Development of the CFD model to simulate VIA
A CFD model was initially developed for troubleshooting the excessive product aggregation in the VI operation of the pilot lot, S100_1. The operating condition is presented in Table III. As previously reported (Jin et al., 2019), the lot was performed in a SUM-100 mixer containing 66.5 L of PAE (28 g L-1IgG4-N1, pH 4.6) with a 50 rpm agitation speed and 0.1 N HCl addition at 2.62 L h-1 L-1 to achieve pH 3.6 (i.e., VIA). The acidified VIA pool was held at room temperature for 60 min for viral inactivation and followed by neutralization to pH 5.5 with 2 M Tris-base (i.e., VIN). The HMW levels of the VIN pool was 7.1%, which exceeded the level specified for in-process pool and led to lot rejection of the final drug substance produced. In the preliminary troubleshooting investigation, experiments and the CFD model simulation suggested that the localized low-pH (≤ pH 3.3) zone during VIA was the root cause for the product aggregation.
The CFD model for VIA simulation was modified from the well-establishedk ‐ε turbulence model coupled with the species transfer model in literature (Spann et al., 2019). Unlike the original model that used a single bolus tracer addition to simulate mixing time, the modified model mimicked continuous addition of 0.1 N HCl to simulate VIA. To develop the new model, simulations were performed to screen different volume fractions and the screening criterion was based on the agreement between the model predictions and the experimental pH values at the end of the VIA step (EOVIA).
The pH of the acidified PAE resulted from the amount of 0.1 N HCl added, i.e. 0.1 N HCl mass fraction (MF0.1N HCl ) of the acidified PAE pool as shown in Figure 2A. The data showed a linear correlation between MF0.1N HCl and the pH of the acidified PAE pool. Furthermore, this correlation was protein concentration dependent. As the protein concentration increases, so does the requirement of 0.1N HCl for a given pH adjustment. For example, it required MF0.1N HCl of 0.078 to bring pH down from 4.45 to 3.65 for the protein centration of 22 g L-1, while it required MF0.1N HCl of 0.115 to achieve the same pH adjustment for the protein concentration of 35 g L-1.
The simulation results for screening volume fraction of outflows are shown in Figures 2B-2D. Since the overall volume fractions was the unit for the two outlets, for a given outflow-1 volume fraction (VFout1 ), the outflow-2 volume fraction (VFout2 ) equals to 1 -VFout1 . Therefore, onlyVFout1 needed to be tracked for data analysis. The screening was first performed with simulations for lot S100_1, of which EOVIA MF0.1N HCl was 0.1215. TheVFout1 value was initiated from 0.1, and stepwise (0.1 per step) increased to 0.9 in the simulations. As shown in Figure 2B, there was a clear linear correlation between the simulated EOVIAMF0.1N HCl and VFout1value. The best agreement between the simulation and the experimental EOVIA MF0.1N HCl was observed atVFout1 = 0.7 (0.1214 of simulation vs. 0.1215 of measurement). Using VFout1 = 0.7 for lot S100_1, the simulated MF0.1N HCl profile appeared to overlap with the experimental profile as shown in Figure 2C.
To consolidate the CFD model with VFout1 = 0.7 for VIA simulation, the model was applied to 5 additional VI runs in 2L or 20L Applikon reactors. The conditions for VI experiments are presented in Table III. The experimental and simulation results are shown in Figure 2D. The variations of the simulations and the experimental EOVIA MF0.1N HCl values were within 10%, demonstrating that the CFD model was generally applicable to mixers of different types and scales for VIA simulation.
In summary, the modified CFD model for VIA simulation was thekε turbulence model coupled with the species transfer model, which facilitated continuous tracer addition and the constant working volume during simulation using two outflows. The preferredVFout1 and VFout2 are 0.7 and 0.3, respectively.