To Make a Plant Enzyme Work in a Bacterium, Change Almost Everything

`★ Insight ─────────────────────────────────────` The piece leans on a "hushed register" budget — `gently`, `almost`, `quiet`, `honest` — that recurs once per paragraph. That cadence repeating is itself the tell, so the fix is mostly deletion plus rationing `gentle` to a single use. `─────────────────────────────────────────────────`

Plants are better chemists than we are. Over millions of years they evolved to build pigments, medicines, and flavors that we still struggle to make in a factory. So for decades biologists have wanted to steal the recipe: take the protein a plant uses to perform some bit of chemistry, drop it into common gut bacteria, and let those bacteria churn it out cheaply by the vatful. The trouble is that a plant's proteins are finicky tourists in a bacterial cell. A protein is a long chain that has to crumple into one precise three-dimensional shape to work at all, and in a foreign cell these chains crumple wrong. They clump together and quit. The standard fix has always been to intervene as little as possible: change a few of the protein's building blocks, add a helper, coax the original to behave. The instinct is reverence. This thing took evolution ages to perfect, so touch it gently.

A team of researchers tried the opposite. They used an AI tool called ProteinMPNN, built to rewrite a protein's sequence while keeping its three-dimensional skeleton intact. The structure comes first; the AI reimagines everything else around it. So they asked it a colder question: what is the smallest part of this thing that actually has to stay the same? The answer was startling: next to nothing. The parts that truly mattered were the tiny pocket where the protein grips the molecule it works on, a few struts that hold that pocket in its precise shape, and the handful of positions that evolution has kept identical across distantly related species. Everything else, the overwhelming bulk of the protein, was open for rewriting. So they rewrote it. Not adjusted, rewrote. The new version barely resembled the plant's original.

It worked far better. The redesigned proteins crumpled into their working shape cleanly inside bacteria instead of into the wrong one, stayed dissolved, and did their chemistry at levels the conventional fixes had never reached. In the big stirred tanks used for industrial production, two redesigned proteins each made a target compound at commercially serious levels, between two and four grams per liter of broth depending on the variant. The product here is what chemists call a glucoside, which only means a molecule with a sugar stuck onto it; you can picture the protein as a molecular stapler whose single job is to fasten that sugar in place.

The implication is the unsettling part. If you can throw away most of a protein's sequence and have it work better, then most of that sequence was never doing the real job. It was evolutionary luggage, fine-tuning for life inside a plant cell and useless, even harmful, anywhere else. The "identity" of the protein, the long string of building blocks we spent years trying to preserve intact, turns out to be mostly noise around a small functional core.

The limit is that this was shown on two plant proteins. Two. Whether it holds across the staggering variety of life's enzymes is simply unknown. But the possibility is worth sitting with. If it does hold, we have spent decades guarding sequences that were never load-bearing, treating the packaging as the gift.