Design and Implementation of a Low-Cost Bioreactor System for Synechocystis sp. PCC 6803: Integrated Cultivation, Lysis, and Filtration for Sustainable Glucose Harvesting

Abstract

This thesis describes the design, modeling, and fabrication of a three-part bioreactor and biomass processing system designed to cultivate Synechocystis sp. PCC 6803 and extract its intracellular glucose. The resulting glucose can support sustainable biomanufacturing for diverse downstream applications, including serving as a feedstock for K. rhaeticus to produce cellulose, as a precursor for biofuel production, or as an ingredient in food supplements. The system incorporates a photobioreactor, a lysis module for acid and ultrasound-based cell disruption, and a pressure-driven filtration setup. The photobioreactor was equipped with a pH, dissolved oxygen, and temperature probe; and optical density was continuously monitored using a custom-built module. The lysis unit contained an ultrasound, a pH, and temperature probe in addition to pumps connected to acid and base chambers. The filtration unit was connected to a compressed air tank and designed with a pressure control valve, safety valve, and syringe filter. Glucose concentration was quantified offline using high-performance liquid chromatography (HPLC). Various light regimes were tested, and under an incident light intensity of approximately 400 µmol m⁻² s⁻¹ at a color temperature of 6500 K, cultures were shown to reach a biomass productivity of 90 mg L⁻¹ day⁻¹, with a specific growth rate of 0.166 day⁻¹ and glucose concentrations up to 5.08 mg L⁻¹. Innovative culture strategies were explored at a small scale, including the cultivation of Synechocystis sp. PCC 6803 in spent K. rhaeticus media to promote economic and sustainable media recycling. When supplemented with additional nutrients, the spent media supported Synechocystis growth up to an OD680 of 0.5. To further characterize the photobioreactor and expected growth based on environmental parameters, both mathematical and machine learning models were built. While the mathematical models were not experimentally validated, the machine learning model model achieved a strong predictive accuracy with a mean absolute error and variance of 0.0009±0.0003 over a 10-fold cross-validation. The system demonstrates up to 65% reduction in cost compared to commercial alternatives.