ARTICLE Fast and Scalable Purication of a Therapeutic Full-Length Antibody Based on Process Crystallization

Benjamin Smejkal,1 Neeraj J. Agrawal,2 Bernhard Helk,3 Henk Schulz,3 Marion Giffard,3 Matthias Mechelke,1 Franziska Ortner,1 Philipp Heckmeier,1 Bernhardt L. Trout,2 Dariusch Hekmat1 t Mu nchen, Boltzmannstr. Institute of Biochemical Engineering, Technische Universita 15, 85748, Garching, Germany; telephone:49 89 289 15770; fax: 49 89 289 15714; e-mail: hekmat@lrz.tum.de 2 Chemical Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts 3 Novartis Pharma AG, Basel, Switzerland
scalable, fast, and inexpensive alternative to key steps of a standard purication process for therapeutic antibodies. Biotechnol. Bioeng. 2013;110: 24522461. 2013 Wiley Periodicals, Inc. KEYWORDS: therapeutic monoclonal antibody; purication process; stirred process crystallization; scale-up; CHO cell culture harvest; alternative to chromatography
1

ABSTRACT: The potential of process crystallization for purication of a therapeutic monoclonal IgG1 antibody was studied. The puried antibody was crystallized in nonagitated micro-batch experiments for the rst time. A direct crystallization from claried CHO cell culture harvest was inhibited by high salt concentrations. The salt concentration of the harvest was reduced by a simple pretreatment step. The crystallization process from pretreated harvest was successfully transferred to stirred tanks and scaled-up from the mLscale to the 1 L-scale for the rst time. The crystallization yield after 24 h was 8890%. A high purity of 98.5% was reached after a single recrystallization step. A 17-fold host cell protein reduction was achieved and DNA content was reduced below the detection limit. High biological activity of the therapeutic antibody was maintained during the crystallization, dissolving, and recrystallization steps. Crystallization was also performed with impure solutions from intermediate steps of a standard monoclonal antibody purication process. It was shown that process crystallization has a strong potential to replace Protein A chromatography. Fast dissolution of the crystals was possible. Furthermore, it was shown that crystallization can be used as a concentrating step and can replace several ultra-/dialtration steps. Molecular modeling suggested that a negative electrostatic region with interspersed exposed hydrophobic residues on the Fv domain of this antibody is responsible for the high crystallization propensity. As a result, process crystallization, following the identication of highly crystallizable antibodies using molecular modeling tools, can be recognized as an efcient,

Introduction
Investigations on crystallization as a means of purication of proteins were started quite early in history. Already in 1840, the protein hemoglobin was crystallized by Huenefeld from a relatively crude mixture (Margolin and Navia, 2001). Other proteins were puried by crystallization from impure material, for example, lysozyme and ovalbumin from egg white (Alderton and Fevold, 1946; Hopkins, 1900; Judge et al., 1995), fungal lipase from fermentation broth (Jacobsen et al., 1998), ferritin from fresh slices of horse spleen (Granick and Michaelis, 1942), or urease from jack bean extract (Sumner, 1926). However, crystallization as a protein purication step was almost forgotten after the development of preparative chromatography methods using a variety of media with different selectivities (Przybycien, 1998). For years, the larger portion of the manufacturing costs of therapeutic proteins, in particular monoclonal antibodies (mAbs), were caused by upstream processing. The concentrations of antibodies in mammalian cell cultures were typically in the range of a few milligrams per liter. Due to signicant improvements of upstream processing in recent years, these concentrations have increased up to multi-grams per liter (Wurm, 2004). As a result, downstream processing, in particular preparative chromatography, now represents a bottleneck of the manufacturing process and the proportion of the manufacturing costs for downstream processing has

Additional supporting information may be found in the online version of this article at the publishers web-site. Correspondence to: Dariusch Hekmat Contract grant sponsor: Novartis Pharma AG Contract grant sponsor: Singapore-MIT Alliance Received 19 November 2012; Revision received 27 February 2013; Accepted 15 March 2013 Accepted manuscript online 26 March 2013; Article rst published online 22 April 2013 in Wiley Online Library (http://onlinelibrary.wiley.com/doi/10.1002/bit.24908/abstract). DOI 10.1002/bit.24908

