Whole-Enzyme Reactive Free-Energy Simulations with a Universal Machine Learning Interatomic Potential: Catalytic Mechanism and Mutational Effects of PETase

Universal machine learning interatomic potentials (uMLIPs) have emerged as powerful tools for atomistic simulations that achieve quantum mechanical accuracy at significantly reduced computational cost. Yet, their direct application to complex enzymatic reactions remains challenging owing to the immense size of solvated enzyme systems and the scarcity of transition-state structures in training data. In this study, we demonstrate that PFP (PreFerred Potential), a pretrained uMLIP across 96 elements in nature, simulates the complete hydrolytic cycle of polyethylene terephthalate (PET) catalyzed by PETase as a single reactive domain of approximately 20,000 atoms, without multiscale partitioning or system-specific fine-tuning. Umbrella-sampling simulations revealed stepwise mechanisms for both acylation and deacylation through distinct tetrahedral intermediates. The computed rate-limiting deacylation barrier of 20.5 ± 0.3 kcal/mol is in agreement with the range of 18.0–18.7 kcal/mol inferred from experimental turnover numbers. The PFP-based simulations further capture catalytic consequences of point mutations from loss of function (D177N, M132A) to predicted activity enhancement (D157N) and reveal that simultaneous neutralization of three distal acidic residues which are located more than 10 Å from the active site raises the activation barrier due to a collective long-range electrostatic effect that emerges naturally in the partition-free treatment. These findings demonstrate that the pretrained uMLIP, PFP, can provide physically plausible free-energy profiles for enzymatic reaction mechanisms and capture mutational effects, providing a powerful platform for rational enzyme design.