(331e) Prediction Framework and Experimental Validation of Peptide Binding to Three MHC Class II Alleles

Major histocompatibility complexes (MHC’s), specifically human leukocyte antigens (HLA’s), are intimately involved in human immune response and autoimmunity. Aberrations in the function of HLA’s result in a variety of relevant, difficult-to-treat autoimmune diseases in humans. Examples include rheumatoid arthritis, induced myasthenia, and schizophrenia. In normal function, HLA’s bind to antigen peptides within the cell and present these peptides to the surface of the cell so T cells can recognize them. The T cells then destroy the cell or activate some other immune response to eliminate the foreign pathogen. Autoimmune diseases are characterized by overactivation of HLA binding, resulting in immune response against tissues normally present in the body. So the development of peptidic antagonists to the HLA’s are essential to the development of safe and effective treatments for such diseases. However, there are many different alleles of HLA, which makes designing peptidic therapeutics that target HLA a challenging problem. This work presents two methods for predicting binding affinities and applies the methods to predict binders and non-binders to three different MHC Class II HLA alleles. The peptides have known IC50s, which are used to determine the success of the prediction capabilities.

The first method [1-2] employs Monte Carlo simulations to first predict the structure (using RosettaAbinitio [3]) of the peptide sequences then to perform docking simulations (using RosettaDock [4]) between the sequence and the target HLA allele. A rotamerically-based ensemble of structures for the peptide, the HLA allele, and the peptide-HLA complex is generated using RosettaDesign [5], and is used to calculate approximate molecular partition functions of the peptide, the HLA allele, and the complex. Using these approximate partition functions, an approximate binding affinity is calculated [6]. The more accurate the partition functions are, the more precise the binding affinity will be, and so this process is iterated, sampling more and more conformations, until the partition functions converge.

The second method uses an identical method as the first method to construct a large number of peptide and complex structures for each sequence and target HLA allele. However, instead of using the built in Rosetta energy function [7] in the binding affinity calculation, a more detailed energy function FAMBE-pH [8] is used. This energy function efficiently calculates electrostatic effects that are not taken into account  by the Rosetta energy function. The peptides are ranked by binding affinity using both methods and compared to experimental results to assess the potential improvement of the first method by using a more detailed energy function.

            The work detailed in this presentation shows success in predicting binding affinity and IC50 results for two different methods. Both methods use RosettaAbinitio, RosettaDock, and RosettaDesign to generate a large number of structures for the candidate sequence and the target Class II MHC allele. The first method uses the built in Rosetta energy function to calculate an approximate binding affinity, while the second method uses a more detailed and time consuming energy calculation that takes into account electrostatic effects. Both methods show success in differentiating between sequences which are able to bind to the HLA allele from those that have no detectable experimental binding affinity. This work details an important step in our ability to predict effective peptidic therapeutics for the wide variety of medically relevant autoimmune diseases.

[1] Bellows, M. L., H. K. Fung, C. A. Floudas, A. Lopez de Victoria, and D. Morikis, 2010. New Compstatin Variants Through Two De Novo Protein Design Frameworks. Biophys J 98:2337–2346.

[2] Bellows, M. L., M. S. Taylor, P. A. Cole, L. Shen, R. F. Siliciano, H. K. Fung, and C. A. Floudas, 2010. Discovery of entry inhibitors for HIV-1 via a new de novo protein design framework. Biophys J 99:3445–3453.

[3] C. A. Rohl, C. E. M. Strauss, K. M. S. Misura, and D. Baker. Protein structure prediction using Rosetta. Methods in Enzymology, 383:66-93, 2004.

[4] J. J. Gray, S. Moughon, C. Wang, O. Schueler-Furman, B. Kuhlman, C. A. Rohl, and D. Baker, Protein-protein docking with simultaneous optimization of rigid-body displacement and side-chain conformations. Journal of Molecular Biology, 331:281-299, 2003.

[5] Y. Liu and B. Kuhlman. RosettaDesign server for protein design. Nucleic Acids Research, 34:W235-W238, 2006.

[6] R. H. Lilien, B. W. Stevens, A. C. Anderson, and B. R. Donald. A novel ensemble-based scoring and search algorithm for protein redesign and its application to modify the substrate specificity of the Gramicidin Synthetase A Phenylalanine Adenylation enzyme. Journal of Computational Biology, 12:740-761, 2005.

[7] B. Kuhlman, G. Dantas, G. C. Ireton, G. Varani, B. L. Stoddard, and D. Baker. Design of a Novel Globular Protein Fold with Atomic-Level Accuracy. Science, 302 (5649), 1364-1368, 2003.

[8] Y. N. Vorobjev, J. A. Vila, H. A. Scheraga. FAMBE-pH: A Fast and Accurate Method to Compute the Total Solvation Free Energies of Proteins. The Journal of Physical Chemistry B, 112 (35), 11122-11136, 2008.