Discussion
We reported here on the CFD-based HMW prediction model. The model
includes two components. The first component is the CFD model for VI
simulation, which provides profiles of ApH3.3 andMF0.1N HCl to calculate ILPZ. The second
component is the HMW predicting equations from ILPZ. The model was
established using a training data set of 6 VI runs and verified with a
validation data set of 12 VI runs.
The CFD model for VIA simulation was described in our previous report
(Jin et al., 2019). We report here the
details of the modeling methodology. Furthermore, in the previous work,
the volume fractions of the outflows in the CFD model were fixed, whereVFout1 of 0.1 was used. The model provided
acceptable accuracy to simulate VIA (≤ 5% variation between the
predictions and the experimental EOVIA MF0.1N
HCl ) in SUM-100 and SUM-50 mixers (n = 3)
(Jin et al., 2019). Being screened whole
spectrum of VFout1 values (0.1 - 0.9), theVFout1 of 0.7, appeared to applicable to broader
types/scales of VI mixing vessels, including SUM-100, SUM-50, XDM-200,
Applikon 20L, and Appilkon 2L mixers (n = 18).
The CFD model for VIA simulation was modified from the original species
transfer model (Spann et al., 2019), of
which the tracer was patched to the fluid and distribution of the tracer
was then tracked in a duration for mixing time calculation. This bolus
addition model is not the case of VI condition at pilot/production
scale, where acid is typically continuously added during VIA. To
simulate volume expansion due to acid addition during VIA, one thought
was to use the gas-liquid two-phase model and patch air in headspace
that would enable liquid expansion during simulation. However, the ANSYS
FLUENT software is incapable to run the species model coupled with the
multiple phase model. To have continuous acid addition in the liquid
phase model while avoid volume expansion, two outlets boundary was
introduced into the modified model. With this modification, the CFD
model is capable to simulate the VIA processing with the continuous acid
addition.
This work aims to mitigate the risk of product aggregation in the VI
operation. This would ensure the designed protein loading at subsequent
chromatographic polishing steps while achieving satisfactory step yield,
final product quality, and overall robust downstream performance. A
purification platform for mAbs typically includes two chromatographic
polishing steps after VI operation (Fahrner
et al., 2001; Shukla et al., 2007),
e.g., AEX/HIC, AEX/CEX, and etc. (Shukla,
Leslie, Wolfe, Mostafa, & Norman, 2017). In the case here, a polishing
step yield was positively correlated the protein loading but inversely
correlated with HMW%, VIN (data not shown). The
reliable control of HMW%, VIN to <
2.5% significantly improved downstream throughput, resulting in
substantial reduced cost of goods of the manufacturing.
We demonstrated here on the application of the CFD-based HMW prediction
model to assist troubleshooting and guide scale-up of the VI process.
Like other common situations during commercial process development,
representative scale down model (SDM) was not established yet when the
initial scale-up run was operated. The at-scale test using PAE was
impractical due to prohibitive study cost and large material requirement
(Kateja, Kumar, Sethi, & Rathore, 2018).
In this work, the CFD-based HMW prediction model was used to quantify
and minimize the localized extreme low-pH zones, especially for
large-scale VI operation, to reduce the mAb aggregation risk. Agreed
with the model predictions, the validation data set of 12 pilot runs
(50-200 L scale) achieved 100% success rate. The results suggested that
the CFD-based HMW prediction model may be general applicable to optimize
the scale-up parameters of VI process from bench to pilot/production
scale for mAbs and potential biologics of other modalities.