De novo discovery of synthetic peptide binders to protein-protein interfaces

Abstract

Protein-protein interactions (PPIs) play crucial roles in mediating normal cellular physiology, but their modulation has been historically challenging. PPIs tend to be intractable to small molecule inhibition, due to their wide and relatively featureless interfaces, and biologics are generally not viable for the approximately two-thirds of PPIs that take place in the intracellular milieu. Therefore, this class of target has been in many circles deemed undruggable. Peptides are an emerging therapeutic modality for disrupting PPIs. With proper engineering, they can engage proteins over large surface areas and in some cases be modified to access the cytosol. PPI-disrupting peptides are often discovered from highly diverse combinatorial libraries, through either genetic or chemical means. Genetically encoded approaches can reliably investigate enormous libraries (10⁸-10¹³ members), typically via selection.

However, despite progress in this area, these libraries are generally limited to natural chemical space. Peptides identified from such approaches therefore require extensive engineering to improve proteolytic stability and promote cell penetration, at the potential cost of potency. Synthetic libraries, on the other hand, are highly amenable to non-canonical amino acid incorporation and a wide variety of chemical modifications. However, these libraries are typically examined by screening, which in practice limits the diversities that can be explored to ~10⁶. In this thesis, a magnetic bead-based affinity selection-mass spectrometry (AS-MS) workflow was developed to interrogate fully randomized, chemically accessed peptide libraries comprising up to 10⁸ members with high fidelity.

This approach takes advantage of recent advances in nano-liquid chromatography-tandem mass spectrometry (nLC-MS/MS)-based peptide sequencing, which facilitates high-confidence decoding of complex peptide mixtures. Starting with a model selection target, an anti-hemagglutinin monoclonal antibody, it was demonstrated that high enrichments of true binders could be achieved from a library comprising 10⁶ members. The number of binders identified scaled in proportion to library diversity, as diversity was then increased from 10⁶-10⁸. Beyond 10⁸, the complexity of isolated pools from single-pass selections became too great for reliable decoding. These results were applied to selections against biomedically-relevant targets, enabling the identification of p53-like binders to the oncogenic ubiquitin ligase MDM2, and a family of low-nanomolar affinity, [alpha]/[beta]-peptide-based binders to 14-3-3. Both sets of binders engage these targets at PPI epitopes.

Finally, machine learning methods were developed to distinguish nonspecific from true binders identified by AS-MS, which we anticipate will greatly streamline future discovery efforts.