Domain insertion scanning to study and engineer ion channels

Abstract

New proteins primarily evolve through recombining modular protein domains with discrete structure and function. Often domain recombination combines a catalytic functioning domain with a sensing domain, so protein function can be regulated by different stimuli. This form of domain recombination-based evolution underlies the intricate signaling networks that allow our cells and by extension our bodies to sense and respond to stimuli. Protein engineers mimic nature by combining domains with desirable properties into new useful combinations never seen in nature. This approach for generating synthetic multi-domain proteins has yielded groundbreaking therapeutics and tools for biology, most notably Car-T cancer therapies and GCaMP calcium sensors. However, domain recombination is challenging and requiring years of iterative optimization but unlike evolution we don’t have millions of years to spare. Both, our basic understanding of the biophysical principles of how proteins evolve and how to better engineer proteins are limited by a lack of domain compatibility rules. In the work presented in this thesis, we sought out to apply massively parallel domain insertion experiments and learn rules for domain compatibility. As our target protein, we used ion channels as they are an attractive engineering target and ion channels evolved through extensive domain recombination. Initially we started with a small set of 3 inserted domains inserted into all amino acid positions of a potassium channel kir2.1. We successfully engineered a light-switchable potassium channel that could be used by neuroscientists, however we found a tremendous amount of variability that necessitated expanding out to a broader sample of domain recombination space. Before we could achieve this goal, we needed to improve experimental pipelines because the methods the domain insertion field used at the time were not scalable nor generalizable. We developed a new domain insertion library generation method, SPINE, that yielded near perfect libraries. SPINE allowed us to expand out to over 700 different inserted domains with which we exhaustively sampled insertional space and developed a mechanistic model of domain recombination. We then expanded outward to several additional recipient channels to benchmark our work. Overall, we made major strides towards the goal of a mechanistic model for assembling protein domains. We expect this body of work will provide a foundation that will make domain-based engineering more effective and improve our understanding of the fundamentals of how proteins evolve.