ahamed_ (13.2 MB) - TU Delft Institutional Repository

High-Throughput Technologies

for

Bioseparation Process Development

PROEFSCHRIFT

ter verkrijging van de graad van doctor aan de Technische Universiteit Delft,

op gezag van de Rector Magnificus prof. dr. ir. J.T. Fokkema voorzitter van het Hoofdleften College voor Promoties,

in het openbaar te verdedigen

op donderdag 09 oktober 2008 om 15:00 uur

door

Tangir Ahamed Technologisch Ontwerper in Bioproces Technologie

Geboren te Jessore, Bangladesh

Dit proefschrift is goedgekeurd door de promotor:

Prof. dr. ir. L.A.M. van der Wielen

Copromotor Dr. ir. M. Ottens

Samenstelling promotiecommissie:

Rector Magnificus Voorzitter Prof. dr. ir. L.A.M. van der Wielen Delft University of Technology, promotor Dr. ir. M. Ottens Delft University of Technology, copromotor Prof. dr. P.D.E.M. Verhaert Delft University of Technology Prof. dr. ir. R.M. Boom Wageningen University and Research Centre Prof. dr. J. Hubbuch University of Karlsruhe, Germany Prof. dr. A.M. Lenhoff University of Delaware, USA Dr. ir. M.H.M. Eppink Schering-Plough Corporation, adviseur

The research described in this thesis was performed at the Department of Biotechnology,

Delft University of Technology, The Netherlands.

This project is financially supported by the Netherlands Ministry of Economic Affairs and the

B-Basic partner organizations (www.b-basic.nl) through B-Basic, a public private NWO-

ACTS program (ACTS = Advanced Chemical Technologies for Sustainability).

ISBN 978-90-9023391-8

ii

Contents

Nomenclature iv Chapter 1 1

Introduction Chapter 2 11

A generalized approach to thermodynamic properties of biomolecules for use in bioseparation process design

Chapter 3 37

Design of self-interaction chromatography as an analytical tool of predicting protein phase behavior

Chapter 4 63

Phase behavior of an intact monoclonal antibody

Chapter 5 81pH-gradient ion-exchange chromatography: An analytical tool for design and optimization of protein separations

Chapter 6 97

Optimization of pH-related parameters in ion-exchange chromatography using pH-gradient operations

Chapter 7 111

Towards establishment of database for bioseparation process development parameters

Chapter 8 129

Outlook

Summary 132 Samenvatting 135 Curriculum Vitae 138 Publications 139 Acknowledgements 140

iii

iv

Nomenclature 2D Two-dimensional AC Affinity chromatography AEX Anion-exchange chromatography ATPS Aqueous-two phase separation BCA Bicinchoninic acid BHS Hard sphere/excluded volume contribution to Bmm Bicine N,N-Bis(2-hydroxyethyl)glycine Bmm Osmotic second virial coefficient BSA Bovine serum albumin CA Carbonic anhydrase CCS Cell culture supernatant CEX Cation-exchange chromatography CHES 2-(Cyclohexylamino)ethanesulfonic acid FF Fast flow GE Gel electrophoresis HIC Hydrophobic-interaction chromatography HTE High-throughput experimentation HTS High-throughput screening IEX Ion-exchange chromatography IEF Isoelectric focusing LALLS Low-angle laser-light scattering MAb Monoclonal antibody MINLP Mixed integer nonlinear programming MO Membrane osmometry MS Mass spectrometry NHS N-hydroxysuccinimide NMWCO Nominal molecular weight cut-off PAGE Polyacrylamide gel electrophoresis PAN Primary access number in UniProtKB/Swiss-Prot database PEG Polyethylene glycol RPC Reversed-phase chromatography SANS Small-angle neutron scattering SAXS Small-angle X-ray scattering SELDI Surface-enhanced laser desorption/ionization SEC Size-exclusion chromatography SIC Self-interaction chromatography SLS Static light scattering SMB Simulated moving bed

Chapter 1

Commercially implemented techniques for purification of biological macromolecules include ion-exchange chromatography (IEX), hydrophobic-interaction chromatography (HIC), affinity chromatography (AC), size-exclusion chromatography (SEC), precipitation, and crystallization [13]. Successful modeling of chromatography operations requires critical stationary phase properties of the column and isotherm parameters of all major components present in the crude mixture (Table 1.1). The stationary phase parameters are actually constant for a particular column and are usually provided by the column supplier. On the other hand, different chromatography operations require concentration, diffusivity, and respective class of isotherm parameters of the product and impurity components (Table 1.1). For example, the steric mass action model of IEX requires effective charge, equilibrium constant, and steric factor as isotherm parameters [45]. Both HIC and AC can be modeled through a competitive equilibrium isotherm framework [46], which requires hydrophobicity or affinity and the maximum binding capacity as input parameters. The hydrophobicity in this approach can be described as an elution-salt concentration from a HIC column, because only surface hydrophobic groups of a molecule take part in HIC. Modeling of SEC is relatively straightforward using molecular size and gel fiber diameter [47]. Table 1.1. Critical bioseparation process modeling parameters.

Separation method Critical process modeling parameters

Col

umn

para

met

ers

All chromatography Column length, column diameter, particle diameter, pore diameter, bed porosity, particle porosity, flow rate, and maximum pressure drop limit

IEX Ionic capacity SEC Diameter of gel fiber

Isot

herm

pa

ram

eter

s (o

f all

maj

or

com

pone

nts)

All chromatography Concentration and diffusivity IEX Effective charge, equilibrium constant, and steric factor HIC Hydrophobicity and maximum binding capacity AC Affinity constant and maximum binding capacity SEC Molecular size

Crystallization Solubility and osmotic second virial coefficient Precipitation Solubility In contrast to various chromatography systems, the prediction of a precipitation process is rather simple when solubility is available. However, successful prediction of crystallization requires solubility and optimized process conditions, which can be translated into osmotic second virial coefficient [48]. Estimation of all isotherm parameters listed in Table 1.1 requires experimentation, although there is a possibility of interrelating different parameters through simple equations. For example, diffusivity of a component can be estimated from the correlation with molecular radius or molecular mass [49].

6

Introduction

1.3.3. Objective of the research

The objective of this study is to develop experimentation techniques for fast acquisition of critical bioseparation process development parameters (Table 1.1) from complex biological mixtures. Since the properties of auxiliary materials or phases are easily obtainable from the supplier, this study is focused on developing HTE tools for process condition optimization, and estimation of chromatographic isotherm parameters and phase separation properties. 1.4. High-throughput technologies

The number of critical parameters, to be determined experimentally, is first minimized in Chapter 2 by exploiting thermodynamic interrelation among different parameters. A set of novel HTE technologies is then developed for fractionation and characterization of crude biological mixtures and to acquire critical bioseparation process development parameters. The technologies developed to achieve this ambitious goal include a miniaturized self-interaction chromatography (SIC) technique, a novel pH-gradient IEX, and a multi-dimensional fractionation and characterization system (Fig. 1.2). The number of critical parameters obtained in this framework and their accuracy are good enough to implement in state-of-the-art models of different bioseparation methods.

Fig. 1.2. High-throughput technologies for estimating critical process development parameters. As one of the high-throughput technologies, SIC is designed for rapid measurement of osmotic second virial coefficient (Bmm) of proteins. In Chapter 3, a miniaturized SIC technology is designed in details with respect to theoretical framework, experimental methodology, data analysis, troubleshooting, and application in different kinds of proteins. The approach requires retention data from an immobilized protein column as well as from a protein free column for determination of a Bmm value. The novelty in this approach, unlike other SIC techniques, is the freedom to use non-identically packed columns in terms of column volume and packing integrity. The developed methodology is then successfully applied in Chapter 4 to predict the phase behavior of an intact monoclonal antibody.

