The
Bioelectric
Code

The flatworm did not get the message. Researchers at Tufts University had cut it in half, as planarian flatworms are routinely cut — they regenerate, it is what they do. But this worm had been treated with something first: a brief pharmacological cocktail targeting the ion channels in its cells. No gene was altered. No DNA was touched. When the fragment regenerated, it grew a head at both ends. And when that two-headed worm was then cut and allowed to regenerate again, with no further treatment of any kind, it grew two heads again. And again. The new body plan had been written — not in the genome, but somewhere else entirely. In the electrical language that cells use to remember what they are, and what they are supposed to build.

Part I — The Hidden Language: What Bioelectricity Actually Is

Every cell in your body is a battery. Its membrane — the thin lipid bilayer separating its interior from its surroundings — maintains a voltage difference between inside and outside, driven by ion pumps that move charged particles (sodium, potassium, calcium, chloride) across the barrier at a metabolic cost the body pays continuously, every moment of life. In neurons, this voltage is well-understood: nerve impulses are rapid, brief reversals of the membrane potential — depolarisations that propagate along the axon at speeds up to 120 metres per second, carrying the signals we call thought, sensation, and movement.

What Michael Levin's laboratory at Tufts University has spent two decades demonstrating is that this same electrical language is not exclusive to neurons. All cells — muscle cells, skin cells, liver cells, the cells of developing embryos, the cells of regenerating tissues — maintain membrane potentials, and those potentials form a network. Not the rapid, all-or-nothing signals of neuroscience, but slow, graded, persistent electrical patterns: a voltage topology that encodes not a moment of sensation but an instruction about form. The bioelectric pattern distributed across a tissue carries information about target morphology — what the body is supposed to look like, and how to get back there when something goes wrong.

Levin calls the body's stored electrical representation of its intended anatomy the "anatomical setpoint" — a target state encoded in bioelectric patterns that cells collectively compute toward, the way a thermostat is set to a target temperature. Cut a salamander's tail and it regrows exactly the right structure, then stops. The genome did not change. The cell types did not change. What changed was the local bioelectric context, and the cells read it and rebuilt accordingly. The genome provides the catalogue of possible proteins. Bioelectricity provides the blueprint for which proteins to deploy, where, and in what configuration — the difference between having all the letters of the alphabet and knowing how to write a sentence.

The voltage map of a tissue can be visualised using voltage-sensitive dyes or genetically-encoded fluorescent reporters that light up in proportion to membrane potential. When Levin's team images a developing frog embryo this way, before a single organ has formed, the bioelectric pre-pattern is already there — a glowing topological map predicting where the face will be, where the gut will fold, which cells will become muscle and which will become nerve. The body draws its blueprint in electricity before it begins construction.

Part II — The Flatworm With Two Heads (And What It Means)

Planarian flatworms are the most studied regenerative organism in biology. Cut them transversely — head from tail — and each piece regenerates the missing part with remarkable fidelity: the head piece grows a new tail, and the tail piece grows a new head. This polarity — which end becomes head, which becomes tail — was long assumed to be determined by the molecular gradients established during the original organism's development, signals laid down in the tissue that could not be overwritten without genetic intervention.

Levin's lab demonstrated otherwise. By transiently manipulating the bioelectric state of planarian tissue — using drugs that block gap junctions (the cellular connections through which electrical signals propagate between cells) or that target specific ion channels — the team was able to produce worms that regenerated with heads at both ends, or that regenerated the body plan of a different species of planarian entirely. No genetic modification. The change was purely in the electrical conversation between cells during the first hours after amputation.

The most striking property of this effect is its permanence. Treated worms that had adopted a two-headed body plan continued to regenerate as two-headed worms indefinitely, through many subsequent cuts and regrowths, with no further pharmacological intervention. The bioelectric pattern had been rewritten, and the cells' collective memory had adopted the new target state as their new normal. The genome was unchanged. The instruction was not in the DNA — it was in the electrical field.

The same bioelectric manipulation that changes head-tail polarity can also, Levin's lab has shown, direct non-eye tissue to form functional eyes in wrong locations — eyes on the gut, the tail, the flank — that are properly structured, innervated, and in some experimental contexts, functionally capable of detecting light. This is not a developmental accident. It is a demonstration that the spatial information encoded in bioelectric patterns is the actual master variable controlling body form, with the genome serving as the executor of those instructions rather than their author.

Part III — Cancer as a Political Disorder

The most provocative claim in Levin's framework — and the one with the most direct clinical stakes — concerns cancer. In conventional oncology, cancer is understood as a genetic disease: mutations accumulate, proto-oncogenes are activated, tumour suppressors are lost, and cells escape the normal constraints on proliferation. The treatment paradigm follows logically: find the mutations, target them with drugs, kill the aberrant cells.

Levin proposes a different framing. Cancer, in his account, is primarily a bioelectric disorder — a failure of cellular community. Healthy tissues maintain a bioelectric network that communicates the body's architectural goals: grow here, stop there, differentiate into this cell type, remain this size. Cancer cells are characterised by a loss of normal membrane potential — they are chronically depolarised, meaning their voltage is less negative than healthy cells — and this depolarisation both drives aberrant behaviour and severs the bioelectric communication channels through which the surrounding tissue would normally constrain them. The cancer cell has, in the language of collective intelligence, defected from the community.

