Simplified approaches to mid-infrared spectrometers calibration for bioprocess monitoring applications

This thesis aimed at developing new, simple methods of calibration for on-line monitoring of bioprocesses by Fourier-transform mid-infrared spectroscopy (FTIR). The conventional calibration approach implies the preparation and measurement of a large number of standards, usually around 50, and involves advanced mathematical tools, such as Principal Component Regression or Partial Least Squares for data treatment. This procedure is time-consuming, and requires, in addition, a fair level of expertise in linear algebra. For these reasons, there is a big need for calibration approaches that can be performed rapidly, by non-experts, in order to allow a routine operation of mid-infrared spectroscopy for the on-line monitoring of cultures. Mid-infrared spectroscopy deals with the vibrational energy of molecules, which means that a vast majority of compounds can be detected in this wavelength range. Data treatment is therefore complex, due to strong peak overlapping, but also because of the low concentrations involved in culture media (< 3%) and unavoidable drift of the signal instrument. Multi-level calibration designs have been developed to tackle these problems, and they proved to achieve a robust process modeling. However, in the case where a method to correct for signal drift is available and compounds do not interact with each other –which is likely to happen in dilute solutions as such as culture media– multi-level calibration designs are clearly oversized, since a 2-level design provides the same quantity of information for a much lower work load. It was shown in this research project that, for most common substances found in culture media, the absorbance is linear with respect to concentration (i.e. that the Lambert-Beer law is followed) and that species do not significantly interact with each other in solution. Based on this evidence, a relatively simple calibration method has been developed, which consists in correcting for signal drift through an anchorage method, and calculating the concentration using a library of pure component spectra as calibration set. Pure component refers here to a single compound dissolved in water at a concentration around 0.1 mol L-1. Anchorage points were set in the spectra in regions where none of the compounds of interest absorb, but close enough from the calculation range to insure an appropriate correction. Batch cultures of the yeast S. cerevisiae were carried out to validate this approach, and it was shown that the results were as accurate and as robust as if they were found using the conventional approach (standard error of calibration of 0.86, 0.98, 0.15 g L-1 for glucose, ethanol and ammonium respectively). Pulse additions of compounds during the culture could be successfully monitored by the calibration, which therefore proved to be truly predictive. A second calibration approach was developed and tested with batch cultures of the bacteria E. coli. While a spectra library was also used in this method to c