(568e) Dynamic Sequence Specific Constraint-Based Modeling of E. coli Cell-Free Protein Synthesis

Cell-free protein expression has emerged as a commonly used tool in systems and synthetic biology, and a promising technology for personalized point of care medicine. Systems derived from crude cell-extracts have shown remarkable utility as a protein synthesis technology. Cell-free systems offer many advantages for the study, manipulation and modeling of metabolism compared to in vivo processes, such as direct access to metabolites and the biosynthetic machinery, without the interference of a cell wall or the complications associated with cell growth. However, if cell-free platforms for on-demand biomanufacturing are to become a reality, the performance limits of these systems must be defined and optimized. Toward this goal, we used a dynamic stoichiometric modeling approach to simulate the time-course of cell-free expression of a model protein, chloramphenicol acetyltransferase (CAT). A core E. coli metabolic network, describing glycolysis, the pentose phosphate pathway, energy metabolism, amino acid biosynthesis and degradation was augmented with sequence specific descriptions of transcription and translation, and effective models of promoter function. Model parameters were largely taken from literature; thus, the constraint-based approach coupled the transcription and translation of the protein product, and the regulation of gene expression, with the availability of metabolic resources using only a limited number of adjustable model parameters. Furthermore, the approach constrained the concentration of a subset of metabolites in the metabolic network using time-course concentrations predicted by a previous kinetic modeling approach, itself identified from a comprehensive metabolomic data set including the substrate glucose, protein product CAT, organic acid intermediates, energy species, and 18 of the 20 proteinogenic amino acids. The time evolution of metabolites, scaled enzyme activities, transcription, and translation were modeled using ordinary differential equation material balances, with reaction rates determined by a convex optimization subproblem. This linear programming subproblem minimized the sum of reaction rates, subject to material balance constraints and flux bounds determined from global kinetic parameters. The model was implemented in the Julia programming language while the underlying linear program was solved at each time point using the GLPK solver. The simulation captured biphasic protein production: in the first phase, glucose was consumed as the carbon and energy source, while organic acids accumulated as byproducts. These organic acids, especially pyruvate and lactate, take over as substrates during the second phase following glucose depletion. The model captured the rate of protein production, which occurred at 8 μM/h in the first phase and 5 μM/h in the second. The first phase was characterized by utilization of the Entner-Doudoroff pathway (81% of flux), with glycolysis playing a lesser role (18%), and pentose phosphate being negligible (1%). In the second phase, the Entner-Doudoroff pathway and glycolysis were equally utilized (49% each). To determine the uncertainty in the estimated flux distribution, flux variability analysis (FVA) was conducted, with protein production constrained to the base rate. The FVA analysis suggested CAT production was not greatly affected by the choice of glycolysis, pentose phosphate, or Entner-Doudoroff; thus, substrate utilization is robust. However, reactions more directly associated with protein synthesis (translation, tRNA charging, mRNA degradation) were tightly constrained by the protein synthesis requirement. We also systematically varied, using simulated annealing, which dynamic metabolite concentrations were used as constraints in the flux calculation, to address which measurements were most important for characterizing cell free protein synthesis. Variation of the constraint set revealed central carbon metabolites, specifically upper glycolysis, TCA cycle, and pentose phosphate were the most important to the prediction of unmeasured metabolites. Thus, while choice of substrate utilization pathway may not matter significantly for protein production, it seems to affect prediction of the metabolism as a whole. Meanwhile, measurements of the more downstream metabolites (certain amino acids and metabolites of lower glycolysis and TCA cycle), as well as those connected to many fluxes (ATP, PEP, pyruvate) were seen to be the least important to characterizing the system. Predictive power was measured using two metrics: error between model predictions and synthetic data, and flux uncertainty obtained via flux variability analysis. Interestingly, these were not seen to be competing objectives. Rather, while error could be reduced by over four orders of magnitude by optimizing the constraint set using simulated annealing, flux uncertainty was virtually unchanged. Taken together, this finding suggested that metabolic fluxes in CFPS reactions remain largely unidentifiable, despite being constrained by a comprehensive metabolomic dataset. Further, it suggests that labeling approaches commonly used for in-vivo flux estimation, will likely need to be adapted to cell free systems to quantitatively estimate metabolic flux.