Monitoring bacterial growth in liquid cultures through the bulk optical parameters in the near-infrared region extracted using the radiative transfer theory

Abstract

Near infrared (NIR) spectroscopy offers promise as a monitoring tool for fermentation reactions by its potential to provide information about the physical and chemical state of given system. However, issues associated with confounding effects due to light scattering variations that occur during the fermentation process due to changes in the state of the microorganisms makes it difficult to obtain robust estimation of the analytes of interest. An approach for removing multiple scattering effects and separating absorption from scattering using the radiative transfer equation (RTE) is proposed in this work. In order to investigate its feasibility and to aid in the development of the technique, the method was applied to a simple system consisting of Bacillus subtilis growing in an aqueous solution. In this study optical properties (absorption coefficient μa, scattering coefficient μs, and anisotropy factor g) were estimated and their changes during growth, stationary, and decline phases of Bacillus subtilis culture were examined. It was found that the greatest changes were seen during the growth phase, predominantly manifested in the scattering spectra. Effect of sample thickness on the estimation of the optical properties throughout the cultivation was investigated. The extracted absorption and scattering spectra were found to be fairly consistent even though they were obtained from measurements from different sample thicknesses. The extracted optical properties were used to develop Partial Least Squares (PLS) models for prediction of biomass and glucose concentrations. The performance of these models was compared with those obtained using the raw measurements. As a result of the comparison it was revealed that scattering coefficient based models demonstrate good performance while predicting biomass. The accuracy of these models was equal or in some cases greater then the accuracy of the models built on transmittance and reflectance measurements. Glucose, the second analyte of interest in this work, was modeled with limited success and models were able to distinguish only between low and high level of glucose concentration. The effect of sample thickness on PLS models performance was also studied. Results have shown that sample thickness need to be chosen based on the specific analyte of interest in order to achieve good performance of the models.