University of Warsaw researchers took what looked like an artificial intelligence hallucination seriously, and it led them to a previously unknown protein family. In the prestigious journal PNAS, they describe how these proteins form a knot so complex that scientists had not believed one could tie itself.
Proteins are built from long chains of amino acids. To perform their function, such as acting as a receptor to which a specific molecule attaches, they must take on a particular shape. During this folding process, parts of the protein’s structure wrap around one another, forming a knot. Tools from topology, a branch of mathematics, can describe these knots and find similarities among proteins.
Knots can be characterized by their complexity, meaning the number of crossings they have when simplified as much as possible and drawn on a flat sheet of paper. A crossing is a place where one piece of string passes under another. The simplest knot, the trefoil knot known as 3_1 (the Celtic braid), has three crossings. The 4_1 knot, the figure-eight knot popular in climbing and sailing, has four crossings when drawn on paper.
Now a team led by Prof. Joanna Sułkowska of the Centre of New Technologies at the University of Warsaw has discovered the most complex protein knot observed to date: the 8_3 knot, which has as many as eight crossings. In theory, 21 different knots with eight crossings can be drawn. Topology assigns them consecutive numbers, and the one observed in proteins is known to topologists as 8_3, or the eight-crossing knot number 3.
The Polish team named the new protein family EIGHT fold (Extraordinarily Intricate Globular Helical Tangled fold).
“The occurrence of this particular knot in proteins was predicted by artificial intelligence, the AlphaFold program (its creators were awarded the Nobel Prize in chemistry in 2024). At first it seemed to us like a hallucination by the model, because according to the existing theory of protein folding, such a fold had no right to exist. I even wrote in a publication once that this knot cannot exist in proteins because it is too difficult to tie,” Prof. Joanna Sułkowska told PAP. It turns out that evolution found a way.
Despite the apparent absurdity of the structure proposed by the AI, the researchers took the risk of looking for such proteins in nature. Preliminary analyses indicated that proteins from the knotted family are found in various species of anaerobic bacteria living in the human gut. The researchers found those bacteria, produced the characteristic protein from them in quantity and determined its structure experimentally. After several years of searching, the knot turned out to be exactly as the AI model had predicted.
Sułkowska explained that the idea intrigued her because the AI program proposed the new structure on its own, even though its training had not taught it that such complex arrangements were possible. On the contrary, the theories of the time forbade it from constructing such forms.
Until now, it was assumed that proteins could contain only knots that can be tied through a so-called single threading. Forming the 8_3 knot, however, required the end of the chain to be threaded twice. Nature manages that, too.
How this knot forms in proteins is not yet known. With a double threading, the process must be complex, and the pace at which proteins are “woven” in ribosomes may change so that the knot has time to form. The question remains unresolved, but if humans want to learn to synthesize any protein, they must be able to recreate those conditions.
So far, only the process of unknotting the protein has been studied. “Our research shows that this protein has the most complex unfolding process described in the literature so far. It consists of as many as six phases and takes 72 hours. Despite such complicated topology, this time is within the norm,” Sułkowska noted.
Knotted proteins are in the minority, making up only 0.4% of known protein types. That may not sound like much, but with hundreds of millions of proteins, it means as many as 600,000 protein types contain at least one knot. Sułkowska added that without the ability to synthesize such proteins, organisms on Earth would not be able to cope, because in every organism they perform important functions.
The research question is not only how proteins tie knots, but also why most proteins have no knots at all.
In everyday life, disordered structures form on their own. Long hair tangles without any effort on our part, and charger cables tossed into a backpack end up knotted without trying. “Evolution had to do an enormous amount of work to avoid forming knots in 99.6% of protein types, and to create tangles only where they serve a specific, unique role. How is that possible? We are looking into it in our research,” Sułkowska concluded.
Ludwika Tomala (PAP)
