NOTES.

Metamorphic proteins make up a small, odd corner of the protein universe, and that may be exactly why they matter. Structure predictors learned mostly from static crystal and cryo-EM models, so they tend to return a single confident shape.

For a protein with two native states, one confident shape is half the truth. If fold-switching is more common than the databases suggest, some of our confident answers are incomplete in a way the confidence score cannot tell us. Researchers have coaxed both of KaiB's states out of AlphaFold2 by clustering its input sequences, but that takes deliberate effort.

In tryptophan synthetase, Charles Yanofsky's group found that a damaging amino acid change could be repaired by a second change somewhere else in the protein. The two positions were not neighbors in the sequence.

Protein modelers now mine that same logic statistically, across millions of sequences. In 1964 it took one enzyme, a lot of bacteria and a great deal of patience.

Christian Anfinsen unfolded ribonuclease A with urea and a reducing agent, then removed them by dialysis. The enzyme slowly refolded and regained its activity with no help from the cell.

The experiment fits in a paragraph. It also set the terms for the next half century of structural biology, because it implied that the information for a fold is already in the sequence.