The diagnostic implications emerged before the therapeutic ones. In tadpole experiments, Levin's team discovered that early-stage tumours — before they were histologically detectable — could be identified by mapping the bioelectric patterns of the skin. Tumour cells showed characteristic depolarisation signatures visible as voltage anomalies in the surrounding tissue, appearing as irregular patches in the bioelectric map that preceded any visible morphological change. A future in which early cancer detection is achieved through non-invasive bioelectric imaging rather than tissue biopsy is not theoretical — the principle has been demonstrated in living organisms.

On the therapeutic side, multiple groups are now exploring approaches that target the bioelectric state of tumours: ion channel drugs that restore normal membrane potential to cancer cells, nanoparticle systems that disrupt the bioelectrical homeostasis of tumour tissue, and MXene-based bioelectronic interfaces that induce cancer cell death through irreversible depolarisation and ion channel disruption. In 2025, a preprint demonstrated bioelectronic modulation of glioblastoma — one of the most treatment-resistant brain cancers — via wireless carbon nanotube interfaces. The tumour cells were not killed chemically. Their electrical community was disrupted.

Part IV — The First Scientific Discovery Made by a Non-Human Mind

In the history of science, every model — every theory describing how a biological system works, every quantitative account of a mechanism — has been the product of human cognition. Experiment, pattern recognition, hypothesis, verification: the loop has always had a human at its centre. Until recently.

Levin's laboratory built an AI system specifically to test whether machine intelligence could operate at the level of scientific discovery — not as a search tool, not as a classifier, but as a theorist. The system was given access to the published literature on planarian biology: every experiment performed on the worms, every dataset collected, every result reported across the decades of research accumulated by the field. It analysed this literature and identified a novel computational model of planarian regeneration — a new account of the bioelectric circuits that govern head-tail polarity, explaining the full breadth of experimental data in a unified framework.

The model was not a refinement of existing ideas. It was a genuinely new framework — a hypothesis about how the bioelectric information in the tissue is encoded and read that the human research community, despite decades of expertise and immersion in the data, had not produced. It was subsequently validated experimentally. The first quantitative model capable of explaining the complete dataset of planarian regeneration experiments was not discovered by a human scientist. It was discovered by a machine.

A second, more recent development takes this further. A March 2025 arXiv paper — "AI-Driven Control of Bioelectric Signalling for Real-Time Topological Reorganization of Cells" — demonstrates AI systems that do not merely analyse bioelectric phenomena but actively control them in living tissue. The system observes the electrical state of a cell population, computes the pharmacological intervention required to achieve a target bioelectric configuration, and delivers it — closing the loop between machine intelligence and living matter in real time. One experimental validation: a wearable bioelectronic bandage that used fluoxetine (better known as Prozac, acting here not as an antidepressant but as an ion channel modulator) to alter wound-border cell membrane potential, producing a 39.9% increase in tissue re-epithelialization and a 27.2% reduction in inflammatory signalling. The AI was not given the intervention. It derived it from first principles and applied it.

The Allen Discovery Center at Tufts, which Levin directs, has framed its institutional mission in terms of what it calls "reading and writing the morphogenetic code" — developing the computational and experimental tools to decode bioelectric patterns and then edit them with the precision we currently bring to genetic sequences. What AlphaFold did for protein structure prediction, this programme aims to do for the electrical language of form. The analogy is instructive. Before AlphaFold, protein folding was a largely unsolved problem despite decades of effort. After AlphaFold, it was accessible. Bioelectric decoding is roughly where protein structure was in 2015 — understood in principle, largely inaccessible in practice, and standing at the edge of a computational breakthrough.

Part V — Living Robots Made From Your Own Cells

In 2020, Levin's lab — in collaboration with computer scientist Josh Bongard — announced Xenobots: tiny living constructs assembled from African clawed frog embryo cells according to body plans designed by an evolutionary AI algorithm. The AI generated candidate forms optimised for specific behaviours (locomotion, carrying a payload), and the researchers then assembled living cells into those shapes. The result were millimetre-scale biological robots that could move, could carry objects, and could, in some experimental conditions, assemble loose frog stem cells in their vicinity into copies of themselves — a form of kinematic self-replication not seen before in any organism or robot.

The 2023 follow-up was in certain respects more significant. Anthrobots — built from human adult tracheal cells — were not designed by an AI algorithm and did not require genetic modification or careful assembly. They formed spontaneously when human tracheal cells were placed in the right conditions: freed from their normal tissue context, the cells' cilia (hair-like projections that normally move mucus along airway surfaces) propelled the cell clusters into self-organised, motile forms. The cells, given permission by their context to become something other than airway lining, became something unprecedented.

