3. Results
3.1. Column chromatography and filtration for mAb with aggregate spike
The effect of column chromatography and nylon prefilter treatments on mAb filterability is shown in Figure 1A and 1B. Protein aggregate characterization by SEC analysis is shown in Figure 1C and a comparison of dimer and trimer or larger aggregate and HCP content are shown in Figure 1B. From the figures, we see that the purified mAb (reference solution) with 175 ppm HCP and 1.3% dimer content has a stable flow rate on Planova BioEX. The control, which is aggregate-spiked mAb (spiked at 1.0%), has markedly lower throughput from shortly after the start of filtration, showing that an increase in larger aggregate (trimer or larger species) content from 0.2% to 1.2% is the likely cause for the marked decrease in filterability. Further, aggregate-spiked mAb processed in flow-through mode with normal AEX has similarly high trimer or larger aggregate content and filtration behavior that is almost the same as the control, showing that normal AEX column chromatography has no impact on improving filterability. The moderate improvement by the nylon prefilter on filterability of the aggregate-spiked mAb can be attributed to the reduction of trimer or larger aggregates to 0.5%, which is less than half of the level in the control. The output from mixed-mode AEX1 and mixed-mode AEX2 shows markedly higher filterability, surprisingly higher than the reference. These chromatography processes effectively reduced HCP and aggregate, particularly reducing trimer or larger aggregates to below detectable levels. HCP was also decreased for both outputs, but the similar improvement in filterability increase observed for both despite mixed-mode AEX2 output having twice the HCP as mixed-mode AEX1 output suggests that the cause of decreased filterability for aggregate-spiked mAb is the trimer or larger aggregates more so than HCP. These results suggest that filterability was not greatly impacted by the HCP or mAb dimer content at these concentrations.
The molecular weight distribution profiles based on SEC analysis are shown in Figure 1C for the reference, control, mixed-mode AEX1 output and nylon prefilter output. The aggregate-spiked mAb (control) has increased dimer and especially more trimer or larger aggregate content, which is believed to have a great impact on filterability in the virus filtration step. For mixed-mode AEX1 output, which shows improved filterability at the virus filtration step, the molecular weight distribution profile clearly shows that this processing almost completely removed trimer or larger aggregates and reduced the dimer content. Filtering the solution with a nylon prefilter moderately improved filterability and it decreased the proportion of trimer or larger aggregates, but its effect on improving filterability is very small compared to mixed-mode AEX1.
3.2. Column chromatography and filtration for plasma IgG with aggregate spike
Figure 2 shows the effect of column chromatography on the filterability of polyclonal plasma IgG isolated from plasma derivatives with 0.5% aggregate spike. Aggregate content by SEC analysis is shown in Figure 2A. Plasma IgG spiked with 0.5% IgG aggregate (control) had increased trimer or larger aggregate content (from 0.3% for the reference solution to 0.5%) and 7.8% dimer content as shown in Figure 2A. The figure clearly shows that the 0.5% IgG aggregate spike causes a marked decrease in filterability as evidenced by achieving flux of 100 LMH at nearly 80 L/m2 for the reference, while the flux of the control had become nearly zero and the run was ended by 12 L/m2. Normal AEX processing of aggregate-spiked plasma IgG produced no aggregate removal and there was no improvement in filterability.
Mixed-mode AEX1 shows more than double the throughput of the control and high aggregate removal with dimer content decreasing from 7.8% to 5.8% and trimer or larger aggregate content decreasing from 0.5% to below detectable level. On the other hand, while trimer or larger aggregate content was also reduced to below detectable level by modified CEX1 and modified CEX2 processing, dimer was not reduced markedly from the control and remained at 7.3% and 7.7%, respectively. Interestingly, modified CEX output shows significantly higher filterability than the mixed-mode AEX1 output and even exceeds the flux of the reference in the early phase of filtration (Figure 2B). Modified CEX1 and modified CEX2 both show high filterability, but modified CEX2 output shows a greater flux decay than does modified CEX1 output.
3.3. Determination of clogging factor k for model analysis
For clogging model analysis, clogging factor k is determined by applying the obtained filtration data (filtration volume and flow rate) to Equations 2, 4, 6 and 8 for the cake filtration model, intermediate blocking model, standard blocking model and complete blocking model, respectively. Graphical results of applying the control (aggregate spike) filtration results of both mAb and plasma IgG to each clogging model and finding the line of best fit using the least squares method are shown in Figure 3. Based on the overlap between experimental and calculated data, the mAb is best fit to the standard blocking model (Figure 3C) and plasma IgG is best fit to the complete blocking model (Figure 3D).
3.4. Clogging model analysis for aggregate-spiked mAb
Evaluation of filtration behavior for the reference (no spike) and control (aggregate spiked) mAb solutions in Figure 1B with each of the models is shown in Figure 4A and 4B, respectively. For the reference (no aggregate spike), values calculated based on each of the models overlap with experimental values, indicating that pronounced clogging was not observed for this purified protein (Figure 4A). In contrast, the control (aggregate-spiked mAb) had distinct plots for each clogging model, and the standard blocking model plot overlapped the most with experimental values (Figure 4B).
