Multiplex Methods for Profiling and Programming Protein Degradation

Abstract

This thesis focuses on the development of multiplex experimental methods for profiling and programming intracellular protein degradation. In addition, it explores the synergistic application of deep-learning models for understanding and engineering protein function at scale.Chapter 1 provides an overview of the mechanisms governing proteome homeostasis and outlines the technological gap that has historically limited our ability to study these systems. I review how the convergence of multiplex cellular assays and deep learning models offers a new paradigm for dissecting endogenous regulatory networks and engineering novel protein functions. Chapter 2 presents COMET (COmbinatorial Mapping of E3 Targets), a pooled assay for mapping E3 ubiquitin ligases to their substrates. By coupling combinatorial libraries of dual-fluorescent reporters with E3-targeting CRISPR guides, we enabled the many-by-many measurement of E3-dependent changes in protein abundance within a single experiment. I apply COMET to identify substrates of SCF complex E3 ubiquitin ligases as well as map the E3s mediating degradation of short-lived transcription factors, revealing that proteolytic regulation is often characterized by complex, many-to-many connectivity rather than simple one-to-one relationships. Finally, I demonstrate the use of deep-learning-based structural prediction models for the in silico validation of COMET-nominated E3–substrate pairs, pointing toward a future where computational nomination guides experimental validation. Chapter 3 shifts from mapping endogenous degradation to programming it. Here, we demonstrate a multiplex framework for the discovery of functional de novo designed “proximity handles.” We designed a library of binders targeting various effector proteins and characterized them using LABEL-seq, a multiplex RNA-barcoded protein abundance assay. This approach identified hundreds of designs capable of mediating the stabilization or degradation of a target protein. We further validated a subset of designs in orthogonal assays, demonstrated handle-mediated degradation of the endogenous oncoprotein MCL1, and applied these handles to remodel mitochondrial organization. This study establishes a generalizable pipeline linking computational protein design to high-throughput cell-based readouts. To conclude, Chapter 4 discusses future directions at the interface of multiplex technology and deep learning, specifically focusing on the potential for closed-loop design-build-test-learn cycles and the expansion of these methods to enzyme engineering.