Team:UCopenhagen/Engineering

Introduction

This page documents our success with iterative engineering of novel G protein alpha subunits in yeast Saccharomyces cerevisiae. Protein modeling in Rosetta took us most of the way, finally accompanied by wet lab experiments. Our efforts followed the traditional engineering design cycle (see figure 1), with multiple phases of research followed by testing of our models and continuously refining them. Individual steps are noted chronologically as they happened, thus proceeding from relatively simple to more advanced and specialized methods. Finally, the results challenged our initial assumptions about protein interactions.

Fig. 1: Expanded engineering cycle.
We closely followed the expanded engineering cycle in our engineering process.

As described on our project design page, our most ambitious design called for modification of the alpha subunit of endogenous yeast G protein. The challenge here was that we wanted to maintain its natural functionality while expanding it to serving our purposes.
This meant we needed to design a G-alpha that:

A) has a natural-like affinity for the beta/gamma complex of the yeast G protein to stay bound to the beta/gamma complex when no signal was present, thus inhibiting leaky activation of the pathway.
B) has a low enough affinity for the beta/gamma complex after cleavage by our TEV-protease to enable dissociation from the beta/gamma complex.

Our work on G-alpha hinged on finding a mutant that preserved natural function when there was no signal (i.e. no cytokine), and also dissociated appropriately in the presence of a signal.

This page presents all stages of G-alpha engineering in a chronological order. The progress followed the engineering cycle (design-build-test-learn) as we iterated our designs several times. As we utilized primarily dry-lab experiments, the whole cycle often spans over each stage. Every stage was informed by all the previous ones.

Stage One - Design

We based our design choices on knowledge about the interaction interface of the alpha and beta subunits in bovine G protein, as we found no crystal structure of yeast G protein. Immediately, we noticed that the yeast G protein contained an ‘insertion sequence’ (highlighted in figure 2) that other G proteins did not. This indicated that the sequence was not essential for the general functionality of the G-alpha subunit. As such, we chose to search this sequence for a mutagenesis site.

Based on literature review, we identified the residue sequence that the TEV-protease cleaves: EXLYΦQφ, where X is any residue, Φ is any large or medium hydrophobic residue, and φ is any small hydrophobic or polar residue. For the sake of interpretability of experiments, however, it was necessary to introduce a single sequence type at all times, as the TEV protease may display different proteolytic rates depending on the recognition site (Dougherty et al., 1989). Following extensive literature search, we opted for a widely used sequence: ENLYFQG.

Figure 2: Amino acid sequence of the S. cerevisiae G protein alpha subunit.
The highlighted sequence was initially selected as an optimal region for mutagenesis.

Stage Two - Design-build

We received a file with the empirically observed binding interface of bovine G-alpha and G-beta through our supervisors. We could translate known characteristics of the bovine G protein into what we expected to see in the yeast G protein because of the similarity of their sequences. Based on the biophysical properties of different amino acids, we identified what we believed were the most essential residues for the binding between the alpha and beta subunits (see figure 3).

Figure 3: Crystal structure of the G protein complex.
The red-colored residues are the ones we identified as the most important for binding of the alpha and beta subunits to each other.

Stage Three - Test & Learn

Based on suggestions from the Sinisens (Aalto-Helsinki iGEM2020 team), we chose to use Rosetta for our protein engineering. Rosetta is a powerful tool for mapping and characterizing protein structures and interactions, and, we theorized, good for mutagenetic analyses of our G alpha protein.
Using Rosetta, we supplemented our own analysis with an alanine scan, which provided us with quantitative predictions of the role of selected residues in the G-alpha in terms of binding to G-beta. Rosetta's alanine scan is a script that calculates how the Gibb's free energy of two proteins change upon changing a single amino acid residue to an alanine (see modeling for more details).

The results of the alanine scan confirmed our intuitions to a good extent, and offered several valuable insight points that led to the refinement of our G-alpha designs. As illustrated on figure 4, residue number 227 was predicted as the single most important residue in the protein-protein interaction between the alpha and beta subunits. This result, among mutagenetic data found in literature about what the most functionally important amino acid residues in G-alpha are , guided us in discerning between which ones we could change, and which ones were critical for the correct functionality of G-alpha.

