Project Inspiration and Description

We aimed to harness the naturally occuring yeast mating pathway to detect custom peptides by evolving the yeast alpha-mating GPCR (Ste2) to detect alternate ligands. For this we were inspired by an already established approach for the directed evolution of Ste2 by gradually diversifying the ligand by incrementally changing the alpha mating factor and performing repetitive rounds of mutagenesis (diversification) and subsequent selection.

The reported approach used in-vitro error-prone PCR, we improved on this by postulating an in-vivo approach using a yeast strain with a hyper-faulty engineered DNA polymerase with a staggering error rate 100000 times higher than the endogenous DNA polymerase. The strain also harbours a linear cytoplasmic plasmid with the specific ORI for the orthogonal error-prone polymerase. The error-prone polymerase was validated to exclusively replicate the orthogonal cytoplasmic plasmid sparing the endogenous chromosomes. Integrating target genes into the orthogonal P1 plasmid renders them a susceptible to elevated mutation rate and accelerated evolution.

Integrating the mating receptor into the orthogonal plasmid gives rise to an in-vivo workhorse to drive the evolution of Ste2 and screen for variants that bind the desired ligands, while also maintaining the intracellular mating-inducing downstream interactions.

In addition to the in-vivo approach we also devised an improved in-vitro mutagenesis pipeline by harnessing the remarkable recombination capacity of yeast to perform DNA shuffling of gene-blocks with diversified ligand-interacting segments. For any directed evolution approach to yield receptors with desired ligand specificity within humanly acceptable time-frames the following three variables have to be optimized;

1) The types and steepness of selective pressures. 2) The increments of changing the ligand (size of evolutionary steps) 3) The intracellular ques used to couple desired ligand-receptor affinity to an evolutionary advantage (reward signal). To tackle these issues we used a complementary combination of experimental and in-silico approach to determine the optimal levels and type of each variable.

The simple setting of mutagenizing the Ura3 gene was used to validate the orthorep system and characterize the dynamics of mutagenesis/evolution. Ura3 gene allows biphasic selection by alternating between drop-out medium and Ura3 conveyed toxicity using 5-FOA treatment.

To render the orthorep strain suitable for Ste2 directed-evolution and ligand affinity selection we deleted the genomic Ste2 gene and integrated the Ste2 into the P1 orthogonal plasmid.

We also developed an array of reporter constructs to enable the selection for successful evolutionary variants that bind the desired ligands. These include fluorescent proteins under two different mating promoters in addition to receptor fusions with other proteins that can serve as indicators of mating (e.g fluorescent proteins with NLS) or to modulate the outcome of successful binding (transcriptional modulators with NLS). In both cases the binding of the receptor to a suitable ligand renders linker of the fusion protein susceptible to protease cutting and sets the fused protein free to translocate from the memrane to the nucleus.

A rational design approach was also implemented, in which chimeras of the intercellular parts of the yeast GPCR with the ligand binding domains of particular receptors were created.

We also conceived a full in-silico toolbox to support experimental runs of directed evolution.

In order to guide the evolution of the STE2 receptor, we lay out the mathematics for finding the shortest evolutionary path between the native and a target variant, allowing in the best case scenario binding to a custom ligand.

To ensure that the requirements of viable evolution are being met along this journey, the fitness landscape must be simulated. We suggested the use of machine learning to solve this problem by leveraging a recently developed neural network - MaSiF test .

To lay the groundwork for this machine learning approach, we used well-established protein folding algorithms to create a database of modified mating receptors, as well as an array of potential “theoretical” ligands. By using topological and chemical information, MaSiF should gives a measurement the affinity of evolved receptor to diverse ligands.

To complement this general model of directed evolution. Once the optimal evolutionary path has been found, we are able to use game theory to provide insights into the optimal duration and steepness of the selection steps necessary to arrive at the desired outcome. As an example of this, we developed a model for the evolutionary landscape of the Ura3 gene under various concentration of 5-FOA.

The aforementioned tools are the base for a comprehensive toolbox to implement directed-evolution of yeast mating receptors. This contributes towards our goal of recognizing and bind alternate peptide ligands, hence supporting and complementing our experimental approach and biological parts.