PURIFYING THERAPEUTIC MONOCLONAL ANTIBODIES Amit Mehta, Martha Lovato Tse, Jace Fogle, Amy Len, Roshan Shrestha, Nuno Fontes, Bénédicte Lebreton, Bradley Wolk, Robert van Reis Genentech Eliminating the affinity chromatography step in the mab purification process trims equipment requirements, shrinks the plant footprint and reduces costs. The market for therapeutic monoclonal antibodies (mabs) has grown tremendously in the last decade, and it is estimated that mabs and their derivatives account for almost 36% of the biopharmaceuticals under development, including vaccines and gene therapy (1). These and other therapeutic proteins are produced at the industrial scale using various recombinant cell lines, such as bacteria (e.g., E. coli), yeast, and mammalian cells (e.g., Chinese Hamster Ovary (CHO) cells (2)), with CHO cells being the most popular choice because they offer several benefits (3). Biopharmaceutical products must have very high purity, with the concentration of host cell proteins and DNA reduced to the range of parts per million relative to the desired product, or lower. The final product must also be sterile (no viable micro-organisms present), and should contain less than 1 ng of DNA per dose and less than one virus per million doses. This stringent purification of mabs produced in CHO cells is typically accomplished using a three-column chromatography process (Figure 1) that consists of protein A affinity chromatography as an initial capture step, followed by cation exchange (CEX) and anion exchange (AEX) chromatography as polishing steps and a virus filtration (VF) step (3). The protein A ligand has a high affinity for one area of the mab, specifically the crystallizable fragment (Fc), which enables the mab s capture from the cell culture fluid (thus the term affinity chromatography). Even though these chromatography steps are able to meet the stringent purification requirements, they are expensive, especially the protein A affinity step, which accounts for almost 35% of the total raw material costs for downstream purification (4). With growing demand for therapeutic mabs and increasing market competition, significant attention is being focused on reducing manufacturing costs and improving process efficiency for industrialscale production. As a result, there has been much interest in the development of cost-effective non-affinity purification processes that do not involve protein A. In one of the early studies on the purification of industrial-scale mab feed streams using a non-affinity process (4), Fahrner and Follman integrated different chromatography steps into three-column non-affinity purification sequences and evaluated their performance. Three different non-affinity sequences were identified that provided product quality comparable to the affinity process. This threecolumn non-affinity process with ultrafiltration/diafiltration (UF/DF) has since been streamlined into a two-column non-affinity process with the integration of high-performance tangential-flow filtration (HPTFF) (5). This twocolumn non-affinity process utilizes cation exchange as a capture step, followed by anion exchange as a polishing step, followed by HPTFF (which allows concentration, formulation and purification in a single step). This eliminates the most expensive chromatography step (protein A chromatography). In addition to reducing raw material costs, the two-column non-affinity process can provide several other benefits, including a smaller plant footprint and the need for fewer pieces of process equipment (e.g., tanks, columns and control systems) due to the smaller number of chromatography steps. This article addresses several aspects of the two-column non-affinity process: its development and optimiza- S14 SBE Special Section Bioprocessing CEP
tion using a manufacturing-scale feed stream; process yield and product quality; virus removal; and application to multiple mab feed streams. Purification process development Industrial-scale production of therapeutic mabs is complex, involving process development, process characterization and validation, process scale-up, automation, quality assurance, and regulatory compliance. The development of purification processes (typically at the laboratory scale) involves optimization and integration of several unit operations that provide the desired purity, product quality, throughput and yield. Although a comprehensive discussion of the different unit operations associated with the two-column non-affinity process is beyond the scope of this article, the key aspects related to the development and optimization of each step are addressed in the following sections. Process development is illustrated using, as an example, the proposed two-column non-affinity process to purify a mab A feed stream. mab A has an iso-electric point (pi) of 9.3 and molecular weight of approximately 145 kda. HPTFF experiments were performed on a