Figure 4: Change in Gibbs free energy (in kcal/mol) generated by replacing the given residues with alanine. As Gibbs free energy correlates with a change in affinity, it appears that residue 227 is the most important one for the binding interface of G-alpha and G-beta proteins.

Stage Four - Redesigning, Building, Testing & Learning

By compiling the information we had received so far, we isolated fitting mutation points in G alpha, where we

1. Avoid changing residues that lead to loss of function in G alpha (Gladue, D. P. 2008).
2. Avoid changing residues identified as very important through Rosetta's alanine scan.
3. Change residues located in the insertion sequence and other non-conserved residues, by comparing the yeast G-alpha to the bovine G-alpha.
4. Change residues located in loops based on visual analysis in Pymol.

Figure 5: Localization prediction of modified alpha subunits.
Cello was used to predict the localization of our three GPA1-versions. For all three versions, an increased cytoplasmic reliability score was observed compared to the wildtype.

With this information in mind, we designed three version of G-alpha, each with varying degrees of modifications. Using SwissProt protein modeling, we then confirmed that none of our edits lead to unexpected conformational changes. In addition, we used Cello to predict protein localization to the plasma membrane, as this is crucial for the functionality of our biosensor. As shown on figure 5, our designs yielded even better localization scores than the wildtype.
As shown on the figure, some cytoplasmic localization was also predicted, but this was desirable as certain enzymes in the cytoplasm are known to modify G-alpha for it to go to the membrane.

Stage Five - Design, Build, Test & Learn

Here we faced the reality of laboratory experimentation. Three G-alpha mutants with many diverse mutations could lead to lots of ambiguity as the laboratory results could be hard to interpret due to inability to monitor what mutation had a profound impact. Time constraints in the wet lab motivated us to limit the number of possible mutation points. We eliminated point mutations and kept only cleavage sites for TEV protease. To improve the quality of future experiments, we expanded the range of G-alpha candidates from three to seven by introducing all combinations of the three selected mutations. We checked folding predictions for all seven novel G-alphas which resulted in a significant overlap (as visible in the figure). None of the proteins seemed likely to take on a faulty conformation according to the SwissProt analysis.

Figure 6: Alignment of predicted structures of our G-alpha mutants. Apart from the intrinsically disordered region on the left, the predicted structures overlap fairly well. This indicates that the conformation of none of the G-alpha mutants was severely impacted by their mutations.

Stage Six - Test & Learn

In order to predict the behavior of the fragments generated by cleavage of our mutant G-alpha by the TEV protease, we decided to run a simulation in Rosetta (see modeling for more details). However, as this was an extremely demanding computational analysis we could not perform ourselves, we reached out to Assoc. Prof. Kurt V. Mikkelsen of Theoretical Chemistry at the University of Copenhagen, who has a supercomputer capable of such demanding calculations. With a help from Andreas Erbs Hillers-Bendtsen, MSc student of chemistry at Copenhagen University, we were able to simulate the change in Gibbs free energy of the seven mutant G-alphas (see figure 7), and the nine fragments formed after cleavage (figure 8). Our goal was to identify a G-alpha mutant candidate with binding affinity similar to the wildtype, but with the lowest possible binding affinity of its fragments after cleavage. Our simulation lead to some unexpected findings:

The majority of the fragments seemed to have a higher affinity towards G-beta than the non-cleaved G-alphas had. Only mutant m124 displayed a higher affinity than its corresponding fragments.

Figure 7: Changes in Gibbs free energy of our G-alpha mutants relative to the wildtype. Numbers in their labels signify what cleavage sites were included in each mutant. As all changes in energy are negative, it indicates that all mutants have a higher affinity to G-beta protein.

Figure 8: Change in Gibbs free energy of fragments produced by cleavage of G-alpha mutants. The x-axis denotes the length of each fragment. There seems to be a slight trend towards a higher affinity in smaller fragments, most likely caused by more degrees of freedom for a binding region constituting a majority of such fragments.