The next developed high-throughput technology is a novel pH-gradient IEX technique to rationally select the optimum IEX operating conditions, thereby eliminating need for

7

Chapter 1

traditional empirical screening. In Chapter 5, the pH-gradient IEX technology explains the previously non-understood behavior of neutral proteins in IEX and dismisses the use of pI as the key parameter for selection of optimum IEX operational conditions, in contrast to traditional assumptions or rules-of-thumb. The pH-gradient IEX is eventually applied in Chapter 6 for the optimization of IEX operational parameters in a case study on the capture of a monoclonal antibody. This fast and rational approach of optimizing pH-related parameters in an IEX, can be identified as a key HTE technique for replacing presently existing pH-scouting and robotic screening.

The final high-throughput technology developed is a multi-dimensional fractionation and characterization system for fast acquisition of chromatographic isotherm parameters from crude biological mixtures. Chapter 7 shows that IEX parameters (effective charge, equilibrium constant, and steric factor) of the product and all major impurities can be obtained by model fitting from two salt-gradient IEX runs at the optimized pH. The collected IEX fractions are further analyzed by HIC, gel electrophoresis (GE), and other techniques (if required) in parallel to estimate HIC parameters, size, and other parameters (such as affinity), respectively. Finally, the GE bands are analyzed by mass spectrometry to reveal the identity of the impurities. This multi-dimensional fractionation and characterization system is applied to acquire bioseparation process development data from an Escherichia coli lysate and a hybridoma cell culture supernatant. All the parameters required for modeling and rational process synthesis can be obtained within an experimental period of maximum two weeks.

The data obtained through this fractionation and characterization system are mostly not case specific and readily usable for other products produced from the same host organism. For example, the profile of impurities and their properties are the same for any protein overexpressed in the same E. coli strain. The same E. coli lysate impurities database can eventually be used for developing separation process of any protein produced in E. coli. This work, thus, emphasizes the need and importance for a ‘database for bioseparation processes’ for common host cells. The foundation of such a database, which does not currently exist, is provided through this research. References [1] A.K. Pavlou & J.M. Reichert. Nat. Biotechnol. 22 (2004) 1513-1519. [2] D.R. Headon & G. Walsh. Biotechnol. Adv. 12 (1994) 635-646. [3] G. Walsh. Nat. Biotechnol. 18 (2000) 831-833. [4] C. Smith. Nat. Methods 2 (2005) 71-77. [5] K.A. Thiel. Nat. Biotechnol. 22 (2004) 1365-1372. [6] G. Sofer & L. Hagel. In G. Sofer & L. Hagel (Eds.) Handbook of Process Chromatography: A

Guide to Optimization, Scale-up, and Validation, Academic Press, London, 1997, p. 27-113. [7] E.N. Lightfoot & J.S. Moscariello. Biotechnol. Bioeng. 87 (2004) 259-273. [8] FDA. Innovation or stagnation: Challenge and opportunity on the critical path to new medical

products, Food and Drug Administration, USA, 2004. [9] G. Hodge. BioProcessing J. March/April (2004) 31-35. [10] S.D. Barnicki & J.J. Siirola. Comput. Chem. Eng. 28 (2004) 441-446.

8

Introduction

[11] N. Nishida, G. Stephanopoulos & A.W. Westerberg. AIChE J. 27 (1981) 321-351. [12] A.W. Westerberg. Comput. Chem. Eng. 28 (2004) 447-458. [13] B.K. Nfor, T. Ahamed, G.W.K. van Dedem, L.A.M. van der Wielen, E.J.A.X. van de Sandt,

M.H.M. Eppink & M. Ottens. J. Chem. Technol. Biotechnol. 83 (2008) 124-132. [14] J.A. Asenjo & I. Patrick. In E.L.V. Harris & S. Angal (Eds.) Protein purification applications a

practical approach, Oxford University Press, Oxford, 1990, p. 1-28. [15] J.A. Asenjo, L. Herrera & B. Byrne. J. Biotechnol. 11 (1989) 275-298. [16] D.P. Petrides, C.L. Cooney & L.B. Evans. In Chemical Engineering Problems in

Biotechnology, American Institute of Chemical Engineers, 1989. [17] N.J. Samsatli & N. Shah. Comput. Chem. Eng. 20 (1996) S315-S320. [18] Y.H. Zhou & N.J. Titchener-Hooker. Bioprocess Eng. 14 (1996) 263-268. [19] Y.H. Zhou, I.L.J. Holwill & N.J. Titchener-Hooker. Bioprocess Eng. 16 (1997) 367-374. [20] Y.H. Zhou & N.J. Titchener-Hooker. Biotechnol. Bioeng. 65 (1999) 550-557. [21] Y.H. Zhou & N.J. Titchener-Hooker. Bioprocess Biosyst. Eng. 25 (2003) 349-355. [22] Y.H. Zhou & N.J. Titchener-Hooker. J. Chem. Technol. Biotechnol. 74 (1999) 289-292. [23] J.A. Asenjo, J.M. Montagna, A.R. Vecchietti, O.A. Iribarren & J.M. Pinto. Comput. Chem.

Eng. 24 (2000) 2277-2290. [24] M.A. Steffens, E.S. Fraga & I.D.L. Bogle. Biotechnol. Bioeng. 68 (2000) 218-230. [25] E. Vasquez-Alvarez & J.M. Pinto. J. Biotechnol. 110 (2004) 295-311. [26] M.H.M. Eppink, R. Schreurs, A. Gijsen & K. Verhoeven. BioPharm Int. 20 (2007) 44-50. [27] J.L. Coffman, J.F. Kramarczyk & B.D. Kelley. Biotechnol. Bioeng. (2008) In press. [28] K. Rege, M. Pepsin, B. Falcon, L. Steele & M. Heng. Biotechnol. Bioeng. 93 (2006) 618-630. [29] M. Bensch, P. Schulze Wierling, E. von Lieres & J. Hubbuch. Chem. Eng. Technol. 28 (2005)

1274-1284. [30] J. Thiemann, J. Jankowski, J. Rykl, S. Kurzawski, T. Pohl, B. Wittmann-Liebold & H. Schluter.

J. Chromatogr. A 1043 (2004) 73-80. [31] H. Charlton, B. Galarza, K. Leriche & R. Jones. BioPharm Int. June, Supplements (2006) 20-

26, 42. [32] B. Kelley, M. Switzer, P. Bastek, J. Kramarczyk, K. Molnar, T. Yu & J. Coffman. Biotechnol.

Bioeng. (2008) In press. [33] J.F. Kramarczyk, B.D. Kelley & J.L. Coffman. Biotechnol. Bioeng. (2008) In press. [34] D.L. Wensel, B.D. Kelley & J.L. Coffman. Biotechnol. Bioeng. (2008) In press. [35] V. Brenac, V. Ravault, P. Santambien & E. Boschetti. J. Chromatogr. B 818 (2005) 61-66. [36] S.R. Weinberger, E. Boschetti, P. Santambien & V. Brenac. J. Chromatogr. B 782 (2002) 307-

316. [37] V. Brenac, N. Mouz, A. Schapman & V. Ravault. Protein Expr. Purif. 47 (2006) 533-541. [38] J. Shiloach, P. Santambien, L. Trinh, A. Schapman & E. Boschetti. J. Chromatogr. B 790

(2003) 327-336. [39] O. Kokpinar, D. Harkensee, C. Kasper, T. Scheper, R. Zeidler, O.-W. Reif & R. Ulber.

Biotechnol. Prog. 22 (2006) 1215-1219. [40] T. Londo, P. Lynch, T. Kehoe, M. Meys & N. Gordon. J. Chromatogr. A 798 (1998) 73-82. [41] G. Denton, A. Murray, M.R. Price & P.R. Levison. J. Chromatogr. A 908 (2001) 223-234. [42] P. Schulze Wierling, R. Bogumil, E. Knieps-Grunhagen & J. Hubbuch. Biotechnol. Bioeng. 98

(2007) 440-450. [43] FDA. Guidance for industry PAT - A framework for innovative pharmaceutical development,

manufacturing, and quality assurance, Food and Drug Administration, USA, 2004.

