Comparing knockout strategies from in silico rational strain design tools

In the last decades, in silico constraint-based modeling became an important tool for engineering of microbial cell factories for industrial purposes. A number of computational methods currently allow determination of efficient intervention strategies for most metabolic engineering problems [1][2][3]. Although useful, some of these methods rely on the assumption that wild-type and mutant cell metabolism is optimal for a certain objective, which may not always be true. Alternative methods based on elementary modes only recently became available for genome-scale metabolic models with the MCSEnumerator methodology [4].The concept of constrained minimal cut sets (cMCS) exploited by this method defines a minimal intervention strategy that always blocks an undesired phenotype while keeping a desired one [5]. Intervention strategies from optimization-based methods guarantee product coupling given there is growth maximization. With cMCSs, product coupling occurs with any cell objective. These types of product coupling have been called, respectively, strong and weak coupling [6]. While both methods have proven useful, there is still a gap regarding the comparison between the two methods in realistic metabolic engineering case studies. In this work, an in-depth analysis of both solution types is presented for a case study regarding succinate production in Saccharomyces cerevisiae using the iMM904 model reconstruction. An analysis pipeline has been developed to assess productivity, robustness and metabolic pathway usage with use of parsimonious flux balance analysis (pFBA) and flux variability analysis (FVA). All subsets of a solution are generated and run through the pipeline. Solutions are determined using an evolutionary algorithm (EA) and an in-house implementation of MCSEnumerator's formulation. Solution similarity between both methods is also assessed. While biomass-product coupled yield is higher for EA solutions, product coupling requires a minimum fraction of biomass flux to occur. With cMCSs, maximum productivity is generally lower but product synthesis always occurs, making these solutions more robust. Their minimality ensures all knockouts are needed for strong product coupling. However, some subsets of cMCSs exhibit the same phenotype as EA solutions. Similarity analysis further proves that there are knockouts common to both solution types. cMCSs are a useful tool for understanding what interventions are needed for strong product coupling, but their in vivo implementation can be difficult due to lower biomass and product yield. Even so, as we show here, one can build weak coupled solutions from their subsets. Overall, it is shown that there are shortcomings for both methods. For some microbial cell factories, the optimal growth assumption may stand true and traditional optimization methods provide intervention strategies with higher biomass-product yields. More complex cells may not exhibit the predicted phenotypes and thus, cMCSs provide an unbiased, optimal answer for any metabolic engineering problem