2452

Biotechnology and Bioengineering, Vol. 110, No. 9, September, 2013

2013 Wiley Periodicals, Inc.

signicantly increased (Low et al., 2007). Therefore, the pharmaceutical industry has developed an increased interest in alternatives to protein purication by preparative chromatography (Giffard et al., 2011; Gottschalk, 2008). A standard purication process of a therapeutic antibody consists of several steps in order to reach the statutory high purity requirements. The antibody is usually captured by Protein A chromatography (ALC) from claried CHO cell culture supernatant. Then, a viral inactivation step at low pH (VIN) is performed. Two further subsequent chromatography steps, typically anion exchange chromatography (AEC) and cation exchange chromatography (CEC) follow in order to remove host cell proteins (HCPs), DNA, leached Protein A, and aggregates. At last, a viral clearance step is performed by nanoltration followed by ultra-/dialtration (UF/DF) (Fahrner et al., 2001; Kenney and Chase, 1987; Shukla et al., 2007). From a cost-efciency point of view, the reduction of costly chromatography steps as well as UF/DF steps is desired. Process crystallization appears to be a viable alternative since it can be performed inexpensively with large volumes (Harrison et al., 2003; Judge et al., 1995; Takakura et al., 2006). However, crystallization of full-length antibodies had little success in the past even on the rather intensively investigated mL-scale for structure determination by X-ray diffraction. This was due to the facts that the constant Fcregion is of little interest for structure determination and that Fab-fragments crystallize easier. To the knowledge of the authors, only one crystal structure of a human full-length antibody is published in the Protein Data Bank (Saphire et al., 2001) but hundreds of structure hits exist for human Fab-fragments. Three recent studies reported on the potential of mL-scale crystallization for purication of whole antibodies: Lewus et al. (2011) found crystallization conditions for the mAb IDEC-152 and suitable values of the osmotic second virial coefcient were identied where crystallization could take place. In another work, 4 out of 16 IgG2s formed diffraction-quality crystals and 4 others formed crystal-like particles (Trilisky et al., 2011). Zang et al. (2011) published results on the purication efciency of crystallization of another whole antibody. Model protein contaminants (lysozyme or BSA) were added to the antibody solution. These contaminants were excluded from the antibody crystals by more than 90% as it was observed before (Judge et al., 1998, 1995). Crystallization of the antibody was also possible from a previously conditioned and concentrated cell culture supernatant in a 40 mL vapor diffusion experiment. The resulting crystals had a purity of above 90%; however, the yield was only about 31%. In comparison, the yield from the puried antibody solution was about 95%. Additionally, the crystallization required several days to reach equilibrium. As a consequence of the described previous work, it is obvious that signicant improvements of yield and process time have yet to be achieved and scale-up of the crystallization process has to be performed (Hekmat et al., 2007). In the present work, the purication by crystallization was investigated for a full-length therapeutic IgG1 antibody which had not been crystallized before. The aims of the work

were: (1) perform experiments on the non-agitated microbatch scale and transfer suitable conditions from these experiments to stirred tanks in the mL- and the L-scale; (2) evaluate the scalability of the stirred crystallizers; (3) investigate the purication by crystallization from different impure solutions, in particular from pretreated claried cell culture supernatant; (4) determine the degree of reduction of DNA and HCP by crystallization and the level of biological activity of the redissolved antibody crystals; and (5) perform molecular modeling in order to identify antibody candidates which suggest a potentially high crystallization propensity.

Materials and Methods

Materials Claried cell culture supernatant (harvest) from a CHO cell line secreting monoclonal IgG1 type antibody mAb01 as well as puried and partially puried solutions of mAb01 were kindly provided by Novartis Pharma AG (Basel, Switzerland). All other chemicals were analytical grade purchased from Sigma-Aldrich (Munich, Germany) and from Carl Roth (Karlsruhe, Germany). Pretreatment of Harvest for Crystallization On the mL-scale, the harvest was rst concentrated 6.5-fold by a centrifugal lter unit (No. Z706345, Sigma-Aldrich). Afterwards, the concentrated harvest was dialyzed overnight at room temperature against a 100-fold volume of a 10 mM histidine/acetate buffer at pH 5 using a dialysis tubing (No. 1784.1, Carl Roth). Precipitated substances were removed by centrifugation (3,200 rcf for 15 min). The solution was sterile-ltered (0.2 mm) and stored at 4 C. On the L-scale, the harvest was titrated to pH 5 using 10% acetic acid. Precipitate was removed by centrifugation (3,200 rcf, 15 min). The salt was reduced by UF/DF against 10 mM histidine buffer at pH 5 using a crossow ultraltration unit (Experiment A: seven dialtration volumes with a Hydrosart membrane (Sartocon Slice 200, Hydrosart 30 kD MWCO, Sartorius AG, Gttingen, Germany); Experiment B: ve dialtration volumes with a PESU membrane (Sartocon Slice 200, PESU 10 kD MWCO, Sartorius AG). Precipitated substances were removed by centrifugation (3,200 rcf for 15 min). The solution was sterile-ltered (0.2 mm) and stored at 4 C. Crystallization The antibody mAb01 was crystallized at 10 C unless otherwise mentioned. TRIS base or sodium hydroxide was used to adjust the pH. Crystallization on micro-batch scale was performed in 72-well plates (No. 654102, Greiner BioOne, Frickenhausen, Germany) with a volume of 10 mL per well. Each well was lled with 5 mL of a protein solution and 5 mL of a crystallization reagent solution. The crystallization was performed with a parafn oil overlay (No. 9190.1, Carl Roth). The scale-up of crystallization was performed in

Smejkal et al.: Purification of an Antibody by Crystallization Biotechnology and Bioengineering

2453 2453

geometrically similar stirred tanks with working volumes of 6 mL, 100 mL, and 1 L as described before (Smejkal et al., 2013) (Fig. 1). The ratio of impeller diameter (d) and tank inner diameter (D) was 0.5. The ratio of tank lling height (H) and D was 1.0. A three-bladed segment impeller was used in all stirred vessels (IUL Instruments, Knigswinter, Germany). The ratio of impeller height (h) to H was kept constant at one-third. All vessels were temperature controlled by a refrigerated circulator (No. 1157P, VWR, Darmstadt, Germany). The stirrer speed was 250 rpm in the 6 mL-scale, 200 rpm in the 100 mL-scale, and 150 rpm in the 1 L-scale. Separation and Dissolution of Crystals The separation of mAb01 crystals was performed by centrifugation (6 mL scale: 16,200 rcf, 5 min; 1 L-scale: 3,200 rcf, 15 min) at the respective crystallization temperature. The supernatant was removed and the crystals were resuspended and dissolved in 10 mM histidine/acetate buffer pH 5.0. The pH was adjusted to pH 5.0 by adding a solution of 10% acetic acid when required. Microphotographs Visualization of crystals and precipitate was performed using a polarizing microscope (Nikon Eclipse 50i, Nikon, Dsseldorf, Germany) with either a 10-fold objective (CFI Plan Fluor) or a fourfold objective (CFI Plan Achromat). Microphotographs were taken using an attached digital camera (DS-2Mv, Nikon). The mean crystal length was determined from the microphotographs using an image processing software (NIS Elements v3.2, Nikon).