The control, output from normal AEX and output from nylon prefilter (all three solutions having relatively high proportions of trimer or larger aggregate content) had markedly higher k than column output from mixed-mode AEX (low proportion of aggregate), and the higher the aggregate content, the larger the k (Figure 4C). Further, for all three solutions with trimer or larger aggregates, the k decreased in the order of cake filtration, intermediate blocking, standard blocking and complete blocking models. The differences between experimental and calculated values for each solution with all four clogging models in Figure 4D show that for solutions with high aggregate content and high k, the standard blocking model is the best fit for mAb.
3.5. Clogging model analysis for aggregate-spiked plasma IgG
Evaluation of the filtration behavior for the reference and control for plasma IgG solutions in Figure 2B with each of the four clogging models is shown in Figure 5A and 5B. While the calculated results for all four clogging models overlap with experimental results (Figure 5A), indicating no significant clogging for the reference (no aggregate spike), control (spiked with aggregate) had distinct plots for each clogging model, and the complete blocking model showed the smallest differences between experimental and calculated values (Figure 5B).
Plots of k and mean differences between experimental and calculated values for filtration behavior (Figure 5C and 5D) show that plasma IgG solutions with relatively high proportions of trimer or larger aggregate content (control and normal AEX output) showed the best fit with the complete blocking model, which is a different from that for mAb (standard blocking model). A deeply interesting point is that the output for mixed-mode AEX1, which removed the trimer or larger aggregates through column chromatography and had lower dimer content than the outputs for modified CEX1 and modified CEX2 (5.8% vs. 7.3% and 7.7%, respectively), had a markedly higher k than the output for both modified CEX resins and the reference, which had 0.3% trimer or larger aggregate content as shown in Figure 5C. Furthermore, the differences between experimental and calculated values for each solution in Figure 5D show that mixed-mode AEX1 output has the best fit with complete blocking model. These results suggest that there are components besides aggregates detected by SEC that impede the filterability of plasma IgG on the virus filter, and the components formed in the process of producing aggregates are reduced by processing with modified CEX but not by mixed-mode AEX1.
4. Discussion
Plots of filterability profiles for mAb (Figure 1) and plasma IgG (Figure 2) with aggregate spike and after processing with various chromatography columns clearly show that chromatography processing significantly affects filterability for both solutions, but with different results for various chromatography resins. It should be noted that all runs were conducted with the same solution conditions for control and simplicity of the experimental design, but optimizing solution conditions for each different resin could potentially result in different outcomes. Differences in initial flux, even for constant pressure filtration at the same pressure, may be due to differences in viscosity resulting from the chromatography processes.
Based on analysis of filtration behavior and the addition of column chromatography on filterability, users can consider choosing chromatography resins that will improve the overall performance of their virus filtration process. For aggregate-spiked mAb processing, the output from mixed-mode AEX1 and mixed-mode AEX2 showed improved filterability while normal AEX did not. Based on manufacturer information on the resins, mixed-mode AEX1 has a primary amine and butyl base and mixed-mode AEX2 has a tertiary amine and phenyl group, and as such, these mixed-mode AEX resins do not rely on the strength and weakness of an anion exchange group and hydrophobic group. Similarly, for aggregate-spiked plasma IgG processing, filterability was improved over reference by two resins with a sulfate ligand, modified CEX1 with dextran sulfate and modified CEX2 with cellulose sulfate, indicating that plasma IgG filterability improvement is due to dextran sulfate being more effective for flow-through processing. However, while both mixed-mode AEX and modified CEX column chromatography removed trimer or larger aggregates from plasma IgG, there were differences in removal of dimers by these two methods. Despite greater removal by mixed-mode AEX, modified CEX showed markedly better improvement in filterability with flux at the start of the filtration exceeding that for the reference. This observation suggests that the decrease in filterability of plasma IgG, which is polyclonal, is not dependent solely on the aggregate content determined by SEC, unlike the pattern observed for mAb solutions.
From clogging model results based on filtration behavior, we see that mAb with aggregate was best fit to the standard blocking model (Figure 4) and plasma IgG with aggregate was best fit to the complete blocking model (Figure 5). Appropriate selection of the best-fit model for each molecule was shown as the lowest k for both solutions with aggregate spike (control) and solutions with moderate reduction of aggregate following chromatography processing. Based on selection of the standard blocking model for mAb, the pores of the filter are likely narrowed by molecules adhering to the walls of the pores. In contrast, plasma IgG likely obstructs the pores based on the selection of the complete blocking model.
Although the clogging models assume simplified uniform cylindrical pores, which may not be exactly representative of virus filters, based on the studies and analyses presented here, applying the clogging models to filtration behavior could be an insightful way to characterize filtration processes. Our findings indicate that, by selecting chromatography processes that are compatible with virus filtration and that improve the filterability of the feed stream, the capacity of production processes can be increased. These processes can be conducted at large scales of at least 1000 L/m2, and even larger throughput can be expected, for example, as has already been put into practice (Lute et al., 2020).[12]
Optimizing filterability through consideration of aggregate removal is of great interest for downstream process development. As future work, correlation of clogging model results and analytical results of solution characteristics including aggregate content along with the use of visualization techniques will deepen our understanding of filtration mechanisms. Applying filtration data from higher throughput runs (more than 500 L/m2) to clogging model analysis will provide guidance for scaling up.