Research
Protein electron transfer
The current view of electron transfer (ET) is based on Marcus' theory, which was developed over 50 years ago with respect to ions interacting in solution. In this model, the free energy surfaces are parabolic (owing to the Gaussian transition probability), and can be fully described by three parameters: the reorganization energy (lambda), a measure of the curvature of the free energy surfaces based on the fluctuating energy separating the donor and acceptor electronic states; the Stokes Shift, a measure of the horizontal separation of the minima of the acceptor and donor free energy surfaces (or the horizontal separation of the maxima of the absorption and emission spectra); and the reaction Gibbs energy or driving force, which represents the vertical energy difference between the minima of those free energy surfaces. Marcus theory is formulated using the linear response approximation, which then connects the reorganization energy and the Stokes shift via the following relationship:
where the left-hand side of the equation is the Stokes shift, the right-hand side is the thermal average of the fluctuating energy gap, beta is the inverse of kBT, kB is the Boltzmann constant and T is the temperature.
The figure to the right is an illustration of model protein-protein electron transfer between the membrane-bound cytochrome b6f complex (orange) and the aqueous plastocyanin (purple) under investigation in the lab. Here, an electron is transferred from the cytochrome b6f complex to plastocyanin, while two protons are pumped across the thylakoid membrane via plastoquinone from the stroma to the lumen. When applied to protein ET, however, the typical Marcus approach has several failures. Most notably, the tight restriction that couples the Stokes shift and the reorganization energy is not preserved due to the glassy nature of the protein dynamics and the surrounding biological water. In fact, my earlier work showed that reorganization energies in protein ET can exceed 5 eV (10x any previous estimations) when calculated with a nanosecond observation window. That said, a more fundamental understanding of inter-protein electron transfers as required for ET in both respiratory and photosynthesis chains is needed in light of these recent developments and is one major research focus of this group.
Protein-protein interactions
Amyloidosis and other infamous diseases involving protein aggregation and deposition -- type II diabetes, Alzheimer's, Parkinson's, Huntington's and related prion disorders -- affect tens of millions of people worldwide. Yet no cure exists because the microscopic details underlying the mechanism of protein oligimerization in vivo is relatively unknown, despite the fact that the dewetting and nucleation events involved in aggregation are rooted with basic physical characteristics and likely
follow a general formation mechanism. Experimentally, these systems are extremely difficult to study because amyloid fibrils are intrinsically insoluble and noncrystalline (preventing traditional structure determination using X-ray crystallography or solution NMR). Computational studies of amyloid fibrillation, on the other hand, are marginally easier, albeit in the face of the microsecond self-assembly timescales, large system sizes, and the broad conformational sampling inherent to the aggregation process. Considering the importance of finding a cure for amyloidosis, furthering the ability to study these systems using enhanced sampling techniques is a target of active research in computational biophysics.
Even after the surge of computational research into amyloid fibrillation over the last few years, open questions still exist, in part, because results from various groups seemingly contradict one another. In one case it has been suggested that amyloid nucleation occurs by a dock-lock mechanism, which, for hydrophobic peptides, can be accelerated by the presence of surface waters. However, a more recent account suggests that amyloid nucleation occurs by a general mechanism that requires a multi-dimensional free energy surface for the aggregation process, one that does not require the surface water interactions at all. It has been well documented that surface water guides protein folding, it enhances protein flexibility, it mediates protein-protein binding, it limits the rate of protein-ligand binding, and it organizes into ordered assemblies on the protein's surface in order to drive enzymatic reactions. Therefore, it is reasonable to assume that if a general amyloid fibril nucleation mechanism exists, the reaction coordinate would involve the peptide's surface waters. So to resolve this discrepancy, and to provide a microscopic picture of fibrillation, one must create a multi-dimensional free energy surface for aggregation that directly includes peptide-water interactions as a reaction coordinate. Therefore, another major focus of this lab will be to develop enhanced-sampling methods to study the protein-water interactions that are likely driving protein and peptide aggregation.
High performance computing software development
While a huge effort has been placed on optimizing MD simulations so they run on massively parallel supercomputers and GPU clusters, the same attention has not been paid to the optimization of MD trajectory analysis. MD simulation analysis typically occurs using serial algorithms, and, at best, may include parallelization that is limited to a handful of compute cores through threaded or script-level message passing. Unfortunately, many useful MD trajectory analysis routines consume a large amount of CPU time, and with only a limited amount of parallelization available to the user, computationally intensive yet scientifically relevant analyses are often not performed.
Over the last several years, Dr. LeBard developed Pretty Fast Analysis (PFA), a parallel analysis suite that can analyze MD trajectories on supercomputers and GPU clusters. The software can analyze CHARMM- and AMBER-formatted trajectories from both binary and formatted files, and it scales exceptionally well in hybrid CPU/GPU compute environments. A large number of analysis routines are already available in PFA, including those for geometric analysis (RMSD, distance between atom selections, gyration tensor), energetic analysis (interaction energies including long-ranged electrostatics), cluster analysis, surface water analysis, radial distribution functions, and a host of other routines. The most computationally demanding routines have been off-loaded on the the GPU, and for a large number of particles in the simulation box, the GPU-based analysis can scale up to 16,000x compared to its serial counterpart. With that in mind, an effort towards pushing the ability of parallel analysis routines is a topic of current research.