Analytics Solution purity by SEC, HCP content, DNA content, and biological activity were assessed at Novartis Pharma AG after crystal dissolution in 10 mM histidine pH 5.0. The method for measurement of HCP content is described elsewhere (Rey and Wendeler, 2012). Contamination of drug substance produced in CHO cells by residual DNA is determined by quantitative polymerase chain reaction amplication of a repetitive sequence dispersed throughout the CHO cells genome (Haynes and Jelinek, 1981). Further details on the method used are to date not disclosed by the industrial sponsor besides the lower limit of quantication (2 pg DNA per digest, i.e., 2 pg DNA per mg mAb01). The biological activity of mAb01 is measured by a competitive ELISA assay. Further details on the method used are to date not disclosed by the industrial sponsor. At the Technische Universitt Mnchen (TUM), the concentration of the pure mAb01 antibody in solution was measured photometrically at a wavelength of 280 nm. The extinction coefcient was 1.622 L g1 cm1. The protein concentration of impure mAb01 solutions was measured by SEC (Superdex 200 10/ 300 GL column, GE Healthcare, Freiburg, Germany) at a ow rate of 0.5 mL min1. The amount of mAb01 was determined by the peak area compared to an external reference standard.

Molecular Simulations Structures of the Fab domain of mAb01 and other antibodies were generated and 20 ns of molecular dynamics (MD) simulations in explicit water were performed as outlined in Lauer et al. (2012). For each antibody, from the last 10 ns of MD simulation trajectory, a structure of the Fab domain was extracted at every 0.1 ns and the spatial interaction map (SIM) tool (Agrawal et al., 2012) was applied to these 100 structures to identify clusters of exposed hydrophobic residues on the Fab domain surface. For the SIM tool, a hydrophobic cutoff value of Fcutoff 0.15 was used and the sequence conservation criterion was not used. The residues identied by the SIM tool were visualized on the van der Waal surface of the Fab domain using the PyMOL molecular visualization system (The PyMOL Molecular Graphics System, Version 1.3, Schrdinger, LLC). An electrostatic potential map for each Fab domain was generated on the last frame of the MD simulation using the Adaptive Poisson-Boltzmann Solver (APBS) (Baker et al., 2001) implemented in APBS Tools version 2.1 plugin of PyMOL. Default values of all parameters were used for the APBS calculations. For generating the input le for APBS calculations, the PDB2PQR version 1.7 (Dolinsky et al., 2007) was used with the CHARMM force eld (MacKerell et al., 1998). For histidine residues in the Fab domain, a neutral protonation state of the side-chain was used with dnitrogen protonated and e-nitrogen unprotonated. For all other residues, default protonation states of their side chains at neutral pH were used. The electrostatic potential map was projected onto the solvent-accessible surface of the Fab

Figure 1.
scale.

Stirred crystallization systems. (A) 6 mL-scale; (B) 100 mL-scale; (C) 1 L-

2454

Biotechnology and Bioengineering, Vol. 110, No. 9, September, 2013

domain and visualized using the PyMOL molecular visualization system.

Results and Discussion

Crystallization and Characterization of Pure mAb01 Crystals of the monoclonal antibody mAb01 were incidentally found during a stability study at Novartis. The crystallization window was determined by a large number of experiments on the non-agitated micro-batch scale (Fig. 2A). The ionic strength and the pH of the solution had the main impact on the crystallization. The ionic strength was varied by different concentrations of sodium chloride and the pH was varied by adding different amounts of TRIS base to the crystallization buffer. Crystals were obtained up to a sodium chloride concentration of 60 mM within the observed timeframe of 2 weeks. The solution stayed clear at higher salt concentrations and many small crystals were formed at low salt concentrations. Hence, sodium chloride had an inhibiting effect on nucleation. Thus, low ionic strength seems to be advantageous for the crystallization of mAb01. This also applies to several other antibodies from literature (Harris et al., 1995; Larson et al., 1991; Trilisky et al., 2011). Crystallization of mAb01 took place by adding 616 mM TRIS base with the optimum at 10 mM TRIS base. Since the pH could not be measured in a small volume of

Figure 2. Crystallization of mAb01 on micro-batch scale at 10 C. A: Phase diagram at 10 g L1 mAb01 in 10 mM histidine/acetate buffer; TRIS base was used to vary the pH;  clear well; ^ crystal formation. B: 10 g L1 mAb01 in 10 mM histidine/ acetate buffer; 25 mM NaCl; 7 mM TRIS base; pH 6.2. C: 10 g L1 mAb01 in 10 mM histidine/acetate buffer; 25 mM NaCl; 8 mM TRIS base; pH 6.3. D: 25 g L1 mAb01 in 10 mM histidine/acetate buffer; 20 mM NaCl; 10 mM TRIS base; pH 6.8.

