Guiding Receptor Directed Evolution Using a Physiochemical Specificity Model

We recently evolved of a yeast G-protein coupled receptor (GPCR) to detect a peptide biomarker for chronic kidney disease. The capability to evolve a receptor for any peptide biomarker enables the production of yeast-based biosensor (YBB) diagnostics for nearly any disease. YBBs use yeast to detect biomarkers in readily accessible patient samples such as urine and produce a colorimetric output discernable to the naked eye. Because YBBs use active-dry yeast, they are inexpensive to produce and have a long shelf life at room temperature. These features make YBBs ideal for point-of-care diagnostics and patient self-test applications, especially in resource-limited settings.

A major bottleneck to production of YBBs remains the difficulty of predicting an optimal path for directed evolution of the yeast GPCR, Ste2. Of two peptide targets for which we attempted to evolve receptors, one receptor took approximately 11 months to develop. A receptor for the other peptide target was never evolved, likely due to an infeasible directed evolution strategy. Through our work, it became clear that each receptor has a characteristic evolutionary step size, which is the maximum difference in the ligand for which a responsive receptor can be found in a single round of directed evolution. We also found that increased receptor promiscuity is a hallmark of intermediates along successful directed evolution pathways. However, the mechanisms underlying receptor promiscuity are poorly understood, preventing predictive design of receptor libraries. A better understanding of mechanisms for receptor promiscuity would enable the design of libraries that are more likely to contain successful variants, and of experiments that are more likely to isolate them.

The promiscuity of a receptor can be selected directly. We use fluorescence-activated cell sorting  (FACS) to evolve receptors of varying promiscuity. The receptors were then characterized against a library of single amino acid variants of α-factor. Partial least squares regression (PLSR) was applied to model the responses of each receptor to the peptide library. For each peptide, 8 physiochemical properties of each amino acid were provided as predictors. This modeling approach accurately predicts the response of a receptor to a peptide, and can be used to estimate the ligand range of a receptor. This provides a way to estimate the allowable evolutionary step size for any receptor and enables design of a more efficient directed evolution strategy. Finally, examination of the subset of highly predictive physiochemical properties provides insight into the mechanisms that produce promiscuity.