A New Scale-Down Membrane Adsorber Device for Process Development and Validation

ABSTRACT

Anion exchange membrane chromatography (AEX) is an attractive alternative to flow-through anion exchange column chromatography. Replacing AEX column chromatography with AEX membrane chromatography provides similar output but at a much higher load density, usually greater than 10 kg/L of membrane. The commercially available scale-down model, Sartobind nano, which has a 1 mL membrane volume, requires a significant amount of material for process development and validation whereas a relatively small amount of material is typically available during early clinical development. To overcome this limitation, an ultra scale-down device, Sartobind pico, was developed to reduce material consumption and validation cost. In this article, the development of the new ultra scale-down device is detailed and scalability to Sartobind nano and to a large-scale capsule are demonstrated. Studies using model proteins and industrially relevant monoclonal antibody feedstock are described. The new ultra scale-down device, Sartobind pico, enables process development, characterization, and validation with scalability to large-scale membrane chromatography devices while reducing sample consumption, time, and cost.

Anion-exchange (AEX) membrane chromatography is an attractive technology for monoclonal antibody (mAb) purification because of advantages such as elimination of column packing and unpacking, higher throughput, smaller plant footprint, and considerably less buffer consumption. Compared with AEX resins, which are typically loaded to approximately 100 g/L, AEX membranes can provide orders of magnitude higher loading capacity in flow-through mode with adequate impurity removal. For example, Zhou et al. reported greater than 3000 g/m2 or 10.9 kg/L load capacity with > 5 log reduction value (LRV) for four different model viruses (1). In another study, Zhou et al. showed that a similar LRV for X-MuLV could be obtained at a load capacity of 13 kg/L and at flow rate of 600 cm/hr (2). Glynn et al. recently described the evolution of Pfizer's antibody purification process from three columns to two by replacing the resin-based AEX chromatography step with a membrane adsorber and increasing the load capacity of this step by a factor of 100 (3). The removal of process-related impurities with AEX membrane adsorbers at high load capacity and high flow rate has also been published by Arunakumari et al. (4). Lately, the authors demonstrated virus removal by membrane adsorbers with a LRV greater than 4.5 and 4.4 for X-MuLV and MMV, respectively, at 20 kg/L mAb load capacity (5). Mehta et al. showed that purity and product quality comparable to traditional three-column affinity processes can be achieved with a novel process using a nonaffinity capture step and membrane-based technologies such as AEX membrane adsorbers and high performance tangential flow filtration (6).

It is thus well documented in the literature that an AEX membrane adsorber is a powerful alternative to column chromatography and can facilitate development of new purification strategies for downstream processing in the biopharmaceutical industry (7). However, the high load capacity achieved with membrane adsorbers in the flow-through mode implies the need for a significant amount of material for process development with laboratory-scale devices. For example, a load capacity of 10 kg/L means that 10 g of material is required for each experiment with a 1 mL laboratory-scale device. High material consumption can be a limiting factor, particularly during early stages of drug development where relatively small amount of material is typically available. Reducing the virus validation cost by minimizing the amount of virus spike required is also of significant interest.

To overcome these limitations, a new ultra scale-down membrane adsorber device, Sartobind pico (Sartorius Stedim Biotech GmbH, Göttingen, Germany), with a membrane volume of 0.08 mL has been developed. The 12.5-fold lower membrane volume than the current laboratory-scale device, 1 mL Sartobind Nano, significantly minimizes feedstock and virus spike requirements for development, characterization, and validation studies. The performance of this device was evaluated using model molecules and industrially relevant mAb feedstock and was compared with the current scale-down device, Sartobind nano. Data demonstrating the scalability of the new ultra scale-down device to a manufacturing-scale device are also presented.

MATERIALS AND METHODS

Devices

Sartobind pico, the new scale-down device was provided by Sartorius Stedim Biotech GmbH, Göttingen, Germany. The device consists of 15 membrane layers with polypropylene sealing rings every 3 layers, and is assembled into a molded polypropylene housing with luer lock connectors to enable easy connection to a liquid chromatography system (see

Figures 1

and

2

). The bed height of 4 mm is similar across the entire Sartobind SingleSep family and the frontal surface area of 20 mm 2 gives pico a membrane volume of 0.08 mL. Sartobind nano, (Sartorius Stedim Biotech GmbH, Göttingen, Germany) with 15 layers, 36.4 cm 2 total surface area, and 1 mL membrane volume was used as a reference device (see

Figure 1

). The Sartobind nano has a radial flow and is constructed in the same way as process scale SingleSep capsules, which assures direct scalability to manufacturing scale capsules (7–11). The key attributes of Sartobind pico and Sartobind nano are summarized in

Table I

.

