One Enzyme Learned to Make the Molecule Chemistry Said Shouldn't Win

Biochemists had long treated one particular boundary as a settled fact of nature. Flavin-dependent enzymes, the ones that use a vitamin-derived cofactor to wrench electrons off their substrates, split into two camps a long time ago. One camp, the amine oxidases, strips hydrogen from amines. The other, the Baeyer-Villiger monooxygenases, slips an oxygen atom into a ketone and rearranges it into an ester. The two chemistries were thought to be mechanistically incompatible. Run them in the same active site and they would interfere with each other. The evolutionary record seemed to agree: the two families sit on separate branches of the tree, phylogenetically segregated, as if life had tried the combination, found it wanting, and filed the functions far apart for good.

There was, in principle, a way to test whether the separation was truly load-bearing. Trace both family trees back far enough and you reach ancestral proteins from before the split, scaffolds that had not yet committed to either job. Using ancestral sequence reconstruction, the authors resurrected one, AncFO-221, and it could indeed run both reactions in sequence, carrying an amine all the way to a lactone. Barely. The yield was 4 percent. A proof of principle, not an enzyme.

Then came the engineering. Round after round of it, reshaping the cofactor pocket, tuning the electrostatics around the substrate, disrupting a catalytic triad, each move guided by AlphaFold3 models, molecular dynamics, and quantum calculations. The endpoint was a variant called M15, and it converted amine to lactone at 93 percent yield. The side reaction that had dominated the ancestor, reducing the intermediate to an alcohol, all but vanished: a product ratio that started at 20:80 against the desired lactone flipped to better than 96:4 in its favor, and to roughly 99:1 with a cofactor-regeneration system bolted on. The interference the textbooks predicted never showed up. The obstacle had only ever been engineering.

That alone would be a tidy result. But M15 did something stranger. There is a rule in organic chemistry, old and reliable, about Baeyer-Villiger reactions: when the molecule rearranges, the carbon that migrates is the one best able to feed electrons into the bond, tertiary before secondary, secondary before primary, a lone methyl group dead last. M15 ignores it. Across a wide range of substrates it prefers to migrate the methyl, the weakest migrator of all, producing "abnormal" regioisomers that natural Baeyer-Villiger enzymes rarely make. The merged active site does two enzymes' work back to back — and reaches into a region of chemical space that evolution, having optimized every natural enzyme of this kind for the normal migration, simply never went.

There are limits. The amine-oxidation step throws off hydrogen peroxide as a stoichiometric byproduct, which would have to be engineered away before this runs inside a living cell. Every number here comes from defined conditions in a tube, with cofactor regeneration supplied; what happens at scale is unknown. But the conceptual result already sits upstream of those caveats. The wall between these two oxidative families was never a necessity of chemistry. It was a contingency of history, and the space of single-enzyme cascades is wider than the family tree ever let on.