Xenobots: frog cells that became a new kind of machine

Most of this journal is about rewriting DNA. Xenobots are the opposite dare: leave the genome alone and rewrite the body plan. Michael Levin's group at Tufts, with Josh Bongard's evolutionary robotics laboratory at the University of Vermont and Douglas Blackiston's microsurgery, treated frog cells as a material you could shape. An algorithm proposed millimetre-scale bodies that would move if cardiac muscle — that's heart muscle — twitched in the right place. People then assembled those shapes under a microscope, by hand. The things walked. They survived for weeks in saline. They closed lacerations. They were named after Xenopus, the African clawed frog, because English didn't have a comfortable word for a living millimetre that isn't a tadpole. We've tried other names. This one stuck, and it's fine. The 2020 paper is a pipeline: a computer evolves a shape, a microsurgeon builds it, and you compare the predicted walk with the one in the dish. That's why this object belongs next to CRISPR and organoids on the same shelf.

In short. They left frog DNA alone and changed the body shape instead. A computer suggested millimetre blobs that should crawl; people built them by hand; the blobs walked.

Kriegman, Blackiston, Levin and Bongard, Proceedings of the National Academy of Sciences, 13 January 2020, 117: 1853–1859, is the first paper you want. Title: A scalable pipeline for designing reconfigurable organisms. Co-first authors Kriegman and Blackiston. The significance paragraph said out loud what the field had been circling: computers automatically design new machines in simulation, and the best designs are then built by combining different biological tissues. Skin and heart cells from Xenopus laevis embryos, sculpted into millimetre-scale walkers whose geometry had been evolved, not inherited. The algorithm ran for months on the Deep Green supercomputer cluster at UVM's Vermont Advanced Computing Core, generating thousands of candidate morphologies, rerun ninety-nine times from different random populations. Performant designs were filtered for robustness to random phase modulation of their contractile cells, then manufactured. The Cozzarelli Prize in Engineering and Applied Sciences followed, which is a way of saying the Academy liked the methods, not just the nickname. We've read that paper more times than is strictly professional. The later two papers did not make it less strange.

In short. The first paper, in 2020, used a computer to evolve walker shapes and then built those shapes from frog skin and heart muscle. The DNA was not rewritten.

What walked in the dish was a few thousand embryonic cells, a body smaller than a millimetre, no neurons, no gut, no reproductive tract, powered by yolk platelets the oocyte had already loaded. Cardiac muscle provided the twitch. Epidermis provided a passive chassis and a sealed surface. The organism pushed salt crystals, walked in circles, recovered from being cut, and eventually ran out of yolk and fell apart into dead skin. That last clause is a safety feature as much as a limit, and the authors said so. A tadpole, given the same genome, would have built a nervous system, a heart in the right place, a mouth, a tail, and a feeding programme. These cells were given a different anatomical prompt and built a walker instead. The genome underdetermined the machine. That sentence is the conceptual payload, and it's why a science section is allowed to spend this many words on a frog blob. De-extinction rewrites the hardware. Xenobots rewrite the chassis. Both were true in the same decade, which is a slightly thrilling fact once you've sat with it.

In short. The walkers were tiny, had no brain, lived on egg yolk, and fell apart when the yolk ran out. The same genes would normally have built a tadpole.

An evolutionary algorithm, in Bongard's sense, isn't a catchphrase about artificial intelligence. It's a population of candidate bodies, a physics engine, a fitness function, and a lot of computer time. Each candidate was a voxelated soft body: a three-dimensional grid of little cubes, some passive, some contractile, the contractile ones firing in a cycle. The simulator, in the lineage of Voxelyze and the soft-robot papers Cheney, Lipson and Bongard had already run on silicone-like materials, asked how far the thing would travel. Selection kept the travellers. Mutation and crossover produced the next generation. Months on Deep Green. Ninety-nine independent runs, because a single run is an anecdote about a random seed. Diversity of performant designs was the point of the census: not one heroic shape, a family of shapes that moved. Transferability was the second filter. A design that only works if every cardiomyocyte fires in a perfect simulated phase is a design that will fail in a dish, where cells are noisy. They jittered the phase on purpose and kept the shapes that still walked. That's engineering, written as an algorithm, and it's more honest than a hand-sculpted mascot.

In short. A computer bred thousands of blob shapes in a physics game, kept the ones that travelled farthest, and threw out designs that only worked if every heart cell fired on cue.

The wet half of the pipeline is a construction kit, not a printer. Blackiston took blastula-stage Xenopus embryos — that's the hollow ball of cells just after fertilisation — specifically the animal cap, and microsurgically sorted two progenitor populations: epidermal cells that would make a sealed, pigmented skin, and cardiomyocyte progenitors that would make twitching muscle. Those tissues were layered and carved under a microscope into the evolved silhouette: a few hundred micrometres to a millimetre, holes and joints where the algorithm had put them, cardiac tissue where the algorithm had asked for an actuator. No scaffold. No three-dimensional print of a hydrogel. Living cells, placed by hand, healing into a single epithelium, then differentiating. Manufacturing, in this paper, is a person at a scope with forceps and a map from a supercomputer. That's slower than a headline about robots designing robots, and it's the actual method. Scalability, in the title, means the algorithm can propose many shapes and a microsurgeon can attempt the transferable ones, not that a factory is running. We'll keep the word pipeline and not the word factory. A pipeline with a human in the middle is still a pipeline.

In short. A surgeon cut frog-embryo skin and heart muscle and assembled them by hand into the computer's shape. No 3D printer. Living cells, placed, then left to heal together.

Two tissues, two jobs, and the jobs aren't symmetrical. Epidermis is the body: cadherin-mediated adhesion — cadherins are the stickiness proteins that hold sheets of cells together — a sealed outer surface, pigment that makes the thing visible, a mechanical stiffness the contractile tissue can push against. Cardiomyocytes are the motor: spontaneous electrical activity, calcium transients, sarcomeres that shorten, a twitch you can see. Put the motor in the middle of a symmetric blob and you get a pulse. Put it off-centre, or as a hinge between two passive lobes, and you get a gait. The algorithm discovered hinges. Blackiston built hinges. The dish confirmed that a hinge made of frog heart, attached to frog skin, will walk on a plastic surface in saline. That confirmation is the paper's actual result, and it's smaller and better than scientists create living robots. A predicted behaviour transferred across a sim-to-real gap that usually eats soft-robot papers. The gap here wasn't silicone versus simulation. It was a voxel versus a cell that has its own programme, its own adhesion, its own tendency to round up. Transfer still happened. That's why the pipeline is a method, not a gallery of renders.