The Sartobind SingleSep 10" capsule with a membrane volume of 180 mL was used to further confirm scalability. The devices were assembled with a salt tolerant AEX membrane, Sartobind STIC PA, consisting of a polyallylamine ligand covalently coupled to the cellulose membrane matrix (12).

Equipment

All laboratory-scale chromatography experiments with mAb feedstock, model proteins, and model DNA were performed using an ÄKTA Explorer FPLC system (GE Healthcare Bio-Sciences Corp., Piscataway, NJ, USA). The devices were connected to the ÄKTA Explorer with standard tubing and luer-lock connectors. A flow rate of 10 membrane volume (MV)/min was used. Binding of endotoxin and bacteriophage molecules was performed using a separate experimental setup consisting of a peristaltic pump (Watson Marlow 302S), which allowed proper cleaning of the system. To determine flow rates, membrane adsorber devices were connected to a pressure vessel filled with buffer or protein solution. The filtrate volume was monitored using a balance and the flow rates for different pressures were calculated up to an inlet pressure of 3 bar.

Model systems
Bovine serum albumin (BSA, Lot 50121326) was purchased from Kraeber GmbH & Co. and salmon sperm DNA (DNA, Lot 8087) from Biomol. The protein throughput was determined using γ-globulin (Sigma, γ-globulin from bovine blood, Lot STB0227K9). Endotoxin from Escherishia coli (Lonza LPS E. coli 055:B5 N185 Lot 0000100778) was used as standard. Bacteriophage ΦX174 (ATCC 13706-B1) was produced in a 50 L disposable bioreactor using the E. coli (ATCC 13706) expression system. Subsequently, phage was purified, concentrated, and sterile filtered by several steps including a depth filtration cascade, crossflow filtration, precipitation with polyethylene glycol, and centrifugation.

MAb feedstock
The mAb feedstock was obtained from pilot-scale batches produced at Genentech (a member of the Roche Group). It was expressed in mammalian cells and clarified to remove insoluble impurities. The mAb was processed through a protein A chromatography step and further purified using a cation-exchange chromatography step. Protein concentration was approximately 11 g/L.

METHODS

Dynamic binding capacity

Each device was sanitized with 1 N NaOH for 30 min at 10 MV/min followed by equilibration with 150 MV binding buffer composed of 150 mM NaCl in 20 mM Tris/HCl pH 7.3 ± 0.1, conductivity 16 mS/cm. 150 MV of 1 g/L BSA in binding buffer or 0.1 g/L DNA in binding buffer were loaded. All solutions used were prefiltered with a 0.2 μm membrane filter. All steps were performed at flow rate of 10 MV/min. Breakthrough curves were recorded by measuring the extinction at 280 nm (protein) and 260 nm (DNA) using the ÄKTA Explorer. To compare different devices the void volume of the experimental setup was determined by injection of acetone (2 %). The dynamic binding capacity at 10% breakthrough was calculated as shown in

Equation 1

,

where V10% is volume loaded at 10% breakthrough, Vv is void volume, Vm is membrane volume, and ci is initial concentration.

Protein throughput
Each membrane adsorber device was sanitized with 1 N NaOH for 30 min at 10 MV/min followed by equilibration with 100 MV binding buffer composed of 150 mM NaCl in 20 mM Tris/HCl pH 7.3 ± 0.1, conductivity 16 mS/cm. Protein throughput was determined using the pressure vessel filled with a solution of 20 g/L γ-globulin in binding buffer was used to determine the protein throughput with the membrane adsorber devices. The filtrate volume up to 1000 MV was monitored at a constant pressure of 3 bar using a balance.

Chinese hamster ovary proteins clearance
Chinese hamster ovary proteins (CHOP) clearance was determined using industrially relevant mAb feedstock. Before loading the MAb feedstock onto the membrane adsorber, the membrane was equilibrated with 10 MV of 50 mM Tris buffer at the appropriate pH. The conductivity of this buffer was adjusted by altering the concentration of sodium acetate. After equilibration, the mAb feedstock was loaded onto the devices to a targeted load density of 10 kg mAb/L of membrane at a flow rate of 10 MV/min. Pool fractions were collected during the experiment and analyzed for CHOP concentration.

