UCLA

The amyloid protein fold refers to a multimeric, fibrillar structural motif in which polypeptides of the same sequence fold into relatively planar vertebrae that stack to form fibers. Each layer contributes β-strands to β-sheets that run the length of the fiber, so that the fibers are stabilized longitudinally by backbone hydrogen bonds and laterally by stretches of intercalating, usually hydrophobic sidechains that form "steric zippers." The aggregation of proteins into such fibers is a hallmark of hundreds of protein misfolding disorders, including neurodegenerative diseases such as Alzheimer’s disease (AD), making studies of both disease-associated fiber structures and their protein folding pathways valuable targets in drug discovery. Current computational methods for both protein folding modeling and structure prediction are not suitable for either of these tasks, however. Most proteins are monomorphic, with a single protein sequence producing a single protein structure, usually assumed to be a thermodynamic minimum; amyloid fibers are polymorphic, however, meaning that a single protein sequence will produce many different amyloid fiber structures, currently thought to be kinetically metastable. Subsequently, current computational methods have thus far focused on monomorphic proteins, predicting a relatively limited number of highly similar structures rather than full folding landscapes because of major recent breakthroughs in protein structure prediction methods that make no attempts at protein folding. This dissertation attempts to fill this gap for amyloid fibers by proposing a method for amyloid structure enumeration rather than merely prediction of a singular or limited set of structures that also enables various energetics experiments to explain fiber folding pathways. This is done by modeling amyloid polymorphs as discrete, combinatorial objects, and creating a “steric zipper” representational schema that mimics that of RNA base-pairing. Despite noting some mismatch between observed and recapitulated structures and its causes, our method recapitulates the overall architecture of multiple tau polymorphs while also finding a much broader landscape. Given our structure enumeration is initially energy agnostic and generates a finite but large search space that must be parsed to identify disease-associated polymorphs and their folding pathways, we also assess a knowledge-based energy function used in the field for predicting steric zippers for this purpose and evaluate several explanations for amyloid folding. Our method only generates single layers of amyloid structures, however, leading us to explore determinants of twist between layers, an important global amyloid parameter necessary to reconstruct full fibers. We find that twist is well-predicted by calculating the twist at which inter-layer amino acid sidechain steric clashes are minimal. This work offers an approach that can be further developed and refined to create full, polymorphic folding landscapes of amyloid fibers convertible to atomic models.