In short. Skin made the body. Heart muscle made the motor. Off-centre muscle made a walk rather than a throb. The dish behaved as the simulation had guessed.

Xenopus laevis is the chassis for a reason a developmental biologist already knows and a roboticist had to learn. The embryos are large, they develop in a dish, the fate map is a century old, the animal cap can be explanted without killing a mammal, and the cells are packed with yolk so they don't have to feed. Nieuwkoop and Faber staged this species so that a methods line can say stage 9 and another laboratory can find the same tissue. Animal-cap explants have been rounding up into epidermis since people were asking questions about induction. Blackiston didn't invent the explant. He pointed it at an evolved silhouette and added a second tissue the algorithm had requested. Marc's modified Ringer's, 0.75×, pH 7.8, 14 °C for some of the culture, 20 °C to heal: those are the numbers in the later methods, and they're amphibian numbers, not mammalian incubator numbers. Chassis choice is a methods decision. It's also why this object is a millimetre of amphibian, not a millimetre of person, and why the 2023 human-cell cousins are a different paper. We like the frog for being the right tool, not for being a mascot.

In short. African clawed-frog eggs are large, well mapped, and already full of food. That is why the first living machines were frog, not mouse or human.

Predicted versus observed is the test any sim-to-real paper has to sit. The 2020 figures show designed morphologies next to the manufactured organism and a trace of displacement over minutes. Some designs walked more or less as advertised. Some designs were less transferable: a simulated hinge that a pile of cardiomyocytes didn't quite build, a hole that healed shut, a gait that became a shuffle. The authors reported the transfer, not only the successes, which is why the paper is still readable. Fitness in silico was displacement. Fitness in the dish was also displacement, plus the unsimulated facts of healing, adhesion and a cell that would rather be an epithelium than a voxel. An evolutionary algorithm that cannot survive contact with those facts is a graphics card warming a room. This one survived contact often enough to be a method. We'd rather a method with a messy transfer plot than a render with a confident caption. The later ciliated spheroids, which needed less carving, were in part a response to how expensive a perfect silhouette is to manufacture by hand.

In short. Some computer shapes walked in real life as promised. Some did not, because cells heal holes and prefer to be skin. The paper reported both.

A cardiomyocyte is a muscle cell of the heart, and in a tadpole it would have joined a tube that pumps blood. Here it's an actuator parked in a novel geometry. The contractile apparatus is still a sarcomere: actin thin filaments, myosin II thick filaments, tropomyosin, troponin, a calcium switch. Depolarisation opens calcium channels; calcium binds troponin C; the tropomyosin shift exposes myosin-binding sites on actin; the cross-bridge cycle shortens the sarcomere; the tissue twitches. Spontaneous pacemaking in embryonic Xenopus cardiomyocytes means you don't have to wire a stimulator. The cells beat because that's what they do in a dish, and in a frog, and in a xenobot. Frequency is a function of temperature, of coupling, of how large a syncytium you built. Coupling is gap junctions, connexins, ions and second messengers sharing a neighbourhood so that a field of cells can twitch as a field. Put that field on one side of a passive lobe and the lobe walks. The motor wasn't invented. The address of the motor was. They relocated a muscle the embryo was already going to make. That's a ruder, better sentence than they made muscle from scratch.

In short. Heart-muscle cells already twitch. In a xenobot they are simply placed off-centre so the twitch becomes a walk. The motor was moved, not invented.

Geometry is the gait. A symmetric spheroid with muscle in the core pulses. A two-lobed shape with muscle at the join flexes and inches. A hole through the body, which several of the evolved designs carried, changes stiffness and can let the thing straddle a particle. The 2020 organisms pushed salt, walked in arcs, and sometimes cooperated by accident of crowding. None of that required a nervous system. A twitch plus an asymmetric body plus friction against plastic is a sufficient machine at this Reynolds number — water feels thick, inertia is small. It's a rude sentence if you were hoping for a tiny frog with intentions. It's a precise sentence if you're trying to say what was actually demonstrated. Control theory would like a tunable oscillator and a sensor. Version 1.0 has an oscillator — the cardiomyocyte — and almost no sensor beyond whatever stretch and ion neighbourhood the tissue already had. Version 2.0, the ciliated spheroids, traded the cardiac oscillator for a ciliary one and still didn't grow a nerve. The later photoconvertible-protein experiment is a writable memory, not a nervous system. Hold the distinction. A machine can locomote without representing the world. These ones do.

In short. Shape decides how a twitch becomes movement. A lopsided body walks; a round one just pulses. No brain is required for that.

They healed. A laceration in a millimetre-scale epithelium is a classical wound: actin purse-strings, crawling of marginal cells, cadherin resealing, a scar that is barely a scar because the object is small and embryonic and not yet committed to a fibrotic programme. The 2020 paper showed organisms that had been cut and then walked again. That's the sentence the press kit liked, and it's also a real observation. Embryonic Xenopus tissues heal; everyone who has worked animal caps knows they round up and close. Putting that fact inside a designed walker doesn't make the healing synthetic. It makes the walker robust in a way a silicone robot of the same scale is not. Tear the silicone and you have two pieces. Tear this and, often, you have a smaller walker after a pause. Robustness was one of the original motivations in the significance paragraph: living systems repair, silicone does not. The careful version is narrower. Embryonic amphibian epidermis repairs. Whether a xenobot built from adult human cells would repair is a different construction kit, and Gumuskaya's later anthrobots had to answer it on their own tissues. Don't borrow the frog's wound for a mammalian claim.

In short. Cut them and they often seal and walk again. Frog-embryo skin already knows how to close a cut. That is toughness, not magic, and it may not transfer to human cells.

Yolk is the battery, and it sets the clock. Early Xenopus blastomeres are packed with platelets of lipoprotein the oocyte deposited; the embryo lives on that dowry until a mouth works. A xenobot, being those blastomeres in a novel shape, lives on the same dowry. No feeding. No gut. A week or two in saline, then disintegration. Nutrient-rich amphibian media, in the 2021 work, stretched that to months, the organisms going pale and translucent as the pigment and the yolk were spent. Spontaneous death at the end of the yolk is the containment argument the authors were right to make. These objects don't forage. They don't, in the papers, survive being dried, being put in a pond with predators and microbes, or being asked to complete a tadpole programme they have already missed the stages for. A millimetre of tissue that runs out of yolk isn't a grey-goo scenario. It's a culture that ended. If you inflate the kinematic-replication paper into an ecological threat, you've skipped the feedstock requirement, which is thousands of loose embryonic cells in a clean dish. Ponds don't provide that reagent. Laboratory protocols do.

