Engineering Picomolar Affinity into a Rationally Identified 5 Kda Scaffold for Tumor Targeting

Engineering Picomolar Affinity into a Rationally Identified 5 kDa Scaffold for Tumor Targeting

Max Kruziki, Hong Zhou, Patrick Holec, and Benjamin Hackel, University of Minnesota

Small protein scaffolds, relative to antibodies, can provide superior physiological distribution, stability, production, and ease of conjugation. Yet while many scaffolds have been tested to generate synthetic ligands, the causes of the varying degrees of success or failure have not been systematically explored. We merged computational and experimental tools to elucidate the biophysical factors that dictate evolutionary potential to empower rational identification or design of protein scaffolds. We have identified the 45-amino acid Gp2 domain as an effective protein scaffold to engineer stable, picomolar affinity ligands.

We developed and implemented a computational algorithm to efficiently evaluate naturally occurring protein domains on numerous metrics with potential impact on evolution of binding function. In one implementation, we identified all domains in the Protein Data Bank with two accessible loops (diversified paratope) and scored them on the size, shape, orientation, and target accessibility of their paratopes as well as protein size, stability upon mutation, and dependence on disulfide bonds and/or cofactors. Machine learning was used to preliminarily weigh each metric based on historical scaffold performance.

The top scaffold was the 45-residue T7 phage gene protein 2 with a single alpha helix opposite a beta sheet with two adjacent loops amenable to mutation. We diversified the twelve loop amino acids using complementarity bias and length diversity. The library of 10⁸ mutants was displayed on the yeast surface and screened for binders to several protein targets. Novel ligand discovery and directed evolution yielded binders with affinities as strong as 200±100 pM and no observable nonspecific binding. Circular dichroism showed secondary structure comparable to the wild-type protein and Tm values from 70±4ºC to 80±1ºC for evolved ligands relative to 67±4º for wild-type. We evolved binders to epidermal growth factor receptor and MET that effectively label human tumor cells. Ligands are readily conjugated with diagnostic moieties. Updated progress on tumor targeting in murine models â?? via fluorescence tomography and positron emission tomography â?? will be presented as well as scaffold and evolutionary optimization.

More broadly, libraries from the top 30 scaffolds were created for comparative analysis of the ability to evolve specific, high affinity binding to numerous targets. We will discuss the relative efficacy of each scaffold as well as the scaffold design metrics that correlate most strongly with performance. Implications for future scaffold design, including synthetic protein topologies, will be addressed as well as other designs.