(614f) A Genome Scale Model for Prediction of Growth Rates and Fluxes for Rhodococcus Opacus PD630 Metabolism | AIChE

(614f) A Genome Scale Model for Prediction of Growth Rates and Fluxes for Rhodococcus Opacus PD630 Metabolism


Schenk, C., Basque Center for Applied Mathematics
Anthony, W., Washington University School of Medicine in St. Louis
Carr, R., Washington University in St. Louis
Dantas, G., Washington University in Saint Louis
Tang, Y., Washington University in St. Louis
Moon, T. S., Washington University in St. Louis
Garcia Martin, H., Joint BioEnergy Institute (JBEI)
Problem (background/scope/motivation).

Rhodococcus opacus PD630 is a gram positive bacterium with high tolerance to phenolic compounds and able to produce significant amounts of triacylglycerol (TAG). These characteristics make R. opacus an excellent platform for converting lignin monomers into liquid fuel components, yet the mechanisms underlying its phenolic tolerance and fatty acid production are largely unknown. In this study, we present iGR1576, a genome-scale model (GSM) of R. opacus metabolism to address these questions. The transcriptome and fluxome of this species were studied when grown with sugars and aromatics for a base strain, as well as for adaptively evolved mutants on numerous aromatic carbon sources. A key finding from these studies is that the adaptive mutants improved their growth up to 1900% growth on aromatic substrates, despite lacking many mutations (~8 on average) or significant flux rewiring [1, 2, 3]. The mutant strains, however, show differences in their transcriptomic profiles when compared to the wild type strain, and we believe that these changes account for mutant’s abilities to tolerate and utilize higher concentrations of aromatics.


A novel genome scale model for R. opacus was developed in a reproducible manner using CarveMe. A post-processing algorithm was developed to polish the model, and the GSM was evaluated with MEMOTE, the Metabolic Model Tester. The overall score for the annotated and curated model is 93%. As a reference, a recent E. coli genome scale model, iML1515, has an overall MEMOTE score of 91% [4].

Along with flux balance analysis (FBA) and parsimonious flux balance analysis (pFBA), two methods were used to convert transcript measurements into flux values: E-Flux2 and SPOT. E-Flux2 maps normalized transcripts to flux bound constraints and then solves an FBA problem with these modified bounds. Instead of maximizing growth, SPOT maximizes the correlation between fluxes and corresponding transcripts based on the uncentered Pearson correlation coefficient, a linear correlation measure. These methods were evaluated on their ability to predict growth rates. This was done by setting the carbon uptake flux to the experimentally measured value (units = mmol substrate / g dry cell weight / hr), so that the predicted biomass flux can be compared to the measured growth rate (units = hr-1). The accuracy of GSM flux values was determined using 13C-Metabolic Flux Analysis (13C-MFA) fluxes as ground truth values. ~40 reactions were mapped from the GSM to the core metabolic network used for 13C-MFA, and this allowed for direct comparison of GSM flux states and 13C-MFA values. The quality of GSM fit was determined based on Mean Absolute Error (MAE) and R2 between the MFA flux and the GSM predicted flux.


E-Flux2 made the most accurate growth rate predictions, while the other methods had larger errors. FBA, pFBA, and E-Flux2 all predict faster growth rates for the phenol cultures than the experimentally measured growth rates. Using fluxes from 13C-MFA as the true reaction rate values, the general trend for flux prediction accuracy in order of most accurate to least accurate method was: E-Flux2, SPOT, pFBA, and FBA. For flux predictions of glucose pathways FBA had a mean absolute error (MAE) of 58.6, pFBA: 53.37, SPOT: 44.43, and E-Flux2: 28.05. E-Flux2 had the greatest correlation between 13C-MFA fluxes and predicted fluxes with an R2 of 0.49. A major cause for the inaccuracies of glucose predictions is the preference of the EMP pathway for glucose consumption over the ED pathway. The EMP pathway and ED pathway differ in that the EMP pathway produces more ATP, but requires more enzymatic steps. R. opacus has been shown to metabolize nearly 100% of glucose through the ED pathway, leading to reduced enzyme cost [3]. The ED pathway produces less ATP, so when the models are optimized for growth rate, the EMP pathway is preferred. Both FBA and pFBA predict 0% of glucose flux through the ED pathway while SPOT and E-Flux2 predict ~10% and ~45% of glucose flux through the ED pathway. For phenol the predictions of FBA and pFBA are more accurate than for glucose with R2 values of 0.72 and 0.88. This is likely because there are fewer competing pathways for phenol uptake.


It is not surprising that FBA and pFBA would overestimate growth rates since they predict the maximum theoretical growth rate. In practice, we would expect that the actual growth rate would be less than the theoretical maximum due to other factors. For example, soil bacteria like R. opacus need to consume many carbon sources, and maintaining that ability imposes a cost on the growth rate for any one carbon source. Additionally, FBA and pFBA seek out the most efficient use of carbon resources and do not factor in competing interests including the cost to make the enzymes. A pathway that is carbon efficient, but requires high enzyme cost would allow for a high FBA or pFBA growth rate, but would have low in vivo flux due to its overall resource cost. FBA, pFBA, and E-Flux2 all underestimate the growth rate of glucose cultures which is unexpected because this means that the cells are growing faster than the theoretical maximum growth rate.

This study found that transcript values can be used to improve genome scale model flux predictions. This is true for both phenol and glucose conditions and for growth rate and flux predictions. However, there is still room for improvement of predictions through machine learning approaches.

  1. Henson WR, Campbell T, DeLorenzo DM, Gao Y, Berla B, Kim SJ, et al. Multi-omic elucidation of aromatic catabolism in adaptively evolved Rhodococcus opacus. Metab Eng. 2018;49: 69–83. doi:10.1016/j.ymben.2018.06.009
  2. Yoneda A, Henson WR, Goldner NK, Park KJ, Forsberg KJ, Kim SJ, et al. Comparative transcriptomics elucidates adaptive phenol tolerance and utilization in lipid-accumulating Rhodococcus opacus PD630. Nucleic Acids Res. 2016;44: 2240–2254. doi:10.1093/nar/gkw055
  3. Roell GW, Carr RR, Campbell T, Shang Z, Henson WR, Czajka JJ, et al. A concerted systems biology analysis of phenol metabolism in Rhodococcus opacus PD630. Metab Eng. 2019;55: 120–130. doi:10.1016/j.ymben.2019.06.013
  4. Monk JM, Lloyd CJ, Brunk E, Mih N, Sastry A, King Z, et al. iML1515, a knowledgebase that computes Escherichia coli traits. Nat Biotechnol. 2017;35: 904–908. doi:10.1038/nbt.3956