9

Chapter 1

10

[44] J. Hubbuch & M.-R. Kula. J. Non-Equilib. Thermodyn. 32 (2007) 99-127. [45] S.R. Gallant, A. Kundu & S.M. Cramer. J. Chromatogr. A 702 (1995) 125-142. [46] G. Guiochon, S.G. Shirazi & A.M. Katti. Fundamentals of Preparative and Nonlinear

Chromatography, Academic Press, New York, 1994. [47] D.A. Horneman, M. Wolbers, M. Zomerdijk, M. Ottens, J.T.F. Keurentjes & L.A.M. van der

Wielen. J. Chromatogr. B 807 (2004) 39-45. [48] T. Ahamed, B.N.A. Esteban, M. Ottens, G.W.K. van Dedem, L.A.M. van der Wielen, M.A.T.

Bisschops, A. Lee, C. Pham & J. Thommes. Biophys. J. 93 (2007) 610-619. [49] E.M. Renkin. J. Gen. Physiol. 38 (1954) 225-243.

2

A generalized approach to thermodynamic properties of biomolecules for use in

bioseparation process design

Abstract Bioseparation techniques exploit the differences of physicochemical or thermodynamic properties between the product and the contaminants. Rapid development of a downstream process, therefore, requires physicochemical and thermodynamic characterization of the components to be separated. In this paper, we investigate that a generalized thermodynamic interrelation exists among different parameters. For instance activity coefficient, osmotic virial coefficients and the solubility of macromolecule are interrelated among each other. Experimental determination of any one of these parameters can be translated across the boundaries of different separation techniques. A number of downstream separation processes, including size-exclusion chromatography, hydrophobic-interaction chromatography, reversed-phase chromatography, aqueous-two phase separation, crystallization and precipitation, are found to be explained and designed using this generalized thermodynamics. This generalization of thermodynamic properties together with high-throughput experimentation provides a systematic and high-speed approach to bioseparation process development and optimization. The applicability of this approach for the bioseparation process design was investigated by a case study on nystatin, a medium-sized biomolecules. The distribution coefficients of nystatin in reversed-phase chromatography showed straightforward relationship with the solubilities at various solvent compositions and the experimental data supported the trend of the relationship.

Keywords: Activity coefficient; Osmotic virial coefficient; Bioseparation; Biomolecule; Bioprocess Design ___________________________________________________________________________ Published as T. Ahamed, M. Ottens, B.K. Nfor, G.W.K. van Dedem & L.A.M. van der Wielen. Fluid Phase Equilib. 241 (2006): 268-282.

Chapter 2

2.1. Introduction Purification processes of biomolecules are never straightforward and usually composed of a sequence of operations. Recent advances in various separation methods provide the possibility of generating a number of process options that may satisfy the process and end product constraints (i.e., yield, purity, activity, sterility). A quantitative comparison of the process options is required in order to facilitate the selection of the best process alternative.

Separation methods, in principle, exploit the differences in physicochemical, thermodynamic or molecular properties between the target compound and its contaminants. Therefore, the first step in developing a downstream process is to acquire sufficient knowledge on the properties of the target compound and its contaminants. These properties are then used as input data for designing separation methods and generating process options in a rational manner [1]. In order to quantify molecular or thermodynamic properties and operating condition, process designers usually use intuitive qualitative concepts, based on substantial experience. Generalizing quantitative principles in the determination of thermodynamic properties can assist in translating this valuable knowledge into quantitative tools, thereby making process design systematic and high-speed.

Physicochemical properties of biomolecules important in chromatographic separation processes are molecular size, charge, pI, hydrophobicity and affinity [2]. In addition to these constant physicochemical properties, interactions of biomolecules among themselves and with the environment are the key parameters in partitioning of biomolecules in different phases. For instance, the osmotic second virial coefficient (Bmm) is a thermodynamic property of dilute protein solution, which characterizes pair-wise protein self-interactions including contributions from excluded volume, electrostatic interactions and short-range interactions [3]. This Bmm has been used to model and/or thermodynamically explain a number of separation techniques, including crystallization [4], precipitation [5], aqueous two-phase separation (ATPS) [6], folding/refolding and aggregation [7]. On the other hand, some separation methods are complex to model and require many physicochemical and thermodynamic parameters as input. For example, detailed modeling of partitioning behavior in ATPS process fundamentally requires knowledge of molecular size, charge, hydrophobicity and Bmm [8]. The question arises whether any general relationship does exist among these properties. A previous study was done in this regard and a general relationship was established to correlate limiting thermodynamic properties, which are in use in a wide variety of separation processes [9]. The parameters in this general methodology can be obtained from a limited number of experiments, translated across the boundaries of different separation techniques and can be predicted from data that are commonly found in the characteristics of the final product. However, the study was restricted to small sized peptides, for which conformational changes may not be significant. Larger molecules exhibit separation behavior that is strongly affected by conformational changes that could also be induced by the environment. This complicates any generalized approach for biological

12

A generalized approach to thermodynamic properties of biomolecules

macromolecules. Nevertheless, the general thermodynamic framework may, in principle, be applied to macromolecules.

A generalization of thermodynamic parameters was shown previously in a number of cases, although not for the purpose of process design. For instance, the solubility of proteins may be interrelated to Bmm [10-12]. In this chapter, the physicochemical and thermodynamic parameters of biomolecules and its aqueous solution, needed for separation process design, and the existing high-throughput experimentation (HTE) techniques of obtaining the properties are reviewed. The second part of this chapter shows the interrelationship among different thermodynamic properties and how a limited number of variables can be used to model and predict almost all of the separation methods applied in biotechnology industries today. In order to realize the practical use of the thermodynamic generalization approach for bioseparation process design, a case study on the relationship between solubility and chromatographic distribution coefficient of an intermediately large biomolecule, nystatin, is shown at the last part of this chapter. 2.2. Properties of biomolecules needed in purification process design Selection and design of downstream processing operations for biomolecules have been virtually impossible in a systematic manner due to a lack of fundamental knowledge on the thermodynamic properties of the components to be separated. In order to approach the issue more systematically, it is important to summarize first the separation techniques usually applied in the biotechnology and biopharmaceutical industries and the input thermodynamic data required for designing each of these methods.

Industrial scale downstream processes usually consist of recovery steps followed by purification steps; the recovery steps are rather straightforward and impose fewer complications in design and selection [13]. On the other hand, a wide range of techniques is available for high-resolution purification and many combinations of them may achieve the desired level of purity [14]. We, therefore, focus only on the high-resolution separation techniques and list out the important parameters governing each of these techniques. 2.2.1. Chromatography Chromatography is undoubtedly ubiquitous for separation of biomolecules because of its industrial maturity and very high-resolution [15]. A number of chromatography methods are being used nowadays, for example ion exchange, size-exclusion, hydrophobic-interaction, reversed-phase, and affinity chromatography, all of which differ from each other in their separation principles. Regardless of the type of chromatographic operation, the partitioning behavior of the product and contaminant molecules, in terms of distribution coefficients, is the parameter required for selecting and designing the chromatographic separation processes. Asenjo and co-workers listed out the physicochemical parameters governing different chromatographic methods [2,13,16-21]. In order to identify each of the parameters distinguishably, we elaborate this issue to some extent.

13

Chapter 2

Separation in ion exchange chromatography occurs due to charge differences between product and contaminant molecules. Since the net charge of a biomolecule varies with the pH of the solution, the elution profile depends solely on operational pH and ionic strength. For a given operational condition, charge density, rather than the net charge or surface charge, is the parameter affecting protein-partitioning behavior in ion exchange chromatography [20]. Charge density, defined as net charge divided by molecular weight, as a function of pH can therefore be used to design ion-exchange chromatography processes.

