Life is spectacular at computation. A human immune system runs a distributed search algorithm across billions of antibody variants, finds the one that binds a novel pathogen, and amplifies it by a factor of a million—all without central coordination, all in about a week. A slime mold solves the Tokyo rail network. A single neuron integrates ten thousand synaptic inputs and makes a decision in milliseconds. Biology computes with a sophistication that makes our best engineered systems look like abacuses.
But there is a paradox buried in all this brilliance. Life is extraordinary at running its own computations—the ones that keep it alive, that find food, that evade predators, that replicate. It is terrible at running yours.
Try to put a genetic circuit into a living cell—a simple Boolean logic gate, say, something that should produce a fluorescent protein when two chemical inputs are both present—and you will discover that the cell has its own priorities. It will mutate your circuit. It will evolve around it. It will divert resources away from your foreign DNA toward the metabolic programs that matter to it: growth, division, survival. Within a few dozen generations, your carefully designed gate will be leaky, drifting, broken. The cell treated your computation as an infection and, slowly, won.
A living cell will always prioritize its own survival over your circuit. That is not a flaw. That is what being alive means.
This is what synthetic biologists call the chassis problem. The cell you put your circuit into is not a neutral vessel. It is an active antagonist. Its metabolism is a black box of staggering complexity—Craig Venter’s team at the J. Craig Venter Institute spent years stripping a bacterial genome down to the bare minimum, producing JCVI-syn3.0, a cell with only 473 genes. Even in this minimal organism, roughly a third of the genes have unknown functions. Nobody knows what they do. The simplest living thing we can build still contains mysteries we cannot decode.
Your genetic circuit, meanwhile, is a guest in this house. It competes for ribosomes, for amino acids, for the cell’s limited energy budget. Every protein your circuit expresses is a protein the cell is not making for itself. The cell does not understand your intentions. It understands metabolic burden, and it responds the way three billion years of evolution taught it to respond: by shedding what is not useful. Mutations that silence your circuit are positively selected. The better your circuit works, the more metabolic load it imposes, and the faster evolution dismantles it.
The field has tried everything. Orthogonal ribosomes. Genetic firewalls. Kill switches that destroy cells which lose the circuit. Each solution adds complexity, and each layer of complexity adds new failure modes. The fundamental problem remains: you are trying to run a computation inside something that is already computing, furiously, on its own behalf. Every resource your circuit uses is a resource stolen from the cell’s own imperatives. You are not programming the cell. You are negotiating with it, and it has three billion years of leverage.
Kate Adamala’s laboratory at the University of Minnesota took the opposite approach. Instead of trying to make living cells tolerate foreign computation, she asked: what if the chassis were never alive in the first place?
Her lab builds synthetic cells—lipid vesicles containing only the molecular machinery needed for a specific computation. No genome. No metabolism. No replication. No evolution. The vesicles are biological in their components but have never been alive by any reasonable definition. They are, in Adamala’s phrase, “dead as a doornail.”
The TRUMPET system—Transcriptional RNA Universal Multi-Purpose GatE fToolkit—is how these dead vessels compute. The logic gates are built from single-stranded DNA templates, each containing three functional regions: a T7 RNA polymerase promoter at one end, a logic-gate region in the middle, and an RNA aptamer sequence at the other end. The aptamer, when transcribed, folds into a structure that fluoresces. Glow means output 1. No glow means output 0.
The inputs are short oligonucleotides—single-stranded DNA fragments that are complementary to specific sections of the logic-gate region. When an oligonucleotide binds, it makes that section double-stranded. And here is where the logic happens: a restriction enzyme is present that can only cut double-stranded DNA. If both inputs bind—making the entire gate region double-stranded—the restriction enzyme cleaves the template in two. The promoter separates from the aptamer. Transcription cannot proceed. No fluorescence. Gate output: 0.
