Computational protein design is poised to bring forth many advances in areas ranging from personalized medicine to biofuel production. While protein engineers have made huge strides towards the design of structure, proteins with new functions remain extremely difficult to design in silico, owing largely to the precise geometric requirements they demand. In this dissertation, I present two new methods for the design of functional proteins that focus specifically on enabling the precise sidechain geometries that are required for functions such as binding and catalysis. Pull Into Place (PIP) accurately places key amino acid sidechains by creating and stabilizing new irregular backbone geometries that are compatible with a provided functional interaction, utilizing new fragment-based local structure design and prediction methods. We experimentally validate PIP on a ketosteroid isomerase model system by redesigning its active site loop such that it uses a glutamate, rather than an aspartate, to perform proton extraction on its substrate, a task requiring accuracy on the scale of a carbon-carbon bond.Where PIP enables one or a few functional sidechain interactions requiring irregular backbone structures, Helical ELements Interface eXplorer (HELIX), a new method for de novo protein-protein interface design, instead attempts to connect highly regular secondary structure elements to facilitate many functional sidechain interactions at once. It does this by matching tertiary structural motifs from the PDB with a library of proteins containing systematically reshaped α-helices. HELIX results in computational designs with a high degree of shape complementarity, many of which contain difficult-to-design polar interaction networks. Together, these two works encompass several new tools and approaches that enable the design of new functions.

Protein binding to small molecules is fundamental to many biological processes, yet it remains challenging to predictively design this functionality de novo. Current state-of-the-art computational design methods typically rely on existing small molecule binding sites or protein scaffolds with existing shape complementarity for a target ligand. Here we introduce new methods that utilize pools of discrete contacts observed in the Protein Data Bank between protein residues and defined small molecule ligand substructures (ligand fragments). We use the Rosetta Molecular Modeling Suite to recombine protein residues in these contact pools to generate hundreds of thousands of energetically favorable binding sites for a target ligand. These composite binding sites are built into existing scaffold proteins matching the intended binding site geometry with high accuracy. In addition, we apply pools of rotamers interacting with the target ligand to augment Rosetta’s conventional design machinery and improve key metrics known to be predictive of design success. We demonstrate that our method reliably builds diverse binding sites into different scaffold proteins for a variety of target molecules. Our generalizable de novo ligand binding site design method will lay the foundation for versatile design of protein to interface previously unattainable molecules for applications in medical diagnostics and synthetic biology.

This work presents two separate investigations based on a similar theme: that the topology of

protein-protein interactions should in part be predictive of a protein's sensitivity to dosage. The

rst investigation focuses on individual multi-protein complexes examining whether protein

complex topology is a determinant of dosage sensitivity. Mass balance equations are used to

compute the response of complex formation to varying amounts of each component protein. The

computed response is then related to experimental data in the form of tness of hetezygous

deletions, overexpression phenotypes, and protein expression noise. The second investigation

focuses on larger protein interaction networks and looks specically at proteins that

interchangeably bind multiple other proteins. A series of simple networks is used to gain insight

into changes in the amount of unbound protein upon overexpression, again using equations

describing mass balance. Predictions of the simple models are then validated with experimental

data.

Biology is driven by functional macromolecules, most notably proteins and non-coding RNAs. Learning how to design similarly functional macromolecules is a natural goal. Success will not only bring the ability to create new biological systems, but also the ability to more finely study and manipulate existing biological systems. In this thesis, I will describe two design projects that I pursued over the course of my PhD. The first is a project to remodel the backbone of a protein for the purpose of accurately positioning a catalytic sidechain. The second is a project to ligand-sensitive guide RNAs for the CRISPR-Cas9 system. There is of course much more to be done before we can say that we are able to design functional macromolecules, but the projects described herein move us closer to that goal.

Computational protein design aims to predict protein sequences that will fold into a given three-dimensional structure and perform a desired function. Though significant accomplishments in computational protein design have been achieved in the past several years, including the design of novel enzymes and protein-protein interactions, the accuracy of computational protein design is relatively low and many designed sequences must be experimentally tested in order to obtain a successful design. Moreover, successful designs often require directed evolution to achieve catalytic activities or binding affinities similar to naturally occurring proteins. A major challenge in computational protein design that limits its accuracy is the inability to sufficiently sample protein sequence and conformational space at a high resolution. Sampling is difficult due to the combinatorially large number of possible protein sequences and the inherent flexibility of the protein backbone, which may change its conformation upon changes in sequence. To address the issue of sampling a large number of sequences, I developed a deterministic computational protein design algorithm that identifies all sequences within a given energy of the global minimum energy sequence. To identify an accurate method of modeling backbone flexibility, I created a benchmark that evaluates designed sequences based on their similarity to natural sequences with respect to amino acid covariation. Lastly, I developed a novel method of coupling side-chain and backbone sampling. I applied this method to re-designing enzyme substrate specificity and showed a substantial improvement in accuracy over previous computational protein design methods. Taken together, these results demonstrate the importance of modeling protein backbone flexibility and provide new tools to enable higher accuracy computational protein design.

Genetically engineered therapeutic cells are a new paradigm in treating historically intractable diseases. Chimeric antigen receptor (‘CAR’) T cells have made a tremendous impact in the clinic against blood cancers, but potent signaling through the CAR can result in toxicity and CAR T cell exhaustion. Further engineering efforts to modulate CAR signaling are needed to continue to deploy engineered cell therapies against increasingly complex maladies. The use of protein degradation to ablate CAR signaling is a promising method for improving CAR T cell safety and efficacy. In preclinical models, small molecule-controlled protein degradation to tightly control CAR expression levels was shown to circumvent issues of CAR T cell toxicity and exhaustion. An equally important complement to small molecule modes of control are genetic circuits that allow therapeutic cells to reactively modulate CAR signaling in response to their environment. Using genetic circuits in combination with protein degradation could result in the birth of a new class of engineered cell therapies capable of cell autonomous regulation of therapeutic function. However, programmable targeted protein degradation tools for this purpose do not exist for immune cell engineering. To address this gap in the existing CAR T cell engineering toolbox, we designed a novel bioPROTAC that uses two protein domains to bridge a target protein of interest with machinery from the ubiquitin proteasome system to promote proteasomal recruitment and degradation. We built the bioPROTAC such that it is capable of degrading both cytosolic and membrane proteins while also being as small as 181 base pairs for efficient delivery by lentivirus. We show that the bioPROTAC is capable of degrading cytosolic proteins such as green fluorescent protein (‘GFP’) in multiple mammalian cell lines, as well as degrading and functionally modulating membrane proteins such as second-generation CARs in Jurkat and primary T cells. To demonstrate our bioPROTAC’s ability to be composed into genetic circuits that modulate cell signaling pathways, we engineer an antigen-inducible circuit capable of shutting down CAR signaling through degradation of the tyrosine kinase ZAP70. Through this circuit, we show that this new bioPROTAC can interface with and modulate endogenous signaling networks under user-defined cell autonomous control. Genetic circuits that actuate protein degradation can be a powerful method for engineering self-regulating CAR T cells thereby improving therapeutic safety and efficacy.