scale-down control system using.1-m 2 prototype membranes with 3-kDa molecular-weight cut-off pore size. CEX column chromatography Cation exchange resins have negatively charged functional groups immobilized on their surface. They can be used as an initial mab capture step by loading harvested cell culture fluid (HCCF) under conditions where the mab is positively charged (the ph of the material being loaded is lower than the isoelectric point of the mab). Most of the impurities flow through the column during the loading phase; other impurities are separated from the mab during the wash and the elution phases. During the wash phase, the column is rinsed with a buffer having an appropriate ph and ionic strength, which removes residual impurities but leaves the mab bound to the column. A linear salt gradient is typically used during the elution phase, which allows separation of mab and impurities, such as aggregates, Harvest Protein A CIEX VF AIEX UF/DF This step utilizes centrifuge and depth filters for removal of intact cells and cell debris from cell culture fluid Low-pH Virus Inactivation Removes >98% of CHOP, virus, DNA and small molecules GLOSSARY Chinese Hamster Ovary proteins (CHOP) host cell proteins produced by CHO cells during fermentation, in addition to the desired protein product Diavolume the ratio of total volume of buffer added during the operation divided by the retentate volume Diafiltration (DF) a technique that utilizes a semi-permeable membrane to exchange the product of interest from one liquid medium into another Dynamic Binding Capacity (DBC) amount of target protein that binds to the media under actual flow conditions prior to product breakthrough Harvested Cell Culture Fluid (HCCF) cell culture fluid containing product of interest and devoid of intact cells and cell debris High-Performance Tangential-Flow Filtration (HPTFF) an emerging technology that separates proteins by exploiting the differences in size as well as charge Iso-electric Point (pi) ph at which there is no net charge on the protein Load Density amount of product loaded per volume of media (resin, membrane) Monoclonal Antibody (mab) very specific, identical antibody molecules that are produced by a cellular clone Process Throughput amount of mab processed per unit of time Ultrafiltration (UF) a technique that utilizes a semipermeable membrane to concentrate the product of interest Removes product variants, CHOP, DNA, virus, leached protein A and small molecules > 4 log virus removal Removes trace levels of CHOP, DNA and potential viruses Used for product concentration and formulation Figure 1. A three-column affinity process with ultrafiltration/diafiltration (UF/DF) is used to purify monoclonal antibodies expressed from CHO cell lines. CEP SBE Special Section Bioprocessing S15
Binding Capacity, g/l 1 1 Resin A Resin B Resin C Resin D Resin E Table 1. Yield and CHOP levels in the elution pool at the optimal wash and elution conditions for CEX resins. Resins were loaded to % of their DBC. CHOP levels in the initial load were approximately 1, ng/mg of mab. CEX Resin Load g/l Yield, % CHOP, ng/mg Resin A 77 97 6, Resin B 81 93 1, Resin C 69 98 2, Resin D 87 98 6, Resin E 41 5, 4 6 8 1 12 14 16 Conductivity, ms/cm Figure 2. Dynamic binding capacity for cation exchange resins used to purify the mab A feed stream as a function of load conductivity at ph 5. product variants and CHO protein (CHOP), due to their differing electrostatic affinities for the charged resin. Thus, the development of a CEX step is very rigorous and requires a thorough study of several process parameters. It involves: the selection of appropriate chromatographic resin; the optimization of load density, load flowrate (residence time at constant bed height), load ph, and ionic strength; and the optimization of the wash and elution phases. The selection of an optimal CEX resin entails the study of the inter-relationships between load ph, ionic strength, load density (mass of mab loaded divided by resin volume), impurity removal, load flowrate, and processing time required for a given feed stream. For example, loading the HCCF at lower ph and ionic strength may provide higher dynamic binding capacity (DBC) (6), but may significantly increase the processing time if a large dilution volume is required to adjust the ph and ionic strength. Lowering the load density might enhance impurity removal, but would increase costs due to lower utilization of chromatography resin. In addition to process parameters, the performance of CEX resin can be strongly affected by product characteristics, so different resins would be optimal for different products. An ideal CEX resin should provide sufficient DBC (typically > 5 g/l), high yield (> 9%) and acceptable product purity in the eluent (i.e., low levels of impurities such as host cell proteins, DNA, aggregates, etc.) to ensure product stability and reasonable cycle time. To