If only one input binds, or neither, the template remains at least partially single-stranded. The restriction enzyme cannot cut. T7 RNA polymerase transcribes the full template, the aptamer folds, and the readout glows green. Gate output: 1. Two inputs on, output off. One or zero inputs on, output on. This is a NAND gate—the universal gate from which any Boolean function can be composed.
| A | B | Out |
|---|
The deeper principle here is not about DNA or enzymes. It is about the relationship between autonomy and reliability.
Adamala’s synthetic cells cannot replicate. They cannot evolve. They cannot self-repair. From the perspective of the living, these are devastating limitations. From the perspective of computation, they are the entire point. Non-replication means non-evolution means non-drift. The gate you designed on Monday is the gate you have on Friday. And next year. The TRUMPET circuits have been shown to remain stable for three years on paper at room temperature. Try that with E. coli.
Every capability removed is fidelity gained. Strip away replication and you eliminate mutation. Strip away metabolism and you eliminate resource competition. Strip away the cell’s evolutionary imperatives and you eliminate the pressure that dismantles foreign circuits. What remains is a chassis that does exactly what you designed and nothing else—not because it chooses to cooperate, but because it lacks the machinery to defect.
This is a trade-off that nature never makes. Living systems are robust, adaptive, self-healing, self-replicating—and precisely because of those properties, they are unreliable hosts for externally imposed computation. The robustness is the problem. A system that can adapt will adapt away from your design. A system that can evolve will evolve to shed your burden. The only way to get a biological chassis that faithfully executes your program is to build one that cannot do anything else.
The Build-a-Cell community—the international consortium working toward creating synthetic cells from scratch—has recognized this as a design principle, not merely a limitation. The goal is not to approximate life. The goal is to build something better than life at the specific task of hosting computation. Better because it is simpler. More reliable because it is less capable. More programmable because it is less autonomous.
There is a cautionary thread woven through all of this, and it matters enough to trace.
Adamala’s lab has the expertise to build mirror-image cells—synthetic biology using the opposite chirality of amino acids and sugars from what natural life employs. All terrestrial biology is built from L-amino acids and D-sugars. Mirror life would use D-amino acids and L-sugars. Chemically equivalent. Functionally identical. And completely invisible to the existing biosphere.
No predator would recognize mirror-life organisms as food. No immune system would detect them as foreign. No bacteriophage could infect them. No protease could digest them. Mirror life would exist in a parallel biochemical universe, occupying the same physical space as natural life but interacting with none of its regulatory mechanisms. Nothing would eat it. Nothing would control it. Nothing would stop it.
The Adamala lab chose not to build it. So did other groups with the capability. The scientific community self-organized a moratorium on mirror-life research that could produce replicating organisms—not because the science is impossible, but because the risk calculus is straightforward. An organism that grows unchecked because the biosphere cannot recognize it is not a tool. It is a catastrophe with no off switch.
This restraint is itself a form of computation—the calculation of what not to build. The same principle that makes dead synthetic cells reliable makes mirror life terrifying: removal from biological regulation. In one case, you remove autonomy from the chassis, and the result is fidelity. In the other, you remove the chassis from the biosphere’s immune system, and the result is existential risk. The variable is not the technology. The variable is what you subtract and from what.
We assumed, for a long time, that life and computation were natural allies. That the extraordinary computational power of biological systems meant biology was the ideal substrate for engineered computation. That the path forward was to harness life’s capabilities—its adaptability, its self-repair, its molecular precision—and redirect them toward our purposes.
The evidence points the other way. Life does not enable externally imposed computation. Life constrains it. The very properties that make living systems magnificent—replication, evolution, adaptation—are the properties that make them hostile to circuits designed by anyone other than natural selection. The cell is not your ally. The cell is a three-billion-year-old optimizer that will convert your carefully designed logic gate into raw materials for its own reproduction the moment the selective pressure permits.
The most reliable biological computer, it turns out, is one that was never alive at all. A vessel of lipids and enzymes and DNA that computes with perfect fidelity precisely because it cannot grow, cannot change, cannot want anything. Dead as a doornail. And better for it.