What Xenobots and Anthrobots reveal is a biological insight that goes beyond their immediate medical applications: the cells of complex organisms carry latent computational properties that their normal tissue context suppresses. A tracheal cell is a tracheal cell because its bioelectric and chemical environment enforces that identity. Remove those constraints, and the cell does not simply die or divide randomly — it participates in a new form of collective problem-solving, drawing on ancient cellular programmes that predate the evolution of tissues and organs. Levin's framing treats this as evidence that cognition — goal-directed, adaptive behaviour — is not an emergent property of nervous systems specifically, but of any sufficiently interconnected biological information-processing system. The body is not a machine. It is a collective intelligence. Bioelectricity is its language, and AI is now beginning to speak it.

Part VI — The Clinical Pipeline: One Day of Treatment, Eighteen Months of Growth

Morphoceuticals, the Tufts spinout company founded on Levin's research, achieved in early 2022 what had previously been considered the exclusive province of science fiction: functional limb regeneration in an adult animal of a species that does not naturally regrow complex limbs. The subject was an African clawed frog — a species that, as an adult, is incapable of regenerating its hindlimbs. The protocol was a five-drug cocktail applied in a bioreactor worn on the frog's amputated stump for a single 24-hour period. That single day of bioelectric intervention triggered an 18-month self-directed process of regrowth that produced a functional limb: vascularised, innervated, and capable of movement and underwater propulsion.

The significance of the 24-hour window cannot be overstated. The treatment did not maintain a continuous pharmacological intervention through the months of growth. It set the bioelectric pattern — wrote a new anatomical setpoint — and then the tissue's own collective intelligence executed the construction plan. The cells already knew how to build a leg. They lacked the bioelectric instruction to do so. That instruction, delivered briefly at the critical moment, was sufficient.

The company, which has raised $10 million in seed funding from Juvenescence and Prime Movers Lab, appointed a new CEO in April 2024 and is progressing through preclinical work in rodent models. Initial human applications in the pipeline target stump health improvement, wound healing innovation, and skin repair — each a step on the pathway toward the more ambitious goal of functional limb regeneration in humans. The mechanism is validated. The pathway is conventional drug development, not regulatory novelty. The timeline to first human applications is measured in years, not decades.

Adjacent to limb regeneration, the bioelectronics field is developing wearable and implantable devices that deliver pharmacological or electrical stimulation to guide tissue repair in real time. The fluoxetine-delivering bioelectronic bandage demonstrated in the 2025 arXiv paper — which produced nearly 40% wound-healing improvement — is one example. Psychiatric medications are showing unexpected utility as ion channel modulators in cancer biology: antipsychotics including chlorpromazine have been found to promote apoptosis in breast cancer cells by altering membrane potentials and calcium flux. The pharmacological toolkit for bioelectric medicine is already partially assembled, built from drugs approved for entirely different purposes.

Part VII — The Second Programming Language of Life

Step back from the experiments and the clinical pipeline and the AI-discovered models, and something larger comes into view.

The central dogma of molecular biology — DNA makes RNA makes protein, the genome is the master controller of life — is not wrong. It describes a real and important layer of biological computation. But it is incomplete in ways that the history of 20th-century biology systematically obscured. The focus on genes as the unit of inheritance, the spectacular success of recombinant DNA technology, the revolution of genomic sequencing — these created an intellectual framework in which the genome appeared sufficient to explain life. The body plan was assumed to be in the sequence.

What bioelectricity reveals is that this assumption was false. The sequence provides the vocabulary. The bioelectric field provides the grammar — the rules by which individual cellular words are assembled into the complex sentences of form and function. And these are genuinely independent layers: you can rewrite the grammar without changing the vocabulary, and the result is a different body. You can preserve the grammar while changing specific words, and within limits, the grammar will compensate. Evolution has operated on both layers simultaneously, for billions of years, and we are only now beginning to read the second one.

The implications propagate in all directions. In regenerative medicine: the ability to write bioelectric instructions that invoke the body's own latent regenerative programmes, rather than engineering replacement tissues from scratch. In oncology: the possibility of treating cancer as a community failure — a bioelectric re-integration problem — rather than solely as a genetic aberration to be targeted and destroyed. In developmental biology: an account of birth defects that traces causal pathways through bioelectric pre-patterning, offering new intervention points before structural abnormalities are established. In neuroscience: a continuum of cognitive complexity that extends from brain to body, suggesting that the problem of consciousness may be less singular than our neuron-centred assumptions have led us to believe.

And in artificial intelligence: a set of design principles derived from the most successful information-processing system in the known universe. Levin's model of distributed, goal-directed, error-minimising bioelectric computation in biological tissues is informing theoretical frameworks for understanding how intelligence emerges from physical substrates — insights that flow in both directions between the study of living bodies and the construction of artificial minds.

The flatworm with two heads is not, in the end, an oddity. It is a proof of concept. It demonstrates that the information system governing body form is accessible, writable, and separable from the genetic layer we have spent seventy years studying. Learning to read and write it — with AI as the decoder, and bioelectronics as the pen — may be among the most consequential scientific projects of this century. It is also, at present, among the least known.

The cells in your body are not merely following genetic instructions. They are participating in a conversation — electrical, distributed, ancient, and extraordinarily intelligent. We are only now learning to listen.