Columbia redesigns E. coli proteins
- Columbia researchers and collaborators reported in Science that they rebuilt E. coli ribosomal proteins to remove every isoleucine, creating cells with core machinery running on 19 amino acids. - Simple valine-or-leucine swaps worked in only about 43% of tested proteins, but AI-guided redesign pushed all 52 essential ribosomal proteins above 90% relative fitness. - It matters because protein AI is doing more than prediction now—it is starting to rewrite living cells’ biochemical rules. (science.org)
Proteins are built from a 20-letter chemical alphabet. Every known free-living organism uses the same 20 amino acids, and biology has looked pretty locked-in on that point. But a Columbia-led team just showed that one of those letters is not as untouchable as it seemed. They redesigned core E. coli proteins so the bacteria could run essential ribosome machinery without isoleucine — one of the standard 20 amino acids. ### What actually changed? The team did not make a totally “19-amino-acid organism” in one shot. (science.org) The narrower, more important move was this: they took E. coli’s ribosome — the machine that builds proteins — and redesigned all 52 essential ribosomal proteins so none of them contained isoleucine. Those redesigned versions still worked inside living cells, which is the real milestone. ### Why pick isoleucine? Isoleucine was a plausible target because it often behaves a lot like valine and leucine — two chemically similar amino acids. (science.org) Evolutionary analyses had already hinted that isoleucine is one of the less conserved amino acids, meaning biology sometimes swaps it out more easily than others. So if you were going to try deleting one “letter” from the alphabet, this was a reasonable place to start. ### Why wasn’t simple swapping enough? Because proteins are not just strings of letters. (science.org) They are folded 3D objects, and one residue change can ripple through the whole structure. The team first tried the blunt approach — replace every isoleucine with valine or leucine in 39 essential or highly expressed proteins. Only about 43% of those variants still functioned in vivo. Basically, chemistry similarity alone was not enough. ### So what did the AI actually do? The models proposed smarter replacements than one-for-one swaps. (science.org) The group used sequence-based models including ESM2 and MSA Transformer, plus structure-based tools including ProteinMPNN and AlphaFold2, to generate new isoleucine-free protein variants that could preserve fold and function. That matters because the job was not “predict this protein.” It was “redesign this protein so the cell still lives.” ### Why focus on the ribosome first? Because the ribosome is one of the most conserved and essential complexes in the cell. (science.org) If you can rewrite that machinery and keep the bacterium viable, you are testing the hardest version of the idea. It is a bit like changing parts inside an engine while the car is still driving — if the engine keeps running, the redesign is probably real. ### How good were the redesigned cells? Pretty good, which is why this result landed in Science. Using iterative design-build-test cycles, the researchers replaced all isoleucines in each ribosomal protein — along with extra compensatory mutations where needed — and kept relative cellular fitness above 90% of wild type. (science.org) That is not “barely alive.” That is close enough to normal growth to show the proteins are genuinely doing their jobs. ### Does this mean all of E. coli now uses 19 amino acids? Not yet. The current result is about core ribosomal proteins, not the whole proteome. (science.org) Nature’s coverage makes the point clearly: the cells run key machinery on 19 amino acids, which is different from proving that every protein in a free-living cell can do without isoleucine. The catch is that moving from one essential complex to an entire organism is a much bigger systems-engineering problem. ### Why does this matter beyond a neat stunt? Because it pushes protein AI from analysis into intervention. (science.org) Researchers have used these models to predict structures and suggest designs before, but here the output was wired into a living cell and selected by survival. That opens two doors at once — basic science about why life settled on 20 amino acids, and synthetic biology that could build organisms with simplified or altered chemistries. ### Bottom line? The big idea is not just “AI designed a protein.” It is that AI-guided redesign changed a fundamental rule inside living bacteria — at least for one of the cell’s most important machines. (nature.com) That is a meaningful step toward organisms whose chemistry is engineered, not just inherited. (science.org)
