Analysis of fouling and breakthrough of process related impurities during depth filtration using confocal microscopy

Abstract Titer improvement has driven process intensification in mAb manufacture. However, this has come with the drawback of high cell densities and associated process related impurities such as cell debris, host cell protein (HCP), and DNA. This affects the capacity of depth filters and can lead to carryover of impurities to protein A chromatography leading to early resin fouling. New depth filter materials provide the opportunity to remove more process related impurities at this early stage in the process. Hence, there is a need to understand the mechanism of impurity removal within these filters. In this work, the secondary depth filter Millistak+ X0HC (cellulose and diatomaceous earth) is compared with the X0SP (synthetic), by examining the breakthrough of DNA and HCP. Additionally, a novel method was developed to image the location of key impurities within the depth filter structure under a confocal microscope. Flux, tested at 75, 100, and 250 LMH was found to affect the maximal throughput based on the max pressure of 30 psi, but no significant changes were seen in the HCP and DNA breakthrough. However, a drop in cell culture viability, from 87% to 37%, lead to the DNA breakthrough at 10% decreasing from 81 to 55 L/m2 for X0HC and from 105 to 47 L/m2 for X0SP. The HCP breakthrough was not affected by cell culture viability or filter type. The X0SP filter has a 30%–50% higher max throughput depending on viability, which can be explained by the confocal imaging where the debris and DNA are distributed differently in the layers of the filter pods, with more of the second tighter layer being utilized in the X0SP.

Depth filtration is a popular choice for primary recovery of CHO cell culture due to the reduced shear rates compared to centrifugation and especially its suitability for single-use facilities. 2000L single-use bioreactors are a popular choice with some companies choosing to run several in tandem. 1 At this scale, depth filtration is the typical choice for primary recovery. It has the advantage of being easier to operate compared to a centrifuge, highly flexible and requires less capital investment as well as being fully disposable. 5 A filtration train is used at harvest to make the process more efficient. It is typically comprised of three filter trains, a coarse depth filter to remove the cells, a finer depth filter to remove colloidal matter and a 0.2 μm bioburden filter. 6,7 Each depth filter is also made up of several layers of media with different nominal pore sizes. The first layer is a coarse media that traps larger particles, followed by a tighter depth filter that clears colloidal and sub-micron particles. This arrangement of the media layers helps to increase the holding capacity of the filter at fixed loading.
Depth filters have been traditionally made from cellulose fiber backbone, a porous filter-aid such as diatomaceous earth and an ionic charged resin binder. 7 Recently manufactures have begun moving away from natural materials to ensure better consistency during filter manufacture 8,9 and future supply as DE is a finite resource. In comparison to membranes used in bioprocessing that are specified with a single nominal retention rating, depth filters are often given a range instead given their wide pore structure. 10 In addition to removal of material based on size exclusion, depth filters have been shown to remove soluble impurities by adsorption through hydrophobic, ionic, and other interactions. 7 They have been used to remove endotoxin from water, 11 and DNA from cell culture supernatant. 12,13 It has been shown that positively charged depth filters can reduce HCP and reduce the turbidity of Protein A chromatography eluate. 14,15 However, depth filters are struggling to cope with the high cell density and current set-ups are reaching maximum holding capacity for solids removal. Holding capacity expansion is only available by an increase in filter number and facility footprint. There is also an increased interest to capture more soluble impurities before the chromatography stage. Therefore, organizations are looking to make their current processes more efficient, rather than expanding manufacturing space with cost being the major driver. 1 Compared to other unit operations, such as chromatography, depth filtration has not been characterized in as much detail. Hence, there is a need to better understand the separation mechanisms.
The objective of this study is to examine the performance of Millistak+ depth filters, in order to develop an improved understanding of the mechanism of DNA and HCP removal when challenged with a high cell density CHO cell culture. This work investigates the effect of flux, cell culture viability and secondary depth filter materials during the harvest of CHO cell culture. Additionally, a novel confocal laser scanning microscopy (CLSM) method was developed to identify the distribution of foulant within the depth filter structure after they are utilized in the harvest experiments.

