Outside of the lab

Genome Scale Metabolic Model


Due to the slow growth of the high fatty acid producing strain developed by the Systems and Synthetic biology division at Chalmers University of Technology (SysBio)(Yu et al., 2018) , a native yeast strain was instead used in the lab. Therefore, we decided to create a genome scale metabolic (GEM)-model of the base strain of the high fatty acid producing strain (Zhou et al, 2016) and investigate how its growth i.e biomass production differed from native yeast.

GEMs & Flux Balance Analysis

A Genome Scale Metabolic model (GEM) is a computational reconstruction of the metabolism of an organism. Most, if not all, metabolites and metabolic reactions of the organism are included in the model, with a boundary containing most reactions, erected around it. The model itself consists of a stoichiometric representation of all compounds across each and every reaction. This can be described using a matrix (typically referred to as the S-matrix), where rows represent compounds, columns represent reactions, and each cell contains the stoichiometric coefficient of a particular reactant. Generally, solving the metabolic flux (the “speed” of which a reaction occurs at) distribution given a series of input fluxes, i.e. substrate compounds used up by the organism, is impossible using the S-matrix alone, as the this matrix is typically overdetermined. Nonetheless, one can utilize linear programming to solve this flux distribution given an objective function that the system wants to maxime. This is the fundamental concept of Flux balance analysis (FBA) , where one looks at the metabolic in- and output-fluxes of different reactions, under the assumption of steady-state and a given objective function. For microbial organisms, this objective function is commonly set to maximizing growth, i.e. optimizing for the reaction that produces biomass. This can then be used to gain insight into the metabolic state of an organism (Wang et. al, 2018).

  Experimental Procedure

RAVEN, a toolbox for genome scale modeling and reconstruction developed by SysBio, was used to create the GEMs. To obtain a GEM for the intermediate YJZ045 strain we started the latest native yeast-GEM developed by SysBio(Yeast8)(Lu et al.,2019) and added, as well as removed reactions based on what genomic changes had been done to the native yeast in (Zhou et al, 2016). A total of 4 genes were added: ME (Malic Enzyme) from Rhodosporidium toruloides , ACL (ATP citrate lyase) from Mus musculus, and tesA (Thioesterase) from E. Coli. Additionally, 4 genes were deleted: POX1, FAA1, FAA4 and HFD1.

Based on this, 3 different models were created in addition to the already present native yeast GEM, seen in the table below.
Table 1. Three different models in addition to the native yeast GEM.
Gene Addition No Gene additions
Gene Deletions YJZ045 II
No Gene Deletions III Native Strain
In the flux balance analysis of the GEMs we decided to use a maximum input of 1000 mmol/gDWh oxygen and 1000 mmol/gDWh glucose because we did not have any reference values from experiments. The objective function was set to maximize biomass production.

At first the resulting metabolic output suggested that the yeast did not produce any ethanol, which is something it should do naturally due to the Crabtree effect (Badford et. al, 1979). Therefore we decided to implement enzymatic constraints using GECKO. This works by introducing enzymatic rates into the reactions as well as having a maximum amount of enzymes that can be used in them. Overall this limits the metabolic fluxes, as metabolic rates can’t proceed infinitely fast in vivo due to a limited protein pool in the cell and a limited catalytic efficiency for each enzyme (Domenzain et al., 2021). When we incorporated enzymatic constraints in our GEM, the yeast produced ethanol as expected. Biologically this can be explained by the fact that fermentation is a more optimized route for ATP/biomass generation, i.e. fewer enzymes are required to produce one unit of ATP compared to respiration. The TCA cycle and electron transport chain contain a multitude of large and complex proteins, which take up a significant chunk of the cell's protein pool.


