Modeling redox-dependent cysteine states in proteins using AlphaFold 3 and geometry minimization

Accurate modeling of redox-dependent cysteine chemistry remains a major challenge in protein structure prediction, particularly for states involving disulfide bond formation. Here, we present a generalizable strategy that combines AlphaFold 3 (AF3) predictions with targeted constraint engineering and geometry minimization to model oxidized cysteine states. Using human peroxiredoxin 1 and 2 (hPrdx1 and hPrdx2) as examples, we introduce a shared persulfide sulfur directly in the AF3 input file as a transient constraint to bring distant cysteine residues into proximity, thereby enabling subsequent formation of canonical disulfide bonds during refinement. This approach enables the generation of structurally coherent models of disulfide-bonded states that are otherwise inaccessible from standard AF3 predictions. Geometry minimization then converts these constrained intermediates into chemically realistic disulfide bonds while accounting for the major structural rearrangements. Reduced-state models most accurately reproduce experimentally observed active-site geometries, while persulfide-constrained intermediates can be refined into high-quality disulfide-bonded structures through iterative geometry optimization. Notably, AF3 predictions of decameric assemblies recapitulate the experimentally observed preference of peroxiredoxins for reduced states, highlighting the importance of oligomeric context in redox modeling. Importantly, this workflow is broadly applicable beyond peroxiredoxins. The transient use of a shared persulfide constraint provides a simple and adaptable framework to model disulfide bond formation and other cysteine-based modifications in diverse proteins. Together, our results establish a practical computational approach to explore cysteine redox-dependent structural states and generate testable hypotheses for cysteine-mediated regulation.