| Cell culture conditions
A mAb feedstock produced in Chinese Hamster Ovary (CHO) cells was provided by FUJIFILM Diosynth Biotechnologies utilizing their Apollo X™ platform. Several batches of the material were produced in 2 L shake flasks with cell density approx. 30 million cells/ml and harvested on varying days to achieve a difference in viabilities.

| Depth filtration
The filtration experiments were carried out in two parts. Figure 1 shows the scale down process used to mimic a 2:1 primary to secondary filters used at the manufacturing scale. Two Millipore Millistak+ D0HC 23 cm 2 pods were operated in parallel and both feeding into one X0HC 23 cm 2 pod. All experiments were operated at a constant flux of 150 LMH unless stated otherwise.
In the second part of the depth filtration experiments, the primary filter was scaled up to a 270 cm 2 D0HC filter pod. activated. 16 Reconstruction was performed using Nikon X-Tek software.

| Confocal laser scanning microscopy
Filters were removed from the pods after use in the experiments described above and a 10 mm disc was cut from the middle of each filter. Filter discs were fixed in 4% PFA overnight and then frozen at À20 C. Slices were cut using a cryostat and transferred to a 24-well plate with PBS buffer and 10 μl PicoGreen ® (ThermoFisher Scientific) and 10 μl Nile Red (ThermoFisher Scientific) fluorescent dyes. The samples were placed on a shaker, protected from light, for 30 min. Image analysis was performed using the software ImageJ. From each image, three areas were selected for sampling and the average integrated density was calculated, which is defined as the total sum of pixels in the sample.

| Scanning electron microscopy
Images of the depth filters were acquired with a Zeiss Supra 55 VP electron microscope. The voltage was set at 5 kV. Each filter layer was imaged from the inlet side at 250, 500, and 1000Â magnification.

| Scale-down model and depth filter composition
A scale-down depth filtration train was created to mimic the 2:1 ratio of primary to secondary filters. This was carried out using the Millipore Millistak+ HC series, which is composed of a cellulose-based backbone and diatomaceous (DE) filler. Figure 1 shows the flowsheet of the scale down model used. Briefly, a peristaltic pump is used to feed the cell culture directly into two primary D0HC filters, which feed into one secondary X0HC system. A valve is used to sample the filtrate to create the DNA and HCP breakthrough curves.
X-ray computed tomography images were taken of the unused D0HC and X0HC filters. It can be seen in Figure 1 Figure 3a shows the pressure profile of the three conditions and as expected there is an earlier and sharper pressure increase with higher flux as described by the interception mechanism.
At higher flux the inlet side of the filter gets blocked faster, reaching max pressure and the throughput is reduced. Reduced flux is a common method used in manufacturing to increase the solids capacity of the filter. However, there is often a balance between the filter capacity and processing time available in a manufacturing setting.
The HCP concentration was calculated by subtracting the IgG concentration from Total protein concentration, as measured by a BCA assay. 17 This was found to be a more sensitive method for measuring HCP compared to a CHO HCP ELISA. As the sample at harvest is so crude, the HCP ELISA was not able to give reliable measurements.
There was an almost immediate breakthrough of HCP during the filter loading process and also little overall reduction of HCP. As can be visually observed in Figure  Though perhaps a small improvement in DNA capture was seen by operating at 75 LMH. As size exclusion affects DNA retention, 25 it is plausible that there would be a delayed breakthrough at lower flux.
The difference in final throughput reached is a factor of the flux, that is, how early the filter reached max pressure.

| Filter type, cell culture viability, & DNA breakthrough
As the DNA breakthrough did not reach 100% and the secondary filters also had not reached the max pressure in Figure  This work aimed to look at the secondary filters and their ability to remove DNA and HCP, hence the primary filter was not changed.
The synthetic equivalent filter, D0SP, has an additional two layers in the pod, which would have altered the composition of the intermediate pool. HCP breakthrough data is not shown as there was no difference seen due to filter type or viability. A cholesterol assay was performed on the breakthrough samples however the data was inconclusive. It is believed this was because the samples were frozen due to the logistics of the project.
The characteristics of the cell culture material are outlined in Table 1, with cell density reaching similar levels for all viabilities. With the change to the synthetic secondary filters, there was an increase in the throughput of 28% for viabilities 37%-67%, while at 87% viability the throughput was 51% higher. The pressure profile was lowest for the highest viability at 87%, which was harvested on day 9. While this gave the best pressure profile and the lowest DNA breakthrough, it is unlikely to be used in a manufacturing setting as the IgG titer was still low, only half compared to harvesting on day 13 or 14.
The outlier in terms of pressure is the condition at 37% viability which performed better than expected. This was the case for both the cellulose and synthetic filters. This was also seen in other work (data not shown here).
At present it is not clear what is causing this behavior. One explanation might be that the harvest material is undergoing an aggregation/flocculation behavior at this low viability, which was removed by the primary filter. The

