According to the first author of the study, one of the research’s continuity paths is to understand how changes in protein dynamics contribute to the onset of human diseases (image: Shutterstock)
Published on 08/10/2026
By Luciana Constantino | Agência FAPESP – Researchers have developed a groundbreaking technique that can simultaneously map, on a large scale, how thousands of proteins change shape over time. The results, published in the journal Nature, reveal that proteins with similar structures can exhibit completely different dynamic behaviors.
Moreover, these changes do not occur only at isolated points on the molecule; entire sections can unfold, relax, or become temporarily less stable. This insight improves our understanding of how these molecules function in the body. Thus, this study paves the way for training artificial intelligence models that can predict not only the shape of proteins, but also how they “fold” (i.e., coil and assume their three-dimensional form) and change.
Traditionally, studies have focused on determining the static three-dimensional structure of a protein, obtaining a sort of “snapshot.” However, the molecule’s function depends directly on how it moves and transitions between different states over time. Experimentally capturing these dynamic fluctuations is a slow, costly process generally limited to analyzing one or a few proteins per study.
Instead of analyzing one molecule at a time, the scientists developed a strategy combining hydrogen-deuterium exchange (a process in which hydrogen atoms in a protein are replaced by those from a special liquid in which the protein is immersed) and multiplexed mass spectrometry (mHDX-MS). This advanced technique precisely measures variations in molecular weight to track protein movements. This innovative approach made it possible to map the modifications and energy barriers (i.e., the energy required for the molecule to change shape) of 5,778 protein domains (i.e., smaller modules or “building blocks” of the molecule), ranging in length from 28 to 64 amino acids and belonging to ten distinct protein families, under identical experimental conditions.
“We developed a multiplexed experimental strategy that can map the energy landscape [the different energy states and conformations] of hundreds of these molecules in a single experiment. We’re observing a fundamental aspect of structural biology that was previously difficult to access: the various conformations that proteins can adopt beyond their most common structure. We now have a set of tools to create a massive database of protein dynamics and, with it, train models capable of predicting those fluctuations directly from the amino acid sequence. That’s important for understanding, for example, the molecular mechanism by which point mutations subtly alter local stability and trigger disease phenotypes [symptoms and clinical manifestations], without necessarily denaturing the entire protein,” Professor Allan Ferrari of the Department of Pharmacology at Northwestern Medicine in Chicago, United States, and first author of the article, explained to Agência FAPESP.
Illustration of a protein changing its shape from compact or coil-like (left) to unfolded (credit: Allan Ferrari)
Ferrari began the study while still a postdoctoral fellow at the Institute of Chemistry at the State University of Campinas (UNICAMP) in Brazil, with support from FAPESP, having worked for more than five years in the laboratory of the study’s corresponding author, researcher Gabriel Rocklin.
To illustrate how these changes impact human health, Ferrari cites lysozyme, a defensive enzyme (protein) found in secretions such as tears and saliva. Lysozyme’s main function is to break the chemical bonds that form bacterial cell walls, thereby protecting the body. However, mutations can make lysozyme less stable, leaving parts of the protein exposed and allowing them to stick together. This process generates abnormal aggregates that accumulate in the liver and kidneys, progressively impairing their function and potentially causing organ failure.
The three-dimensional structures of the two forms of the protein – the original (unmutated) and the mutant – are virtually identical when analyzed using traditional methods of structural biology. However, mapping the dynamics reveals that the mutation facilitates the local opening of the protein’s structure.
“To map these transient conformational nuances, traditional structural biology relies on techniques such as protein crystallography, X-ray diffraction analysis, nuclear magnetic resonance, or cryo-electron microscopy. However, what researchers often want to observe is invisible to these methodologies. Furthermore, they require each system to be analyzed individually, which takes a long time to characterize just a few variants – in contrast to the scale achieved by the mHDX-MS platform,” Ferrari adds.
The importance of databases
For over 50 years, the international scientific community has deposited experimentally resolved three-dimensional structures in the Protein Data Bank.
Combined with billions of known sequences, this public database enabled the creation of AlphaFold in 2018. AlphaFold is an artificial intelligence model that helped solve the complex problem of protein folding, thereby revolutionizing the field. AlphaFold can predict the three-dimensional structure of biomolecules, such as proteins, DNA, and RNA, based on their genetic sequence. This innovation was recognized with the 2024 Nobel Prize in Chemistry, awarded to John Jumper and Demis Hassabis of Google DeepMind in London for developing the tool and to David Baker of the University of Washington for his work in computational protein design.
Although AlphaFold can predict a protein’s static “selfie” based on its amino acid sequence, it still faces limitations in capturing the molecule’s dynamic fluctuations over time – a sort of “movie.” The problem is that there was no large-scale database capable of describing these dynamics systematically.
“There’s an unlimited number of possible combinations of amino acids [organic molecules that bind together to form proteins]. If you’re designing a new drug or a new sensor for biotechnology, what’s the best amino acid sequence for a specific function, or how will modifications to that sequence alter the protein’s behavior? The mHDX-MS approach now allows us to examine these conformational fluctuations for thousands of different protein sequences,” Rocklin explained in a press release.
The mHDX-MS approach resulted in the first publicly available experimental database of large-scale protein dynamics. This repository can be used to train and refine the next generation of machine learning models, teaching them to predict complete energy landscapes directly from amino acid sequences.
In addition to mapping the dynamics of thousands of natural and designed structures, the researchers demonstrated the application of the platform in protein engineering. Using the experimental data generated by the technique, the group identified two mutations that could stabilize a naturally flexible region of one of the proteins characterized in the study. Subsequent experiments confirmed that the modifications produced the expected effect.
To understand the method
The technique involves immersing the protein molecule in a special water-based solvent that contains deuterium, a heavier form of hydrogen. When the molecule comes into contact with this liquid, a natural substitution occurs, and the hydrogen atoms in the protein are replaced by those from the solvent.
This process acts as a tracer. Parts of the protein hidden within its structure (“closed”) remain protected and exchange atoms more slowly. In contrast, regions that move or are on the surface (“open”) rapidly exchange atoms, gaining mass. By measuring this exchange rate under identical conditions, the scientists can precisely map where and how the molecule fluctuates in space and calculate the energies associated with the local movements of each protein. This is impossible to detect using traditional structural biology techniques.
As a result, it was observed that changes in the amino acid sequence alone are sufficient to significantly alter the dynamics and stability of these molecules. In many cases, proteins from the same structural family exhibited more distinct dynamic behaviors than those from different families.
Regarding structural oscillations, the study showed that among the most dynamic proteins, entire structural blocks (such as alpha-helices and beta-sheets) are less stable and unfold more easily together.
Despite these advances and the ability to predict the three-dimensional structure of molecules with high precision, the scientists note that understanding how protein sequences determine transitions between states remains a challenge in molecular biophysics.
According to Ferrari, his research is proceeding along two paths. One path involves expanding data generation for specific families and exploring how much can be “learned” about the dynamics of these proteins using a large dataset. The other path involves understanding how changes in protein dynamics contribute to the onset of human diseases.
The article “Large-scale discovery, analysis, and design of protein energy landscapes” can be read at www.nature.com/articles/s41586-026-10465-z.
Source: https://agencia.fapesp.br/58927