10 mL, 1 mL of the crystallization solution was prepared containing 10 mM TRIS base. Here, a pH of 6.8 was measured. This pH was identical with the isoelectric point (pI) of the antibody. A pH equal to the pI leads to a low solubility of the protein and therefore to a high supersaturation. More nucleation events appear at a higher supersaturation level. Therefore, it is comprehensible that only few large crystals were formed at a lower pH of about 6.2 using 7 mM TRIS base (Fig. 2B) and a medium amount of crystals were formed at a slightly increased pH to 6.3 using 8 mM TRIS base (Fig. 2C). The supersaturation was increased even more by increasing the mAb01 concentration from 10 to 25 g L1 and adjusting the pH to the pI. As a result, a signicantly higher amount of crystals was formed (Fig. 2D). The crystallization was transferred to the stirred 6 mL-scale with the aim to achieve a scalable crystallization process (Smejkal et al., 2013). The crystallization process was fast and reproducible (Fig. 3A). A crystallization yield of 92% was reached within 2 h. The yield at equilibrium was 93%. Microphotographs conrmed that only crystals and no amorphous aggregates were formed (Fig. 3B). The rod-like morphology was comparable to the morphology of the crystals obtained in the micro-batch experiments. However, the crystal size of the stirred experiments was smaller. A scaleup was successfully performed from the 6 mL-scale to the 1 L-scale. In this experiment, the supersaturation level was increased by using a higher antibody concentration of 25 g L1 instead of 10 g L1. In addition, no sodium chloride was added since it had a nucleation inhibiting effect, as mentioned above. Both measures led to a signicantly higher nucleation rate. As a result, the crystallization process was extremely fast (Fig. 3C). After just 3 min, the crystallization yield was already as high as 95.8%. A high yield of 97.8% was reached within 2 h. The remaining antibody concentration in solution after 2 days was 0.4 g L1 which was equivalent to an equilibrium yield of 98.3%. The successful transfer to the 1 L-scale allows to expect that a further scale-up will pose no problems. This particular crystallization process is outstanding in view of the extremely fast kinetics and the high yield. The stability of the antibody crystals was measured at different temperatures and pH values. Prior to this, the antibody was crystallized on the stirred 6 mL-scale from a solution containing 25 g L1 mAb01, 20 mM NaCl, and 10 mM histidine/acetate buffer adjusted to pH 6.8 with TRIS base. The experiments were performed for temperatures ranging from 10 to 30 C, the pH was changed stepwise by adding a solution of 10% acetic acid or 1 M sodium hydroxide, and the dissolved protein concentration was measured. The condition was dened as stable when the antibody concentration in solution was less than 2 g L1 and constant in time (Table I). At 10 C, the mAb01 crystals were stable and did not dissolve over a wide pH range from 5.5 to 7.6. The stable pH region decreased at higher temperatures. At 30 C, the crystals were not stable any more (Fig. 4). Additionally, amorphous aggregates were formed at a pH higher than 8. The crystals dissolved very fast at a pH below 5.5 at all investigated temperatures. From this observation, it

Smejkal et al.: Purification of an Antibody by Crystallization Biotechnology and Bioengineering

2455 2455

Figure 3. Stirred batch crystallization of pure mAb01 at 10 C. TRIS base added to adjust the pH to 6.8. A and B: 6 mL-scale; 10 g L1 mAb01; 20 mM NaCl; 10 mM histidine/acetate buffer. C and D: 1 L-scale; 25 g L1 mAb01; 52 mM trehalose; 10 mM histidine/acetate buffer. A and C: Time courses of the crystallization process. B and D: Robust rod-shaped crystals.

could be deduced that the preparation of a highly concentrated mAb01 solution from crystals is feasible. In fact, a highly concentrated solution containing 200 g L1 antibody was obtained from a centrifuged crystal pellet at pH 5 within 0.5 h. In previous work, dissolution of protein crystals required signicantly more time (patent application No. WO2009085765 A1). The study of mAb01 crystallization from pure solution could be applied to drug product formulation as a means to achieve stable, highly concentrated, injectable suspensions.
Table I. Stability of mAb01 crystals at different temperatures and pH. Stable pH region (protein solubility <2 g L1) 5.57.6 6.07.2 6.47.2 Not stable Solubility at pH optimum, g L1 0.85 1.2 1.3 4.9

Temperature ( C) 10 20 25 30

pH optimum 6.8 6.7 6.6 6.5

Of trehalose, 250 mM was added to the suspension formulation in order to adjust the osmolality of the formulation, as trehalose is commonly used as a stabilizer for protein formulation (Andya et al., 2003; Shenoy et al., 2001). The mean crystal length was controlled in an appropriate range of 1040 mm by decreasing the protein concentration at the crystallization start and thereby decreasing supersaturation (Fig. 5). These measurements conrmed the classical nucleation theory which states that a disproportionately lower number of nucleation events take place at a lower supersaturation level (Garcia-Ruiz, 2003). However, mAb01 assays performed after crystal dissolution revealed that the amount of water in the crystalline lattice is approximately 70%. Due to this high water content, an injectable crystalline suspension of mAb01 is not likely to deliver a dose signicantly greater than what is currently achieved with a highly concentrated liquid formulation (125 mg for 1.0 mL). This observation would to date limit the interest of mAb01 injectable crystalline suspension development to a potential extension of the serum pharmacokinetics

2456

Biotechnology and Bioengineering, Vol. 110, No. 9, September, 2013

Figure 4.
(


)T 10 C; () T 30 C.


