Application of Dielectric Spectroscopy Combined with Simple Linear and Multivariate Modelling for In-Line Monitoring of Biomass Concentration in E. Coli Cultures

Biomass concentration is a crucial parameter while monitoring Escherichia coli bioprocesses. Real-time monitoring of this attribute is preferable to discrete measurement, as it immediately detects deviations, reduces sampling efforts and thereby minimizes the risk of contamination. Therefore, the potential of dielectric spectroscopy for continuous monitoring of E. coli cultures was investigated. Two datasets were acquired across culture batches in a benchtop bioreactor: ∆ permittivity and the dielectric spectrum. For each batch, variations in ∆ permittivity followed the OD600nm trend, highlighting the E. coli growth phases. The dielectric spectra also evolved throughout the batches, and Principal Component Analysis (PCA) revealed at least two main contributors to spectral variability. Both datasets were then used to estimate the OD600nm using either simple linear or multivariate Partial Least Squares (PLS) modelling, with the aim of comparing these two strategies. For four calibration batches, both simple linear and PLS models generally predicted the OD600nm accurately, with only minor differences. For a validation batch, both strategies showed good predictions during exponential growth but were slightly less accurate in the other bacterial culture phases. Prediction noise was also evaluated, and PLS provided slightly less noisy predictions than the simple linear model. Overall, this indicates that simple linear and PLS predictions are not considerably different, and both modelling strategies can be used indistinguishably to calibrate the dielectric signals for estimating the OD600nm in E. coli cultures.