Computational methods potentially offer a powerful design tool for the growing industry of therapeutic peptides (typically up to ~40 amino acids in length). There has long been a related interest in inferring the three-dimensional structures of globular proteins using computers. The vast majority of protein structure prediction algorithms, and the most successful to date, have been bioinformatics-centric methods that draw on extensive knowledge databases of known protein structures [1]. These methods work remarkably well when a homologous protein can be found to provide a structural template for a candidate sequence; however, in cases where homology to database proteins is low or where conformational flexibility is substantial—as in short, often synthetic peptides—these methods have been shown to perform poorly [2]. Moreover, there is a growing demand for peptide structure prediction algorithms: in recent years, these molecules have targeted as potential new therapeutics due to their low toxicity, high selectivity as inhibitors, and ability to be synthesized as well as fermented [3].

An alternative approach to bioinformatics in peptide modeling, pursued here, is a physics-based, database-free method, relying on proper thermal sampling and molecular potential energy functions. Relative to bioinformatics methods, a physics-based approach can capture full conformational ensembles, solvent and surface effects, binding and induced-fit interactions, and noncanonical residues, in particular D-amino acids, which are often used to reduce in vivo degradation by serum proteases. Here, we use the AMBER96 force field and a sampling methodology called ZAM (Zipping and Assembly Method) [4, 5] to determine the conformational ensembles of short peptides. ZAM works in two parts: (1) the full polypeptide chain is broken into small fragments that are first simulated independently and then successively re-assembled into larger segments with further sampling, and (2) consistently stable structure in fragments is detected and locked into place, in order to avoid re-sampling those degrees of freedom in subsequent steps. We use this approach to predict the conformational ensembles of a series of permeation-enhancing therapeutic peptides designed by a company called Nastech. We show that the generated ensembles can be used in a simple model of membrane insertion [6], based on free energies derived from structural features, to predict the permeabilities of the peptides. This type of analysis enables both the prediction of sequences with enhanced function and the identification of the dominant driving forces behind the mechanism of action.

[1] J. Moult, A Decade of Casp: Progress, Bottlenecks and Prognosis in Protein Structure Prediction, Curr. Opin. Struct. Biol. 15, 285 (2005).

[2] A. Thomas, et al., Prediction of Peptide Structure: How Far Are We?, Proteins (2006).

[3] P. W. Latham, Therapeutic Peptides Revisited, Nature Biotech. 17, 755 (1999).

[4] S. B. Ozkan, G. H. A. Wu, J. D. Chodera, and K. A. Dill, Protein Folding by Zipping and Assembly, Proc. Nat. l. Acad. Sci. U. S. A. accepted, (2007).

[5] M. S. Shell, R. Ritterson, and K. A. Dill, A Test on Peptide Stability of Amber Force Fields with Implicit Solvation, under review (2007).

[6] N. Ben-Tal, B. Honig, C. Miller, and S. McLaughlin, Electrostatic Binding of Proteins to Membranes. Theoretical Predictions and Experimental Results with Charybdotoxin and Phospholipid Vesicles, Biophys. J. 73, 1717 (1997).