In short. They live on food already packed in the egg, for about a week in salt water, longer if you feed the dish. They do not hunt, and they fall apart when the stores run out.

Blackiston, Lederer, Kriegman, Garnier, Bongard and Levin, Science Robotics, 31 March 2021, 6: eabf1571, is version 2.0. Title: A cellular platform for the development of synthetic living machines. The construction logic inverted. Instead of carving two tissues into an evolved silhouette, they took animal-cap explants, let them round up, and watched a spheroid of epidermis cover itself with motile cilia and swim. No cardiomyocytes. No hand-built hinge. Self-organisation did the manufacturing. Explants formed spheres of 3,026 ± 180 cells, differentiated over four days into ciliated epithelium, and became mobile at three days, at speeds exceeding 100 micrometres per second. Fully differentiated spheres ran from 487 ± 39 micrometres for the smallest cuts to 602 ± 30 micrometres for the largest. Those are the numbers another laboratory can take and try. Bottom-up, in the molecular blurb of this page, means that sentence: the cells knew how to be a ciliated ball if you stopped forcing them into a tadpole and stopped carving them into a walker. The genome was still wild-type. The prompt was geometry and isolation. We find that slightly wonderful, still.

In short. In 2021 they stopped carving heart muscle into shapes. Frog skin balls grew tiny hairs, swam on their own, and needed no heart cells at all.

A motile cilium is a different motor from a sarcomere, and the diagram above is there so nobody fuses them. The axoneme is the 9+2: nine outer microtubule doublets, a central pair, radial spokes, nexin links, and axonemal dynein arms that walk on the neighbouring doublet, sliding it, converting the slide into a bend because the doublets are cross-linked. ATP at the dynein heads. A beat. On a tadpole epidermis those beats move mucus and materials along the surface; the animal is going somewhere else, by muscle. On a xenobot the same beats, covering a millimetre-scale spheroid, move the spheroid. The appendage was repurposed by context, not by sequence. FOXJ1 and the RFX transcription factors are the ciliogenesis programme the wild-type genome already writes in this lineage; multiciliated cells of amphibian epidermis are a textbook object. Notch signalling restrains how many of them you get. Overexpression of the Notch intracellular domain, in the 2021 paper, cut the ciliated census and, with it, the swimming. That's a loss-of-function control a physiologist can like: remove the hairs, lose the gait, keep the ball. The motor is named. The gene list is the frog's.

In short. The swim motor is a tiny hair with a nine-plus-two skeleton of tubes inside, already used by tadpole skin to move mucus. Here the same hairs row the whole ball along.

They contain no neurons. Blackiston and colleagues bisected small and large xenobots, stained for acetylated tubulin, and found multiciliated cells on the surface and no nerve cells inside or out, n = 24. Young tadpoles, same stain, show cilia and also the axons you would expect. The comparison is the point. A swimming millimetre with a beat and no nervous system is a preneural locomotor, which is how the paper framed it, and which is a more useful sentence than mindless robot. Plenty of living things locomote without neurons — ciliates, sponges' larvae, a spreading epithelium. Version 2.0 joins that list at a multicellular, designed-adjacent, frog-genome address. Coordinated beating across a spheroid still wants a coupling story: hydrodynamic, mechanical, maybe ionic. It doesn't want a spinal cord. People who watched the movies and asked where the brain was had brought the wrong phylum. We brought the stain. The stain is why this paragraph is allowed to be this sure, and why you can be sure too once you've seen it.

In short. Staining found the swimming hairs and no nerve cells. A millimetre of frog skin can row itself without a brain. Tadpoles of the same age already have nerves.

Collective behaviour arrived without being programmed as a personality. In a dish with debris or silicone-coated beads, ciliated xenobots clustered and pushed particles into piles. Garnier's group built a computational model of the swarm to ask which pile-ups were predictable from locomotion and adhesion rather than from a hidden controller. Emergent group behaviours, in the abstract, means that sentence. Crowded self-propelled particles with a sticky payload will aggregate that payload; the interesting frog-specific bit is that the particles are living, healing, ciliated epithelia whose adhesion molecules are cadherins the embryo already used to be an animal cap. You can sculpt the collective by changing density, dish geometry, and how sticky the loose cells are. You cannot yet sculpt it by writing a program into a nucleus, because the nucleus is still a wild-type frog's. The 2021 paper is a platform paper in the strict sense: a way to generate many individuals, a way to watch them interact, a model that predicts some of the interactions. Platforms are for other people's next experiments, which is a generous kind of result.

In short. In groups they shove loose particles into piles. That comes from swimming plus stickiness, not from a plan. A model of the crowd predicted some of the piles.

A writable molecular memory was the other platform clause, and it's easy to oversell. They expressed a photoconvertible protein — a fluorescent protein that changes colour when you hit it with a defined wavelength — and showed that a xenobot could record having been illuminated. Proof of principle, the paper said, which is the adult phrase. It isn't a hippocampus. It isn't even a phosphorylation cascade that stores a state for a behaviour. It's a pigment you can switch, in a living millimetre, so that later you can ask which individuals saw the light. As a sensor-and-memory story for a robot, it's a first diode. As a biology story, it's a transgenic reporter in an otherwise wild-type anatomical experiment, and that mix is worth keeping honest: the body plan wasn't transgenic; this particular readout was. Most of the work in the paper didn't need the reporter. Swimming, healing, swarming, the Notch control, the cell counts, the speeds, those stand without it. The reporter is a hook for people who want a robot to remember. We'll leave it at the size of a colour change, which is already a neat trick.

In short. They added a protein that changes colour in light, so a bot could be marked as having seen a flash. That is a dye with a memory, not a mind.

Lifespan, counted, because headlines skip it. In 0.75× Marc's modified Ringer's, wild-type explants lived about nine to ten days after formation, moving across the testing window, metabolising yolk, then coming apart. NotchICD explants, fewer cilia, similar chronological life, less useful locomotion. Nutrient-rich Xenopus media pushed survival past ninety days; the organisms bleached toward the look of a tadpole tail-fin epidermis, which is a visual that tells you the yolk and the pigment were spent and the tissue was still an epidermis. Ten days without feeding is the number to carry in your head. Months with media is the number that says the clock is metabolic, not a programmed apoptosis at day ten. Neither number is a generation time in the evolutionary sense. Neither is a product lifetime. Both are culture parameters. A methods line that cannot say which medium, which temperature, which day post-explant, isn't yet a xenobot paper. It's a movie. The 2021 supplement is the place the ninety-day cultures live. Read the supplement. The abstract's living machines clause isn't a licence to skip it.

