It all began when the scientists were studying a computer model of the structure of S-adenosylmethionine synthetase (MAT), an enzyme that helps make a protein that contributes methyl groups during many biochemical reactions. Proteins are apt to glom up like loose yarn clamped in a fist, but the structure of this one also seemed to contain a knot.

Knot sightings have been claimed before in proteins. Princeton University chemist Kurt Mislow and colleague Chengzhi Liang last year reported a half-dozen knot structures from an analysis of a large database of protein structures. But the Kansas researchers argue that Mislow's structures are only "pseudolinks" and "pseudoknots" because they include bonds with metal atoms (as in metalloproteins) or sulfur-sulfur linkages that are not strictly part of the protein's backbone.

So whose knots really are knots? The mathematical definition of a knot is a closed curve that never intersects itself, Mislow points out. His structures fit that definition, but the newly reported MAT structure does not. "Novelty is lacking here," he says. "End of story."

Takusagawa says the MAT structure's knot is more like what you get when you form a loop with a string and then pull one end through the loop-except it's at the end rather than in the middle. "Ours is more of a common knot," he says.