Project Team: Christopher J. Marx, Ankur Dalia (Indiana University), Sergey Stolyar, Jeremy Draghi (Brooklyn College), Norma Martinez-Gomez (Michigan State University)

Epistasis – non-linear interactions between genotypic changes upon phenotypes –represents a critical challenge to optimization of biological systems, whether by evolution or engineered via synthetic biology. When mutational effects upon growth or product generation depend on the genetic background, assessing performance across the entire parameter space of any system of realistic size quickly becomes impossible. This is especially problematic when there is sign epistasis – mutations that change from beneficial to deleterious depending upon the other loci – as this creates ridges and peaks on the fitness landscape that can restrict stepwise optimization via either synthetic biological changes or beneficial mutations. Development of kinetic computational models of metabolism can provide guidance, but unfortunately these models are dogged by numerous free parameters. There is an immediate need for two linked developments: empirical techniques that can rapidly generate and assess rational, combinatorial variants, and modeling techniques to incorporate these data and predict where in parameter space further rounds of generation and assessment of variation would be most effective. The test-bed for our novel approach is to optimize the function of the high-efficiency ribulose monophosphate (RuMP) pathway that the team has successfully introduced into the model methanolconsuming organism, Methylobacterium extorquens. First, in vivo gene editing of a plasmid-encoded suite of enzymes will be performed, and deep sequencing used to rapidly assess the fitnesses of a quartermillion genotypes with combinatorial variation in nine dimensions of expression. Second, this massive volume of data about epistasis – combined with direct measurement of intracellular metabolite concentrations for select combinations – will be used to infer the numerous parameter values in our kinetic model. Third, the model will be utilized to predict which regions of parameter space would be more or less evolvable and these will be targeted and compared in the second round of editing, fitness assays, and experimental evolution.