In short. In plain salt water they last about ten days. With extra food in the dish they can last months. The clock is their stored yolk, not a built-in expiry date.

High-throughput, in the 2021 claim, is relative to carving walkers by hand. Animal caps self-organise; you can cut many explants in a session; you don't have to place a cardiac patch on each. That's a manufacturing improvement, not a factory. Genomic editing is unnecessary for the basal swimmer, which is the sentence that makes this object different from almost every other synthetic-biology machine in this journal. No Cas9. No transgene required, except if you want the colour-change reporter. No scaffold. No microprinting. Surgical, genetic, chemical and optical perturbations can still be applied during assembly, which is how NotchICD got in, and how a future user would put a drug or a morpholino into the platform. Amenable to high-throughput projects, the abstract said. Amenable is doing honest work. A few dozen to a few hundred individuals in a paper isn't a screen of ten thousand genotypes. It's a step away from one-at-a-time sculpture. Version 1.0 was sculpture. Version 2.0 is a culture method that yields a motile spheroid. Both are xenobots in the literature. They aren't the same object, and headlines that treat them as one blob haven't looked at the actuator.

In short. Making swimmers is faster than carving walkers, because the skin balls build themselves. You still need a microscope and a frog embryo. It is not a factory.

The genome is still Xenopus laevis. That sentence has to sit at the top of this heading, not in a footnote, because the coverage will keep implying a rewrite. No CRISPR. No plasmid driving a developmental transcription factor, except the NotchICD and reporter experiments that were controls and readouts. No synthetic chromosome. The nucleotides are a frog's nucleotides, with a frog's repeat landscape, a frog's allotetraploid history, a frog's maternal transcripts already in the egg. What changed is the anatomical context those nucleotides were asked to interpret. A blastula cell in a tadpole is on a fate map: this blastomere contributes to epidermis, that one to neural plate, that one to mesoderm. The same cell, explanted, dissociated, reaggregated, carved, or placed next to a cardiac patch, is on a different map. Levin's larger claim is that the map was always more plastic than the textbook arrows, and that bioelectric and mechanical coupling are how cells negotiate a body plan on top of the genome. Xenobots are the demonstration object for that claim. They aren't a proof that DNA doesn't matter. They're a proof that DNA isn't a blueprint with one building on it.

In short. The DNA is ordinary frog DNA. Nobody rewrote it. What changed is the shape the cells were asked to build, and they built a machine instead of a tadpole.

A fate map is a statistical claim about a normal embryo, not a law about a cell in every context. Vogt's dyes, then fluorescent lineages, then single-cell atlases: this blastomere, in this stage, in this intact embryo, tends to become that tissue. Take the animal cap out and it becomes epidermis, which is the oldest induction control in amphibian embryology, Holtfreter and the rest. Add a cardiac progenitor patch and some of the cells twitch. Dissociate and reaggregate and you get a spheroid that ciliates. The genome didn't acquire a new gene between those experiments. The cell-cell conversation did. Cadherins, connexins, secreted BMPs and FGFs, a membrane voltage, a mechanical strain — the morphogenetic conversation the Levin laboratory has spent a career measuring in planarians, in frog faces, in bioelectric patterns that predict an eye. Xenobots don't require you to accept every bioelectric claim in that career. They require you to accept a narrower, already-true sentence: embryonic amphibian cells, given a novel geometry and isolation from the rest of the embryo, produce a stable, motile anatomy the fate map didn't list. Plasticity is the word. Magic is not.

In short. In a normal egg, each early cell has a likely future tissue. Take those cells out and group them differently, and they take a different future without changing their genes.

Transcription is still the machine. A cilium doesn't appear because a computer asked for swimming. It appears because FOXJ1, RFX2, RFX3 and the downstream axonemal genes are transcribed, spliced, exported, translated, and assembled into a basal body and a 9+2. A sarcomere doesn't twitch because a voxel was painted contractile. It twitches because cardiac myosin heavy chain, actin, troponin C and the calcium-handling proteins were written from the wild-type loci in a cell that took a myocardial path. The diagram above is the Cell-desk transcription unit, parked here so that no genome was rewritten isn't misread as no genome was used. Promoters, enhancers, chromatin, Mediator, polymerase II. The follicle essay in this journal uses the same diagram for FGF5. The pig-organ essay uses it for human CD46 sitting in a porcine nucleus. Here the twist is the absence of an edit. The letters were already correct for a tadpole. The anatomical prompt decided which letters got read in which cell, and what those cells, together, amounted to. Underdetermination is a transcriptional and a multicellular fact at once.

In short. The swimming hairs and the heart twitch still had to be built from genes being switched on. The genes were the frog's. The new part was which body those genes were building.

This journal's neighbouring CRISPR essays are the contrast, and the contrast is the point of putting them on the same shelf. Cas9, base editors, prime editors: those rewrite letters, then wait for a cell to interpret the new text. Colossal's woolly mouse is a seven-locus coat cassette in Mus. The Maryland pig heart is ten modifications in Sus. Casgevy is a BCL11A-enhancer edit in a patient's stem cells. Xenobots skip the rewrite. They take the unedited text and change the setting it is performed in. That isn't a superiority claim. A sickle-cell patient needs a letter changed, not a new body plan in a dish. A mammoth programme needs alleles, not a millimetre of frog. The claim is narrower and, to us, more interesting for being narrower: some of what you thought was locked in the genome was locked in the geometry, and you can demonstrate that without a nuclease. People who only have a hammer see every developmental question as a knock-in. Levin has been saying the quiet part for years. The xenobot is the object you can point at when the conversation gets abstract.

In short. Other work in this journal rewrites DNA. Xenobots do not. They show that some of what a body becomes is decided by shape and neighbours, not only by the genetic text.

