Fast, Atomic-Level AFM and Magnetic Tweezers Simulations of the Unfolding of Membrane Proteins Using a New Membrane Burial Potential with H-Bonding

We conduct unfolding simulations using our Upside MD algorithm that can reversibly de novo fold some soluble proteins in CPU-hours (Jumper et al. (accepted) PLOS Comp Biology). Our model has an authentic polypeptide backbone and Ramachandran maps along with a 6-position side chain bead. Here we incorporate a knowledge-based membrane burial potential that includes energies for unsatisfied H-bond donors and acceptors (Wang et al., submitted). By accounting for peptide group burial, helices are allowed to unfold within the bilayer. In 100 AFM unfolding simulations of bacteriorhodopsin, we reproduce the characteristic experimental features including the unfolding of individual and pairs of helices, and even the back-and-forth unfolding of single helical turns with comparable or more resolution than in experiment (Yu et al. (2017) Science 355:945). We also emulate unfolding with magnetic tweezers by pulling laterally on GlpG, a six-helix bundle (Min et al. (2015) Nat Chem Biol 11:981). We observe a variety of intermediates and unfolding pathways with sequential unfolding from either terminus or the center. The bias for N-terminal unfolding can be increased by a mutation that disrupts the H-bond network at this end. How force is applied can alter the observed landscape. For unfolding simulations of GlpG by magnetic tweezers, the force remains nearly constant after the initial rip, and only a few intermediates are observed. In contrast for unfolding by AFM, the force immediately drops to near-zero after a rip and over a dozen intermediates are observed. Hence, the experimental mode strongly affects the experimental detail one observes and the apparent folding cooperativity, issues that should be considered when interpreting forced unfolding data and designing experiments.