In 2015, at UT Southwestern Medical Center in Dallas, a researcher named Arthur Prindle was watching a Bacillus subtilis biofilm under a microscope when the interior of the colony started pulsing. Not growing. Pulsing, in waves, like something with a heartbeat. Working with Gurol Suel, whose lab studies collective bacterial behavior, Prindle traced the pulses to a potassium ion channel called YugO. Cells starved for glutamate at the biofilm’s core were releasing potassium, hyperpolarizing their neighbors, and sending that hyperpolarization outward in a wave that reached cells hundreds of microns away. Bacteria were talking to each other the same way neurons do: with ions and voltage, not just diffusing chemical signals.
That finding, published in Nature, is the reason this article exists. It reframed bacterial biofilms as electrical networks capable of coordinating a decision, in that case, whether peripheral cells should keep consuming a limited nutrient or back off so the starving interior could survive. Coordinating a shared resource under conflicting local incentives is, functionally, a computation. It has inputs, a rule, and an output. Researchers have spent the decade since asking how far that idea extends, and whether “how far” includes something as concrete as an AND gate or an OR gate built out of living tissue instead of silicon.
Bacteria Already Compute, They Just Do It With Potassium
The 2015 paper wasn’t really about logic gates. It was about survival. But the mechanism it uncovered has the right shape for computation. A cell in the interior of the biofilm senses a local condition, metabolic stress from glutamate limitation, and converts that condition into an electrical signal via an ion channel. The signal propagates through the community at a measurable, repeatable speed. Peripheral cells receive it and change their behavior in response. Swap “glutamate limitation” for “input bit” and “growth arrest” for “output bit,” and you have most of what a logic gate needs: a rule that maps conditions to a determined response.
Suel’s group followed up in 2017 with a Science paper, led by Jintao Liu, showing that two physically separated biofilm colonies could synchronize their electrical oscillations and effectively time-share a limited nutrient supply between them, taking turns growing rather than competing simultaneously and starving each other out. That’s a step past a single decision rule. It’s distributed coordination between independent units using nothing but ion channel signaling as the communication layer. Biofilms don’t have neurons. They don’t have a genome-encoded blueprint for “electrical computation.” They evolved a control system for resource allocation that happens to look, mechanistically, like the thing neuroscientists spent a century studying in animal brains.
Bacteria don’t have neurons. They still learned to compute with voltage.
What makes this relevant to the idea of programmable bioelectric circuits is that the mechanism is tractable. YugO is a single, identifiable ion channel gene. Researchers can knock it out, overexpress it, or replace it with a channel of different voltage sensitivity, and watch the biofilm’s collective behavior change accordingly. That is the raw material of an engineered circuit: a component with known input output behavior, sitting inside a living, self-replicating chassis. The gap between “we found a natural bioelectric decision system” and “we built a synthetic one with a defined truth table” is exactly the gap synthetic biologists are now trying to close.
Michael Levin’s Argument That Anatomy Is Also a Decision Being Held in Voltage
The other major thread in this story runs through Michael Levin’s lab at Tufts, home since 2016 to the Allen Discovery Center for Molecular Structure of Regeneration, funded by the Paul G. Allen Frontiers Group. Levin’s argument, made across two decades of work in planaria and Xenopus embryos, is that tissues maintain bioelectric states, patterns of resting membrane voltage across groups of cells, that function as a layer of stored information sitting on top of the genome. Gap junctions link cells into electrical networks the same way wires link circuit components. Change the voltage pattern, and you change what the tissue builds, without touching a single gene.
The demonstration that made this hard to dismiss involved planarian flatworms. Cut a planarian in half and each half regenerates the missing piece, tail grows a new head, head grows a new tail, using the same genome throughout. Levin’s group showed that briefly altering the bioelectric state of the wound site, using gap junction blockers or ion channel drugs, could cause a worm to regenerate two heads instead of one, or a head where a tail should be. The genome never changed. The instructions encoded in the voltage pattern did. Cut that same reprogrammed worm again, with no further intervention, and it keeps producing two headed offspring for multiple rounds of regeneration. The bioelectric state is stable. It’s memory, not a one time nudge.
The instructions for anatomy live partly in voltage, not only in DNA.
Read as a logic problem, a cell at a wound site is integrating multiple inputs, injury signal, neighbor voltage, gap junction connectivity, and producing one of a small number of discrete anatomical outputs: head, tail, nothing. That’s a decision tree implemented in bioelectric hardware. Levin’s group has since used optogenetic tools, light activated ion channels expressed in specific tissues, to flip these voltage states directly and predictably, rather than relying on pharmacology that affects the whole organism. Light on, channel opens, voltage state changes, anatomical outcome follows. That is about as close as biology gets to flipping a switch and reading a bit.