Identity, then, is anatomical. A taxonomist asked what species a xenobot is will say Xenopus laevis, because the genome and the cell lineage say so. A roboticist asked what machine it is will say a designed, motile, biodegradable millimetre with a frog actuator. A philosopher asked whether it is alive will be sent to the 2020 significance paragraph and to the fact that the cells metabolise, heal, and, in the next heading, assemble copies. We refuse to pick only one of those answers, which is the FAQ this page already wrote, and which we're restating as physiology rather than as a catchphrase. Both words — robot, organism — are doing work. The genome is a frog's. The shape is engineered, either by a carving hand following an algorithm or by a self-organising explant following a protocol. Novel multicellularity, Levin's phrase, is the compromise term that hasn't caught on, and perhaps shouldn't, because compromise terms are how a result becomes a brand. Call it a xenobot. Then say what it is made of, how it moves, and that the DNA is wild-type. The rest of the argument will still be there after the noun fight ends.

In short. It is frog tissue in a designed shape. Calling it only a robot, or only an animal, misses half of what was shown. The genes are frog. The body plan is not.

Kriegman, Blackiston, Levin and Bongard, PNAS, 7 December 2021, 118: e2112672118, is the paper that still sounds fake after you've read it. Title: Kinematic self-replication in reconfigurable organisms. Stem cells taken from an early blastula, dissociated, placed in saline, cohere into spheres of about three thousand cells, grow cilia in three days. Drop a swarm of those mature spheroids into a 60-millimetre dish containing about sixty thousand loose stem cells, and their collective motion pushes some of the loose cells into piles. Piles of at least fifty cells develop into ciliated offspring that swim. Give those offspring more loose cells and they build further offspring. All living systems, the opening says, perpetuate themselves by growth in or on the body, then splitting, budding or birth. Here, synthetic multicellular assemblies replicate by moving and compressing dissociated cells in their environment into functional self-copies. Previously unseen at this scale. Arises over days, not millennia. No genetic manipulation required. That's the abstract, almost in their words. We've checked. It's still the strangest methods result in this journal.

In short. In late 2021 they showed that swimming skin-balls can shove loose frog cells into piles, and the piles become new swimming balls. Copying without eggs or cell division of the parent.

Spheres do it spontaneously and badly. Left as balls, the parents make some children and the system peters out. Kinematic replication, at molecular scale, is familiar: a crystal face, a templating RNA, a prion, something that moves and gathers rather than grows and splits. At cellular and organismal scale it hadn't been seen, or at least not named and quantified like this. The spheroids gather because they swim and because loose embryonic cells are adhesive. Most piles are too small. Fifty cells is the threshold the paper measured for a pile that will ciliate and locomote. Below that you have a clump that doesn't become a useful offspring. Spontaneous, in the title of the first figure, means the authors didn't have to evolve a gathering behaviour into the genome. The behaviour fell out of motility plus adhesion plus a dish of feedstock. That's already a result. It isn't yet a reliable replicator. Reliability is the next paragraph, and it's where the evolutionary algorithm comes back, this time not to invent walking but to invent a mouth. We like a mouth that is just physics.

In short. Round bots do this poorly and then stop. They do not need a new gene to pile cells; swimming and stickiness are enough. Most piles are too small to become a new bot.

The algorithm, again on Deep Green, was asked to vary parent geometry so that kinematic replication would last longer. Triangles, squares, pyramids, starfish, a census of silly silhouettes, months of simulation, fitness defined as postponing the loss of replicative ability. Some geometries beat the spheroid. The winner that made the photographs was a semitorus, a C in three dimensions, an open-mouthed Pac-Man. Blackiston carved that cavity into the living spheroids. The mouth gathered loose cells more effectively than a ball, packed them, and released a progeny pile that, if large enough, ciliated. Four generations, in the conditions they reported, if you kept supplying dissociated cells. Walls in the dish that constrained movement also helped, which is an environmental clause the self-replicating robot headlines skipped. Shape plus arena plus feedstock. Three variables. The AI didn't write a replication gene. It found a bucket. We find that both more impressive and more ordinary than the coverage: impressive because a body plan that doesn't exist in the tadpole is a better kinematic replicator; ordinary because once you've seen the cavity, you can explain the physics without a mystery. Pac-Man is a bucket with cilia.

In short. A computer searched body shapes and found that a C-shaped, Pac-Man mouth piles cells better than a ball. With extra loose cells, copies ran to four generations.

Numbers, because a Pac-Man isn't a methods line. Parent: on the order of three thousand cells. Feedstock: on the order of sixty thousand dissociated stem cells in a 60 mm circular dish. Offspring threshold: at least fifty cells. Generations: a few, four in the reported conditions, then failure unless you replenish the loose cells. Time: days, not hours and not weeks per generation. Medium: saline, embryonic Xenopus cells, no serum required for the basal observation. The parent doesn't split, bud a hydra-style clone, or reduce a genome toward a gamete. It's a motile adhesive object that concentrates a resource that, being embryonic frog cells, will self-organise into another motile adhesive object. Kinematic means that sentence. Mitotic would mean the xenobot's own cells ran a cell cycle and partitioned a copy. They didn't, not as the replicative mechanism. Embryonic cells can and do divide; that isn't how a new xenobot is originated in this paper. Origin is assembly. Growth of the pile into a ciliated spheroid is then ordinary amphibian self-organisation, the 2021 platform pointed at a clump the parent made.

In short. A parent is about three thousand cells. A new one can start from fifty. The parent does not split in two. It gathers spare cells, and those spare cells build the copy.

Four generations isn't an ecology, and the feedstock clause is the whole safety analysis. Each round consumes loose blastula cells that you provided from other embryos. Stop providing them and replication stops. Nothing in these papers shows a xenobot harvesting cells from a tadpole, a pond, a person, or a leftover of itself, nor making a gamete, nor reverting to a frog. The stages it skipped aren't waiting in reserve; they were a context it is no longer in. Containment, here, is biology rather than a lock on a freezer: a metabolic clock, a feedstock that doesn't exist outside a laboratory protocol, a millimetre-scale object that desiccates, and a genome that still wants, if returned to an intact embryo at the right stage, to be a tadpole, not a replicator. The authors said unique and useful phenotypes can be reached from wild-type organisms without selection or genetic engineering. That's the scientific sentence. Self-replicating living robots is the press sentence. We'll keep both, labelled, because the first is the result and the second is how you lost the feedstock clause.

In short. Copying only continues if you keep adding loose embryo cells. A pond does not provide those. This is not a wild breeding population, and they do not turn back into frogs.

