biology actually deploys active countermeasures to prevent this. Research has identified specific safeguard proteins, such as DPPA2 in mice. DPPA2, what does it do? The primary function of DPPA2 is to bind to the promoter regions of developmental genes in the early embryo and actively recruit chromatin remodelers to establish a naive, totipaned epigenetic state. So it aggressively strips away any acquired marks from the parents? Yes. Life at the mammalian level invests massive biochemical energy into forgetting. So if the biological wetware is this fiercely guarded and the physical footprint of memory keeps getting erased by proteins like DPPA2, Yes. how do we ever conclusively map how structure dictates behavior? Well, we have to step back. If we cannot find the physical hard drive of generational memory because the biology keeps wiping it clean, We have to find a biological hard drive that we can map entirely. We have to digitize life. Which brings us to the computational pivot of this deep dive. If biology is too messy, we map it digitally. Yes, and this brings us to the monumental achievements in our source stack. A massive consortium of researchers, Specifically, the Flywire Consortium, bringing together scientists from Princeton, Cambridge, and dozens of other institutions, decided to map the complete physical structure of a single brain. The first complete connectome of the adult Drosophila melanogaster, the fruit fly. We're talking about mapping over 125,000 neurons. and physically tracing 50 million individual synaptic connections using high-resolution electron microscopy. Just the sheer scale of the data. They sliced a fly brain into thousands of microscopic sections, imaged them, and used machine learning to trace the wiring diagram. But a map is just a static photograph. To prove that physical structure actually dictates behavior, they had to make the digital map think. And to do that, they utilized a leaky integrate and fire model, or LIF, to simulate the flow of electrical activity through the connectome. Right. Let's visualize how the LIF model works for the listener because it's a brilliant concept. Imagine a bucket with a small hole in the bottom sitting under a dripping faucet. Okay, I'm with you. The bucket represents a single neuron, and the water dripping from the faucet represents the electrical potential coming from upstream synapses. Now, if the water drips from the faucet fast enough, the bucket fills up faster than it leaks out the hole, right? Right. Eventually, it tips over and spills. That spill is the action potential. That is the neuron firing. Got it. But if the faucet drips too slowly, the water just leaks out the bottom, the bucket never reaches the tipping point, and nothing happens. The neuron stays quiet. That bucket logic was applied to all 125,000 neurons in the digital model. But they didn't just use standard buckets. They weighted every single connection based on the physical number of synapses between the neurons in the electron microscopy data. So a stronger connection means a wider faucet pouring more water. Exactly. Furthermore, they utilized machine learning to predict the specific neurotransmitters at every single synapse. They figured out whether the connection was excitatory pouring water into the bucket or inhibitory. Inhibitory meaning it's effectively drilling a larger hole in the bottom of the bucket to prevent it from firing. Exactly. But I am curious how a model of this rigid accounts for neuromodulators. I mean, a bucket model works perfectly for a direct synaptic connection. But what about neuropeptides that bathe entire regions of the brain and change the baseline firing threshold of thousands of neurons simultaneously? That is a phenomenal limitation to point out. The current LAF model primarily simulates fast, direct synaptic transmission. It does not fully capture this slower volumetric release of neuropeptides. It's a work in progress. It is. Yet even without those broad chemical modulators, the sheer predictive power of the structural wiring alone was staggering. Their goal was to predict sensorimotor transformations. Meaning how a sensory input physically routes through the brain to create an action. Right. How a sensory input, like the taste of sugar on the fly's leg, physically routes through the 50 million synapses to trigger a specific motor output, like extending its proboscis to feed.- So they literally stimulated the specific digital neurons corresponding to the fly's sweet taste receptors, and they just watched the digital electricity cascade through the network.- And the model accurately predicted the activation of the exact motor neurons responsible for feeding behavior.- It's amazing.