Solubility of mAb01 crystals depending on pH and temperature.

prole (Yang et al., 2003). It should be also noted that neither the usual sterilization process for biopharmaceuticals (sterile ltration) nor the usual sterilization process for small molecule microparticles (terminal sterilization) could be used for biopharmaceutical crystalline suspensions. Crystallization of mAb01 From Partially Puried Solutions The purication capability by crystallization was investigated from solutions obtained from different intermediate steps of

the existing purication process of mAb01. The purication process started from claried cell culture harvest. Protein A chromatography (ALC) was followed by virus inactivation at low pH (VIN). Final polishing was performed by AEC and CEC. The antibody could be crystallized directly from the combined ALC/VIN step as well as from the AEC step. This was possible by simply adjusting the pH of the solutions. The crystallization conditions of the 6 mL stirred tank were: 8 g L1 mAb01; 10 mM histidine/acetate buffer adjusted to pH 6.7 with TRIS base. The crystals were separated, dissolved, and recrystallized. All crystallization steps had a high yield between 93% and 97% (Table II). Crystallization directly from the CEC step was not possible because of the high salt concentration required for product elution. No aggregation or degradation products were formed during the crystallization and dissolving processes and the protein was kept in its native state. A HCP reduction was observed in all crystallization steps; however, the HCP reduction level of the AEC step was not reached by crystallization. On the other hand, the crystallization achieved a comparable HCP reduction level of the CEC step. Crystallization offers an advantage compared to CEC since the salt content after dissolving the crystals is signicantly lower and as a result, a potential dialtration step could be avoided. Crystallization of mAb01 From Claried Cell Culture Harvest The antibody mAb01 was produced in CHO cells. It was secreted among other substances into a complex serum free cell culture medium. A direct crystallization was not possible from claried cell culture harvest. Crystallization was not possible even after concentration of the harvest to 21.5 g L1 whereas pure mAb01 could be crystallized at concentrations as low as 1 g L1. The presence of contaminants like HCP, peptides, DNA, RNA, and various salts might inhibit the crystallization reaction in the claried cell culture harvest. Crystallization was successfully performed on a micro-batch scale after the concentrated harvest was dialyzed against a 10 mM histidine/acetate buffer pH 5.0 (Fig. 6). Crystals were observed already after 1 day at 10 C. This pretreatment removed most of the interfering cell culture media and fermentation products and enabled a relatively fast crystallization. The conductivity of the solution was lowered by the pretreatment from 10 to 0.7 ms cm1. It is likely that the high
Table II. Purication of mAb01 from the ALC/VIN and AEC step by crystallization. Step yield, % 94.4 96.8 93.1 95.5 Purity by SEC, % 98.8 98.8 99.0 99.1 99.2 99.2 HCP, ppm 2,656 1,935 1,290 29 8 5 Bioactivity, % 89.3 93.8 96.4 88.7 93.0 91.9

Sample ALC/VIN ALC/VIN crystallized ALC/VIN recrystallized AEC AEC crystallized AEC recrystallized

Figure 5. Mean crystal length as a function of mAb01 concentration. The solid curve represents an optical t of the measured data.

Smejkal et al.: Purification of an Antibody by Crystallization Biotechnology and Bioengineering

2457 2457

Figure 6. Crystals from micro-batch experiments at 10 C using concentrated and dialyzed mAb01 harvest. Crystallization conditions were 5 mM histidine/acetate buffer, 10 mM NaCl, 6 mM TRIS base added, (A) 4.0 g L1 mAb01, (B) 6.5 g L1 mAb01. Microphotographs were taken after 24 h.

salt concentration in the harvest was the main inhibiting factor, since high concentrations of sodium chloride inhibited the crystallization of pure mAb01. The crystallization was transferred to the stirred 6 mL-scale and 1 L-scale. Two crystallization experiments were performed at the 1 Lscale from two different cell culture harvests. The HCP concentration in the starting material varied by a factor of 3 (Table III). Experiment A comprised a pretreatment of the claried cell culture harvest, a crystallization, and a recrystallization step. Experiment B comprised the same steps as Experiment A; however, several steps of a common mAb purication process and a nal crystallization step for product formulation were added. The mAb purication process steps were: virus inactivation at low pH, AEC, and nanoltration. In previous experiments on the 6 mL-scale, it was observed that the process was accelerated by addition of small amounts of PEG 10000 (data not shown). This

conrmed observations reported in literature (Crisman and Randolph, 2010; Jion et al., 2006; Weber et al., 2008). Therefore, 2% w/v PEG 10000 was added to the pretreated harvests to accelerate the rst crystallization steps. The mAb01 start concentrations were 2.2 g L1 for Experiment A and 4.0 g L1 for Experiment B. First crystals were visible already after 1 h in Experiment A and after 1.5 h in Experiment B. The crystals were separated by centrifugation after crystallization overnight at 10 C. Figure 7 shows a microphotograph of crystals from Experiment A. Only 0.3 and 0.4 g L1 of the antibody remained in solution for Experiments A and B, respectively. The rst crystallization step of Experiments A and B had a yield of 88% and 90%, respectively. Thus, a reproducible crystallization of mAb01 from different batches of pretreated harvest was possible in the 1 L-scale. A very high purity by SEC of 98.5% and 97.3% was obtained after recrystallization for Experiments A and B,

Table III. Purication on the 1 L-scale starting from harvest. Purity (SEC), % Step Cell-free harvest Pretreatment Crystallization Recrystallization Virus inactivation Anion exchange chromatography Nanofiltration Final crystallization Experiment A ND ND 92.9 98.5 Experiment B ND ND 96.9 97.3 97.7 99.0 ND ND Experiment A 81,752 ND 11,259 4,733 HCP, ppm Experiment B 266,719 222,830 39,070 17,354 13,864 1,506 1,289 88

Experiment A was stopped after recrystallization. ND, not determined.

