A replica-exchange approach to computing peptide conformational free energies

We describe a methodology for computing relative free energies of peptide conformational states. The method uses replica-exchange molecular dynamics (REMD) simulations with backbone restraints that hold a peptide within the conformational space basin surrounding a given structure of interest. Restraints are scaled from full strength to zero strength as one moves up the replica cascade, such that the highest temperature becomes a conformation-unspecific reference state. Then, weighted histogram analysis is used to compute the free energy of the structure of interest relative to the reference state. We test our method on a 14-residue sequence that adopts a variety of alpha and beta structures at equilibrium. We show that computed free energies bear close theoretical agreement with structure populations and have the expected relationship with structure unfolding temperatures evaluated in non-equilibrium heating simulations. By comparison, purely potential energy measures are not able to correctly rank populations, even with harmonic free energy corrections. These results suggest that the methodology can compute free energies underlying structural stability and populations. They further suggest that entropic contributions to free energies and populations are significant, and that force-field potential energies alone cannot be used as structure prediction scoring functions, at least for small peptides.