On the Use of Hybrid Models and Latin Hypercube Sampling to Develop a Chemically Defined Culture Medium Formulation for pDNA Production in E. coli

The demand for pharmaceutical-grade plasmid DNA (pDNA) is projected to outpace current manufacturing capacity due to the expanding clinical pipeline of Advanced Therapy Medicinal Products (ATMPs). While regulatory bodies (FDA, EMA) advise against complex media in an evolving regulatory landscape, existing chemically defined formulations remain proprietary, commercial, or sub-optimal. Herein, we present a combined experimental and modelling approach enabling resource-efficient media design. Starting from a basal M9 minimal medium, a Latin Hypercube Sampling approach was used to define a first set of eight experiments that efficiently explore a design space comprising six factors (primary carbon source concentration, molar C/N ratio, magnesium concentration, the addition of K2HPO4, trace elements solution and iron). Data from the first set of experiments was used to calibrate hybrid mechanistic–data-driven models which were in turn used to define the next set of experiments via model-based optimization. Data from the second set of experiments was used to further refine the hybrid models, creating a robust iterative cycle. Sensitivity analysis of the refined models indicated that carbon, iron, and phosphate concentrations had the strongest effect on pDNA yield. A 3-fold increase in pDNA titre was obtained after 16 experimental runs, highlighting the significant time and cost gains afforded by combined experimental and modelling approaches.