Mitosis, meiosis and CRISPR are the three copying methods a biologist usually brings to a dish, so that a fourth can be seen. Mitosis is a cell cycle, a spindle, a partition of a genome already replicated in S phase. Meiosis is a reduction division toward a gamete. CRISPR is a rewrite you might then breed. Kinematic self-replication is mechanical assembly of other cells into a body that can do the assembling. von Neumann imagined kinematic replicators as machines that gather parts. Molecular biology found them, arguably, in crystals and in some autocatalytic sets. Multicellular biology hadn't put one on a stage you could photograph next to Pac-Man. The 2021 paper did. Whether you file it under origin-of-life, under robotics, under morphogenetic plasticity, or under a cute cavity is a reading-list choice. The measurement is the same: a parent geometry, a feedstock density, a threshold pile, a generation count, a failure when the loose cells run out. Cite Kriegman. Then argue. Arguing before the citation is how this topic became a genre rather than a result. We'd rather the result.

In short. Ordinary copying is cell division, sex, or gene editing. This is a fourth trick: a moving body that builds another body from spare parts lying around.

Scale is the reason a xenobot is a surprising machine and a modest organism. A single blastomere, tens of micrometres, is a cell with a nucleus, a cortex, yolk, and a set of maternal transcripts. Three thousand of them, half a millimetre, is a tissue that can seal, ciliate and swim. A carved millimetre with a cardiac hinge is a walker. A tadpole is centimetres, a feeding larva, a nervous system, a future frog. People who saw the movies at the millimetre and wrote new species of amphibian had jumped two scales without paying. People who saw a cell and wrote just a culture had jumped the other way. The interesting scale is the one in the middle: enough cells for a body plan, not enough for the tadpole programme to run as the textbook draws it. Fifty cells, the kinematic offspring threshold, is near the floor of that middle. Three thousand is a comfortable room in it. A whole embryo is the ceiling the xenobot never reaches, on purpose. Isolation from the rest of the embryo is the prompt. Hold the millimetre. That's where the machine lives.

In short. One cell is not a xenobot. A tadpole is not one either. The machine lives in between: a few thousand cells, about half a millimetre, not a whole embryo.

Hydrodynamics at this scale is a different conversation from a fish. Reynolds numbers for a half-millimetre spheroid in water, moving at a hundred micrometres per second, sit in a viscous-dominated regime: inertia is small, drag is large, coasting is brief, a beat that stops is a body that stops. Ciliary swimming is the right motor for that regime; a sarcomeric twitch against plastic is a different, frictional gait, closer to crawling than to swimming. Version 1.0 walked on a surface. Version 2.0 swam in a volume. The Pac-Man's gathering is a low-Reynolds pushing of adhesive particles, not a predator chasing prey. We mention Reynolds once, then drop it, because the number is a warning not to import fish intuitions, not a mascot. Particle image, dish diameter, wall effects, the 60 mm arena: those are the experimental objects. A model that ignores walls will miss the result that walls helped replication. A headline that says swim out into the world has ignored viscosity, desiccation and the absence of feedstock. The millimetre is large for a cell and small for an animal. It's just right for a movie. Movies aren't hydrodynamics.

In short. At this tiny size, water feels thick. They do not glide like fish. Stop the hairs or the twitch and they stop. Walls of the dish matter to how they pile cells.

Time, at this scale, is days. Cilia appear at about three days. Swimming is scored across a nine-to-ten-day life in saline. Kinematic generations take days, not minutes. A tadpole would be feeding in a similar window; a frog would be months. Xenobots are a fast anatomical experiment, which is why they could exist as a pipeline at all. You can fail on Tuesday and recut on Wednesday. That speed is a privilege of yolk-rich amphibian cells in a dish, not a general law of living machines. Anthrobots, built from adult human tracheal cells, run on a mammalian clock and a mammalian medium, and they're slower to explain and harder to source. The Deep Green months sit on a different clock again: computer time, not developmental time, paid once per shape family. Wet time is the scarce reagent in a microsurgery lab. The algorithm's job was to spend computer time so that wet time was spent on shapes worth carving. That bargain is the pipeline. A millimetre that takes days is the reason the bargain could close in a single grant cycle. Speed is a methods gift. It isn't a superpower.

In short. A xenobot experiment takes days, not months. That speed is why the team could try many shapes. Human-cell versions are slower and harder to feed.

Levin's claim, held at the size of the data, is that cells contain morphogenetic plasticity far beyond the textbook fate map, and that anatomy is in that sense a software layer: a set of computations in bioelectric, biochemical and mechanical media that the genome enables but doesn't uniquely specify. Planarian regeneration, frog face bioelectrics, an eye induced in the wrong place by a voltage pattern — that's the career the xenobot sits on. You don't have to buy the most ambitious versions (an anatomical compiler, a regenerative medicine that sets voltage and waits) to buy the xenobot. The xenobot is the modest version made visible: same genome, different anatomy, stable enough to walk or swim or gather, for days. Software, as a metaphor, earns one use. It means the body plan can be re-prompted without rewriting the firmware, which is the DNA. What it does is negotiate. Cadherins, gap junctions, secreted factors, membrane voltage, strain. The negotiation, isolated from the rest of the embryo, settles on a spheroid or a walker rather than on a tadpole. That settlement is the result. We find the modest version plenty.

In short. The same genes can support more than one body plan. In a dish, frog cells settle on a swimmer or a walker instead of a tadpole. That is flexibility, not a rewritten code.

Bioelectric networks are the part of Levin's map that a sceptical reader will want named, not waved at. Resting membrane potential differs by cell type and by region of an embryo; gap junctions couple neighbours so that voltage patterns can be larger than a cell; ion pumps and channels write the pattern; a disruption of the pattern can scramble or redirect an anatomy. In frogs, Levin's group has shown that voltage landscapes predict aspects of face and eye, and that manipulating those landscapes can produce anatomical surprises. Xenobots were not, in the three papers, a voltage-mapping project first. They were a geometry-and-composition project. The bioelectric conversation is in the background: these cells are coupled, they have Vmem, they are the same cells whose voltage the laboratory knows how to measure. A reader who wants the voltage paper should read the voltage papers. A reader who wants the living-machine paper shouldn't be told that membrane potential invented the Pac-Man. The cavity was evolved in a physics engine and carved by a microsurgeon. Honesty about which layer did which job is how a research programme stays a research programme. Layers can be real without each paper being about all of them.

In short. Levin's lab also studies electrical patterns in tissues that help steer anatomy. The xenobot papers themselves are mostly about shape and cell type, not about those voltages.

