Iterative algorithm-guided design of massive strain libraries, applied to itaconic acid production in yeast

• Published in: Metabolic engineering Vol. 48; pp. 33 - 43
• Main Authors: Young, Eric M., Zhao, Zheng, Gielesen, Bianca E.M., Wu, Liang, Benjamin Gordon, D., Roubos, Johannes A., Voigt, Christopher A.
• Format: Journal Article
• Language: English
• Published: Belgium Elsevier Inc 01-07-2018
• Subjects: Algorithms; Gene Library; Metabolic Engineering - methods; Multivariate statistical methods; Pathway engineering; Saccharomyces cerevisiae - genetics; Saccharomyces cerevisiae - metabolism; Succinates - metabolism; Synthetic biology; Yeast metabolic engineering; Multivariate statistical methods; Synthetic biology; Pathway engineering; Yeast metabolic engineering
• ISSN: 1096-7176; 1096-7184
• DOI: 10.1016/j.ymben.2018.05.002

Metabolic engineering requires multiple rounds of strain construction to evaluate alternative pathways and enzyme concentrations. Optimizing multigene pathways stepwise or by randomly selecting enzymes and expression levels is inefficient. Here, we apply methods from design of experiments (DOE) to guide the construction of strain libraries from which the maximum information can be extracted without sampling every possible combination. We use Saccharomyces cerevisiae as a host for a novel six-gene pathway to itaconic acid, selected by comparing alternative shunt pathways that bypass the mitochondrial TCA cycle. The pathway is distinctive for the use of acetylating acetaldehyde dehydrogenase to increase cytosolic acetyl-CoA pools, a bacterial enzyme to synthesize citrate in the cytosol, and an itaconic acid exporter. Precise control over the expression of each gene is enabled by a set of promoter-terminator pairs that span a 174-fold range. Two large combinatorial libraries (160 variants, 2.4 Mb and 32 variants, 0.6 Mb) are designed where the expression levels are selected by statistical methods (I-optimal response surface methodology, full factorial, or Plackett-Burman) with the intent of extracting different types of guiding information after the screen. This is applied to the design of a third library (24 variants, 0.5 Mb) intended to alleviate a bottleneck in cis-aconitate decarboxylase (CAD) expression. The top strain produces 815 mg/l itaconic acid, a 4-fold improvement over the initial strain achieved by iteratively balancing pathway expression. Including a methylated product in the total, the strain produces 1.3 g/l combined itaconic acids. Further, a regression analysis of the libraries reveals the optimal expression level of CAD as well as pairwise interdependencies between genes that result in increased titer and purity of itaconic acid. This work demonstrates adapting algorithmic design strategies to guide automated yeast strain construction and learn information after each iteration.
•Novel yeast itaconic acid pathway with rewired central carbon metabolism and export.•Cytosolic pathway with bacterial citrate synthase and acetylating aldehyde dehydrogenase.•Design-of-experiments used to design combinatorial libraries that vary enzyme level.•Three rounds of expression tuning improved itaconic acid titer 4-fold.