2458

Biotechnology and Bioengineering, Vol. 110, No. 9, September, 2013

Hypothesis on Why mAb01 Crystallizes Easily MAb01, which is a human IgG1 with k light chain, was found to be easily crystallizable while many other human IgG1 with k light chains could not be crystallized. Since all human IgG1 with k light chains have highly identical constant-domain sequences, Fv domains of mAb01 along with ve other human IgG1 with k light chains mAbs (mAb02, mAb03, mAb04, mAb05, and mAb06) were studied using the SIM tool and the electrostatic potential map to identify distinctive physico-chemical features on the Fv domain of mAb01 that makes it easily crystallizable and the other ve mAbs not. In general, crystallizing a full-length IgG antibody has proved to be an exceedingly difcult task and hence understanding of distinctive physico-chemical features that makes an antibody highly prone to crystallization is limited. For a particular IgG antibody, which displayed high crystallization propensity, Hasegawa et al. (2011) proposed that a cluster composed of ve exposed negatively charged residues on the CDR played a critical role in facilitating its self-assembly leading to high crystallization propensity. Indeed, IgG variants with simultaneous mutations of these ve residues to Alanine or Serine formed no observable crystals. As seen in Figure 8, the Fv region of mAb01 has a large patch of negative electrostatic potential while the Fv regions of mAb02, mAb03, mAb04, mAb05, and mAb06 have smaller or negligible patches of negative electrostatic potential. A negative electrostatic potential indicates a predominant presence of exposed negatively charged residues and a general lack of exposed positively charged residues. Furthermore, the electrostatic negative patch on mAb01 is distinctly interspersed with few exposed hydrophobic residues. These observations on six IgG1 mAbs along with the previous work of Hasegawa et al. (2011) suggest that a presence of a large negative electrostatic patch along with neighboring exposed hydrophobic residues in the Fv domain can serve as a distinctive feature to identify IgG1 with potentially high crystallization propensity. We hypothesize that among other interactions, weak interaction of the Fv domain of an antibody with its Fc domain can restrict the mobility of Fab and Fc fragments around the hinge domain, a feature that can likely lead to the mAbs orderly crystallization. In fact, in the crystal structures of a human IgG1 (PDB ID: 1HZH, Saphire et al., 2001) and a murine IgG2a (PDB ID: 1IGT, Harris et al., 1997) mAb, Fv domain and Fc domain share an extensive interface (see Fig. S1). Consequently, for mAb01, we postulate that the negative electrostatic patch on its Fv domain can potentially interact with the positive electrostatic patch on its Fc domain (see Fig. S2B). This interaction is further strengthened by the interaction of exposed hydrophobic residues around these patches on the Fv and Fc domains. This suggested interaction of Fv domain of mAb01 with its Fc domain is likely responsible for its high crystallization propensity. In the future, this proposed feature responsible for high crystallization propensity of mAb01 can be validated by observing the crystallization propensity of

Figure 7. Stirred batch crystallization of mAb01 from pretreated harvest on the 1 L-scale (Experiment A). 2.2 g L1 mAb01 in pretreated harvest; 2% w/v PEG 10000 added; TRIS base added to adjust the pH to 6.8. Microphotograph was taken after crystallization overnight at 10 C. Robust rod-like crystals (the dark spots represent crystal agglomerates containing native antibody).

respectively (Table III). The purity of the crystallization reactions was similar to afnity purication by Protein A chromatography. The HCP content after recrystallization was reduced 17-fold for Experiment A and 15-fold for Experiment B. The HCP content was further reduced by AEC. An additional HCP removal step like hydrophobic interaction chromatography or CEC would be required to deliver robust removal of HCP prior to the nanoltration step. The chromatography steps could also be replaced by ltration steps with the respective functionalities like quaternary amine, sulfonic acid, or phenyl (Knudsen et al., 2001; Kuczewski et al., 2010; Liu et al., 2010). In Experiment B, the DNA content could be reduced after recrystallization by a factor of more than 39,000. The DNA content of the cell culture harvest of about 78 ppm was reduced below the detection limit of 2 ppb already after the recrystallization step. The process was highly efcient for removing DNA. The nal crystallization step was performed as a drug product formulation step. For this purpose, the crystallization buffer of 10 mM histidine/acetate was kept but sodium hydroxide was used instead of TRIS base to adjust the pH to 6.8. In addition, 250 mM trehalose was added for osmolality adjustment before crystallization in the 100 mL-scale at 10 C. The starting concentration was 11.6 g L1 mAb01. Only 0.79 g L1 mAb01 remained in solution after crystallization for 4 h. The yield of this step was 93%. It was shown that crystallization was compatible with subsequent purication process steps like virus inactivation at low pH, AEC, and nanoltration. Crystallization could successfully replace Protein A chromatography in a standard purication process. Furthermore, crystallization could be used for mAb01 formulation to enable the preparation of samples for preclinical pharmacokinetics studies.

Smejkal et al.: Purification of an Antibody by Crystallization Biotechnology and Bioengineering

2459 2459

Figure 8. For each IgG1 antibody, the electrostatic potential is shown on the left side while the SIM is shown on the right side of the panel. Each Fab domain is oriented such that the CDR is oriented towards the viewer and the constant region is oriented away from the viewer. Exposed hydrophobic residues identied by the SIM are shown in yellow. Electrostatic potential is shown in units of kT/e, where k is the Boltzmann constant, T is temperature in Kelvin and e is the charge of an electron.

mAb01 variants lacking one or all of the following features: (1) negative electrostatic patch on the Fv domain; (2) exposed hydrophobic residues in neighborhood of this negative electrostatic patch; (3) positive electrostatic patch on the Fc domain; (4) exposed hydrophobic residues in neighborhood of this positive electrostatic patch.

