| This thesis includes three studies involving different aspects of modeling protein structure. The first study illustrates the levels of insight available from atomic-resolution protein structures. The second study derives general trends of protein geometry from atomic-resolution structures and shows their implications for modeling. The third study creates a model of a protein and uses it to derive new biological insights.In the first study, a series of structures were analyzed from human glutathione reductase, a biomedically relevant enzyme. Newly accessible at atomic resolution is structural evidence showing the catalytic importance of active-site compression, which additionally causes distortions from standard geometry that further enhance catalytic power. Another aspect of geometry visible at atomic resolution is the remarkably ideal positioning of atoms for catalysis. The stereoelectronic control displayed by compression and geometric preorganization provides insight into the origins of catalytic power.The second study builds upon quantum-mechanics calculations and empirical analyses of protein structure from the 1990s that showed the concept of a single ideal value for backbone geometry was wrong. Here, a nonredundant set of protein structures at atomic resolution is probed to better define the dependence of backbone geometry upon the conformation of the backbone torsion angles phi and Psi. The set was taken from the Protein Geometry Database created here (http://pgd.science.oregonstate.edu/). The trends seen make structural sense and lay the groundwork for a paradigm shift in the concept of ideal geometry. A conformation-dependent library accounting for these trends has the potential to improve modeling accuracy.In the third study, a model of the tumor-suppressor merlin is created and used to gain new understanding of merlin's function. Merlin is the only known cytoskeletal tumor suppressor, and loss of functional merlin results in neurofibromatosis 2, characterized by nervous-system tumors, cataracts, and skin tumors. Clear errors were evident in available automatically created models, driving the need for a reliable structure. Merlin and its homologs have distinct functions, so the differences between them were probed, suggesting critical functional clusters. A new technique developed here for discovering gains and losses of function should be generally applicable to any two protein subfamilies with distinct functions. |