The separation principle in size-exclusion chromatography (SEC) is simply based on molecular size. However, concentration dependence of retention appears to be a non-negligible parameter in SEC when chromatographic operation is done at high concentration of biomolecules [22-24]. Molecular size is usually approximated simply from the molecular weight.

Two other chromatography techniques, hydrophobic-interaction and reversed-phase, exploit the variable hydrophobic nature of biomolecules. Hydrophobic-interaction chromatography (HIC) is based on the reversible interaction between the hydrophobic patches on the biomolecules and the mildly hydrophobic stationary phase at high salt concentration [25]. Retentions of biomolecules in HIC systems largely depend on the environment, for instance type and concentration of salt [26,27], density and type of hydrophobic ligand in the stationary phase [28]. The type of matrix and salt effect do not, as a rule, alter the elution order of proteins despite the hydrophobic moieties in different matrices interact differently with proteins [29]. For a defined system, separation occurs due to differences in hydrophobicity of biomolecules. The term hydrophobicity covers average surface hydrophobicity as well as location and size of hydrophobic patches on the biomolecule surface, called surface hydrophobicity distribution [30]. The basic retention process in reversed-phase chromatography (RPC) is principally the same as in HIC. RPC matrices are generally more hydrophobic than HIC matrices. The hydrophobic-interaction in RPC is therefore so strong that the elution is accomplished by organic solvents rather than aqueous electrolyte solutions [31].

One more chromatography method, used largely in pharmaceutical protein purification, is based on the biospecific affinity between the ligand attached on the stationary phase and biomolecules in the liquid phase. Modeling of affinity chromatography requires specific information on the ligand-biomolecule binding affinity, in term of association/dissociation constant or adsorption isotherm. 2.2.2. Liquid-liquid extraction Liquid-liquid extraction in aqueous two-phase systems has been widely used for separation of proteins and removal of contaminants from fermentations as an initial separation. Factors and mechanisms that cause the distribution of biomolecules over the different phases are poorly understood. The value of the overall distribution coefficient depends on the molecular size [32], charge [33-36], hydrophobicity [37-39], solubility and

14


affinity [40]. However, not all these parameters are equally important as this depends on the chosen system.

Beyond ATPS, precipitated material is accumulated at the interface between an organic liquid phase and an aqueous solution of a salting out salt to form a third interphase. This three-phase partitioning method has been used recently for the purification of some proteins [41-43] and alginate [44]. The parameters important for designing three-phase partitioning system are essentially the same as for ATPS. 2.2.3. Crystallization and precipitation Most biological macromolecules at ambient conditions are solids when pure. Solid-liquid phase separation occurs, as a consequence, when the biomolecule concentration is high enough to exceed the solubility limit. Formation of solids may occur in the form of either crystals or precipitates, depending on the Bmm value of the biomolecule at that solution conditions chosen [4]. Therefore, Bmm and solubility are two important thermodynamic parameters governing crystallization/precipitation processes.

Commonly used high-resolution purification techniques and their respective physicochemical or thermodynamic parameters are listed in Table 2.1. In addition to these properties, the stability of biomolecules is also of the utmost importance and may dictate conditions to be used and the viability of given separation steps. Design of a bioseparation process requires the availability of these physicochemical and thermodynamic properties. Table 2.1. Most commonly used purification methods and respective properties of biomolecules on which design of these methods are based.

Separation method Parameter by principle Process design parameter Chromatography Partition coefficient

Ion-exchange Charge density as a function of pH Size-exclusion Molecular size Hydrophobic-interaction Surface hydrophobicity Reversed-phase Hydrophilic and hydrophobic

interactions Affinity Specific binding affinity

Liquid-liquid extraction Molecular charge, size and conformation, hydrophobicity, affinity, solubility, Bmm

Partition coefficient Aqueous two-phase separation

Triple-phase partitioning Precipitation/crystallization Solubility, Bmm Solubility, Bmm

Centrifugation Molecular density Sedimentation coefficient Membrane separation Molecular size Permeability 2.3. Models of generalization of thermodynamic properties

2.3.1. Activity coefficient and virial coefficients of aqueous solution of macromolecule Addition of a biological macromolecule to a solvent gives rise to thermodynamic non-ideality of the solution. The thermodynamic property of such a macromolecular solution is

15

Chapter 2

explained by the classical virial expansion of the osmotic pressure, π , in terms of molar concentration scale [45], namely ( )21m m mm m mmmRTc c B c Bπ = + + + (2.1)

where R is the universal gas constant, T is the temperature in Kelvin, cm is the concentration of macromolecule in molar scale, and Bmm, Bmmm, are the McMillan-Mayer’s osmotic virial coefficients [46]. For a binary solution of macromolecule and solvent, deviations from thermodynamic ideality are expressed conventionally, in terms of the chemical potential of the macromolecule in the liquid phase ( L

mμ ) as

lnL Lm m

LmRT aμ μ= ° + (2.2)

where μ° is the standard state chemical potential, is the thermodynamic activity of the

macromolecule in solution. The choice of the standard state of the macromolecule depends primarily on the pressure and the chemical potential of the solvent (

Lma

sμ ) as well as on the

choice of the concentration scale (mole fraction, molar or molal) of the macromolecule. According to the McMillan-Mayer solution theory [46], the thermodynamic properties of a multicomponent system are expressed as a power series of solute concentration in molarity scale. In order to comply with the McMillan-Mayer theory, the sμ must be held constant

even as the concentration of biomolecules changes or the expression of Lmμ must be converted

from a state at constant sμ to a state at constant pressure [47]. Under the usual laboratory

constraints of constant pressure and temperature, sμ does not remain constant with changes

of macromolecule concentration. Under these conditions, is most conveniently expressed

as the product of an activity coefficient,

Lma

Lmγ , and a concentration, , in molal (moles of

macromolecule per kilogram of solvent) dimension. Therefore mm

ln lnL L Lm m m mRT RTμ μ γ= ° + + m (2.3)

The chemical potential of the solvent, sμ , due to the addition of the macromolecule

can be expressed as ( )21s s s m mm m mmm mRTM m B m B mμ μ= ° − + + + (2.4)

where sμ° is the standard state chemical potential of the solvent, sM is the molecular mass

of solvent, and mmB , mmmB , are Hill’s osmotic virial coefficients, accounting for

interaction among macromolecules [48]. Differentiation of Eq. (2.4) yields

( 21 2 3ss mm m mm m

m

RTM B m B mm )μ∂

= − + + +∂

(2.5)

Using the appropriate form of the Gibbs-Dühem equation at constant pressure and temperature, it is straightforward to show that

1Lm

m s mm M m ms

m

μ μ∂= −

∂ ∂∂ (2.6)

Combining Eq. (2.5) and Eq. (2.6) gives

16


( )2/ 1 2 3Lm m mm m mmm m mm RT B m B m mμ∂ ∂ = + + + / (2.7)

Integration of Eq. (2.7) gives the relation

23ln 22

L Lm m m mm m mmm mRT m B m B mμ μ ⎛− ° = + + +⎜

⎝ ⎠⎞⎟ (2.8)

We have now two equations expressing non-ideal behavior, Eq. (2.3) in terms of an activity coefficient and Eq. (2.8) in terms of virial coefficients. Combining Eq. (2.3) and (2.8) yields a relationship between an activity coefficient of the macromolecule and the osmotic virial coefficients, for a particular solution condition at a particular macromolecule concentration, as

23ln 22

Lm mm m mmm mB m B mγ = + + (2.9)

Since the solvent chemical potential is not kept constant during the virial coefficient measurement experimentation, Hill’s virial coefficients are the ones often accessible experimentally [49]. There is also a simple and rigorous thermodynamic relationship between Hill’s virial coefficient, B , and McMillan-Mayer’s virial coefficient, B. If both solvent and macromolecule are assumed to be incompressible, which is typically the case for aqueous solution of macromolecules, the relationship is then [47,50] ( )0

mm s mm mB B 0ρ ν= − (2.10)

( ) ( ) 20 02mmm s mmm mm m mB B Bρ ν= − +20ν (2.11)

where 0sρ is the molar density of pure solvent (mol/ml) and 0

mν is the partial molar volume of

the macromolecule (ml/mol).