identify the optimal CEX resin for the mab A feed stream, several resins were screened. The DBC of each resin was evaluated by loading the HCCF while varying the ph, conductivity and flowrates. Figure 2 shows the DBC as a function of load conductivity for different CEX resins at ph 5. The DBC varied from 2 to 129 g/l at 1% product breakthrough. The DBC increased with decreasing load conductivity due to enhanced electrostatic interaction between the positively charged mab and the negatively charged resin. The binding capacity also varied significantly from one resin to another at any given ph and conductivity. This can be attributed to differences in resin characteristics, such as pore size, ligand density, functional group and spacer arm. A maximum DBC for Resin A was observed at approximately 8 ms/cm, and can be attributed to the reduction in protein uptake by the CEX resin due to charge exclusion (6). The DBC maximum depends on media characteristics, such as charge density and pore size, and solution conditions, such as ph and ionic strength (6). It is likely that a similar trend would be observed with other resins if a wider conductivity range (i.e., less than 6 ms/cm) were explored. The wash and elution phases were designed by loading the resins to % of their DBC and evaluating the effect of parameters such as ph, ionic strength and buffer species on impurity removal. CHOP levels in the elution pools were determined using an enzyme-linked immunosorbent assay (ELISA), while mab concentration was determined by measuring absorbance at 2 nm on a UV/visible spectrophotometer. Table 1 summarizes the CEX step yield and CHOP levels in the elution pools at the optimal wash and elution conditions for each resin. Most of the resins provided more than 95% product yield, but CHOP clearance varied significantly from one resin to another. The best resin was selected based on binding capacity, pool purity, load flowrate (processing time) and the ability of the downstream unit operations to remove residual impurities in the CEX elution pool. For the mab A feed stream, resin C was chosen, because it provided significantly higher impurity removal than the other resins with adequate binding capacity and desired downstream purification. S16 SBE Special Section Bioprocessing CEP
Pressure, psi 1 Membrane A Membrane B Conditions: ph = 8 Conductivity = 4 ms/cm 1 2 3 4 5 Load Density, kg/l Figure 3. Pressure drop across two commercially available membranes used to purify the mab A feed stream as a function of load density. AEX chromatography Anion exchange resin chromatography is typically used as a second or third chromatography step in traditional mab purification processes to remove trace levels of CHOP, nucleic acids, endotoxins and any viruses that may be present. This step typically operates in a flow-through mode, that is, under conditions where the mab is positively charged and flows through the column while the negatively charged impurities bind to the positively charged resin. Although AEX resins are a powerful purification tool, their main drawback is a low mass-transfer rate due to intra-pore diffusion, which results in lower process throughput (mass of mab processed per unit of time). As a result, in some cases, the process-scale columns need to be oversized simply to meet the throughput requirements. Membrane chromatography is an attractive alternative to conventional resins, because binding sites are located within the convective path of the fluid, which enables higher throughput (7). In addition, the availability of membrane adsorbers in pre-packed formats significantly simplifies manufacturing by eliminating the need for packing and unpacking columns. Because of these key benefits, AEX membrane chromatography was selected as the second step for the non-affinity process. Because it operates in a flow-through mode, process development for AEX chromatography is less intensive compared to CEX. It involves the selection of the optimal media, and the determination of process parameters such as load ph and conductivity, flowrate, and load density. AEX membrane screening was done by loading the CEX elution pool under different solution conditions and analyzing the flow-through stream for impurities, such as CHOP and DNA. The load ph was kept at 1 1.5 ph unit below the mab iso-electric point, as the mab maintains a weak positive charge under these conditions and flows through the membrane. Low load conductivity is usually desired, as CHOP removal typically increases with decreasing conductivity due to enhanced electrostatic interaction between the CHOP and the membranes. However, lower load conductivity involves larger dilution of the load, resulting in larger load volume and longer processing time. (The effect of conductivity on processing time and impurity removal was studied in detail prior to the selection of the process parameters.) In addition, the pressure drop