Organoids are the neighbourhood this journal already wrote, and the comparison is clarifying rather than flattening. A cerebral organoid is a self-patterning millimetre of human neural tissue, layered, sometimes electrically active, avascular, not a mind. Lancaster's protocol, Pașca's assembloids: stop flattening stem cells onto plastic, give them Matrigel and a fate, and they will try to be a cortex. Xenobots are the other direction of the same rude fact — cells want to pattern — pointed at locomotion rather than at layers of neurons. Organoids chase developmental fidelity: can we get a ventricular zone, a cortical plate, a projection. Xenobots chase a function the tadpole didn't list: walk, swim, gather. One is a model of a tissue we already have. The other is a model of a tissue we didn't have, built from a genome we did. Both are millimetre-scale, both aren't organisms in the ecological sense, both attract headlines that skip the methods. Related slugs in the metadata are a reading list, not a claim that a xenobot is a brain. A lentil that fires and a millimetre that rows are two answers to what will cells do if you let them. We hold both, happily.

In short. Brain organoids are cells trying to build a cortex in a dish. Xenobots are cells trying to build a mover. Same rude fact — cells pattern — different target.

The axolotl is the other neighbour, and it's a regeneration story rather than a novel-body-plan story. Ambystoma mexicanum amputates a limb and redraws it from a blastema of lineage-restricted, positionally informed progenitors. Xenobots don't regenerate a tadpole they never were. They heal a millimetre they already are, and they assemble a copy from foreign loose cells rather than from their own blastema. Lineage restriction still applies: epidermis makes epidermis, cardiac muscle makes twitch, a ciliated spheroid doesn't grow a kidney because you wished it would. Positional memory, in the axolotl sense of I am a wrist, isn't what a Pac-Man cavity is. The cavity is a geometry that gathers. The blastema is a developmental state that remembers. Putting them in the same journal is allowed because both are arguments against the lazy idea that anatomy is a frozen readout of a genome. Distinguishing them is required because a reader who came for a limb won't find one in a xenobot, and a reader who came for a living machine won't find a walking axolotl. Neighbourhood, not identity. The metadata already said so.

In short. Salamanders regrow a missing limb from a bud that remembers where it is. Xenobots seal cuts and can pile spare cells into a new ball. Related, not the same.

The software claim, held carefully, is this and no more. A genome specifies a space of anatomies, not a single statue. The normal embryo occupies one region of that space because of a conversation we call development. Change the conversation — isolate, reaggregate, carve, add a novel neighbour, evolve a silhouette in silico and impose it — and you can occupy another region without altering the sequence. How large the space is, how many of its regions are stable, how many are useful, how many are pathological, are empirical questions the three papers only sample. They sample locomotion, healing, a kinematic copy, a colour-change memory. They don't sample a gut, a gonad, a cortex, a pain pathway, a life cycle that returns to an egg. Plasticity is real. Plasticity isn't infinity. A wild-type frog genome can build a tadpole and, given a different prompt, a self-replicating walker. That pair is already enough to make anatomy a layer you can write about as such. It isn't enough to make anatomy a programming language. We like the pair. We won't inflate it.

In short. Genes set a range of possible bodies, not one statue. Xenobots show another body in that range. They do not show that any body you can imagine is in range.

Microplastics, drug delivery, mapping a capillary bed: those are the applications that have been floated, and we won't pretend they are results. A millimetre that gathers particles in a dish is a proof that gathering can happen, not a harbour-cleaning device. A biodegradable object that runs on yolk and dies is an attractive safety profile for something you might, one day, put in a body or a waterway; it isn't a pharmacokinetic file. Delivery would want a payload, a targeting logic, a survival time you had chosen rather than inherited from an oocyte, and a regulator who had been shown the feedstock clause. None of that is in the 2020–2021 papers. The 2021 abstract's numerous practical applications in biomedicine and the environment is a platform sentence, the kind a Science Robotics paper is allowed, and the kind a methods reader should pencil not yet beside. We like the honesty of a living millimetre that already locomotes more than we like a slide about the ocean. Movement, healing and piling cells: those are the results. The rest is a grant paragraph.

In short. Cleaning plastic, carrying drugs, and mapping blood vessels have been suggested. They have not been shown. What has been shown is movement, healing, and piling cells in a dish.

Are they robots or animals? The FAQ is still the right FAQ. They are living multicellular systems designed by an algorithm and assembled from frog cells. The genome is Xenopus. The shape is engineered. A robot, in the Bongard sense, is a machine whose body was a design parameter; these qualify. An organism, in the sense of a metabolising, developing, self-repairing, in some conditions self-copying assembly of cells, also qualifies. A frog, in the sense of a sexually reproducing amphibian with a tadpole stage, does not. The question won't die because English doesn't have a comfortable noun for a designed anatomy running on an undesigned genome. We refuse to pick only one word, not as a pose, as a methods stance: picking one word is how you lose either the evolutionary algorithm or the yolk. Keep both in the sentence. Then go back to the dish. The dish isn't confused. It contains a millimetre of tissue that walks or swims. Our nouns are the confused objects. They'll remain confused after this page. The cell counts will not.

In short. They are living frog cells in a designed shape. Robot and organism both apply. Frog, in the ordinary sense of a tadpole that becomes an adult, does not.

Anthrobots are the human-cell cousins, and they belong here as a boundary, not as a sequel we claim to have read at the same depth. Gumuskaya, Blackiston, Levin and colleagues, 2023, took adult human tracheal epithelial cells, let them self-assemble into motile multicellular spheroids, and showed locomotion and healing without a frog yolk and without an embryonic blastula. Different chassis, different clock, different ethical file: these cells came from adult donors, not from a sacrificed Xenopus embryo, and they carry a human genome, which is the genome this journal's organoid essay is already careful about. If xenobots were a proof that amphibian embryos can be prompted into novel motility, anthrobots are a proof that the prompting isn't amphibian-only. We won't restack that paper here. We name it so that frog-only curiosity doesn't survive contact with the reading list, and so that they will put these in people next year doesn't survive either. A tracheal spheroid in a dish isn't a therapy. A xenobot in a dish isn't a therapy. Platforms get oversold in the same tense. We'll stay in the past tense of what was published.

In short. A later project built similar moving balls from adult human airway cells. That shows the idea is not frog-only. It is still a dish experiment, not a medicine.