It has been shown in Eq. (2.9) to (2.11) that activity coefficient, McMillan-Mayer’s virial coefficients and Hill’s virial coefficients are interrelated to each other. Experimental determination of any one of them can directly be used to access the others. For the description of a dilute solution of macromolecule, three-body or higher order interactions may be neglected from Eq. (2.9), whereas two-body interactions ( mmB ) can easily be

measured experimentally by traditional light scattering technique [48]. Several other analytical techniques, including membrane osmometry [51-53], sedimentation equilibrium [54], self-interaction chromatography (SIC) [55-57] and SEC [24], have been used over past decade for the measurement of second virial coefficient. Except light scattering [48,49,58], no other technique was ever investigated to show how close is the experimentally obtained second virial coefficient to the McMillan-Mayers’s mmB . Indeed all of these techniques have

significant inherent inaccuracy in measuring Bmm, whereas the inherent error limit is minimum in SIC [57]. 2.3.2. Application of thermodynamic models in bioseparation processes Separation of biomolecules is usually based on the difference in thermodynamic properties of different components. Therefore, a number of downstream separation methods,

17

Chapter 2

for instance crystallization/precipitation, ATPS, chromatography, can be described with thermodynamic parameters shown in section 2.3.1 and consequently be predicted, modeled and designed.

2.3.2.1. Size-exclusion chromatography process In SEC, partitioning of a biomolecules species between mobile phase and stationary gel phase can be approximated simply based on their molecular size and assuming thermodynamic ideal behavior of solution [59,60]. At higher concentration of biomolecules, thermodynamic non-ideality becomes a non-negligible term. In order to explain this phenomenon, we assume the equilibrium condition, i.e. the chemical potential of the macromolecule in mobile phase ( L

mμ ) is equal to the chemical potential of the macromolecule

in the gel phase ( Gmμ ), L

mGmμ μ= . Using Eq. (2.3) we have

ln ln lnL G G

G Lm m mm m L

m

mRT m

μ μ γ γ⎛ ⎞° − °

= − + ⎜⎝ ⎠

⎟ (2.12)

where Gmγ and L

mγ are the activity coefficients of the macromolecule in gel phase and mobile

phase, respectively, and and are the macromolecule concentration in the gel phase

and the mobile phase, respectively. Here, the concentration in the gel phase is defined as the amount of injected biomolecule divided by the accessible volume. The term / is

described as the distribution coefficient and denoted as KSEC.

Gmm L

mm

Gmm L

mmLmγ in Eq. (2.12) can be

described in virial expansion terms as shown in Eq. (2.9). However, virial expansion of Gmγ

considers interaction of macromolecules with the matrices as well as self-interactions. Neglecting the interactions with matrices, which is typically the case for SEC systems, Eq. (2.9) can be written as

( )23ln 22

L L Lm mm m mmm mB m B mγ = + + (2.13)

( )23ln 22

G G Gm mm m mmm mB m B mγ = + + (2.14)

Combining Eq. (2.13) and (2.14) yields

( ) ( ) ( )2 23ln 2 1 12

L GL Lm m

SEC mm m SEC mmm m SECK B m K B m KRT

μ μ° − ° ⎧ ⎫= + − + −⎨ ⎬⎩ ⎭

+ (2.15)

Eq. (2.15) describes the concentration dependent retention behavior of macromolecules in the SEC column in term of virial coefficients. The first term on the right hand side of Eq. (2.15) is a constant, independent of concentration, which denotes distribution coefficient of the macromolecule in the limit of infinite dilution [22]. 2.3.2.2. Hydrophobic-interaction chromatography process Retentions in HIC are, in principle, based on the surface hydrophobicity of the macromolecule. Unfortunately, surface hydrophobicity still lacks an absolute definition. A

18


number of hydrophobicity scales were used in literature for computing protein hydrophobicity. For HIC modeling, Asenjo and co-workers [61] estimated the average surface hydrophobicity of proteins from the three-dimensional structure data by calculating the hydrophobic contribution of the exposed amino acid residues as a weighted average. Based on this average surface hydrophobicity, they also proposed a model for predicting retention in HIC [61,62]. However, the model does not consider the possibility of a heterogeneous distribution of hydrophobic patches throughout the biomolecular surface, which may largely affect on protein retention in HIC [30]. A better prediction may be obtained, if the areas and location of the hydrophobic patches that react with HIC resin is known [29]. In addition, it is not at all clear how this approach can be applied for a molecule whose three-dimensional structure has not yet been elucidated. Furthermore, the approach does not account for the possibility of specific interactions between proteins and matrix or for the conformational changes of proteins during the chromatography process [63]. A detailed thermodynamic modeling of retention in HIC was reviewed earlier [64]. However, we approach the issue differently, based on the activity coefficients of macromolecules in different phases.

The magnitude of solute retention in linear elution chromatography is measured under isocratic conditions in terms of the retention factor or the capacity factor, k , which is related to the equilibrium constant, KHIC , for the distribution of solute between the bulk mobile phase and stationary phase as

′

( )m uHIC

u

t tk

tK φ

−′ ′= = (2.16)

where tm and tu are the retention time of solute and an unretained tracer, respectively. φ′ is the phase ratio of the column, i.e. ratio of the volume of the stationary phase to that of the mobile phase. Applying Eq. (2.3) at zero ionic strength (superscript (0)) and ionic strength I (superscript (I)) we have (2.17) ( ) ( ) ( )ln lnL I L L I L I

m m mRT RT mμ μ γ= ° + + m

Lm (2.18) (0) (0) (0)ln lnL L L

m m mRT RT mμ μ γ= ° + +

Now the chemical potential of the macromolecule in the liquid phase at ionic strength I ( ) can be written as ( )L I

mμ

( ) ( )

( ) (0)(0) (0)ln ln

L I L IL I L mm m L

m m

mRT RTm

γμ μγ

⎛ ⎞ ⎛= + +⎜ ⎟ ⎜

⎝ ⎠ ⎝mL

⎞⎟⎠

(2.19)

Since the same column load is applied in all the experiments, it is assumed that the activity coefficients are a function of ionic strength (I) only. At equilibrium the chemical potential of the macromolecule in the mobile phase is equal to the chemical potential of solute on the stationary gel phase, i.e. and . Using Eq. (2.19) for both phases at

equilibrium

( ) ( )L I G Im mμ μ= (0) (0)L G

m mμ μ=

( ) ( ) ( ) ( )

(0) (0) (0) (0)ln ln ln lnG I L I L I G Im m mG L Lm m m

m mm m

γ γγ γ

⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛− = −⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜

⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝mGm

⎞⎟⎠

(2.20)

19

Chapter 2

The concentrations in the gel phase in Eq. (2.20) are defined as the amount of biomolecule bound to the solid phase divided by the volume of stationary phase. Manipulating Eq. (2.20) and using the definition of distribution coefficient, /G L

HIC I IK m m=

(0) ( ) ( )

(0) (0) (0)ln ln ln lnG L Im m

HIC L Lm m

mKm

γ γγ γ

⎛ ⎞ ⎛ ⎞ ⎛= + −⎜ ⎟ ⎜ ⎟ ⎜

⎝ ⎠ ⎝ ⎠ ⎝

G ImGm

⎞⎟⎠

2 Φ

(2.21)

Eq. (2.21) can be used for modeling distribution coefficient in the HIC system. The first term on the right hand side of the Eq. (2.21) is a constant, describing the distribution coefficient at zero ionic strength. The ratio of activity coefficients in the mobile phase (the middle term on the right hand side of Eq. (2.21) can be interrelated with osmotic virial coefficients as shown in Eq. (2.9), consequently can be obtained experimentally. Alternatively, this term can also be modeled by a Debye-Hückel term plus a linear term [65]. The last term on the right hand side of Eq. (2.21) is the ratio of activity coefficients on the stationary phase, consequently most difficult to access because it is virtually impossible to directly measure the activity coefficients of macromolecules in the stationary phase. Some researchers believe that the activity coefficient of macromolecule in the stationary phase remains practically constant for a particular resin/solvent combination while varying the mobile phase composition [66]. In that case, the last term on in Eq. (2.21) can be neglected.