The Engineering Problem Is Turning a Discovered Rule Into a Designed One
Here is where I think the field needs a dose of honesty. Genetic logic gates, built from transcription factors and promoters, are a mature engineering discipline. Christopher Voigt’s lab at MIT published a design automation framework in 2016, called Cello, that lets a user specify a Boolean circuit and get back a DNA sequence that implements it in E. coli, with the same rigor an electrical engineer would expect from a chip design tool. That maturity took fifteen years and thousands of characterized genetic parts. Bioelectric circuits are nowhere near that point. The parts list is shorter. The response times are faster, which is an advantage, ion channels open and close in milliseconds where transcription takes minutes to hours, but the signals are noisier and harder to insulate from the rest of a cell’s native electrophysiology.
The tools that make bioelectric engineering possible at all come mostly from optogenetics. Adam Cohen’s lab at Harvard built an all optical electrophysiology platform, published as Optopatch in 2014, that pairs a light gated depolarizing channel with a genetically encoded voltage indicator, so a researcher can both write a voltage state into a cell with one wavelength of light and read the resulting voltage back out with another. That combination, write and read without an electrode, is what makes “programmable” a plausible word here rather than an aspirational one. You can, in principle, express two different light sensitive channels in a cell, each responsive to a different wavelength, and define a truth table: blue light alone does nothing, red light alone does nothing, both together push the membrane past a threshold that triggers a defined output. That is a coincidence detector. It’s a crude AND gate, and its behavior depends on channel kinetics and expression level rather than a clean digital threshold, which is the honest caveat every paper in this space has to carry.
| Circuit type | Signal basis | Response time | Engineering maturity |
|---|---|---|---|
| Genetic transcriptional logic gates | Transcription factor binding | Minutes to hours | Mature, design automation exists (Voigt lab, Cello) |
| Bacterial biofilm electrical signaling | Potassium ion channel waves | Seconds to minutes | Discovered natural system, engineering emerging |
| Tissue level bioelectric patterning | Gap junction coupled membrane voltage | Hours to days for stable state change | Demonstrated control of outcomes, not yet generalized circuit design |
| Optogenetically driven single cell circuits | Light gated ion channels | Milliseconds | Read and write demonstrated, truth tables still crude |
What a Working Bioelectric Logic Gate Would Actually Be Good For
I keep coming back to why this matters beyond novelty. Genetic logic gates changed synthetic biology because they let engineers build cells that sense multiple conditions and respond conditionally, a bacterium that only expresses a therapeutic protein when it detects both low oxygen and a tumor marker, for instance. Bioelectric logic gates would add something genetic circuits are bad at: speed, and access to a signal that already carries developmental information. If Levin’s model of bioelectric state as a rewritable layer of anatomical instruction holds up under engineering rigor, then a circuit that reads voltage state and writes a corrective voltage pattern back could, in principle, correct developmental errors or trigger regeneration directly, without a genetic edit at all. That’s the pitch behind the Allen Discovery Center’s continued work: use ion channel drugs and modulators as a kind of software patch for anatomy, rather than editing the underlying genomic hardware.
For biofilms, the practical case is biosensing. A living material that changes its own electrical output in response to two or more simultaneous chemical conditions could act as a self-reporting sensor for environmental contamination or fermentation control, cheaper and more self-sustaining than an electronic sensor array, provided anyone solves the reliability problem. Right now, natural bioelectric signaling is closer to an analog circuit with a lot of biological noise than a synchronous digital one. Getting from “the biofilm changes behavior when two conditions coincide” to “the biofilm reliably outputs 1 only when both inputs are true, and 0 otherwise, across generations” is the unglamorous engineering work nobody gets a Nature paper for finishing halfway.
The signal was always there. Learning to write to it is the new part.
What’s real right now, and worth being precise about, is this: living cells demonstrably use voltage as an information channel independent of their genetic circuitry, in organisms as different as soil bacteria and flatworms, and researchers can both read and, with optogenetic tools, write to that channel with some precision. What isn’t real yet is a general purpose bioelectric logic gate with the reliability of a genetic one, or a design automation tool that turns a truth table into an ion channel expression plan the way Cello turns one into a DNA sequence. The distance between those two states is where the next decade of this field will be spent, and it’s a more interesting bet than most of synthetic biology’s current genetic circuit work, because the substrate, voltage, moves at the speed nervous systems move at, not the speed transcription does.
Credit: Bhautik Patel on Unsplash