Conclusions
The therapeutic full-length antibody mAb01 was crystallized in non-agitated micro-batch experiments for the rst time. Subsequently, the crystallization process was successfully scaled-up to the stirred L-scale and shown to be reproducible. Crystallization from a pretreated CHO cell culture harvest was possible. It has been shown that process crystallization in stirred tanks can be an efcient, fast, and inexpensive alternative to several chromatography and UF/DF steps throughout the purication process of a therapeutic fulllength antibody. Compared to chromatography, crystallization shows no mass-transfer limitation and a scale-up is easily possible using simple standard equipment. Use of process crystallization for purication has the potential to allow

signicant cost reductions of the manufacturing process of therapeutic antibodies. A completely chromatography-free GMP purication process may be feasible if crystallization is combined with other alternative purication methods like functionalized ltration membranes, precipitation, or aqueous two-phase extraction. In future, it is conceivable that highly crystallizable antibody candidates can be identied in a high-throughput fashion using molecular modeling tools. Hence, new knowledge can be acquired with regard to engineering a crystallizable antibody. Then, process crystallization can be part of a platform purication process.
The authors are deeply indebted to Prof. Dirk Weuster-Botz, Head of the Institute of Biochemical Engineering, TUM, Germany for the excellent support of this work and for the opportunity of using the outstanding infrastructure at the Institute of Biochemical Engineering for the part of this work performed at TUM. The authors thank Dirk Hebel and Norbert Werth (both Institute of Biochemical Engineering, TUM, Germany) for support during design and production of the stirred tanks. This work was funded by Novartis Pharma AG, Basel, Switzerland. The additional support of Benjamin Smejkal by the TUM Graduate School, Munich, Germany is gratefully acknowledged. Neeraj J. Agrawal and Bernhardt L. Trout acknowledge funding support from the Singapore-MIT Alliance.

2460

Biotechnology and Bioengineering, Vol. 110, No. 9, September, 2013

References
Agrawal NJ, Helk B, Trout BL. 2012. A computational tool to predict the protein-protein interaction hot-spot residues from the structure of the unbound protein. Manuscript submitted for publication. Alderton G, Fevold HL. 1946. Direct crystallization of lysozyme from egg white and some crystalline salts of lysozyme. J Biol Chem 164:15. Andya JD, Hsu CC, Shire SJ. 2003. Mechanisms of aggregate formation and carbohydrate excipient stabilization of lyophilized humanized monoclonal antibody formulations. AAPS PharmSci 5:2131. Baker NA, Sept D, Joseph S, Holst MJ, McCammon JA. 2001. Electrostatics of nanosystems: Application to microtubules and the ribosome. Proc Natl Acad Sci USA 98:1003710041. Crisman RL, Randolph TW. 2010. Crystallization of recombinant human growth hormone at elevated pressures: Pressure effects on PEG-induced volume exclusion interactions. Biotechnol Bioeng 107:663672. Dolinsky TJ, Czodrowski P, Li H, Nielsen JE, Jensen JH, Klebe G, Baker NA. 2007. PDB2PQR: Expanding and upgrading automated preparation of biomolecular structures for molecular simulations. Nucleic Acids Res 35: W522W525. Fahrner RL, Knudsen HL, Basey CD, Galan W, Feuerhelm D, Vanderlaan M, Blank GS. 2001. Industrial purication of pharmaceutical antibodies: Development, operation, and validation of chromatography processes. Biotechnol Genet Eng Rev 18:301327. Garcia-Ruiz JM. 2003. Nucleation of protein crystals. J Struct Biol 142:2231. Giffard M, Ferte N, Ragot F, El Hajji M, Castro B, Bonnete F. 2011. Urate oxidase purication by salting-in crystallization: Towards an alternative to chromatography. PLoS ONE 6:e19013. Gottschalk U. 2008. Bioseparation in antibody manufacturing: The good, the bad and the ugly. Biotechnol Prog 24:496503. Granick S, Michaelis L. 1942. Ferritin and apoferritin. Science 95:439440. Harris LJ, Skaletsky E, McPherson A. 1995. Crystallization of intact monoclonal antibodies. Proteins 23:285289. Harris LJ, Larson SB, Hasel KW, McPherson A. 1997. Rened structure of an intact IgG2a monoclonal antibody. Biochemistry 36:15811597. Harrison RG, Todd P, Rudge SR, Petrides DP. 2003. Bioseparations science and engineering. New York: Oxford University Press. 406 p. Hasegawa H, Wendling J, He F, Trilisky E, Stevenson R, Franey H, Kinderman F, Li G, Piedmonte DM, Osslund T, Shen M, Ketchem RR. 2011. In vivo crystallization of human IgG in the endoplasmic reticulum of engineered Chinese hamster ovary (CHO) cells. J Biol Chem 286:1991719931. Haynes SR, Jelinek WR. 1981. Low molecular weight RNAs transcribed in vitro by RNA polymerase III from Alu-type dispersed repeats in Chinese hamster DNA are also found in vivo. Proc Natl Acad Sci USA 78:6130 6134. Hekmat D, Hebel D, Schmid H, Weuster-Botz D. 2007. Crystallization of lysozyme: From vapor diffusion experiments to batch crystallization in agitated ml-scale vessels. Process Biochem 42:16491654. Hopkins FG. 1900. On the separation of a pure albumin from egg-white. J Physiol 25:306330. Jacobsen C, Garside J, Hoare M. 1998. Nucleation and growth of microbial lipase crystals from claried concentrated fermentation broths. Biotechnol Bioeng 57:666675. Jion AI, Goh L-T, Oh SKW. 2006. Crystallization of IgG1 by mapping its liquidliquid phase separation curves. Biotechnol Bioeng 95:911918. Judge RA, Johns MR, White ET. 1995. Protein purication by bulk crystallization: The recovery of ovalbumin. Biotechnol Bioeng 48:316 323. Judge RA, Forsythe EL, Pusey ML. 1998. The effect of protein impurities on lysozyme crystal growth. Biotechnol Bioeng 59:776785. Kenney AC, Chase HA. 1987. Automated production scale afnity purication of monoclonal antibodies. J Chem Technol Biotechnol 39:173182. Knudsen HL, Fahrner RL, Xu Y, Norling LA, Blank GS. 2001. Membrane ion-exchange chromatography for process-scale antibody purication. J Chromatogr A 907:145154.

