Computational Design of Small Molecule Binding Proteins

Abstract

Protein design is still in its infancy, yet there have been many impressive examples of success in designing proteins to fold into a predictable structure, to catalyse enzymatic reactions, or to bind a specific protein, DNA sequence , or small molecule target . Each of these successes in the field is a major milestone, but protein design still lacks a generalized solution for reliably repeating these successes on future targets. The design of proteins capable of binding small molecules is particularly challenging due to the necessity to accurately understand and computationally model atomic scale physiochemical principles. We work towards this goal because being able to reliably design small molecule binders would allow for faster, and more efficient creation of detection elements for biosensors, sequestration proteins to aid in dialysis, and orthogonal binding tags for use in biotechnology applications. Even a modest advantage gained through computational design would allow for faster results when using more traditional directed evolution search methods. Since control of molecular specificity at the atomic level is essential for diagnostic applications in which multiple similar molecules are present and require discrimination from each other, computational modelling can be especially useful because the desired molecular specificity can be explicitly incorporated into the design. Such cases exist with the detection of tetrahydrocannabinol (THC) from the non-psychoactive cannabidiol and downstream metabolites present in users of marijuana, and in the detection of 25-hydroxycholecaliferol from 25-hydroxyergocalciferol, a clinically important distinction of vitamin D3 metabolites where the two compounds differ by a single methyl group. With this particular goal in mind, we have developed a computational protocol, using the Rosetta software package, capable of designing protein models with good shape complementarity, favorable chemical environments, and designed molecular specificity for a target protein-ligand interaction. This protocol was optimized over many iterations and incremental successes into a final revision that is capable of creating protein binders for the ligands 25-hydroxycholecaliferol, the hormonally active form of vitamin D3, and tetrahydrocannabinol, the primary psychoactive ingredient in cannabis. In addition to learning how to make successful protein binding designs, we also attempted to recover non-functional designs through stabilization. Using an algorithm for inserting proline substitutions into failed designs, we believe we have identified a lack of stability as one potential cause for failed binding protein designs. The protocol improvements learned from both our successful and recovered function binders should move us towards a more generalizable and reliable method for designing future protein-ligand interactions.