🧬 biology

Dynamic Network Context Predicts Mutation-Induced Reshaping of Protein Conformational Free-Energy Landscapes

This study introduces the ICed-ENM framework, which utilizes structure-derived dynamic sensitivity scores to successfully predict how missense mutations reshape protein conformational landscapes and distinguish between activity-selective and abundance-related functional losses.

Original authors: Laura Orellana, Byung Ho Lee, Domenico Scaramozzino, Serena Piticchio

Published 2026-11-05

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Laura Orellana, Byung Ho Lee, Domenico Scaramozzino, Serena Piticchio

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Proteins are the workhorses of life, tiny molecular machines that fold into specific shapes to perform tasks like building cells, fighting infections, or sending signals. For a protein to work, it must not only have the right shape but also the right flexibility. Imagine a protein not as a rigid statue, but as a living structure that constantly jiggles, breathes, and shifts between different poses. These movements are essential; they allow the protein to open and close like a hinge, grab onto other molecules, or change its form to do its job. When a single building block, known as an amino acid, is swapped for another in the protein's chain, it can disrupt these delicate movements. Sometimes the swap is harmless, but other times it can jam the machine, causing it to malfunction and leading to diseases ranging from cancer to neurodegenerative disorders.

The challenge for scientists has long been predicting which swaps will cause trouble. Traditional methods often look at the protein's static shape, like a photograph, to guess what might go wrong. However, a single snapshot cannot capture the dynamic dance of movement that defines how a protein actually works. While powerful computer simulations can track these movements, they are incredibly slow and expensive to run for every possible mutation. Researchers need a faster way to see how a change in one spot ripples through the entire structure, altering the protein's internal rhythm and potentially breaking its function.

A team of researchers at the Karolinska Institute in Sweden has developed a new approach to solve this problem. They created a computational framework that treats the protein as a network of connected points, similar to a springy mesh, but with a crucial refinement. Instead of just looking at how far apart points are in space, their model accounts for the specific angles and bonds that hold the protein together, preserving the realistic geometry of the molecule. By analyzing how this network vibrates, the team can calculate how much a single amino acid swap disturbs the protein's natural motion. They call this disturbance a change in vibrational entropy, a measure of how much the protein's flexibility is altered.

The researchers tested their method by simulating thousands of mutations across different proteins. They found that the model could identify "hot spots"—specific locations where a change is likely to cause a major disruption to the protein's movement. To verify this, they selected a few of these high-risk spots and a few low-risk spots from three different proteins and ran detailed, high-speed computer simulations to see what actually happened. The results were clear: when they swapped amino acids at the high-risk spots identified by their model, the protein's movement patterns changed significantly. The protein explored different shapes and shifted its balance in ways that the low-risk swaps did not. In some cases, the mutations caused the protein to sample new shapes that were far from its normal working positions, effectively reshaping the landscape of its possible movements.

Beyond just identifying where mutations might cause trouble, the team looked at the patterns behind these disruptions. They discovered that the impact of a mutation depends heavily on the physical properties of the amino acids involved. Swapping a small amino acid for a very large one, or changing a charged part of the protein for a neutral one, was more likely to cause a major disturbance. Interestingly, the model showed that these disruptive changes were not limited to the tightly packed cores of the protein. Many of the most sensitive spots were actually in the looser, more flexible loops on the surface, suggesting that even areas that look unimportant can be critical for the protein's overall rhythm.

To see if this dynamic sensitivity mattered in a real-world disease context, the researchers applied their method to PTEN, a protein that acts as a tumor suppressor. They compared their predictions against a massive dataset of human mutations that had already been tested in the lab. The goal was to distinguish between two types of failure: mutations that simply make the protein disappear (reducing its abundance) and mutations that make the protein stay present but stop working (reducing its activity). This distinction is vital because a drug might be able to fix one type of problem but not the other. The researchers found that their model was remarkably good at spotting the mutations that caused the protein to lose its function while remaining present in the cell. It outperformed other existing methods that focused on stability or general disease risk, successfully separating the "broken but present" mutations from those that simply caused the protein to vanish.

The study suggests that the way a protein moves is a key factor in determining whether a genetic mutation will cause disease. By focusing on the dynamic network of the protein rather than just its static shape, this new framework offers a faster, more accurate way to predict the impact of genetic changes. While the model is not perfect and cannot yet predict every single outcome, it provides a powerful tool for prioritizing which mutations deserve further study. It shifts the focus from simply asking if a protein is stable to asking if it can still move the way it needs to, offering a clearer window into the mechanical causes of genetic disease.

Technical Summary: Dynamic Network Context Predicts Mutation-Induced Reshaping of Protein Conformational Free-Energy Landscapes

Problem Statement
Missense mutations can profoundly alter protein function by reshaping conformational free-energy landscapes, yet predicting these effects at scale remains a significant challenge. While Molecular Dynamics (MD) simulations offer high-resolution insights into mutation-induced dynamic changes, their routine application is hindered by substantial computational costs and dependence on force-field accuracy. Conversely, coarse-grained models like Elastic Network Models (ENMs) provide computational efficiency but often lack the physical realism required to distinguish subtle, mutation-specific dynamic perturbations from artifacts. A critical gap exists in determining whether dynamic perturbations predicted by coarse-grained models correspond to actual reshaping of conformational landscapes and whether these predictions correlate with experimentally observed functional consequences, particularly when mutations occur far from active sites.