Kuczewski M, Fraud N, Faber R, Zarbis-Papastoitsis G. 2010. Development of a polishing step using a hydrophobic interaction membrane adsorber with a PER.C6-derived recombinant antibody. Biotechnol Bioeng 105:296305. Larson S, Day J, Greenwood A, Skaletsky E, McPherson A. 1991. Characterization of crystals of an intact monoclonal antibody for canine lymphoma. J Mol Biol 222:1719. Lauer TM, Agrawal NJ, Chennamsetty N, Egodage K, Helk B, Trout BL. 2012. Developability index: A rapid in silico tool for the screening of antibody aggregation propensity. J Pharm Sci 101:102115. Lewus RA, Darcy PA, Lenhoff AM, Sandler SI. 2011. Interactions and phase behavior of a monoclonal antibody. Biotechnol Prog 27:280289. Liu HF, Ma J, Winter C, Bayer R. 2010. Recovery and purication process development for monoclonal antibody production. MAbs 2: 480499. Low D, OLeary R, Pujar NS. 2007. Future of antibody purication. J Chromatogr B Analyt Technol Biomed Life Sci 848:4863. MacKerell AD, Bashford D, Bellott Dunbrack RL, Evanseck JD, Field MJ, Fischer S, Gao J, Guo H, Ha S, Joseph-McCarthy D, Kuchnir L, Kuczera K, Lau FTK, Mattos C, Michnick S, Ngo T, Nguyen DT, Prodhom B, Reiher WE, Roux B, Schlenkrich M, Smith JC, Stote R, Straub J, Watanabe M, Wirkiewicz-Kuczera J, Yin D, Karplus M. 1998. All-atom empirical potential for molecular modeling and dynamics studies of proteins. J Phys Chem B 102:35863616. Margolin AL, Navia MA. 2001. Protein crystals as novel catalytic materials. Angew Chem Int Ed Engl 40:22042222. Przybycien TM. 1998. Proteinprotein interactions as a means of purication. Curr Opin Biotechnol 9:164170. Rey G, Wendeler MW. 2012. Full automation and validation of a exible ELISA platform for host cell protein and protein A impurity detection in biopharmaceuticals. J Pharm Biomed Anal 70:580586. Saphire EO, Parren PW, Pantophlet R, Zwick MB, Morris GM, Rudd PM, Dwek RA, Staneld RL, Burton DR, Wilson IA. 2001. Crystal structure of a neutralizing human IgG against HIV-1: A template for vaccine design. Science 293:11551159. Shenoy B, Wang Y, Shan W, Margolin AL. 2001. Stability of crystalline proteins. Biotechnol Bioeng 73:358369. Shukla AA, Hubbard B, Tressel T, Guhan S, Low D. 2007. Downstream processing of monoclonal antibodiesApplication of platform approaches. J Chromatogr B Analyt Technol Biomed Life Sci 848: 2839. Smejkal B, Helk B, Rondeau JM, Anton S, Wilke A, Scheyerer P, Fries J, Hekmat D, Weuster-Botz D. 2013. Protein crystallization in stirred systemsScale-up via the maximum local energy dissipation. Biotechnol Bioeng. doi: 10.1002/bit.24845. Sumner JB. 1926. The isolation and crystallization of the enzyme urease: Preliminary paper. J Biol Chem 69:435441. Takakura T, Ito T, Yagi S, Notsu Y, Itakura T, Nakamura T, Inagaki K, Esaki N, Hoffman RM, Takimoto A. 2006. High-level expression and bulk crystallization of recombinant L-methionine gamma-lyase, an anticancer agent. Appl Microbiol Biotechnol 70:183192. Trilisky E, Gillespie R, Osslund TD, Vunnum S. 2011. Crystallization and liquidliquid phase separation of monoclonal antibodies and Fc-fusion proteins: Screening results. Biotechnol Prog 27:10541067. Weber M, Jones MJ, Ulrich J. 2008. Optimisation of isolation and purication of the jack bean enzyme urease by extraction and subsequent crystallization. Food Bioprod Process 86:4352. Wurm FM. 2004. Production of recombinant protein therapeutics in cultivated mammalian cells. Nat Biotechnol 22:13931398. Yang MX, Shenoy B, Disttler M, Patel R, McGrath M, Pechenov S, Margolin AL. 2003. Crystalline monoclonal antibodies for subcutaneous delivery. Proc Natl Acad Sci USA 100:69346939. Zang Y, Kammerer B, Eisenkolb M, Lohr K, Kiefer H. 2011. Towards protein crystallization as a process step in downstream processing of therapeutic antibodies: Screening and optimization at microbatch scale. PLoS ONE 6:e25282.

Smejkal et al.: Purification of an Antibody by Crystallization Biotechnology and Bioengineering

2461 2461