across the membranes was monitored, because membrane adsorbers can be operated safely only within a certain pressure range (specified by the vendor). Figure 3 shows the differential pressure drop across two commercially available membranes as a function of the load density. Unlike Membrane B, the pressure drop across Membrane A increased dramatically with increasing load density. If the safety limit for the pressure drop is psi, the data suggest that Membrane A can only be loaded to 1 kg/l, while Membrane B can be loaded to > 5 kg/l. Membrane adsorber B was selected for purification of the mab A feed stream because of its better pressure performance. Membrane B also provided approximately a twenty-fold CHOP reduction at a load density of 1 kg/l. High-performance tangential-flow filtration HPTFF (Figure 4) is an emerging membrane technology that performs protein purification, concentration and formulation in a single step. It utilizes a positively Positively Charged Membrane CHOP mab Retentate Flow Filtrate (To Drain) Figure 4. HPTFF is a membrane process used to separate mab and CHO proteins. CEP SBE Special Section Bioprocessing S17
Observed Sieving Coefficient, S o charged, large-molecular-weight cut-off (~ 3 kda) membrane (whereas traditional UF/DF uses a 1 3-kDa neutral membrane). The proteins are separated by exploiting differences in their size as well as charge (5 8). HPTFF performance is governed by membrane properties, such as pore size and membrane surface-charge density, and process parameters, such as filtrate flowrate, number of diavolumes, load ph and ionic strength. Experiments were performed with a prototype membrane having optimal pore size and charge density to determine process conditions that would provide a yield greater than 95% for this step and CHOP levels less than 1 ng of CHOP per mg of mab in the final bulk. Solution conditions were such that the positively charged mab was retained by the membrane while neutral or weakly charged Yield, % 1.1.1.1.1 CHOP mab A Conditions: mab A concentration in the bulk fluid = g/l Filtrate flux = 5 L/m 2 -h Load ph = 5.3 Conductivity = 4 ms/cm 2 4 6 8 1 12 14 16 18 Number of Diavolumes Figure 5. Observed sieving coefficients for mab A and CHOP as a function of the number of diavolumes during diafiltration of the mab A feed stream..1.1.1 Observed Sieving Coefficient, S o,mab Figure 6. Effect of the mab sieving coefficient on yield [Y = exp( N S o ) (9), where N is the number of diavolumes and S o is the observed sieving coefficient for the mab] for a diafiltration step with diavolumes. CHOP passed through it. Figure 5 shows the observed sieving coefficients (S o ) of mab A and CHOP as a function of the number of diavolumes at ph 5.3 and conductivity of 4 ms/cm. S o is the ratio of protein concentration in the filtrate to that in the retentate, and serves as a measure of protein transmission through the membrane. Filtrate flux was set at 75 L/m 2 -h during ultrafiltration and at 5 L/m 2 -h during diafiltration. The observed sieving coefficient of mab (~.1) was approximately two orders of magnitude lower than that of the CHOP during the entire ultrafiltration and diafiltration phase, enabling better than 95% product yield to be achieved after diavolumes. The effect of the mab sieving coefficient on yield for a diafiltration step with diavolumes is shown is Figure 6. The product yield decreases dramatically as the sieving coefficient increases. For example, the product yield decreases from approximately 9% at S o =.5 to less than 15% at S o =.1. Figure 7 shows the CHOP levels in the retentate during ultrafiltration and diafiltration. While the mab is electrostatically excluded from the membrane pores, CHOP passes through, so CHOP levels decrease with increasing diavolumes. CHOP concentration in the initial load was approximately 1 ng/mg, and was reduced to 3 ng/mg during ultrafiltration. CHOP levels were further reduced to 1 ng/mg after 4 diavolumes and to less than 5 ng/mg after 1 diavolumes. Virus clearance CHO cell culture fluid typically contains 1 3 to 1 9 retrovirus-like particles per ml, and since the industry standard is less than one retrovirus-like particle per million doses, downstream purification needs to provide several orders of magnitude of retrovirus removal and inactivation. Conventional purification processes typically CHOP, ng/mg 1 1 UF Conditions: mab A concentration in the bulk fluid = g/l Filtrate flux = 5 L/m 2 -h Load ph = 5.3 Conductivity = 4 ms/cm 5 1 15 Number of Diavolumes Figure 7. CHOP levels in the retentate as a function of the number of diavolumes during diafiltration of the mab A feed stream. DF S18 SBE Special Section Bioprocessing CEP
