Proteins are made of chains of amino acids, and a basic tenet of biology is that the sequence of amino acids determines how a protein folds into a 3D shape which, in turn, dictates the protein’s function. But another tenet of biology is that rules are meant to be broken.
“Many proteins have long stretches of amino acids that don't form a well-defined three-dimensional structure,” Kresten Lindorff-Larsen, Ph.D., a biophysicist at the University of Copenhagen, told Fierce Biotech in an interview. “They are very common.”
Lindorff-Larsen estimates that about one-third of human proteins contain these so-called disordered regions.
Computational biologist Francesco Pesce, Ph.D., working with Lindorff-Larsen and others, has created a method to design disordered proteins with specific biologically relevant properties. This advance teaches us more about the intricacies of protein folding and could allow us to build new therapeutics. The technique was published in Science Advances on Aug. 28.
“I think this is an important and exciting study,” Alex Holehouse, Ph.D., told Fierce in an email. Approaches like this and his group’s tool, called GOOSE, are designed to allow people to test whether the various forms that disordered proteins can take relate to their functions.
Though disordered proteins shift between an ensemble of numerous shapes, they aren’t completely chaotic. Lindorff-Larsen said they’re like spaghetti—they “take on floppy structures, but they are not randomly floppy.”
These regions have been implicated in signaling, both within cells and between cells, and can also serve as linkers that connect other stable parts of the protein. For example, an enzyme meant to break down a complicated molecule, like a big sugar, might have a region that binds to the sugar and another region that actually breaks the sugar’s chemical bonds, with a disordered region linking the two.
“The main problem in protein design is that you need to screen several sequences in order to find the protein with the properties that you're looking for,” Pesce told Fierce in a joint interview with Lindorff-Larsen. Because disordered proteins don’t form stable structures, it takes a lot of computational power to predict what properties they have.
To improve this process, Pesce took several known disordered proteins and simulated them in a model, called CALVADOS, that Lindorff-Larsen’s lab had previously developed. The team then shifted single amino acids at a time to see whether they could make the resulting protein more compact—meaning a shorter distance between functional parts of the protein that are linked by the disordered region. Because assessing compactness using modeling is time-consuming, they instead used alchemical free energy as a proxy.
Free energy is directly related to compactness but can be calculated from the amino acid sequence of a protein alone, circumventing the need to simulate each perturbation of the protein and all of its complicated folding.
By systematically moving amino acids around and calculating the resulting protein’s free energy, the model was able to create versions of the proteins that were more compact. The algorithm tended to create compact proteins over time by clustering the positively charged amino acids near one end of the protein and the negatively charged amino acids on the other.
“We didn't tell the algorithm this,” Lindorff-Larsen said. “That really tells us something about how nature operates and how natural disordered proteins behave.” The process, in a sense, mirrors the process of evolution by natural selection.
Focusing on one disordered protein called A1-LCD, Pesce genetically modified bacteria to produce real versions of five more compact variants produced by the algorithm. Measuring the compactness of these variants showed that the model had gotten it right—they were more compact.
“We chose to work specifically on A1 because it was the protein that gave us the results that we weren't expecting,” Pesce said. A1 was the most compact of all the proteins they started with, so they thought the algorithm would be able to make expanded variants; instead, it gave them versions of the protein that were even more compact. “It felt like the most surprising case, so when we went to the lab and we got the results, it was actually super nice to see.”
Lindorff-Larsen said the approach is very general and could allow protein designers to choose other properties they want in a protein and then let the algorithm loose to find an amino acid sequence that fits the bill. This could open the door for new kinds of therapeutic proteins, like bispecific antibodies, that depend on linker regions.
“There's all these things where you need to tether things together,” he said. “Being able to design the tethers in a specific way, I think, adds an extra tool in the toolbox of design.”
Disordered proteins could also be made that target other, clinically relevant disordered proteins, Pesce said.
“Disordered proteins and regions still are considered to be undruggable from a certain point of view,” he said. Some disordered regions bind strongly to other disordered regions, a fact that could be exploited to make new drugs.