396
B. Igne and E. W. Ciurczak
Partial least-squares models have routinely been employed to develop the models
discussed above, but other quantitative approaches and variable selection methods
have been utilized to improve method performance. An example of fermentation
monitoring that employed variable selection was demonstrated with improved performance over full-scale/ selected-region models through the use of interval PLS [27].
Interval PLS approaches divide the full spectra in sub-regions and find which combination of variables give the best results. Models with the full spectra or selected
regions were also developed, but with higher errors than the interval PLS [27].
All the examples provided above show how NIRS has be used during a bioreactor
run. But a large amount of information can be obtained from retrospective analyzes to
compare batch-to-batch variability. A principal component analysis (PCA) applied
on NIR spectra collected in real-time during five batches allowed the characterization
of the batches based on cell densities (scores on the first principal component) and
batch-to-batch variations (scores on the second principal component) [28].
But despite the reported work, a review of on-line monitoring and control tools
for mammalian cell cultures challenged the applicability of NIR for bioprocesses
[29]: “[…], the molar absorptivity in the NIR range is typically quite small; thus,
the method is not ideal for diluted or minor components. This can be a limiting
factor for the application to mammalian cell cultures, since a key process control
objective during fed-batch operation is to keep glucose and glutamine at very low
concentrations to prevent the accumulation of the toxic by-products ammonia and
lactate. Furthermore, the overlapping signals seen in the NIR range results in very
broad peaks, leading to complex spectra and hindering the assignment of specific
features to individual compounds. These characteristics require a chemometric datamining step to relate spectral information with the target compounds”. For these
reasons, NIRS remains underutilized when compared with Raman for the monitoring
of pharmaceutical bioprocesses. But, innovation in sensitivity and chemometrics will
help support the technology in the long term.
18.3.2 Lyophilization
Another area of application of NIRS in bioprocesses is lyophilization. The removal of
water from the final drug product (after purification of the growth medium containing
the molecule of interest) is necessary to ensure that the monoclonal antibodies remain
stable and can be stored and shipped without affecting their therapeutic effects.
Lyophilization, or freeze drying, can be used to achieve that. Since water has a strong
molecular absorptivity in the NIR region, it is a tool of choice for the monitoring and
control of these processes. Publications have shown the suitability of the technique
for the analysis of water content through the container during water removal [30,
31]. The spatial distribution of moisture within vials was also investigated with NIR
chemical imaging [32].
To ensure that the proteins will maintain their therapeutic effect after water
removal, it is necessary to understand whether the water removal process affects the
B. Igne and E. W. Ciurczak
Partial least-squares models have routinely been employed to develop the models
discussed above, but other quantitative approaches and variable selection methods
have been utilized to improve method performance. An example of fermentation
monitoring that employed variable selection was demonstrated with improved performance over full-scale/ selected-region models through the use of interval PLS [27].
Interval PLS approaches divide the full spectra in sub-regions and find which combination of variables give the best results. Models with the full spectra or selected
regions were also developed, but with higher errors than the interval PLS [27].
All the examples provided above show how NIRS has be used during a bioreactor
run. But a large amount of information can be obtained from retrospective analyzes to
compare batch-to-batch variability. A principal component analysis (PCA) applied
on NIR spectra collected in real-time during five batches allowed the characterization
of the batches based on cell densities (scores on the first principal component) and
batch-to-batch variations (scores on the second principal component) [28].
But despite the reported work, a review of on-line monitoring and control tools
for mammalian cell cultures challenged the applicability of NIR for bioprocesses
[29]: “[…], the molar absorptivity in the NIR range is typically quite small; thus,
the method is not ideal for diluted or minor components. This can be a limiting
factor for the application to mammalian cell cultures, since a key process control
objective during fed-batch operation is to keep glucose and glutamine at very low
concentrations to prevent the accumulation of the toxic by-products ammonia and
lactate. Furthermore, the overlapping signals seen in the NIR range results in very
broad peaks, leading to complex spectra and hindering the assignment of specific
features to individual compounds. These characteristics require a chemometric datamining step to relate spectral information with the target compounds”. For these
reasons, NIRS remains underutilized when compared with Raman for the monitoring
of pharmaceutical bioprocesses. But, innovation in sensitivity and chemometrics will
help support the technology in the long term.
18.3.2 Lyophilization
Another area of application of NIRS in bioprocesses is lyophilization. The removal of
water from the final drug product (after purification of the growth medium containing
the molecule of interest) is necessary to ensure that the monoclonal antibodies remain
stable and can be stored and shipped without affecting their therapeutic effects.
Lyophilization, or freeze drying, can be used to achieve that. Since water has a strong
molecular absorptivity in the NIR region, it is a tool of choice for the monitoring and
control of these processes. Publications have shown the suitability of the technique
for the analysis of water content through the container during water removal [30,
31]. The spatial distribution of moisture within vials was also investigated with NIR
chemical imaging [32].
To ensure that the proteins will maintain their therapeutic effect after water
removal, it is necessary to understand whether the water removal process affects the