Developing a Yeast-Derived Detective Biosensor of Kidney Disease

Design

We developed a computational algorithm based on evolutionary data that predicts a set of targeted mutations. Using the recently developed AlphaFold protein structure prediction software, our pipeline models the receptor-biomarker binding and subsequently tests different targeted mutations to computationally optimize this binding. As a result, this algorithm reduces wet lab time and resources, hence allowing for the development of versatile, user-friendly, and far-reaching diagnostic biosensors implemented even in low-resource regions.

In our design process, we first mutated the most similar amino acid then mutated to the most similar evolutionary amino acid and received stronger binding energies. To solve this problem, we developed a computational algorithm based on evolutionary data that predicts a set of targeted mutations.

On the wet lab end, we created a plasmid with a Ste2 receptor DNA sequence integrated in the multiple cloning site (region in a plasmid with multiple restriction sites), an origin of replication (for both E. coli and yeast), and a selectable marker (for both E. coli and yeast). For E. coli, we inserted an ampicillin resistance gene (AMP) and for yeast, we inserted a Ura3 gene.

Learn

Our computational algorithm required us to first come up with a list of candidate mutations by identifying the interacting residues using Prodigy/PYMOL. We are identifying the interacting residues to determine if mutating a residue will optimize binding by reducing the binding energy.

To be more time-efficient, we adjusted our wet lab protocol and used an in-yeast assembly in place of the Gibson assembly, and it was successful. The Gibson assembly involved linearizing the plasmid and inserting sequences carrying the targeted mutations. Within these sequences, homologous regions allow for orientation-specific annealing which provides a complete vector with the targeted mutations which would be subsequently transformed into bacteria then yeast. Our time needed for experimentation went from 14 days to 3 days, significantly reducing the time constraints and also making resource usage more efficient.

Build

Using the recently developed AlphaFold protein structure prediction software, our pipeline models the receptor-biomarker binding and subsequently tests different targeted mutations to computationally optimize this binding. This algorithm reduces wet lab time and resources, hence allowing for the development of versatile, user-friendly, and far-reaching diagnostic biosensors implemented even in low-resource regions.

Gel electrophoresis for in-yeast assembly in order to determine that mutations were integrated into yeast:

The molecular weight of the mutated plasmid is consistent, indicating that our mutagenesis assembly protocol is accurately functioning with high efficiency.

Test

The bulk of our testing involved wet lab work. We integrated the mutated receptor in the plasmid then expressed the plasmid in genetically engineered yeast. Yeast has several unique properties which makes it an ideal candidate for testing our pipeline. Yeast cannot make uracil on its own, does not express Ste2 (so no competition for space with mutated receptor), and has fluorescence protein (YFP) attached to the promoter of FUS1, a protein expressed as a result of the signal transduction of Ste2. Thus, by measuring the fluorescence response to biomarkers in yeast, we can receive results as to whether our receptor was successfully mutated.

Fluorescence response:

Overall, the fluorescence data was an exciting success for our project, indicating the accuracy of our computational model. It showed that we can generate in silico mutant proteins, engineer yeast to express those proteins, and measure how the in silico mutations affect the binding of the engineered protein to the ligand. In response to this result, we moved to make modifications to both our computational and wet lab protocols, implementing the lessons from this round of experimentation. Our iterative workflow is documented in our modeling and proof of concept wiki pages.