Predicting the structure and function of protein mutants

Proteins are essential molecules composed of a linear sequence of amino acids found in all organisms and participating in many physiological processes. The three-dimensional structure a protein adopts, as defined by its unique sequence, determines the biological role of the protein. Knowledge of the three-dimensional structure is vital in understanding how proteins function and this understanding is a crucial step in designing and engineering new proteins with specific functions. The unique sequence of a protein is so important for function, that the mutation of any single amino acid can disrupt the entire fold and thereby function of a protein by upsetting the balance of interacting forces within the protein. Structure based calculations of the relative activity of a protein after a mutation is an important problem since small mutations can lead to complete loss of function of the protein, and subsequently, to disease. However, predicting the functional outcome of mutations is very difficult, even when sufficient structural data are available. To date, few studies exist that have tried to correlate mutant protein function with biochemical measurement, and in fact, those that have been successful examine the stability of the protein, not the functional outcome. In this thesis I developed computational approaches to understand and predict protein structure and function. My most important contributions come from a number of new algorithms I developed for simulating large proteins and predicting the function of mutant proteins. The first algorithm represents the first method for predicting the structures of domain insertion proteins through combinations of rigid-body and torsion angle perturbations. Using my Domain Insertion algorithm, I was able to predict the structures of large proteins (up to 800 residues). From a set of 50 domain insertion proteins, 51% were predicted to within 2 A and 82% within 5 A root mean-squared deviation (rmsd) of their X-ray crystal structures. The second algorithm is the first computational method to predict both the structure and function of point deletions in proteins. This algorithm for assessing the effect of point deletions on a protein's structure and function was tested on a set of deletions from ricin with experimentally determined activity. Deletions with an active site rmsd from the wild-type protein > 1.0 A have a 93% probability of corresponding to an inactive mutation, while those with an active site rmsd < 1.0 A correspond to an active mutation 29% of the time. Incorporated in the methods for predicting the function of deletion mutants is the ability to perform backbone minimization on proteins greater than 150 residues, resulting in a 10-fold speedup in backbone relaxation of a 290 residue protein. This is an important step in structural modeling because current algorithms are limited by the time required to model large proteins. The final algorithm I present was created to predict the function of point mutations in the 17-N-terminal amino acids of the AraC regulatory protein. Sixteen of the seventeen mutant proteins possessing regulatory properties that are directly comparable to folded predictions were correctly predicted to fold to the arabinose-bound crystal structure. My work represents significant progress in the ability of computational methods to model mutations in proteins and predict how these mutations affect activity.