Biomedical Engineering Reference
In-Depth Information
homogenate as well as biomass in Escherichia coli fermentations [1-7]. Arnold et al. [7]
have even developed a temporally segmented model in which they built predictive NIR
models for the early, middle, and end phases of a bioprocess. Since the nature of the
matrix they were working with changed during the course of the fermentation, they
were able to successfully create predictive models for each phase of the process. The
chemometric technique of partial least squares was used to build the models. The models
were built on wavelength regions, rather than on distinct wavelengths, in a more complex
matrix such as fermentation broths [5]. Data segmentation may be a useful technique
when dealing with complex matrices that change physiologic properties over the
duration of the process. Data segmentation is a powerful diagnostic that may spot
differences between runs but will not necessarily tell you why they are different.
NIR has been used to quantify the amounts of metabolic by-products in cell culture.
There are numerous examples reported in literature on the NIR probes' ability to quantify
key metabolites such as carbon source (glucose), lactate, and ammonia, as shown in
Fig. 12.3 [12-18].
Card et al. [17] reported using a first derivative with a Norris smoothing and mean
centering lead to quantify a number of cell culture metabolites and variable including
glucose, glutamine, ammonia, cell density, and cell viability. In addition, they showed
that lactate required a slightly different treatment and showed the best results with a
slightly more aggressive smoothing and a combination of variance scaling and mean
centering. Total cell density calculations were optimized with no spectral pretreatment
except for amild Savitzky-Golay smoothing andmean centering. They foundmost of the
useful cell density signal was contained in baseline information, either offset or slope, or
a combination of both [17]. They used this information to control the fed batch process by
altering the feed components.
FTIR with attenuated total reflectance, as implemented in bioprocesses, is able to
monitor many of the same attributes as NIR. FTIR has been successfully implemented in
both microbial and mammalian bioprocesses. In some instances for online usage, a
thermoelectric cooled mercury cadmium telluride (MCT) detector is used. Analytes that
have been reported to be measured in microbial fermentation include glucose, sucrose,
ethanol, citrate, lactate, glycerol, methanol, yeast extract, ammonia chloride, ethyl
acetate, and a variety of other components. Schustera, Mertens, and Gapes [28] report
using FT mid-IR to monitor the acetone-butanol-ethanol (ABE) fermentation in the
genus Clostridium. The ABE fermentation process is a technology that converts
renewable resources into liquid fuels and basic chemicals. In cell culture, FTIR has
been used to measure relative amounts of DNA, RNA, lipid, protein, and glycogen
throughout the different growth phases of a culture [29].
Like other spectroscopic techniques, FTIR relies on chemometric techniques such
as partial least squares to build quantitative models, but these models assume linearity of
response and thus may introduce errors so significant as to make the spectroscopic
technique unsuitable for implementation. Therefore, ensuring the use of the appropriate
chemometric technique(s) to fit the process being modeled is imperative. For example,
Franco, Per
n, Mantovani, and Goicoechea [30] were able to accuratelymeasure glucose,
glucouronic, and gluconic acid by combining strategies of nonlinear partial least squares,
a wavelength-selective genetic algorithm, and artificial neural networks.
ı