- The journey of analyzing this cascade yielded profound insights into how the brain contextualizes information. For example, they trace the pathways for sugar and water, discovering that while they start at entirely different receptors, they converge to form a shared attractive pathway deep in the brain. But the discovery that fascinated me the most involved a specific set of neurons called IR94E. Oh yes, this is a great example. Previously, biological studies indicated that these neurons were part of an attractive, positive behavioral circuit. In female flies, they light up when presented with male sensory cues. You would logically assume that activating them triggers a positive receptive behavior across the board. But the connectome revealed the danger of looking at neurons in isolation. Because when they activated the IR94E network in the digital simulation, they discovered that it actually strongly inhibits the proboscis extension. It acts as an aversive pathway specifically for feeding. The same neuron can be a green light in the context of mating, but a hard red light in the context of eating. Exactly. The physical wiring diagram proves that context is entirely structural. And this structural determinism is a massive paradigm shift, especially when we consider its implications for artificial intelligence. Huge implications! Because in standard deep learning neural networks, like the ones powering modern large language models, you have millions or billions of free parameters. Right. The system learns a task through massive statistical brute force, adjusting those parameters over countless training iterations until it produces the right output. But the Flywire model didn't do that at all. They didn't train the digital fly on what sugar was. They didn't give it a data set of feeding behaviors to mimic. There was zero training. Zero. The model achieved over 84 percent accuracy in predicting live biological behavior based entirely on its hardwired physical structure. The structure is the function. Yes. This proves that complex, accurate decision-making behavior can emerge purely from the physical wiring of a network interacting with an environmental input. So we have proven that the physical wiring of a brain dictates complex behavior. But as we know, a brain does not exist in a vacuum. It requires a body. It requires metabolism. Right. If we want to truly map the physical footprint of life... Can we simulate not just the electrical firing of neurons, but the actual chemical physical reality of an entire living organism over its lifespan?

And this brings us to the final and arguably most mind-bending paper in our stack: 32:08

the simulation of the JCVI-SYN3A minimal cell. The J. Craig Venter Institute has been working for years to strip down a bacterial genome to the absolute minimum number of genes required for life and reproduction. And they engineered a cell that has only 493 genes. Which is incredibly tiny. A team of computational biologists then took that minimal genetic blueprint and achieved something unprecedented. They simulated the entire 100-minute life cycle of that cell in four dimensions. Three dimensions of space plus the dimension of time. Exactly. But I have to challenge the methodology here. I mean, we know the chemical formulas for cellular metabolism, right? We know the equations for transcription and translation. We do. So why go to the immense computational expense of simulating three-dimensional space? Why not just put all the concentrations into a massive digital spreadsheet, run the differential equations, and calculate the output? Because, to borrow a phrase from the researchers, biology is not a well-stirred soup. That is the perfect way to phrase it. Imagine the inside of a cell is like a crowded, low-gravity Tokyo subway station during rush hour. Okay, I like this. If you have a passenger, let's say an RNA polymerase, who needs to meet a specific person, a DNA promoter region, on the other side of the station, they don't just magically teleport. Right, they have to physically push through millions of other commuters. Exactly. They're being violently jostled by water molecules, bouncing off ribosomes and navigating a viscous, highly structured environment. Space matters. It matters immensely. Spatial heterogeneity dictates reaction rates. If the polymerase is in the top left corner of the cell and the gene is in the bottom right, the chemical reaction cannot happen until they physically travel and collide. A spreadsheet assumes instantaneous reaction. It assumes teleportation. Yes. But the 4D simulation forces the chemistry to obey the actual laws of physics. So to achieve this, they used a software platform called Lattice Microbes powered by massive arrays of GPU clusters. They divided the digital cell into a three-dimensional grid comprised of individual cubes measuring exactly 10 nanometers across. And 10 nanometers is roughly the size of a single ribosome. Right. Within this grid, they had to solve a massive computational bottleneck regarding the speed of reactions. They utilized a hybrid stochastic deterministic methodology. Let's break that down for the listener. Deterministic