Biomedical Engineering Reference
In-Depth Information
The overall analysis depended on the sensitivity of the method and the cell-culture
performance results. If there was no issue with the cell-culture performance, then an
insensitive test would be acceptable. Every sample would pass the test with the benefit of
no operational difficulty. In the hydrolysate example, the appearance of the powder would
be an example of a test that was insensitive, having all batches pass. All the batches looked
the same by eye and the test was easy to perform. Without data to the contrary, such a test
might seem adequate. AAA, however, would be too sensitive, turning up differences that
did not matter and spawning investigations while only providing some additional
information about the hydrolysate. Once a batch failed during cell culture in manufactur-
ing, however, the added information fromAAA became important, but it was not possible
to know how much weight should be given to the results. Taking this strategy to the
extreme, a method that simultaneously measures everything, distinguishes raw material
lots based on performance in the cell culture, and is easy to run would be ideal.
In the hydrolysate example, the use test from small-scale bioreactors was 100%
predictive of the large-scale bioreactor. Based on the correlations in Table 11.2, what
error would result if some combination of analytical tests replaced the use test? In this
example, the answer was not clear. However, a threshold Zn level could be set that was
sufficiently far from the prediction error for the method that some hydrolysate lots could
be rejected based on the analytical result. However, lots with more Zn than the threshold
value might still fail in production, despite the experiencewith the analytical test because
other variables in the hydrolysate, the LV from NMR and m/z 279 (Table 11.2), for
example, are also important. Without some ongoing estimate of these factors, a small risk
of failure still exists. For this reason, a use test was still recommended on hydrolysate lots
that exceeded the threshold level for Zn. The right hand side of Fig. 11.6 shows the testing
strategy adopted following the investigation.
Ideally, Fig. 11.6might consist of two analytical tests in place of the use test where the
second test captures “unknown” compounds in place of the use test. Would NMR be an
effective option? NMR is unparalleled for compound identification and has been applied
extensively to complex solution analysis for metabolomic and food applications such as
comparing urine, plasma, and beer samples [6,21,22]. In addition, NMR was recently
described for the investigation of soy hydrolysates composition as it pertains to use in
mammalian cell culture [2]. Either 1-Dor 2-DNMR can be used for fingerprint analysis of
complex mixtures. These techniques were applied in our hydrolysate example.
Figure 11.7 shows a typical
1
H
NMR and HSQC spectra for a plant hydrolysate.
Each chemical shift in the 1-D spectrum consists of a linear combination of signals
arising from multiple compounds. For example, the methyl peak between 0.9 and
1.0 ppm (green arrow) consists of at least five different chemical moieties that could be
from several compounds. To interpret these overlapping signals, 2-D NMR spectra like
HSQC can be used.
Not all the signals in theHSQC spectra in Fig. 11.7 have a significant impact on cells.
Highlighting only these “important” compounds can reduce the number of compounds
that need to be identified and improve the NMR-based predictions. The first step was to
reduce the number of peaks in the
1
H
NMR. This was done using PLS comparing the
spectra and cell-culture results from several hydrolysate lots. The number of peaks that
trend with cell growth in the 1-D spectra decreased fromhundreds in the raw spectra to 16