| Method development for confocal imaging of depth filters
A novel method was developed for imaging the fouled depth filters.
The work mainly focused on the DNA distribution within the filters.
Previous methods which had been applied to visualizing resin 27,28 under the confocal microscope were not applicable as depth filters were too thick for the laser to penetrate the entire depth, which in F I G U R E 5 (a) Average integrated density of PicoGreen as a function of filter depth, where the error bar represents 1SD of three measurements. Samples (a)-(e) on the x-axis corresponds to images (a)-(e) in Figure 4. Cell culture viability is indicated by different colors: black 87%, red 66%, blue 48%, and green 37%. (b) Average integrated density of Nile Red as a function of filter depth, where the error bar represents 1SD of three measurements. Samples (a)-(e) on the x-axis corresponds to images (a)-(e) in Figure 4. Cell culture viability is indicated by different colors: black 87%, red 66%, blue 48%, and green 37% this case was 4 mm. Literature reports confocal microscopy only being able to visualize at a max. depth of approx. 50-60 μm. 29 Hence in this method, the filter is cut into thin slices using a cryostat.
The CLSM has been used in literature for the visualization of filters, however they use pre-tagged proteins, DNA or viral vectors as the feed material [30][31][32][33] and fouling layers on membranes. 34 It has been used for many other applications, such as micro-plastic detection, biofuel assays, and so forth. 36 In this application, Nile Red signal represents lipids, cell debris particulates, and any aggregates.
Positive controls were done with DNA, BSA, and oleic acid which confirmed there was no unspecific binding. Negative control was done with clean filters where fluorescence signal was not significant, hence there was no background subtraction in the image analysis. However, the integrated density (IntDen) was measured in ImageJ software, which is defined as the total sum of pixels in an area and was measured for the PicoGreen and the Nile Red individually. This data is shown in Figure 5a,b, where the five points on the x-axis correspond to the five slices from each filter layer. As observed in the images, there are some challenges with the sample preparation method (air bubbles, sample breaking, and brighter signal from the edges of the sample) and as such this data aims to provide trends rather than absolute values.

| Image analysis and trends in foulant distribution
F I G U R E 6 Pressure profile of secondary filters X0HC (a) and X0SP (b) as a function of volumetric throughput, and corresponding DNA breakthrough after secondary X0HC (c) and X0SP (d) using cell culture at different viability described in Table 1. Throughput was adjusted for the hold-up volume of the filters In Figure 6, the breakthrough curves of DNA are similar for both fil- filters compared to X0HC filters, also corresponds to an increase of approx. 30% in the Total PicoGreen IntDen (Figure 7), except at 87% viability where the PicoGreen IntDen is similar to the synthetic filter but the throughput is higher.
While viability is important, the harvest day itself may be a factor in the clarification of cell culture, as the 66% and 48% viability gave similar results (harvested on the same day) in terms of pressure, throughput, DNA breakthrough and even PicoGreen and Nile Red IntDen trends. Very early (Day 9 and 87% viability) and very late (Day 14 and 37% viability) are unlikely scenarios for any manufacturing process, however they were chosen as extremes on both ends to see the effect on the depth filtration.

| CONCLUSION
Previous literature has demonstrated the mechanism of DNA and HCP removal during depth filtration by investigating components individually or at cell density <10 million cells/ml. In this work, we have shown that removing process related impurities is very challenging when using CHO cell culture at high cell density. At area. This may prove beneficial, by extending the lifetime of chromatography resin, for example, however a cost-benefit analysis would be required for any given process.
CLSM is an invaluable tool to understand the distribution of the foulant in the different layers. In combination with the breakthrough studies, it gives a better understanding of how complex feed behaves.
This is important as it can help inform process development and also aid in the design of new depth filters.