Determination of log reduction value of bacteriophages
Equipment and membrane devices were sanitized with 1 M sodium hydroxide for 30 minutes. Membrane devices were further equilibrated with 300 MV of binding buffer. The ΦX174 phage solution with a titer of 1.5x10 7 PFU/mL was prepared and loaded onto the devices at a flow rate of 10 MV/min. Flow-through fractions were collected after 100 and 150 MV of load for quantitative analysis.

Endotoxin removal
Pump, tubing, and devices were treated with 1 M sodium hydroxide for 30 minutes at room temperature and at a flow rate of 10 MV/min before performing the experiment. Compatible vessels and materials were heated at 200 °C for 4 hours to destroy naturally occurring endotoxins. After sufficient rinsing with reverse osmosis water, the equilibration was performed with 300 MV of binding buffer. 150 MV of endotoxin in binding buffer were loaded to the membrane at a flow rate of 10 MV/min. The flow-through was divided into fractions of 50 MV each and was analyzed to determine the endotoxin level.

ASSAYS

CHOP quantification

An ELISA was used for CHOP quantification. Samples containing CHOP were incubated in the wells, followed by incubation with anti-CHOP antibodies conjugated with horseradish peroxidase (HRP). The HRP enzymatic activity was detected with

o

-phenylenediamine, and the CHOP was quantified by reading absorbance at 490 nm in a microtiter plate reader. Based on the principles of sandwich ELISA, the concentration of peroxidase corresponded to the CHOP concentration. The assay range for the ELISA was typically 10–320 ng/mL, with intra-assay variability of approximately 10%. CHOP values were reported in units of ng/mL. CHOP values could be divided by the mAb concentration and the results reported in units of PPM (parts per million; ng of CHOP/mg of mAb).

Bacteriophage ΦX174 quantification

Host organismE. coli was used for the detection of infectious ΦX174 phage particles. E. coli cells were incubated on agar plates (Soybean-Casein Digest Agar Medium– Trypticase Soy Broth 211043), which served as a base layer with nutrients. E. coli cells multiplied rapidly and formed a bacterial lawn. Phage particles infect the cells, causing the lysis of E. coli host cells and producing single circular, nonturbid areas called plaques in the bacterial lawn. Each plaque represents the lysis of a phage-infected bacterial culture and is designated as a plaque-forming unit (pfu), and used to quantitate the number of infective phage particles in the culture. Plaques must be clearly defined and samples were then diluted several times (1:10) depending on the phage concentration. During the study, 150 μL of the host cell solution (optical density 2–6) was mixed with 150 μL of sample and top agar (1.3% Tryptikase Soy Agar BD 211043) and the mixture was then distributed to agar plates (4% Tryptikase Soy Agar BD 211043 in 90 mm petri dishes) and incubated for 18 to 24 hours at 37 °C. Plaque forming units were counted and the titer of the sample in PFU/mL (plaque forming units per mL) was calculated using Equation 2,

where P is the number of plaques of all countable dilutions, E is the sum of emphasis, D is the lowest evaluated dilution, and VSample is the sample volume.

The LRV was calculated using Equation 3,where c0 was the titer of the initial solution and cFT the titer in the flow-through fraction.

Endotoxin quantification

The endotoxin level was measured by the kinetic chromogenic method test according to the manufacturer's instructions (Limulus Amebocyte Lysate Chromogen, Charles River endosafe Endochrome-K R1710K, Lot A4992L 10/2012). The quantification principle is based on coloration caused by the contact of a sample containing endotoxin with a mixture of lysate and chromogenic substrate. A β-glucan blocker was added (Lonza N190 Lot 0000132199 01/11). During the 1-hour incubation the extinction coefficient was measured continuously at 405 nm using a temperature controlled (37 °C) plate reader (Tecan Safire). The reaction rate varies with endotoxin level and the samples were quantified for endotoxin by comparing the results with the calibration series. The detection limit of the assay was 0.012 EU/mL. LRV was calculated similarly to phage quantification by measuring the endotoxin level of the initial solution El0 and the level of endotoxin in the collected flowthrough fractions (ElFT) using Equation 4.

RESULTS

Flow rate and protein throughput

Device geometry must allow for linear scalability through the entire device size range. Pressure flow curves were generated with the axial flow Sartobind pico and radial flow Sartobind nano devices with data shown in

Figure 3

. The normalized flow rate (membrane volume (MV)/minute) increased linearly with the increasing inlet pressure and the flow rates were comparable, suggesting effective flow distribution and efficient utilization of membrane area with both pico and nano devices.

