Deconvolution of near-infrared spectral information for monitoring mycelial biomass and other key analytes in a submerged fungal bioprocess

Near-infrared spectroscopy is a promising technique for the rapid monitorin g of submerged culture bioprocesses. However, despite the key role of mycel ial (filamentous fungal and bacterial) micro-organisms in the manufacture o f antibiotics and other valuable therapeutics, there is little information on the application of the technique to monitor mycelial bioprocesses. In pa rt, this is due to the complex and spectroscopically challenging matrices, which result from the growth of these micro-organisms. Moreover, there is a particular lack of any detailed mechanistic information on how models for the prediction of the concentration of key analytes (e.g. biomass, substrat es, product) can be constructed, evaluated and improved using the spectral data arising from such complex matrices. We investigated the near-infrared spectra of culture fluid from a submerged fungal bioprocess, for monitoring the concentrations of mycelial biomass and other key analytes. Several emp irical models were developed for predicting the concentration of the analyt es, using multivariate statistical techniques. Despite the filamentous natu re of the biomass and the resulting complexity of the spectral variations, empirical models could be developed for the prediction of this analyte, usi ng biomass 'specific' information. SEP values of < 1 g/l could be achieved on external validation, for models developed in the concentration range of 0-20 g/l. The concentrations of the substrates, total sugars las glucose eq uivalents) and ammonium, could also be predicted, simultaneously. However, the product (penicillin) and by product (extracellular proteins) levels had to be monitored on the cell free culture fluid, due to their relatively lo w concentration. Here we report upon how the spectral information can be de convoluted for predicting the levels of the analytes and upon how the 'anal yte specific' information in the spectral data can be used to inform and as sist the modelling process, in order to increase confidence in exactly what is being modelled. (C) 2001 Elsevier Science B.V. All rights reserved.