Modeling, optimization, and predictive control for metabolic cybergenetics

Abstract

This thesis outlines a framework for metabolic cybergenetics that employs computational methods to control gene expression of metabolism-relevant enzymes via external signals. This enables dynamic metabolic engineering through the modulation of intracellular metabolic fluxes. The framework systematically integrates concepts from synthetic biology, metabolic engineering, (machine-learning-supported) dynamic modeling, model-based optimization, predictive control, and estimation. The focus is on batch and fed-batch processes, although the framework can be extended to continuous processes. Two modeling approaches are considered: constraint-based dynamic modeling and (Gaussian-process-supported) quasi-unstructured/unsegregated kinetic modeling. The models are used for model-based optimization, i.e., to determine optimal inputs for maximizing production, including, e.g., cybergenetic inputs, feed rates, and initial concentrations. Repeatedly solving model-based optimization problems -model predictive control- can address uncertainties such as model-plant mismatch and disturbances. Model-based control requires information about the current system states. Real-time process monitoring of relevant states such as biomass components can be achieved with soft sensors, e.g., based on full information estimation. The applicability of the framework is outlined considering a case study dealing with enforced adenosine triphosphate (ATP) turnover for enhanced product yield or productivity, focusing on the anaerobic lactate fermentation by Escherichia coli. The ATP turnover is manipulated by modulating the expression of the ATPase (F1-subunit), an enzyme catalyzing the hydrolysis of ATP, using an optogenetic approach. That is, light is used as a control input to fine-tune ATPase expression. Experimental validation involves open-loop control in batch, employing quasi-unstructured/unsegregated kinetic modeling. The presented framework allows full exploitation of all available input degrees of freedom while counteracting disturbances and uncertain model information. It opens the door to advanced biotechnological applications involving dynamic metabolic control; furthermore, its model-based nature can enable cost-effective process development, robust operation, and flexibility in biotechnology.