Establishing equations to predict HMW formation
The HMW prediction equations were then developed using the training data
set of 6 VI runs as presented in Table III. Considering that the HMW
assay variation could cause a relative standard deviation (RSD) of 25%
for samples containing the VIN pool HMW level
(HMW%,VIN ) of 1%, the lots selected as the
training data set had a criterion of > 2.8%HMW%,VIN in order to ensure a RSD of <
10%. The training data set included VI runs using different agitation
speeds and acid addition rates in mixers of various scales and types.
The values of ILPZ and HMWtotal,VIN of the
training runs were plotted in Figure 4B, which suggests an exponential
correlation between ILPZ and HMWtotal,VIN as
defined by Eq. 4. The HMWtotal,VIN was then
converted into HMW%,VIN using Eq. 5, where theCPAE and VPAE represent
protein concentration (g/L) and volume (L) of the PAE pool,
respectively.
\(\text{HMW}_{\text{total.VIN}}=2.364e^{1.179IPTZ}\) (4)
\(\text{HMW}_{\%,VIN}=100\bullet\text{HMW}_{total,VIN}/\left(C_{\text{PAE}}\bullet V_{\text{PAE}}\right)\)(5)
The model predictions and the experimentalHMW%,VIN values of the training runs are
presented in Table III. The model predictedHMW%,VIN values ranged from 2.7% of Lot A20_1
to 7.9% of Lot S100_1, which were in good agreement with the
experimental HMW%,VIN values (2.9% of Lot
A20_1 and 7.1% of Lot S100_1).
Furthermore, the simulation of lot S100_1 not only predicted well theHMW%,VIN value but also demonstrated the time
cause of low-pH zone growth as previously shown in Figure 3A, suggesting
that product exposure to the localized low-pH zone resulted from poor
mixing were the root cause of the HMW formation observed during the VI
operation.