Ethics, kept specific. Sacrificing Xenopus embryos is ordinary developmental biology; these papers sit inside that ordinary, with the extra step that the explant is asked to be a machine. Kinematic replication sounds, to a public ear, like a thing that could escape. The feedstock clause, the yolk clock, the millimetre, the absence of a gamete, are the reasons it isn't an escapee in any scenario the papers support. Dual-use handwringing about living machines should be aimed at the right objects: airborne pathogens, gene drives, things that eat and breed in the wild. A Pac-Man that needs sixty thousand dissociated blastula cells isn't that object. The human-cell version raises a different, quieter question, the organoid question: at what size and complexity does a motile human tissue in a dish deserve a consent and an oversight that a keratinocyte culture does not. We'd rather have that conversation early, as the organoid essay already said, than late. Xenobots, as frogs, don't start it. They do illustrate how fast a cute movie can outrun a protocol. Our job, here, is to put the protocol back in front.

In short. They will not breed in a pond. They need loose embryo cells and they run out of yolk. The harder ethics sit with the human-cell versions.

The peptide neighbourhood is real and shouldn't be overplayed. This catalogue stocks characterised actin-adjacent reagents; thymosin β4 analogues sit on G-actin sequestration, a clerk of the same cytoskeleton the cardiomyocyte sarcomere and the epidermal cortex spend. Binding of that clerk by a 43-mer, and a millimetre of frog that twitches and heals, are two ways to be in conversation with actin. One is a vial with a chromatogram. The other is a paper with a movie. Neighbourhood, not identity. CRISPR, on the same shelf, is the tool xenobots conspicuously didn't need. Organoids and axolotls are the tissues that make the plasticity claim less lonely. A reader who came for a reconstitution protocol is allowed to leave knowing what kinematic self-replication is. A reader who came for a living robot is allowed to leave knowing that a lyophilised peptide and a xenobot aren't interchangeable energy juice. Sequence in, phenotype out, by chemistry, by CRISPR, or by geometry. The century is doing all three. Pretending they are rival religions is how you miss the decade. We stock the chemistry, labelled for research. We write about the geometry because it is true.

In short. Research peptides here can touch the same cell skeleton that lets xenobots twitch and seal. That is a neighbour on a reading list, not a recipe.

Here's the map we'd like you to take home, rather than a nickname. Version 1.0, Kriegman, Blackiston, Levin, Bongard, PNAS 2020: an evolutionary algorithm on Deep Green, millimetre-scale walkers of Xenopus epidermis and cardiomyocytes, hand-assembled, predicted gaits that often transferred, healing, a yolk clock. Version 2.0, Blackiston et al., Science Robotics 2021: animal-cap spheroids of about three thousand cells, half a millimetre, motile cilia, no heart muscle, no neurons, more than a hundred micrometres a second, nine to ten days in saline, months if fed, a photoconvertible colour change as a first memory, swarms that pile particles. Kinematic self-replication, the same four authors, PNAS 2021: Pac-Man cavities, piles of at least fifty cells, a few generations if you keep providing about sixty thousand loose blastula cells in a 60 mm dish, not mitosis. The genome is wild-type Xenopus laevis. The identity is anatomical. Applications remain proposals. Anthrobots are the human-cell boundary. Organoids and axolotls are neighbours, not aliases. If your next sentence is scientists created a new species, you've left the papers. If your next sentence is it is just a clump, you've left them the other way. Stay with the cell counts.

In short. Leave with the map: walkers with heart muscle, swimmers with tiny hairs, Pac-Man copies from spare cells, ordinary frog DNA, no product. Stay with those facts.

The public papers are short enough to actually read. Kriegman 2020, the pipeline, still the construction kit. Blackiston 2021, the platform, still the numbers. Kriegman 2021, the cavity, still the feedstock clause. Levin's reviews on morphogenetic plasticity and bioelectrics, so the career the object sits on is visible, and so you can see where the object ends and the ambition continues. Lancaster 2013 and the organoid essay in this journal, so millimetre-scale self-patterning has another address. Kragl and the axolotl blastema, so regeneration isn't kinematic assembly. Jinek 2012 and the CRISPR essays, so you can feel the absence of a nuclease as a choice. Gumuskaya 2023, if you want the human tracheal version. That's a week of evenings, not a guru. The movies will still be there when you come back, and they'll look more like a methods figure and less like a species announcement. We think that's an improvement. The neighbouring essays — organoids, axolotl, prime editing — are the rest of the shelf. This one is the millimetre you cannot hold, and shouldn't need to, because the papers already did.

In short. Three named papers cover the walker, the swimmer and the Pac-Man copies. Read those before the movies. The rest of the shelf is organoids, salamanders and CRISPR.

The conceptual payload is underdetermination, and it's allowed to be interesting without being a manifesto. Life's hardware, a genome, underdetermines the machine you can build from it. That isn't a licence to ignore the genome; the cilia still come from FOXJ1, the twitch still comes from cardiac myosin, the adhesion still comes from cadherins the frog already had. It is a licence to treat anatomy as an experimental variable, not only as an output. De-extinction rewrites the hardware and hopes the chassis of a living relative will interpret the new letters as a coat or a haemoglobin. Xenobots keep the hardware and rewrite the chassis. Both are true in the same decade, which is why they belong in the same journal, even if one of them looks like a millimetre of frog having a think. A millimetre that walks is a result. A millimetre that copies itself from spare cells is a result. A headline that skips the yolk, the feedstock and the wild-type sequence is a failure of reading. We wrote this so the reading would have a place to sit, next to the movies, at the size of the cell counts.

In short. The same frog genome can build a tadpole or, in a dish, a walking millimetre that piles copies. That is the idea. The movies are evidence, not the idea itself.

Research-use-only is the wrong legal class for a millimetre of frog, and this page isn't a catalogue listing. The objects here are explants, algorithms, and papers. A neighbouring peptide vial, if you came from that shelf, is a characterised laboratory solid and isn't a xenobot. A neighbouring CRISPR medicine is a licensed haematopoietic edit and isn't a body plan. Confusing those classes is how a reader ends up with a protocol they shouldn't have. The physiology in the paragraphs above is public, cited, and already in print, with the applications still in the future tense. Use it to read the next living-machine paper without the press release doing the reading for you. Name the tissues. Name the actuator, cardiac or ciliary. Name the cell counts and the dish. Name the fact that the DNA is wild-type. Name the feedstock if anyone says replicate. We'll tell you what the three papers did, and we'll keep the genome and the body plan in separate clauses until the headlines do too. That's the whole job of this page.

In short. This is not a product page. It is a map of three papers. Read the next living-machine claim against the cell counts, the actuator, and whether the DNA was actually changed.