System development and machine learning for binding protein-based glucose monitoring platform for online bioprocess monitoring.

Abstract

Glucose being one of the most vital nutrients in all living organisms, its monitoring is crucial in both healthcare applications and bioprocesses. Over the years, extensive research has been conducted on glucose sensing technologies, leading to the development of various techniques and materials. One of the major classes of glucose sensors—electrochemical sensors—became more popular and widely available commercially, primarily focused on healthcare applications. However, the development of sensors and systems for glucose measurement in bioprocesses is still emerging, with optical sensors gaining popularity for this purpose. Among the optical methods for glucose quantification, periplasmic binding proteins have garnered significant attention over the past two decades for fluorescence-based optical measurements, primarily due to their high sensitivity. When a fluorophore is introduced into these binding proteins, the conformational change that occurs upon binding to a specific substrate can be detected and used as a biosensor. Although the construction, characterization, sensitivity, and potential applications of such biosensors have been extensively studied in previous works, a systematic sampling technique and an integrated system for automatic and continuous glucose monitoring, particularly for bioprocesses, have not yet been developed.This dissertation aimed to address the existing gaps by developing a novel glucose measurement platform for bioprocesses, utilizing glucose-binding proteins (GBP) as the sensing material. The research led to the creation of the first glucose sensor prototype based on GBP for automatic and continuous glucose measurements. The development was multifaceted. First, it introduced a method for immobilizing GBP on a microfluidic chip optimized for optical detection. It then integrated a microdialysis sampling technique to achieve continuous, automatic glucose monitoring. Microdialysis provided an aseptic sampling method that conserved culture media and was adaptable to small-scale bioreactors. Despite the dilution of samples associated with microdialysis, the high sensitivity of GBP made it an ideal candidate to overcome this limitation. Additionally, a fluidic network was designed to automate the connection between the sampling device and the microfluidic chip, enabling continuous glucose measurement. A custom-built electronic circuit interfaced with the fluidic system, and software developed in LabVIEW automated the sensing process. The prototype was used in glucose monitoring in bioprocesses, such as E. coli cell culture, Yeast fermentation, and Cell-free expression systems. Another critical aspect of the study was the application of machine learning (ML) techniques to enhance the accuracy of glucose measurements. Biosensors often suffer from noise and variability, making it challenging to achieve consistent results. Rather than relying solely on maximum fluorescence for glucose detection, ML algorithms were used to analyze time-series data and estimate glucose concentration. Various ML-based regression algorithms were explored to evaluate their performance in glucose estimation, and a framework was developed to apply these algorithms to the fluorescence data obtained from the fluorometer. The implementation showed promising results in enhancing the detection accuracy of the developed glucose monitoring platform, paving the way for potential commercialization.