For a typical polishing application with an AEX membrane adsorber, the load capacity is very high, exceeding 10 kg of protein feedstock per liter of membrane volume and can thus present the risk of membrane fouling. To assess fouling as a function of load capacity, the Sartobind pico and Sartobind nano devices were loaded with a 20 g/L γ-globulin solution to a load capacity of 20 kg/L at a constant inlet pressure of 3 bar. As seen in Figure 4, while slightly higher flow decay was observed with the pico device, the overall flow decay was minimal with the two devices thus demonstrating the absence of significant membrane fouling at high load density.

Characterization of membrane adsorber devices using model systems

Chromatography media are usually characterized using model molecules, with dynamic binding capacity and impurity clearance reported at specific process conditions. The dynamic binding capacity for Sartobind STIC-PA was determined using bovine serum albumin (BSA) and DNA, and impurity clearance was evaluated using DNA, endotoxin, and bacteriophage.

Dynamic binding capacity: The dynamic binding capacity at 10% breakthrough was measured for the Sartobind pico, the Sartobind nano, and the Sartobind SingleSep 10" capsule using BSA and DNA model systems. All devices were assembled with STIC-PA membranes. The breakthrough curves for the three devices are shown in Figures 5 and 6 for BSA and DNA, respectively. The breakthrough curves are similar for all devices suggesting consistent flow distribution and efficient utilization of the membrane binding sites at the three scales. Table II shows the average BSA and DNA dynamic binding capacity values for several Sartobind pico, nano and 10" devices. At 10% breakthrough, the difference in dynamic binding capacity for all three devices was insignificant. The consistent dynamic binding capacity with BSA and DNA supports a linear scalability from 0.08 mL axial flow pico device to 180 mL radial flow SingleSep 10" capsule.

Removal of bacteriophage: Pathogen clearance was evaluated using the bacteriophage ΦX174, serving as a surrogate for mouse minute virus (MMV), which is typically used as a model virus for virus validation studies. Both ΦX174 (26-33 nm diameter) and MMV (20 nm diameter) are small nonenveloped DNA viruses with an isoelectric point of around 6.7–7.0 and 6.2 respectively (13). At pH > 7, both ΦX174 and MMV are mainly negatively charged and expected to bind to positively charged AEX chromatography membranes, resulting in their clearance from protein feedstock through electrostatic interactions. To compare clearance between Sartobind pico and Sartobind nano, the same ratio of ΦX174 to membrane volume was loaded. Process-scale capsules were not tested because of the large amount of phage material required. Two flow-through fractions were collected with each pico and nano device, and the LRV was evaluated by comparing the phage titers of the fractions with the load solution. As shown in Table III, similar LRVs were obtained at a load of 100 and 150 MV of phage-spiked buffer, demonstrating linear scalability between the devices.

Removal of endotoxin: Endotoxins are lipopolysaccharides found in the outer membrane of various gram negative bacteria, can be present as different forms of micelles and vesicles, and are generally strongly negatively charged. Because of their ability to elicit immunogenic responses in humans, endotoxins must be removed to typically < 0.25 Endotoxin Units per milliliter (EU/mL) where EU is the unit of measurement for endotoxin activity (USP <29>). Table IV shows the results for endotoxin removal with Sartobind pico and nano devices at pH 7.3 in a buffer containing 150 mM sodium chloride. The concentration of endotoxin in the load was 108 EU/mL, which is significantly higher than the concentration of endotoxin typically found in any in-process pools. Three fractions were collected from the flow-through at loading volumes of 50, 100, and 150 MV. All flow-through fractions had an endotoxin concentration below the detection limit of 0.012 EU/mL resulting in a LRV > 3.96 except one fraction at 50 MV with the pico device. However, subsequent fractions at higher load volumes with the same pico device provided LRV > 3.96 which suggests that the anomalous reading at 50 MV was likely due to an assay error or sample contamination. Based on the load volumes tested, the total amount of endotoxin removal was > 1296 EU with the pico and > 16200 EU with the nano device. Significantly larger amount of endotoxin would be required in the load to saturate the membrane with the endotoxin molecules to determine and compare the breakthrough curves for both pico and nano devices.

