Team:Imperial College/Model
Modelling
· · ·
Introduction
Around the same time we began working on this project a paper (Jie Zhang et al., 2019) came out which used parsimonious FBA, the COBRA toolbox in MATLAB, and the 7th version of the genome-scale consensus Saccharomyces cerevisiae model as a discovery tool to predict which genes were good targets for Tryptophan optimisation. Their code was made publicly available under the MIT licence on GitHub (https://github.com/biosustain/trp-scores),
and this code was our starting point for some minor modifications to carry out our modelling. We extend our thanks to Jie Zhang et al. for making this code available.
Potential Target Validation
For our initial analysis and to verify the literature identified targets, we used Zhang et al.’s code to optimise for Tryptophan synthesis flux and documented the changes in reactions of interest:
Table showing the results of the initial modelling analysis.
This data provided a good sanity check that the majority of our potential targets for knockdown (MDH2, CDC19, ARO7, ARO8) had flux downregulated when setting Tryptophan synthesis as the objective function. Similarly all of
our potential targets for overexpression (ARO4, TKL2, and especially TRP2) has flux significantly increased due to setting Tryptophan biosynthesis as the objective function.
Design Validation
Next we hoped to test specific design implementations of our project (2 overexpressions, and up to 4 knockdowns) within FBA. To generate a specific design, we placed flux constraints on target reactions. Determining how to
set constraints was a key challenge, because without any experimental data, we were unable to get a “realistic” value for each flux after an overexpression or knockdown. One approach we could have taken was to arbitrarily
decrease or increase the lower and upper bounds. Instead, we made the assumption that the Tryptophan optimised fluxes presented above, were at least close to what may be realistically possible. We therefore chose to use
those fluxes as the constraints. For an overexpression, this involved setting the lower bound of flux at 0.99 X Trp Opt flux. For a knockdown, this involved setting the upper bound as 1.01 X Trp Opt flux. The ± 1% here
was necessary to ensure that COBRA was able to find a solution.
We then set the objective function back to growth. This was necessary to demonstrate that our introduced changes led to increased Tryptophan yield alone thus validating our optimisation designs. The assumptions made decreased
the accuracy of the model but suited its purpose as a sanity check. We then recorded the output flux for growth and Trp biosynthesis for some designs:
Table showing the results of the second modelling analysis.
Seemingly the only important factor for increasing Tryptophan yields is the overexpression of TRP2. We believe this is due to the assumptions of FBA, where because it is optimised for growth, even increased fluxes or decreased
fluxes of the other enzymes of interest were ultimately directed towards biomass, whereas increased TRP2 flux always led to increased Tryptophan synthesis. Furthermore, it is likely that the only possible solution after
TRP2 flux had been increased also involved increasing the fluxes of the other target pathways, and therefore constraining those fluxes makes no additional difference. Nevertheless, this modelling still has some utility
in that it demonstrates that under our designs we still expect growth, albeit at half the wildtype growth rate. Secondly, it demonstrated that one potential design even led to a reduction in Tryptophan flux, thus allowing
us to rule out that particular design.
Choosing targets
Based on literature searches and the modelling output, we first decided to include TRP2 as an overexpression in our design. TRP2 was the enzyme with the largest increase in flux in the first analysis and in the second analysis,
increasing TRP2 flux was a requirement for increasing Trp synthesis. We decided to exclude MDH2 as a potential knockdown target due to lack of evidence of its impact on Trp synthesis. Moreover, we decided to keep the other
potential knockdowns (ARO7, ARO8, and CDC19) in our design because each of these pathways did show some significant decreases in flux when optimised for Tryptophan synthesis as opposed to growth
The key question left was to decide between ARO4 and TKL2 for the second gene to overexpress. Neither showed large increases in flux when the objective function was set to Trp synthesis. However, ARO4 was switched on compared
to having 0 flux in the wildtype solution, indicating that expressing ARO4 outside of native regulation might increase Tryp synthesis. A further point in favour of choosing ARO4 was that in the literature a mutant version
of ARO4 (ARO4 K229L) which has removed allosteric feedback inhibition and shows similar activity in the absence or presence of the aromatic amino acids (Markus Hartmann et al., 2003). We hoped this would therefore potentially
make it a better target for overexpression.
Markus Hartmann, Thomas R. Schneider, Andrea Pfeil, Gabriele Heinrich, William N. Lipscomb & Gerhard H. Braus 2003, "Evolution of Feedback-Inhibited β/α Barrel Isoenzymes by Gene Duplication and a Single Mutation", Proceedings of the National Academy of Sciences - PNAS, vol. 100, no. 3, pp. 862-867.