2.3.2.3. Aqueous two-phase separation process Several theories have been proposed for correlating and predicting distribution of solutes between two aqueous continuous phases. A detailed review on different theories has been presented earlier [47,67-69]. Here we reproduce the osmotic virial expansion based thermodynamic model for ATPS to show that osmotic virial coefficients are also important parameters for designing ATPS processes. The osmotic virial expansion that is commonly used for ATPS was first proposed by Edmond and Ogston [70,71]. On this basis, Prausnitz and co-workers established a simple theoretical framework to predict phase separation of aqueous polymer solutions [72] and biomolecule partitioning in these phases [73]. When a macromolecule is placed in an aqueous two-phase system, the electrochemical potential of a macromolecule in a four-component system is expressed as

(2.22) 2

1 1 2 2 1 12

11 1 12 1 2 2 2 22 2

(ln

)m m m mm m m m mmm m mm m

m m mm m m m

RT m b m b m b m b m b m m

b m b m m b m m b m z F

μ μ− ° = + + + + + +

+ + + + +

where is the charge of macromolecule, F is the Faraday constant, and is the purely

electric potential. Subscripts 1 and 2 represent two different polymers. Interaction parameters b in Eq. (2.22) are directly related to osmotic virial coefficients as [73]

mz Φ

( )

1 11 2

21000

mm

mB M Mb = (2.23)

( )

12 1 212 3

31000

mm

mB M M Mb = (2.24)

20


where M1 and M2 are molecular weight of polymer 1 and 2, respectively. In thermodynamic equilibrium, the electrochemical potential of each component must be the same in the two phases. Calculation of phase diagram using osmotic virial expansion usually truncated after the second virial coefficient. From this basic thermodynamics, King et al. [6] showed that for a dilute macromolecule solution, the distribution coefficient of macromolecule ( ) into

top (T) and bottom (B) phases could be well described by ATPSK

( ) ( ) ( )1 1 1 2 2 2ln ln

B TTmB T B Tm

ATPS m mBm

z FmK b m m b m mm R

Φ −Φ⎛ ⎞= = − + − +⎜ ⎟

⎝ ⎠ T (2.25)

where the superscripts T and B represents top and bottom phases, respectively. Although Eq. (2.25) was derived for the description of a dilute macromolecule solutions, it can be applied to concentrated solution as long as the distribution coefficient is independent of the macromolecule’s own concentration, which is approximately 30 wt. % [74]. 2.3.2.4. Crystallization and precipitation

Solid-liquid phase separation in macromolecule solutions occurs when macromolecule concentration exceeds the solubility limit. Solid phase can be developed in the form of either crystal or precipitate. George and Wilson [4] observed that solution conditions under which proteins have a tendency to crystallize correspond to a slightly negative Bmm. They correlated protein crystallizations with Bmm values in the form of so called “crystallization slot” [75]. When the Bmm value is more negative than the crystallization slot, the interactions between molecules are so high that amorphous precipitation is more likely to occur, rather than highly organized crystal structure. A positive Bmm value does not completely exclude the possibility of crystallization or precipitation, but typically requires impractically high concentration of macromolecule in order to bring about any kind of phase separation. A thermodynamic understanding why a particular range of Bmm values promotes crystallization was described in the literature [3,4,76-78]. It was later investigated that the Bmm value within the crystallization slot is an essential prerequisite of crystallization, but does not guarantee successful crystal growth [57]. In addition to the Bmm, solubility of the macromolecule is also an important parameter. Indeed, a relationship exists between solubility and Bmm.

Both the solubility and the Bmm of biological macromolecules are determined by the interaction between the protein molecules. In dilute aqueous solutions, both depend mainly on two important solution parameters, pH and ionic strength. Solubility depends on the binding energy between molecules at a short-distance from one another for their very specific orientations. On the other hand, Bmm is a statistical average of the interaction forces over all distances and orientations of two molecules in the liquid phase. Nevertheless, the similarity in the qualitative nature of Bmm trend and solubility trend is obvious from literature [10,12,76]. In order to simply realize the relationship, we extend the activity coefficient model to supersaturated solutions.

21

Chapter 2

The criterion for thermodynamic equilibrium in a solid-liquid phase separation process is that the chemical potential of the macromolecule in the liquid phase ( L

mμ ) is equal

to the chemical potential of the macromolecule in the solid (crystal/precipitate) phase ( Smμ ).

Therefore, for a saturated macromolecular solution ln lnS sat s

m m m m matRT RTμ μ μ γ= = ° + + m (2.26)

where of Eq. (2.3) has been replaced by the solubility of macromolecule, mm satmm and sat

mμ

represents the activity coefficient of the macromolecule in the liquid phase at saturation. Solubility can be correlated with virial coefficients by evaluating Eq. (2.24) with Eq. (2.9) as

( )23ln 22

Ssat sat satm mm mm m mmm mm B m B m

RTμ μ− °

− = + + (2.27)

Eq. (2.27) shows that a straightforward relationship exists between virial coefficients and solubility. When the solubility of the macromolecule is low enough to neglect higher-body interactions, Bmm alone can predict the solubility. However, the solubility of macromolecules is very high in some solution conditions, where Bmmm cannot be neglected. This phenomenon is evident from the observation of Ruppert and co-workers [12] that Bmm does not correlate strongly with solubility data at protein concentrations >30 mg/ml. Eq. (2.27) cannot be used readily to measure solubility from Bmm, because S

mμ is unknown and impossible to measure

experimentally. One way to obtain solubility from Bmm is to assume that the solid phase thermodynamic properties, i.e. S

mμ , are independent of solution parameters. Then

experimental measurement of solubility in only one solution condition will be enough to access first term on the left hand side of Eq. (2.27). This could be the case for a particular morphology of crystal or precipitate. However, different solution conditions produce different types of crystals, which may not have the same thermodynamic properties. Similarly, precipitates may also occur in different forms, for instance concentrated liquid, aggregated solids, or disordered precipitates. Different forms of the solid phase are therefore expected to have different chemical potentials. Another way to obtain solubility from Bmm is to fit Bmm data to the UNIQUAC activity coefficient model. Solubility can then be predicted from this model [79,80]. In order to predict solubility from Bmm without any other variable, Haas and co-workers [11] combined a square-well potential model to Bmm data with Gibbs free energy models for the crystalline and solution phases. 2.4. Generalized approach to bioseparation process design

In designing a bioseparation process, characterization of the starting material and defining the specifications of final products are required first. If the design program starts from the fermentation broth, the separation process can be divided into two main categories, firstly protein recovery and then protein purification. Since recovery operations are relatively straightforward, we focus only on purification operations in this paper.

