Date of Award

Fall 9-2014

Embargo Period


Degree Type


Degree Name

Doctor of Philosophy (PhD)




Maria Kurnikova


Diphtheria toxin translocation (T) domain is a water soluble protein that consists of ten alpha-helices in high pH solution. Experimental studies have determined that T-domain undergoes conformational changes upon decrease of solution pH, which shifts the protein population from its initial water soluble to a membrane-competent in solution. It was hypothesized that conformational changes of the latter state prepare the protein structure for subsequent membrane binding. After binding, refolding of T-domain on the membrane results in the formation of a transmembrane state, which is characterized by the permeation of the lipid bilayer. The function of transmembrane state is to help the translocation of a catalytic domain attached to the protein N-terminal across the lipid bilayer. The first goal of this work is to study the pH-dependent destabilization of T-domain structure in solution and understand the role of protonation of key residues using a variety of computational and experimental methods. The second goal is to study the subsequent membrane binding of T-domain to lipid bilayers and propose a theoretical model of the early steps of T-domain refolding on the membrane interface using a multiscale approach. Modeling of the low pH induced conformational changes and membrane association of T-domain is performed in two stages. In the first stage, protonation of N-terminal histidines triggers conformational changes of the N-terminal helices of T-domain in solution. The role of histidines was confirmed by thermodynamic integration, continuum electrostatic calculations, and microsecond long molecular dynamics (MD) simulations. Two microsecond MD simulations of a low pH model of T-domain sampled similar destabilized protein conformations; however, an N-terminal helix showed dissimilar degree of refolding. To improve the sampling of conformational changes, we proposed and implement a sampling method based on the accelerated molecular dynamics simulations method. Our proposed implementation accelerates the sampling of the conformational landscape of the low pH T-domain model by boosting the direct-space electrostatic interactions of solute-solute atom pairs. In general, this implementation accelerates the sampling of the conformational space of alanine-dipeptide in comparison to the original implementation of accelerated molecular dynamics. In the second stage, a multiscale approach is used to model the membrane association of the low pH destabilized model of T-domain to lipid bilayers of different compositions. Two preferable membrane-bound conformations of the low pH T-domain model are predicted by equilibrium and free energy calculations. The most frequently observed membrane-bound conformation is stabilized by electrostatic interactions between the protein and the lipid headgroups. In contrast, the less frequently observed membrane-bound conformation is stabilized by hydrophobic interactions between the protein and lipid headgroups. These interactions allow for a deeper insertion of T-domain in the membrane interface. The predicted membrane-bound conformations were refined by atomistic molecular dynamics simulations, which show that membrane-bound conformations are stable for several microseconds. Furthermore, atomistic MD simulations suggested that neutralization of glutamate and aspartate sidechains favored a deeper inserted state of T-domain in the membrane interface. This observation is in good agreement with reported pH-dependent insertion of T-domain in the membrane interface. To study the assembly of transmembrane helices, a coarse-grained model based on a residue level of representation and a rigid-body Monte-Carlo sampling method is developed. The scoring energy function is constructed using a knowledge based potential extracted from water soluble protein structures. To compensate the protein interior packing and the solvation differences, an experimentally determined membrane partition scale for all residues was used. This scoring function was tested in a set of three transmembrane homodimers. The proposed scoring function and the associated rigid-body Monte-Carlo sampling method can be applied in the first steps of prediction of near-native structures of transmembrane proteins followed by structural refinement using atomistic MD simulations.