Initially, we were startled by these findings, however after thorough discussions we concluded that the overall G-alpha conformation probably restricted some degrees of freedom of the high-affinity regions. These regions facilitated most interaction between G-alpha and G-beta, as we learned in one of the earlier stages. Our cleavage sites were located between these high-affinity regions, therefore, cleavage effectively freed the high-affinity regions from the previous restrictive conformation and allowed them to bind G-beta more tightly. Obviously, these findings turned our initial conclusions about the protein-protein interaction on its head. We also realized that affinities do not necessarily provide the full answer to successful signal transduction. In accordance with this, further analysis showed that conformational changes of G-beta and their causes could not answer our questions.

Since mutant m124 was the only mutant whose corresponding fragments had a lower affinity to G-beta compared to the mutant itself, we decided it was the most promising candidate for our biosensor based on the data from the simulation.

Stage Seven - Test & Learn

Based on the conclusion from the previous Rosetta simulation, we designed and ran another simulation to obtain more refined details about the m124 fragments (see modeling for more details). The results showed that all generated fragments had a significantly higher affinity to the G-beta compared to m124. Although exploration of the PDB files revealed that spatial assignment of the fragments showed clashes (figure 9) so that the affinity of each fragment could be limited by the presence of the other fragments, ultimately leading to diminished presence at G-beta. Nevertheless, the results of our simulations could not inform us about conformational changes of G-beta which protected us from formulating clear conclusions.

Figure 9: Change in Gibbs free energy of fragments produced by the cleavage of G-alpha mutant m124.
The labels on the x-axis signify the position of each fragment in the original protein.

Figure 10: An example of clashing positions of the fragments. This indicates a situation that cannot happen in the real-world conditions, thus reducing our trust in the prediction.

Wet Lab

As part of our wet lab, we created four different mutant constructs of GPA1, named GPA1m1, GPA1m2, GPA1m3, and GPA1m4. Mutants 1, 2 and 4 correspond to the m1, m2, and m4 denotations used in our dry lab modeling. The Wt GPA1 was replaced with the different GPA1 mutants in an adenosine biosensor strain of Saccharomyces Cerevisiae, previously created by our supervisor Sotirios Kampranis' lab. In addition, an inducible TEV protease was inserted into the same locus of the S. Cerevisiae chromosome as the rest of the biosensor constructs.

As a result, we were able to test out both the maintenance of the original functions of GPA1 in a GPCR receptor system, and to verify that the GPA1 mutants were cleaved by the induced TEV protease, as this would result in higher concentrations of the reporter.
INSERT FIGURE OF A2A.
The experiments showed that the amount of reporter was increased drastically in all biosensors when the TEV protease was induced by growing the cells in galactose. This included the biosensor with wt GPA1 which made it hard to conclude whether some of the increased signal was caused by cleavage of the GPA1 mutant or that the increase in signaling came from other parts of the system.

The experiments further showed a loss-of-function in the sense that none of the biosensors with mutated GPA1 proteins were able to respond to different levels of adenosine whereas we were able to reproduce the A2A adenosine biosensor created in the Sotirios Kampranis lab when we used the wt GPA1 and grew the cells in glucose media.

Future Vision

Dry lab

The next round of simulations could be based on molecular dynamics for us to see whether the results of Rosetta could be trusted. Apart from that, the focus could shift from affinities to conformations, and particularly G-beta could be more explored.

Wet lab

In the wet lab, the next step would be to understand why the signaling also increases in the biosensor with wt GPA1 when the cells are grown in galactose and raffinose media. First, we would create a strain of the biosensor without the inducible TEV protease to see whether the spike in signaling is caused by the induction of the TEV protease or something else related to the change of the media. If the TEV protease does not seem to be the cause we would investigate the promoters that we use, some of these are glycolytic promoters like pPGK1 and the different media might cause slightly different expression levels even though the promoters are normally classified as constituent. To further investigate the loss-of-function of the mutants, we would like to create strains of the biosensor with the mutated GPA1 proteins but without the inducible TEV promoter. This would enable us to see whether leaky expression of the TEV protease in the glucose fed cells might have caused the loss-of-function.

Douglas P. Gladue, James B. Konopka, Scanning mutagenesis of regions in the Gα protein Gpa1 that are predicted to interact with yeast mating pheromone receptors, FEMS Yeast Research, Volume 8, Issue 1, February 2008, Pages 71–80, https://doi.org/10.1111/j.1567-1364.2007.00311.x↩