How the growth for each of the GEMs differ from that of the native yeast is presented in the table below. From analysing the gathered data, model III seems to increase the most in terms of the biomass production closely followed by the YJZ045-strain. Since it has been established in The YJZ045 strain also produces more free fatty acids compared to native yeast (Zhou et al, 2016), it would therefore be a suitable candidate to continue our experiment with. However, due to lack of access to the strain, we could not implement this finding.
Table 2. Comparing growth rate of different strains with that of native yeast.
Strain Growth difference [mmol/gDWh]
Native 0
II -4.5e-3
III +0.5443
YJZ045 +0.5422

Data and Code Availability

All data, model files and MATLAB scripts are freely available to use for anyone interested in simulating yeast metabolism. You can find them over at our iGEM team GitHub repository: LINK

Structural simulations using RoseTTAFold

  Structural Data for Investigating Substrate Specificity

In our project, we designed a yeast strain to carry three different thioesterases on different induction systems, allowing us to tailor the fatty acid output of our yeast. The exact nature of the created profile depends on what fatty acid chain length each thioesterase cleaves the ACP moiety at, and they tend to be somewhat promiscuous, with TesA for instance targeting chains between 8 to 16 carbons long (Steen et al., 2010). By investigating the structure of the protein via methods such as protein/ligand docking, active site determination and molecular dynamics simulations, the key features of their ligand preference can be determined. This data then aids in the design of novel enzyme structures that improve performance, in this case the selectivity. A novel enzyme specific for just a single chain length would allow for unprecedented control over the fatty acid profile produced.

  RoseTTAFold Predictions of Protein Structure

A recent and very potent development to the field of structural prediction is the development of RoseTTAFold and DeepMind predictive AI/machine learning programs to model proteins without a previous structure to model the target protein onto, something which has been a dream of scientists for over half a century (Tunyasuvunakool et al., 2021). As this software is open source and available for the public to use, we decided to model the three thioesterases employed in our system due to, as there weren’t structures of all three thioesterases on which to base a homology modelling strategy.

  A Future Path Towards Engineering Better Thioesterases

The structures we obtained from RoseTTAFold (figures 1, 2, 3) show promise, and would allow us to further develop our project by designing modified thioesterases for improved selectivity by performing ligand-protein docking simulations, followed by molecular dynamics simulations to determine key structural features for selectivity, upon which we could then improve. The goal is to produce several different enzymes with a single chain length specificity, which would enable us to construct a system that can produce a very specific fatty acid profile.
Figure 1: RoseTTAFold structure of FatB.

Figur 2; RoseTTAFold structure of TesA.

Figure 3: RoseTTAFold structure of TesBT

Data Availability

All code can be found on GitHub – iGEM Judging release.

References for Genome Scale Metabolic Modelling

Yu, T., Zhou, Y. J., Huang, M., Liu, Q., Pereira, R., David, F., & Nielsen, J. (2018). Reprogramming yeast metabolism from alcoholic fermentation to lipogenesis. Cell, 174(6), 1549-1558.

Zhou, Y. J., Buijs, N. A., Zhu, Z., Qin, J., Siewers, V., & Nielsen, J. (2016). Production of fatty acid-derived oleochemicals and biofuels by synthetic yeast cell factories. Nat Commun 7: 11709.

Lu, H., Li, F., Sánchez, B. J., Zhu, Z., Li, G., Domenzain, I., Marcisauskas, S., Anton, P. M., Lappa, D., & Lieven, C. (2019). A consensus S. cerevisiae metabolic model Yeast8 and its ecosystem for comprehensively probing cellular metabolism. Nature Communications, 10(1), 1–13.

Wang, H., Marcišauskas, S., Sánchez, B. J., Domenzain, I., Hermansson, D., Agren, R., ... & Kerkhoven, E. J. (2018). RAVEN 2.0: A versatile toolbox for metabolic network reconstruction and a case study on Streptomyces coelicolor. PLoS computational biology, 14(10), e1006541.

Salari, R., & Salari, R. (2017). Investigation of the best Saccharomyces cerevisiae growth condition. Electronic physician, 9(1), 3592. Badford, J.P.,& Hall, R. J. (1979). An examination of the Crabtree effect in Saccharomyces cerevisiae: the role of respitory adaptation. Microbiology, 114(2), 267-275

Domenzain, I., Sánchez, B., Anton, M., Kerkhoven, E. J., Millán-Oropeza, A., Henry, C., Siewers, V., Morrisey, J. P., Sonnenschein, N., & Nielsen, J. (2021). Reconstruction of a catalogue of genome-scale metabolic models with enzymatic constraints using GECKO 2.0. BioRxiv.

Reference for RoseTTAFold

1: Tunyasuvunakool, K., Adler, J., Wu, Z. et al. Highly accurate protein structure prediction for the human proteome. Nature 596, 590–596 (2021).