Combining evolution and machine learning-guided pathway optimization to engineer a novel methylsuccinate module for synthetic C1 metabolism in vivo

Abstract

De novo metabolic pathways open possibilities for sustainable biotransformations in microbes. However, the in vivo-implementation of such new-to-nature pathways is highly challenging and heavily relies on adaptive laboratory evolution (ALE) of the hosts native metabolic network. Here, we assess how much this need for host-centric ALE can be overcome and/or complemented through the informed design of the newly introduced pathway. Exemplifying for a synthetic CO2-fixation module via methyl-succinate, we established methylsuccinate-dependent growth of Escherichia coli over six months by ALE of E. colis native metabolism. In parallel, we developed a machine-learning guided workflow (MEVIS) for the automated engineering of the synthetic pathway, resulting in methylsuccinate-dependent growth within three weeks. Critically, performing MEVIS in the background of the ALE-evolved strain is necessary to further approach wild-type like growth, demonstrating how ALE in combination with machine-learning-guided lab automation holds great potential to accelerate and improve design-build-test-learn cycles in contemporary metabolic engineering.