Automated Measurement and Monitoring of Bioprocesses: Key Elements of the M3C Strategy - Measurement, Monitoring, Modelling, and Control of Bioprocesses - page 24

very rapid coffee roasting processes [ 97 ] or proton transfer reaction mass
spectroscopy (PTR-MS) for analysis of odorous volatile organic compounds [ 98 ].
Matrix-assisted laser desorption time-of-flight mass spectrometry (MALDI-
tof-MS) has even been reported as ''intact cell mass spectrometry, ICM,'' being
not online but ''rapid'' [ 99 ]. With rapid cell pretreatment, viability could be
estimated from mass spectral data, and discrimination among cell lines with dif-
ferent recombinant proteins or different productivities could be achieved.
3.3.5 Infrared Spectroscopy
Most molecules can be excited with infrared light to perform typical molecular
vibrations. This is exploited in infrared (IR) spectroscopy. The far IR is not very
informative since water—the solvent of most bioprocesses—absorbs massively in
this range and most of the primary or fundamental molecular vibrations are excited
in the mid-infrared (MIR), which lies in the wavenumber range from 200 to
4,000 cm -1 . The respective absorption bands are relatively narrow and can, in the
case of pure compounds, be attributed to typical chemical bonds such as -C-O-C-,
[C = O, -CHO, -O-H, or -S-H groups and so on. At higher wavenumbers
(4,000-13,000 cm -1 ), the near-infrared (NIR), vibrational combinations or simply
overtones are excited. In this region, the absorption bands are much broader due to
the smaller energy separations of the excited states. NIR and MIR are used in
process monitoring. This is much simpler for the case of chemical processes with
fewer components at higher concentrations—as already established by the phar-
maceutical industry in production processes (e.g., [ 100 - 102 ])—when compared
with biological processes: the various absorption bands of many components must
necessarily overlap, which makes the (qualitative) identification procedure quite
cumbersome. Some components in low concentration will contribute to a minute
absorption on top of another band from a major component, and, therefore,
deconvolution of the spectral information and estimation of concentrations
therefrom are not easy. In bioprocesses, we always work with multicomponent
solutions, and therefore we need multivariate evaluation algorithms, typically
chemometric procedures (see Chaps. 2 and 9 ). Furthermore, near-infrared spectra
of mycelial biomass appear to be influenced by morphological changes of the cells,
too [ 103 ].
Calibration models are typically built from spectrometric analyses of solutions
of pure substances or (synthetic) mixtures of a few components in known con-
centrations. Spiking of spent culture supernatants with known amounts of one
single component is a good method to test the calibration model: if the model
predicts a change in only this single concentration while the predictions for the
other components remain constant, the calibration model reflects an analytically
crisp fact. If, however, other predictions move too, although the concentrations of
those components have not been changed, the calibration model is likely to have
learnt a nonanalytic correlation, most often a biological one; for instance, when
substrate concentration is high, the product concentration is low, and vice versa.