Performance of Sartobind pico with an industrially relevant mAb feedstream

In a mAb purification process, AEX chromatography is typically operated in a flow-through mode to bind trace levels of impurities such as DNA, putative viruses, endotoxins, and host cell proteins, while the mAb product flows through. The load capacity is indicated as the mass of product loaded per unit volume of chromatography membrane (kg mAb/L membrane) such that the purity level in the product pool is acceptable. To assess the performance with an industrially relevant feedstream, both pico and nano devices were loaded with an in-process mAb pool post Protein A and cation exchange chromatography step. Subsequently, CHOP levels were monitored in the flow-through as a function of load density. The devices were loaded to 10 kg/L load density at two different solution conditions (pH 7.0 and 8.0 at 11 mS/cm). CHOP clearance as a function of load density is shown in Figure 7. Comparable CHOP clearance was obtained with the pico and the nano device at both solution conditions using an industrially relevant mAb feedstock, suggesting that the Sartobind pico is scalable to the Sartobind nano device. Additionally, at pH 7.0 and 11 mS/cm, a load capacity ≥ 10 kg/L could be achieved with pool CHOP levels < 10 ppm.

The CHOP clearance results are consistent with the earlier data where comparable BSA and DNA dynamic binding capacity was observed between the pico, nano, and process scale 10" devices. Comparable clearance of endotoxin and the bacteriophage further demonstrated the scalability of Sartobind pico to the Sartobind nano.

CONCLUSION

It is well documented in the literature that AEX membrane adsorbers are an attractive alternative to columns for polishing applications in a flow-through mode. Because of its hydrodynamic benefits, load capacity greater than 10 kg/L of membrane can be achieved with membrane chromatography. Such high load density necessitates a significantly large amount of protein feedstock for process development and validation, which could be cost prohibitive. To overcome this limitation and also to reduce validation cost particularly for virus spiking studies, an ultra scale-down device, Sartobind pico, having a membrane volume of 0.08 mL was developed. Using model molecules and an industrially relevant mAb feedstock, Sartobind pico was compared to the existing commercial scale-down device Sartobind nano. BSA and DNA breakthrough curves, CHOP, bacteriophage, and endotoxin clearance data demonstrate the scalability of Sartobind pico to the Sartobind nano. The new scale-down pico device will facilitate the development of flow-through polishing applications for recombinant proteins and monoclonal antibodies by reducing the sample consumption by 10-fold and providing substantial cost savings for process characterization and virus validation studies.

Nathalie Frau, PhD*, is a senior scientist in R&D process technologies at Sartorius Stedim North America, Bohemia NY; Martin Leuthold, PhD, is a scientist in R&D product development at Sartorius Stedim Biotech, Goettingen, Germany; Amit Mehta, PhD, is a senior engineer in purification development and Kome (Kevin) Shomglin, PhD, is a senior research associate in purfication development at Genentech, South San Francisco, CA; and Rene Faber, PhD, is vice-president, R&D process technologies at Sartorius Stedim, North America, Bohemia NY. *To whom correspondence should be addressed, nathalie.frau@sartorius-stedim.com.

REFERENCES

1. J.X. Zhou and T. Tressel, Biotechnol. Prog. 22, 341–349 (2006).

2. J.X. Zhou et al., J. Chromatogr. A 2006,1134, 66–73.

3. J. Glynn et al., "Downstream Procecessing 2010" supplement to Biopharm Int. 22, s16–s20 (2009).

4. A. Arunakumari, J. Wang, and G. Ferreira, "Advances in Process Chromatography" supplement to Biopharm Int. 22 s36–s40 (2007).

5. A. Arunakumari, J. Wang, and G. Ferreira, "Downstream Procecessing 2010" supplement to Biopharm Int. 22 s22–s26 (2009).

6. A. Mehta et al., "SBE Supplement–Bioprocessing" supplement to Chemical Engineering Progress 104, 14–20 (2008).

7. N. Fraud, Bioprocessing J. 7, 34–37 (2008).

8. A. Pastor, M. Hirai, and S. Fischer-Fruehholz, presentation at GVC/Dechema, (Osnabrück, Germany, May, 2007).

9. U. Gottschalk, Pharma Focus Asia 7, 60–65 (2008).

10. J. Zhou et al., Biotechnol. Bioeng, 100, 488–496 (2008).

11. Sartorius Stedim Biotech GMbH, "Scale up with Sartobind SingleSep," Application Note, SL-4042-e07081.

12. R. Faber, Y. Yang, and U. Gottschalk, Biopharm Int. 22 (10), 11–14 (2009).

13. D. M. Strauss et al., Biotechnol. Bioeng. 104, 371–380 (2009).