Deriving Consistent Core Models from Genome-Scale Metabolic Networks and Their Application for Computing Metabolic Engineering Targets

Stoichiometric and constraint-based modelling has become an essential tool for analysing properties and capabilities of metabolic networks and also for identifying suitable metabolic engineering targets.  Up to now, more than 100 genome-scale metabolic networks have been reconstructed for different organisms and the process of compiling these models becomes faster and faster. However, with increasing size and complexity of genome-scale models, the computational effort for their analysis increased as well and some methods, such as elementary-modes analysis, ¹³C flux analysis, or kinetic modelling cannot be applied to networks consisting of several thousand reactions. Furthermore, simulation results obtained with genome-scale models are sometimes difficult to interpret and studying basic principles of an organism’s (central) metabolism might be easier with metabolic core models. An automated and unbiased reduction procedure delivering meaningful core networks from well-curated genome-scale reconstructions is therefore highly desirable.

For this purpose we have recently developed the NetworkReducer algorithm which derives core models (sub-networks) of genome-scale reconstructions by applying network pruning and network compression techniques [1]. NetworkReducer delivers reduced models that i) are consistent with the genome-scale network with respect to the parts maintained, ii) preserve defined key functionalities, and iii) are accessible for comprehensive and detailed analyses including full enumeration of elementary modes.

We demonstrate scope, predictive capabilities, and utility of a core model obtained by applying NetworkReducer to the latest genome-scale reconstruction of E. coli (iJO1366). We specified desired behaviours that have to be preserved by the reduced network. For example, the core model must allow for 99.9 % of the genome-scale optimal yields for biomass and selected metabolites and products.

The derived core model EColiCore2.0 comprises 499 reactions and 519 metabolites. We present results of constrained-based analyses of EColiCore2.0 (including flux variability, gene essentiality, and elementary-modes analysis) and relate them to results from the full model demonstrating the benefit of a reduced model being consistent with its genome-scale counterpart.

Further we analysed EColiCore2.0 with respect to its capability to identify targets for metabolic engineering purposes. We used the concept of constrained minimal cut sets (MCSs) and enumerated in the genome-scale and in the core model reaction knockout sets for different engineering goals. These knockout sets were further analysed for commonalities and differences between the intervention targets found. Finally, we demonstrate how identified knockout sets of the core model can be used as seeds for computation of MCSs in genome-scale networks. This approach reduces the computational burden in large metabolic models considerably and enables the identification of intervention strategies with higher cardinalities in genome-scale networks with reasonable runtimes.

In conclusion, we demonstrate that consistent core models derived from genome-scale metabolic models may complement or/and simplify genome-scale analysis. EcoliCore2.0 provides a valuable tool to analyse features of the central metabolism of E. coli and also supports the computer-aided search for intervention strategies for rational metabolic engineering.

[1] Erdrich P, Steuer R, Klamt S (2015) An algorithm for the reduction of genome-scale metabolic network models to meaningful core models. BMC Systems Biology 9:48.