Methodology
The authors developed a mutation-impact framework centered on an internal-coordinate, essential-dynamics-refined elastic network model (ICed-ENM). The methodology proceeds through several key stages:

Model Development and Parameterization: ICed-ENM represents proteins as networks of nodes (backbone heavy atoms and side-chain centers of mass) connected by harmonic springs. Unlike Cartesian-coordinate models, it operates in internal coordinates to preserve covalent geometry. The model's interaction parameters were optimized against effective stiffness patterns derived from extensive all-atom MD simulations (using AMBER, CHARMM, and OPLS force fields) to ensure the model captures realistic collective motions.
Mutation-Scanning and Scoring: The framework performs systematic in silico alanine scanning. For each substitution, the absolute change in vibrational entropy ( $|\Delta S|$ ) is calculated to quantify the substitution-level dynamic perturbation. These scores are then aggregated across substitutions and spatially smoothed using a Gaussian kernel over neighboring residues to generate a position-level dynamic sensitivity score ( $\Omega$ ). This score reflects the dynamic sensitivity of a residue within its surrounding structural network.
Validation via MD Landscapes: To test if high-sensitivity sites truly reshape conformational landscapes, the authors selected three conformationally diverse proteins (RBP, 5′NTase, and SERCA). They identified "hot spots" (high $\Omega$ ) and "cold spots" (low $\Omega$ ) with comparable local structural contexts. Alanine substitutions were introduced, and all-atom MD trajectories were projected onto experimentally derived Principal Component Analysis (PCA) subspaces to construct free-energy landscapes.
Functional Benchmarking: The framework was evaluated against a curated dataset of PTEN variants, distinguishing between "activity-selective loss" (reduced phosphatase activity with preserved abundance) and "abundance-related loss" (reduced abundance). Performance was compared against ENCoM, DynaMut2, and Rhapsody-2(dyn).

Key Results

Physical Realism: ICed-ENM demonstrated improved agreement with experimental conformational transitions compared to standard ENMs (ANM, edENM) and the internal-coordinate model iMOD. It achieved higher overlap scores ( $O_1$ ) with experimental transition vectors and concentrated transition-relevant motion in lower-frequency modes. Crucially, ICed-ENM preserved stereochemical integrity (backbone geometry and contact networks) significantly better than Cartesian-based models during mode deformation. While the improvement relative to iMOD was moderate, consistent with both methods sharing the fundamental advantages of internal-coordinate representations, the benchmark established ICed-ENM as a physically grounded dynamic reference.
Conformational Landscape Reshaping: In the three validation systems, alanine substitutions at predicted high- $\Omega$ sites produced stronger redistribution of conformational sampling in MD-derived landscapes compared to low- $\Omega$ controls. For instance, in RBP and SERCA, high- $\Omega$ mutants shifted dominant free-energy basins or broadened sampling, whereas low- $\Omega$ mutants largely retained the wild-type landscape. Notably, these redistributions did not always align with canonical functional transitions, indicating that $\Omega$ identifies sites capable of perturbing dynamics rather than predicting specific functional directions.
Structural and Physicochemical Correlates: Analysis of over 560,000 modeled substitutions revealed that high $|\Delta S|$ scores are systematically associated with large-magnitude side-chain volume changes and specific physicochemical transitions (e.g., positive-to-hydrophobic). Position-level $\Omega$ scores showed a tendency toward higher solvent accessibility and coil secondary structures, though high sensitivity was observed across all structural classes.
Functional Benchmarking (PTEN): In the PTEN benchmark, the position-level $\Omega$ score showed the strongest observed preference for activity-selective positions. ICed-ENM achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.745, significantly higher than ENCoM-derived $\Omega$ (0.530). The comparison with Rhapsody-2(dyn) regarding top-group association remained inconclusive, though the full-ranking AUROC of ICed-ENM was significantly higher than both comparators after Holm correction. While substitution-level $|\Delta S|$ showed limited separation, the position-level aggregation successfully prioritized residues associated with functional impairment despite preserved protein abundance.

Significance and Claims
The paper claims that structure-derived dynamic sensitivity provides a functionally informative basis for prioritizing mutations. The authors demonstrate that ICed-ENM offers a computationally tractable framework that bridges the gap between coarse-grained efficiency and the physical realism needed to capture mutation-induced dynamic perturbations.

The primary significance lies in the ability of the $\Omega$ score to identify residues where substitutions alter conformational dynamics in ways that correlate with specific functional phenotypes (activity-selective loss), distinct from stability-driven effects. The authors emphasize that this approach does not merely predict pathogenicity but offers mechanistic insight into how mutations propagate through dynamic networks to reshape free-energy landscapes. The work establishes that dynamic sensitivity, derived from internal-coordinate models, is a distinct and valuable component of mutation impact analysis, complementing stability-based and evolutionary approaches. The framework is presented as a scalable tool for generating testable hypotheses regarding the role of conformational dynamics in mutation effects.

More like this