chemistry is the spread. Yes. For highly abundant molecules and fast metabolic reactions, they use deterministic differential equations because simulating millions of fast moving water molecules individually would crash any supercomputer on Earth. Obviously. But for the rare, slow and critical events like a single RNA polymerase finding a single gene, they use stochastic article by particle simulation. They calculated the random Brownian motion of individual molecules moving from one 10 nanometer cube to the next over time. So they tracked ribosomes physically translating proteins. They simulated the insertion of individual lipid molecules into the cell membrane, watching the physical structure of the cell grow spherically from a 200 nanometer radius to a 250 nanometer radius over the 100-minute cycle. And those spatial constraints forced them to confront biological mysteries they hadn't anticipated at all. Oh, like what? Well, for instance, when the minimal cell divides, the newly replicated chromosomes have to physically separate into the two daughter cells. Now, in wild bacteria, complex protein machineries handle this, but those genes were stripped out of the Syn3a minimal genome. Ah. If the machinery is gone, how does the DNA separate in this simulation? It didn't. It didn't. No. In the 3D space, the massive strands of DNA literally tangled up and physically blocked the cell from dividing. Oh, wow. The simulation revealed a critical gap in our biological understanding. So, to allow the simulation to actually complete the cell cycle, the programmers had to artificially code a 12-pi-knewton spatial repulsion vector. Essentially a Leonard Jones potential to physically push the chromosomes apart. Yes. They had to code a digital tractor beam into the physics engine because the spatial reality proved that our chemical understanding of the cell was incomplete. That is just incredible. And this is why simulating life in 4D is so crucial for the evolution of technology, particularly in the realm of robotics and embodied artificial agents. organizations developing open source robotics like the teams behind open claw our co-work they are moving away from disembodied AI because a large language model doesn't know what a body is exactly but if we want an artificial agent to truly learn and interact with a physical environment it has to understand embodiment it has to understand that moving an arm takes time expense energy and occupies physical space Just like an RNA polymerase navigating the crowded subway station of a cell. The 4D minimal cell and the 50 million Synax fly connectome provide the foundational blueprints for how complex coordinated behavior emerges, not from abstract code, but from slow moving, spatially separated physical components. They teach us how to build interdependent agents whose intelligence is inextricably linked to their physical structure. It really is an incredible journey. We started with a brilliant 10-year-old boy in Japan, using empathy and lavender oil to prove that a caterpillar's memory can survive the biological meltdown of metamorphosis. And then we navigated the fiercely guarded bottleneck of mammalian epigenetics, where the biology actively fights to erase the trauma of the past, challenging whether our anxieties are truly inherited or simply learned. And we ended inside the glowing GPU arrays, mapping the exact structural wiring of a fly's decision. and tracking the physical collision of molecules inside a living digital cell. The unifying thread through all of this, from the Y-maze to the connectome, is the realization that memory is always a physical footprint. It really is. The boundary between an individual and its environment, and even the boundary between a parent and its offspring, is constantly blurring. The physical structure of the past dictates the behavior of the future. Whether we are looking at the transportation of small RNAs in a nematode worm, the hypermethylation of olfactory genes in a mouse, or the specific synaptic weights in a leaky integrate and fire model, we are witnessing the mechanics of how experience becomes architected. It's the translation of life into structure. Which leaves you with one final provocative thought to mull over as we sign off. If we are now capable of mapping the precise biological wiring of memory in insects, and we can simulate the exact physical life cycle of a cell in four dimensions, how long until we can artificially engineer an ancestral memory into a synthetic organism? That is a terrifying and brilliant question. If mammalian biology fights so aggressively to wipe the slate clean between generations generations to protect us from the baggage of the past, maybe our machines and our digital simulations will be the only entities capable of truly holding onto a perfect, unbroken, transgenerational memory. Maybe the next chrysalis isn't biological at all. Thank you for joining us on this deep dive.