An evolutionary basis for protein design and structure prediction

Sammanfattning: The sequence diversity of protein families is a result of the biophysical selection pressures that shaped their evolutionary history. Among the dominant pressures is selection for protein thermostability, which in itself is an attractive target in protein engineering because of its importance for various biopharmaceutical properties, the performance of industrial enzymes, and the ability to design new protein functions. In the first part of this thesis, we use models of evolutionary dynamics and biophysical fitness functions to derive the relationship between amino acid frequencies in sites of proteins and the stability effects of mutations. This analysis suggests that a commonly applied assumption (that amino acids frequencies are Boltzmann distributed) is inaccurate, and we provide a new relation consistent with the current understanding of evolutionary dynamics and protein fitness. Next, we study the extent to which the evolutionary pattern of amino acid substitutions can be explained by protein stability, as predicted using all-atom models of protein energetics. We show that at least 65\% of the substitution pattern can be explained by thermostability. With the same model, we show that functional sites (e.g. active sites or binding sites) can be predicted when the apparent evolutionary site-rate deviates significantly from that of a stability-only null-model of evolution. Finally, we study how the strength of selective pressure affects the evolutionary behavior of proteins, again using the same models, but this time generating evolutionary trajectories. We find that energetic coupling between amino acids (coevolution) and the detriment of mutation increases as the strength of selection increases. Antibodies are a key molecular component of the adaptive immune system of vertebrates and an important biopharmaceutical molecule. In the second part of the thesis, we predict and design the structure of antibodies by using energetics derived from sequence alignments and following the evolutionary encoded modular segmentation of the molecule. Through multiple design and test iterations, we were able to design antibodies, which express stably and, in some cases, bind target antigens. The developed structure prediction algorithm performs as well as other methods, is in some cases more accurate, and produces models with lower chemical strain. We use the structure prediction method to study a tumor-associated carbohydrate binding antibody.Finally, we also review the literature on design of symmetrical protein self-assembly, and study the dynamical properties of a partially disordered chaperone protein, calreticulin.