RICHLAND, Wash.-Two parallel experiments in protein self-assembly produced strikingly different results, demonstrating that protein designers should consider incorporating physical forces now missing from even Nobel-prize-winning protein design algorithms.
The results, published in Nature Communications, also demonstrate that AI and machine learning have evolved to become essential tools for analyzing complex datasets to produce meaningful results.
The multidisciplinary research team, led by a trio of researchers from the Department of Energy's Pacific Northwest National Laboratory and composed of colleagues from the Center for Science of Synthesis Across Scales (CSSAS), a DOE Energy Frontier Research Center, as well as collaborators from multiple institutions, elegantly showed that complex order arises from effects outside the designed framework.
AtomAI proves to be a game changer
The research team studied how protein "nanoribbons" designed by Nobel-prize-winning scientist David Baker and his team, participants in CSSAS at the University of Washington, assemble on mineral surfaces.
Led by co-authors Sergei Kalinin, Amy Stegmann, and Maxim Ziatdinov, the scientists applied a previously developed machine learning tool called AtomAI to analyze the images produced by atomic force microscopy. The trained model allowed the team to track where each nanorod was, how it was oriented, and how aligned the rods became over time, quantifying both orientation and organization.
"This study simply would not have been possible without the machine learning algorithm," said co-lead author James De Yoreo, a Battelle Fellow at PNNL and co-lead of CSSAS.
The original plan was simply to track how negative charges on the nanoribbons arranged themselves to match the regular lattice of positively charged potassium ions on the surface of the natural mineral mica. In principle, the lattice matching should have allowed the protein to array in any of three different directions. But that's not what happened.
Orderly arrays of ribbons, all pointing in one direction, appeared in the AI-assisted microscope analysis.
It's in the water
To better understand the unexpected organization, the researchers compared protein assembly on two types of mica. Both have identical lattices of potassium but possess other differences that change the way water arranges itself on the mica surface, leading to a hexagonal pattern of water on one and a striped pattern on the other.
Experimentally, the two surfaces produced strikingly different outcomes that resembled the arrangement of the water: On one kind of mica, the nanoribbons pointed randomly in three different directions, while on the other, they aligned in a single direction and organized into parallel rows. Thus, the results suggested that it is actually the water on the mica guiding the protein alignment.
The team then confirmed the experimental results with a computational simulation of the experiments. The simulation reproduced the experimental results when the interactions that guide the nanoribbons reflected the arrangement of the water layers rather than the underlying lattice of potassium.
"Water molecules at solid-liquid interfaces display many important behaviors, well beyond those we initially considered," said co-lead Shuai Zhang, a materials scientist at PNNL. "It does play a quite critical role in defining the materials' behaviors."
The study authors, led by De Yoreo, Zhang, and PNNL Laboratory Fellow Chris Mundy, concluded that proteins designed to assemble on surfaces must explicitly include the role of solvents. Some examples of proteins arrayed on a solid surface include industrially important catalysts, biosensors, and biomedical devices. Additional study authors included former PNNL postdoctoral scientist and study first author Sakshi Yadav Schmid, Benjamin Helfrecht, and Benjamin A. Legg from CSSAS and PNNL. The full list of authors is available here.
"Right now, there are a lot of people working on biomineralization, trying to adapt strategies by nature to form inorganic materials," said De Yoreo, who recently co-authored a review article with Baker on the subject. "They are trying to mimic the hierarchical structures that are found in biological systems that endow them with incredible properties."
De Yoreo pointed to the structure of the mantis shrimp shell as an example of a natural composite of nanofibers, proteins and minerals that achieves extreme toughness, making it a key model for bioinspired lightweight, crash-resistant materials.
In practical terms, the new study suggests that to design protein assemblies on inorganic surfaces, researchers must use physics-informed machine learning to account for solvent effects when designing proteins.
The research was supported by the DOE Basic Energy Sciences program through CSSAS, which is located at the University of Washington.