22


Obtaining physicochemical and thermodynamic properties

Modeling of separation methods

Generation of process alternatives

Selection of the best process/es

Expert System

Thermodynamic generalization

HTE

Optimization of the process

Detailed design of the process

HTE

Design of experiments

Fig. 2.1. An integrated way to bioseparation process design. 2.4.1. Process synthesis, optimization and design The first step in designing a purification process is to make decision on whether a particular unit is applicable for separation of the target component from its contaminants. The feasibility of a separation method can be studied by modeling. As it is shown in section 2.3.2, modeling of a separation method requires information on constant physicochemical properties as well as solution condition dependent thermodynamic properties of the components present in the mixture. Once modeling is done, process options can be generated based on the standard rules [18] as well as on the end product requirements. An Expert system may also be of significant help for choosing the sequence of operations [1,81-84]. Block diagrams of different process alternatives with some knowledge on the prospective yield and purity can be obtained in this way. The best process option may also be selected based on this primary knowledge and purity. However, optimization of each separation unit is further required for the detailed design of the process. For instance, we know from an Expert system that the first unit operation of a process is a HIC. We still have a number of variables to be optimized in the HIC unit, including the resin to be used (which supplier, which resin), eluent composition (which salt in which concentration), pH and mode of elution (stepwise or gradient), and sensitivity. The basic model based on thermodynamic and physicochemical properties may not provide the precise data required for optimization. This is also due to the fact that all resin specifications and its behavior in different eluents is not readily provided by

23

Chapter 2

its suppliers. The same holds true for other separation methods. Therefore, experimentation cannot be avoided for process optimization or detailed design of the process. However, traditional laboratory scale analytical experimentation can be replaced by rapidly emerging HTE approaches together with design of experiments. The steps in this approach of process design are shown in Fig. 2.1. 2.4.2. High-throughput experimentation as a bioseparation process design tool HTE can be used as a process design tool for the determination of thermodynamic and physicochemical properties of biomolecules needed for modeling of each separation method. We first identify the parameters needed for the modeling of different separation methods (Table 2.2). Some physicochemical parameters listed in Table 2.2 are constant and independent of the solution composition, for example molecular size, density, hydrophobicity and charge. For common biomolecules some of these properties may be obtained from published results and databases. Throughput of the presently existing techniques is quite high for the determination of molecular size and charge. However, no HTE technique has been developed yet for the determination of hydrophobicity, which relies still on an analytical HIC or calculation from three-dimensional structure of the molecule. Table 2.2. Presently existing analytical methods to obtain process design parameters of biomolecules.

Property Type of property Analytical method to obtain Molecular size/weight Constant Bioanalyzer, gel electrophoresis, mass spectrometry Charge Dependent on pH Capillary isoelectric focusing, capillary

electrophoresis, titration Partial specific volume Constant Analytical centrifugation, SEC, from primary

structure Hydrophobicity Constant HIC, from primary structure Solubility Dependent on

solution condition Solubility measurement experimentation

Bmm Dependent on solution condition

SIC, membrane osmometry, light scattering

Other parameters, for example solubility and Bmm, depend largely on the solution composition, pH, temperature and pressure, eventually may not be obtained from published work and databases. Laboratory-scale analytical experimentation is usually done to measure these parameters. However, throughput of all the conventional techniques for solubility and Bmm (section 2.3.1) measurement is quite low. SIC is the most recently developed technique for the Bmm measurement, which also shows great possibility of miniaturization and parallelization to increase the throughput by at least an order of magnitude [85,86]. Similarly, conventional low-throughput solubility measurement process can be replaced by faster processes with the help of micro-dialysis and interferometers [87]. However, no real HTE technique exists so far for the measurement of protein solubility.

24


Another scope of HTE is for the process optimization. This HTE approach would quickly screen out the best auxiliary materials and process operating conditions using minimal amount of sample, time and effort. Most of the HTE developmental work is being done for chromatography, possibly because chromatography is still the main process in biotechnology industry for downstream purification of their product. Optimization of a chromatography method often entails the initial screening of various classes of stationary phase materials and mobile phase compositions in order to identify the chromatographic conditions with sufficient selectivity. A 96-well microtiter plate based technique for high-throughput screening of different chromatography modes, combinations of resins, excipients and process parameters is recently getting somewhat popular [88-90]. The screening process utilizes 50-100 µl of resin per well, with a robotics system dispensing and mixing the resins, buffers, and protein load. Batch binding, followed by sequential washes and elutions of each well, is used to mimic the behavior of chromatography columns. This 96-well microtiter plated based high-throughput screening approach is amenable to multivalent experimental design. Additional applications for use in development and characterization of chromatography purification processes include assessing solubility, selectivity between the product and impurities and evaluation of the resins equilibrium binding capacities [90]. Therefore, microtiter plate based high-throughput screening methodology shows enormous promises for pharmaceutical process development for biotechnology industries, although it has the limitation that each well had an efficiency of only one theoretical plate which may not mimic the process in column chromatography.

Behind the 96-well microtiter plate based technology, further miniaturized scale microchip based technologies are also being developed. Huang et al. [91] presented an approach for rapid prototyping of microchips in order to evaluate different types of stationary phases. These chips are based on fibers to which particles of different ion exchange groups or antibodies are anchored. The labeled proteins are then microscopically observed with the respect to the retention behavior. Their work describes the rapid assembly of different types of stationary phases required for separation, methodologies for the rapid evaluation of the observed fractionation and the way how the observed properties can be used to quickly define the most appropriate stationary phase, and scale parameters. In another work, Kornmann et al. [92] have developed a technology platform that allows rapid performance mapping of chromatographic processes. This platform includes micro-scale chromatography experimental design and analysis using response surface algorithms. Chromatography operating conditions and excipients, such as buffer composition, pH, temperature, protein quantity and the type of resin, could be included in the mapping. Using this platform, a chromatographic process performance can be described under several process conditions within a short time frame. 2.5. A case study: Solubility and chromatographic distribution coefficient of nystatin Solubilities and chromatographic distribution coefficients are both crucial properties for designing feasible and optimal chromatographic separation processes. This is important for gradient applications in fixed and simulated moving bed (SMB) applications [93,94]

25

Chapter 2

where distribution behavior of the solutes to be separated as well as their solubilities vary with varying solvent composition. In case of chromatographic separation of components with low solubilities, this commonly leads to a very high solvent consumption. Separating the components in a gradient SMB system can reduce the solvent consumption considerably [94].

Rational design of gradients in solvent composition in fixed and SMB chromatographic systems requires an increased amount of quantitative information about the relation of the two mentioned physical properties and the organic solvent volume fraction,

oφ , in the solution. The relation between distribution coefficient and oφ is used for the flow

rate selection in chromatographic SMB systems and cannot be omitted in the design procedure, while the relation between the solubility and oφ defines maximum concentrations.

In this case study, we provide the basic thermodynamic equations that describe these relations. This may be beneficial when only information about one of the two parameters is known or when limited amount of the solute is available.

To illustrate the practical relevance, we have used antibiotic nystatin in an aqueous-methanol system as an experimental model system with low solubilities. Nystatin (Fig. 2.2) is a medium-size biomolecule (926 Da) that inhibits the cell’s cross-membrane transport system. Michel [95] has presented solubility data for nystatin in this solvent system but unfortunately with rather limited detail of the experimental procedure. In this chapter, we present new solubility data for the same system in combination with chromatographic distribution coefficients for a commercial reversed-phase chromatographic material.

Fig. 2.2. Structure of nystatin A.

2.5.1. Relating solubility and chromatographic distribution coefficients

The retention time of a hydrophobic solute, like nystatin, in RPC decreases when the organic solvent fraction in an aqueous-organic solution increases. This is partially due to the energetically more favorable solvation of the solute in the liquid phase at increased organic solvent content, relative to the solute-stationary phase interaction, which is reported to remain fairly constant.