Table 2. Removal and inactivation of retroviruses by different unit operations in the affinity and non-affinity processes. Affinity Process Non-Affinity Process Virus Virus Clearance Clearance Unit Operation (LRV)* Unit Operation (LRV)* Protein A 2. Cation Exchange 3.8 Low-pH Inactivation 6. Heat Inactivation 4. Virus Filtration 4.6 Virus Filtration 4.6 Anion Exchange 5.5 Anion Exchange 5.5 Process LRV 18.1 Process LRV 17.9 * LRV = log reduction value = orders of magnitude reduction achieved LRV across the CEX and the viral heat inactivation steps was obtained using mab A feed stream. LRV across the other steps is based on internal R&D data or data supplied by vendors. Table 4. Amount of CHOP removed (ng/mg) by different steps in the non-affinity process with different mab feed streams. Step mab B mab C mab D HCCF 1,47, 2, 5, CEX 145, 53 45 AEX 41 15 19 HPTFF 2 < 1 2 the non-affinity processes provide approximately 18 logs of retrovirus removal/inactivation. In the affinity process, removal of adventitious virus is achieved in the protein A and anion exchange chromatography steps. Adventitious-virus removal in the non-affinity process can be obtained by utilizing new technologies, such as parvovirus filtration and UV inactivation, along with traditional anion exchange chromatography. target a 12 18 log reduction value (LRV) for retrovirus removal (i.e., the level of retroviruses is reduced by 12 18 orders of magnitude). Several technologies, such as chromatography, virus filtration, low-ph inactivation and heat inactivation, are used to meet the viral safety requirements. Table 2 lists the degree of retrovirus removal and inactivation achieved by the various steps of the affinity and the two-column non-affinity processes. The use of cation exchange as the initial capture step and the incorporation of a viral heat inactivation step in the two-column non-affinity process compensate for the loss of retrovirus removal that would have been achieved by the protein A chromatography and the low-ph inactivation steps that were eliminated. Analytical tests were performed to ensure that viral heat inactivation had no adverse impact on product quality. Both the affinity and Table 3. Process yield and CHOP clearance data (ng/mg) for different steps of the two-column non-affinity process. Step Run 1 Run 2 Run 3 Run 4 Run 5 Run 6 Run 7 Initial Load 162, 14, 14, 1, 76, 76, 67, CEX 3, 1, 1,3 3, 1, 1,15 95 AEX 9 1 1 23 15 1 7 HPTFF 4 4 5 3 4 3 3 Overall Process Yield 82% 79% 79% 76% 78% 81% 82% Process yield and purity After the individual unit operations were developed and optimized at the laboratory scale, the two-chromatography-step non-affinity process was tested at the pilot scale using the mab A feed stream. The process yield and the CHOP removal data across different steps are summarized in Table 3. Process yield was approximately %, and CHOP levels in the final pool were less than 5 ng/mg for all of the experimental runs. Both process yield and CHOP levels in the final pool are comparable to those obtained with the conventional affinity process. Other impurities, such as DNA, aggregates and small molecules, also met purity specifications for therapeutic mabs. The non-affinity process with other mabs The data presented so far illustrate the application of the two-column non-affinity process to a single mab feed stream. In order to study the broad application of this process, different mab feed streams were obtained and purified. Although the overall purification sequence (CEX as a capture step, AEX as a polishing step, and HPTFF for concentration, formulation and purification) remained the same from one mab to another, individual media and process parameters were tailored for different mabs. Table 4 summarizes the CHOP removal across different steps with three other mab feed streams. The CEP SBE Special Section Bioprocessing S19
CHOP concentration in the final pool for all mabs was 2 ng/mg or less. Even though the CHOP level in the mab B HCCF was almost fivefold to sevenfold higher than those in other feed streams, the CHOP concentration in the HPTFF pool was, nevertheless, reduced to 2 ng/mg, illustrating the robust purification capability of the non-affinity process. Concluding remarks The two-chromatography-step non-affinity process that uses cation exchange as a capture step and HPTFF as an additional purification tool shows much promise for the purification of therapeutic monoclonal antibodies. This process reduces the number of processing steps as well as total costs by eliminating the most expensive step in the current antibody manufacturing process protein A affinity chromatography. Yield, purity and product quality comparable to the traditional affinity process were obtained at pilot scale using a commercial-scale feed stream. The non-affinity process was successfully employed for multiple feed streams, illustrating its potential as a future platform process for mab purification. Furthermore, this process uses technologies that can be retrofitted to existing manufacturing plants, eliminating capital expenditure for high-volume production. CEP Literature Cited 1. Sinclair, A., A Practical Guide to Biopharmaceutical Manufacturing, Scrip Reports, London, U.K., and Informa Health Care USA, New York, NY (6). 