In this post, we highlight research on intrinsically disordered proteins published in the Journal of Chemical Theory and Computation by researchers at the University of Edinburgh and Cresset, that advances the application of alchemical free energy calculations to proteins containing intrinsically disordered regions (IDRs) — an important but underexplored frontier in computational drug discovery.
The challenge of drug resistance and intrinsically disordered proteins
Drug resistance arising from protein mutations is a major obstacle in the development of durable therapeutics across oncology, virology, and infectious disease. Alchemical free energy perturbation (FEP) calculations have emerged as powerful tools for predicting how amino acid mutations affect ligand binding affinities, but their application has largely been confined to well-folded, globular proteins. Proteins containing IDRs present a considerably harder problem: their inherent conformational flexibility makes adequate sampling difficult, and standard FEP protocols are not well established for such systems.
This work focuses on MDM2, an oncoprotein and important cancer drug target, whose N-terminal “lid” is an IDR that undergoes ligand-dependent folding. Experimental data show that mutations in this lid region can dramatically reduce the binding affinity of the clinical candidate AM-7209, with the I19G mutation alone causing a three-orders-of-magnitude increase in dissociation constant. Reproducing these effects computationally is a stringent test of any FEP protocol.
Establishing robust FEP protocols for IDRs
The research team systematically compared equilibrium (EQ) and nonequilibrium (NEQ) alchemical protocols for a panel of MDM2 lid mutations against two ligands, AM-7209 and Nutlin-3a. A key finding was that the EQ protocol substantially outperformed the NEQ approach in precision: EQ calculations achieved a ΔΔG RMSE of around 1 kcal/mol with just 30% of the total invested sampling time, while the NEQ protocol required the full sampling budget. The precision deficit of the NEQ protocol was traced to poor overlap between forward and reverse work distributions for mutations involving slowly varying degrees of freedom — a problem that is difficult and computationally expensive to remedy by simply extending switching times.
The study also demonstrated that force field choice is consequential for IDR simulations. Switching from the AMBER FF14SB/TIP3P combination to FF99SB*-ILDN paired with the OPC water model — known to produce more realistic structural ensembles for disordered proteins — reduced the overall RMSE from 0.79 to 0.39 kcal/mol, and crucially improved prediction of the experimentally large loss of AM-7209 binding affinity under the I19G mutation. The work also benchmarked the recently released Boltz-2 foundation model for this task, finding that it was unable to reproduce experimental trends for this IDR system, highlighting the continued importance of physics-based FEP methods for challenging drug resistance predictions.
OpenBioSim’s contribution
BioSimSpace 2024.2 from OpenBioSim was used to prepare all hybrid structures and topologies for the alchemical simulations, providing the interoperable layer that enabled seamless integration of AMBER force field parameterisation, GROMACS simulation engines, and downstream free energy analysis tools. This kind of workflow — where diverse software components need to work together reliably across a large simulation dataset — is precisely the use case that BioSimSpace was designed to address.
Outlook
This study establishes that accurate, reproducible FEP calculations are achievable for proteins with highly dynamic disordered regions, provided that appropriate force fields and sufficient equilibrium sampling are employed. The findings are expected to generalise beyond MDM2 to other IDR-containing drug targets, opening new possibilities for computationally guided drug resistance campaigns in this important and growing class of proteins. All structural inputs, simulation scripts, and analysis code are openly available on GitHub.
Drafted by Claude and edited by Julien Michel
