Monte Carlo simulations of protein folding

Monte Carlo simulations are employed to study how a protein folds from a random coil to its native state. In Part I, we investigate the kinetics and thermodynamics of a polypeptide modeled as a chain of 125 beads restricted to a lattice. The behavior of the model is found to be more complex than that of smaller systems. The diverse trajectories that lead to the native state can be classified into a small number of average pathways: a “fast track” in which the chain forms a stable core that folds directly to the native state and several “slow tracks” in which particular contacts form before the core is complete and direct the chain to misfolded intermediates. Rearrangement to the native state is slow because it requires breaking stable contacts that involve primarily surface residues. Increases in temperature destabilize the intermediates and shift the transition state in a Hammond manner. The mechanism is mapped to two coordinates that are based on a comparison of folding and non-folding trajectories. The free energy in terms of those variables is in good agreement with the observed kinetics, which indicates that they provide an adequate description of the folding reaction. The generality of the results is confirmed by statistical analysis of a 200 sequence database.; In Part II, the study is extended to higher resolution (all-atom) models. We generalize a procedure for local deformation of a polymer by concerted rotation of several sequential rotatable main chain dihedral angles and evaluate its usefulness as an elementary move. A Monte Carlo module that includes this move is implemented for the program CHARMM and is applied to sampling the accessible configuration space of a 16-residue peptide that has been shown experimentally to adopt a hairpin structure in solution. A non-canonical weighting scheme is employed to accelerate the escape from local free energy minima. Overall, there is only a relatively small number of distinct conformations. The results suggest that Monte Carlo methods will be capable of finding the native states not only of peptides but also of proteins in the relatively near future.