2. SBE Special Supplement: From Chinese Hamsters to Therapeutic Proteins, Chem. Eng. Progress, 13 (1), pp. 33 52 (Oct. 7). 3. Fahrner, R. L., et al., Industrial Purification of Pharmaceutical Antibodies: Development, Operation and Validation of Chromatography Processes, Biotech. and Genet. Eng. Rev., 18, pp. 31 327 (1). 4. Follman, D. K., and R. L. Fahrner, Factorial Screening of Antibody Purification Processes Using Three Chromatography Steps without Protein, J. Chromatogr. A., 124, pp. 79 85 (4). 5. van Reis, R., Charged Filtration Membranes and Uses Therefor, U.S. Patent No. 7,1,55 (6). 6. Harinarayan, C., et al., An Exclusion Mechanism in Ion Exchange Chromatography, Biotech. Bioeng., 95, pp. 775 787 (7). 7. Zhou, J., and T. Tressel, Basic Concepts in Q Membrane Chromatography for Large-Scale Antibody Production, Biotechnol. Prog., 22, p. 341 (6). 8. van Reis, R., et al., High-Performance Tangential Flow Filtration Using Charged Membranes, Journal of Membrane Science, 159, pp. 133 142 (1999). 9. van Reis, R., and S. Saksena, Optimization Diagram for Membrane Separations, Journal of Membrane Science, 129, pp. 19 29 (1997). AMIT MEHTA is a group leader and an engineer in the Late Stage Purification Dept. at Genentech, Inc. (South San Francisco, CA; Phone: (65) 225-187; E-mail: mehta.amit@gene.com). His research interests include membrane separations and biopharmaceutical process development. He holds a PhD in chemical engineering from the Pennsylvania State Univ. MARTHA LOVATO TSE is a scientist in the Early Stage Purification Dept. at Genentech, where her work focuses on developing purification processes for molecules in early stages of development. She received a PhD in chemistry from The Scripps Research Institute under the direction of Prof. Paul Schimmel and was a National Science Foundation Postdoctoral Fellow at Stanford Univ. under the direction of Prof. Chaitan Khosla. JACE FOGLE is an engineer in the Process Research and Development Dept. at Genentech. He received a BS in chemical engineering from the Pennsylvania State Univ. and an MS and a PhD from the Univ. of Virginia. AMY LEN is an engineer in the Late Stage Purification Dept. at Genentech. She received BS and MS degrees in chemical engineering from Stanford Univ. and is a member of AIChE, ACS and the Society of Women Engineers (SWE). ROSHAN SHRESTHA is an associate engineer in the Manufacturing Science and Technology Dept. at Genentech. He received a BS in chemical engineering from Stanford Univ. NUNO FONTES is a group leader and senior engineer in the Late Stage Purification Dept. at Genentech. His research interests include protein purification and biopharmaceutical process development. He received a PhD from the New University of Lisbon in Portugal. BÉNÉDICTE LEBRETON is a senior group leader and scientist in the Early Stage Purification Dept. at Genentech. She received a PhD in chemical engineering from the Univ. of Birmingham, U.K., is a member of AIChE and ACS, and has coauthored 12 publications. BRADLEY WOLK is a distinguished engineer and director of process development engineering at Genentech. His research interests include equipment design and the evaluation of new technologies in the areas of cell culture, purification and filling. He received a BS in chemical engineering from the Univ. of California at Davis, and is a member of AIChE and NACE International. ROBERT VAN REIS is a distinguished engineer and senior group leader in the Late Stage Purification Dept. at Genentech. His research interests include the development of protein purification technologies to enable costeffective industrial-scale implementation of manufacturing processes that produce pharmaceuticals to meet critical unmet medical needs. He received his education in chemical engineering at the Royal Institute of Technology in Stockholm, Sweden, and conducted his thesis research at the American Red Cross Biomedical Engineering Laboratory in Bethesda, MD. He holds 14 patents and has coauthored journal articles and seven book chapters. He has given lectures at AIChE, ACS and North American Membrane Society (NAMS) meetings, as well as at Rensselaer Polytechnic Institute, Stanford Univ., Univ. of Wisconsin, and Washington State Univ. Acknowledgements We thank Genetech s Analytical Operations for performing the CHOP assays, and Bin Yang and Min Zhang in the Virus Testing and Development group for providing the virus removal data. We also thank Arick Brown, Jerry Bill, Ai Lin and Rachel Dinges in Late Stage Purification for assistance on different unit operations during pilot-scale runs. And we are grateful to the scaledown and pilot plant groups for supporting the runs. Finally, we acknowledge Millipore, Inc. (Bedford, MA) for providing the HPTFF membranes. S SBE Special Section Bioprocessing CEP