The starting point for this expression is the chemical potential of nystatin ( Lnμ ) in a

multi-component mixture of solvents as shown in Eq. (2.3). For supersaturated solution of nystatin, we reproduce Eq. (2.26) as

26


( )

ln lnSn n L

nsatnx

RTμ μ

γ− °

= + (2.28)

where subscript n denotes nystatin. Since the solvent itself is a mixture of two components of non-identical density, concentration of nystatin is presented in Eq. (2.28) in mole fraction unit (xn). In order to avoid inconsistency, xn is described here as moles of nystatin per mole of solvent. Eq. (26) is valid for different solvent compositions, for instance methanol at different volume fractions. Considering Eq. (2.26) for methanol volume fraction 0.7 (superscripts (0.7)) and for an unknown methanol volume fraction (superscript ( oφ )), we have

( ) (0.7) (0.7)ln ln

Sn n L sa

n nxRT

μ μγ

− °= + t (2.29)

( ) ( ) ( )ln lno

Sn n L sa

n nxRT

otφ φμ μ

γ− °

= + (2.30)

Assuming ‘pure’ crystals of nystatin with a constant composition and crystal structure, i.e. constant S

nμ , while varying the solvent composition, leads to

( ) ( )

(0.7) (0.7)ln lno osat L

nsat Ln n

xx

φ φγγ

⎛ ⎞ ⎛− =⎜ ⎟ ⎜

⎝ ⎠ ⎝n ⎞

⎟⎠

(2.31)

Eq. (2.31) shows the relationship between activity coefficients and solubilities of nystatin in a bicomponent solvent of methanol and water. Solute retention thermodynamics in RPC are principally the same as HIC. Eq. (2.19) is, therefore, applicable for describing the distribution of nystatin in a RPC column at solvent compositions (0.7) and ( oφ ). Reproducing Eq. (2.19)

in mole fraction scale we have

( ) ( ) ( )(0.7)

( ) (0.7) (0.7) (0.7)ln ln ln lno o

o

G GGn n n nL L G Ln n n n

x xx x

φ φ

φ

γ γγ γ

⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛= − +⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜

⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝

oL φ ⎞⎟⎠

o

(2.32)

The last term in equation (2.31) and (2.32) may differ because the activity coefficient of nystatin is a function of its concentration and the concentration in RPC may not be the same as solubility. Assuming the saturation concentration of nystatin in RPC and using the definition of distribution coefficients, i.e. ( ) ( ) ( )/o oG L

n n nK x xφ φ φ= and , Eq.

(2.32) yields

(0.7) (0.7) (0.7)/G Ln n nK x x=

( ) ( )

( ) (0.7)(0.7) (0.7)ln ln ln ln

o oo

G san n

n n G san n

xK Kx

φ φφ γ

γ⎛ ⎞ ⎛

= − −⎜ ⎟ ⎜⎝ ⎠ ⎝

t

t

⎞⎟⎠

(2.33)

Eq. (2.33) shows a direct relationship between chromatographic distribution coefficients and solubilities of nystatin as a function of the methanol volume fraction in the solvent. The first term on the right hand side of Eq. (2.33) is a constant, which describes the distribution coefficient of nystatin when the methanol volume fraction is 0.7 in the solvent. The second term on the right hand side of Eq. (2.33) is the ratio of the activity coefficients of nystatin in the stationary gel phase. During chromatography, the effects of a change in the stationary phase composition as sensed by nystatin immersed in the stationary phase are quite small (<5-10%) when compared to the large change in the solute environment in the mobile

27

Chapter 2

phase [66]. This statement may be interpreted such that the concentration of adsorbed species is not sufficiently high to alter the adsorbent environment and the activity coefficient of nystatin in the stationary phase remains practically constant while continuously varying the solvent composition. This assumption simplifies Eq. (2.33) to

( )

( ) (0.7)(0.7)ln ln ln

oo

satn

n n satn

xK Kx

φφ ⎛ ⎞

= − ⎜⎝ ⎠

⎟ (2.34)

Therefore, the distribution coefficient remains only as a function of solubility.

RPC system behavior at varying solvent composition is almost exclusively affected by liquid phase properties. Distribution coefficients of nystatin can be plotted as a function of solubilities, if the first constant term in Eq. (2.34) is known. Eq. (2.31) and (2.34) also indicate that the negative logarithms of the solubility and the chromatographic distribution coefficient have a similar, if not identical, dependence on the solvent composition. It was also previously observed that the relations between the solvent composition of the mobile phase and logarithm of the Kn value as well as the negative logarithm of the solubility ( ln sat

nx− ) are

identical and linear [96].

In practice, the slope of the distribution coefficient of larger molecules such as proteins versus the organic solvent fraction can be very steep. This limits the window of operation for an isocratic mobile phase considerably. On the other hand, gradient operation is a powerful tool for this type of separations. Often, the window of operation of smaller biomolecules typically ranges from 70% (v/v) to 95% (v/v) methanol in water [97-100].

The relation between the distribution coefficient and the solvent strength as represented by oφ is determined by pulse experiments on an analytical HPLC column. The

distribution coefficient is calculated from the retention time of the solute found in the HPLC analysis. The relation is given as follows [101]

1

n u

u

t tKt

εε

−=

− (2.35)

where tn is the retention time of nystatin and ε is the overall bed void fraction. The average liquid residence time in the column, tu, can be found by injecting an unretained solute, e.g. NaCl. Using this value, also the overall bed void fraction,ε , be found as

ntV

ε′⋅Φ

= (2.36)

where V is the column volume and ′Φ is the volumetric flow rate. 2.5.2. Experimental determination of solubility and distribution coefficient

2.5.2.1. Materials

Nystatin was supplied by Alpharma (Oslo, Norway), while the nystatin standard (5010 USP units/mg) for the analytical HPLC was purchased from Sigma. Potassium Dihydrogen Phosphate was purchased from Mallinckordt Baker (Deventer, Netherlands).

28


Methanol was purchased from Acros (Geel, Belgium). Ultra pure water was obtained by means of a Millipore MilliQ UF Plus (Bedford, ME, USA). 2.5.2.2. Equipment and procedure The solubility experiments were performed in a thermostatic water bath at 298 K, using a Lauda RM6 cryostat. The temperature was controlled with a normal laboratory thermometer (±0.2 K). A multi-sample magnetic stirrer (Variomag telesystem) is placed underneath the water bath for proper and homogeneous stirring of the samples. Each sample was prepared in a 12 ml glass sample flask with a rubber slip in duplicate. First step is to weigh in the accurate amount of methanol (±0.0001 g). The amount of water is exactly weighed in and the solution is stirred in the water bath for about ten minutes. After that, the solution is saturated with nystatin. The samples are stirred for an hour in the water bath. It was found in a preliminary experiment that one-hour was sufficient to reach equilibrium. A sample is withdrawn from the flask, filtered with a 0.45 µm syringe filter (Gelman Acrodisc 32 Supor) and diluted in pure methanol to approximately 2200 USP units/ml. The sample is weighed and eventually stored in a freezer before analysis by HPLC.

Nystatin concentrations were determined by analytical HPLC. The nystatin from Sigma was used to obtain a standard curve for determination of activity and concentration of the Alpharma nystatin. A Waters Alliance HPLC system (Melford, ME, USA) was used. The HPLC analysis was performed on a Hypersil PEP 100 C18 column, which was packed with particles of 5 µm (4.6 id. x 250 mm). The mobile phase was a 3 mM KH2PO4 and methanol solution at 30.7 % (v/v), at a flow rate of 0.9 ml/min. The injection volume was 6 µl (~ 13 USP units). The column temperature was kept at 298 K, and the sample was kept at 278 K to prevent degradation. The chromatograms were recorded at 306 and 385 nm.

Distrubution coefficient experiments were carried out on the same HPLC and with the same HPLC method, except that pure water was used in place for the phosphate buffer and that the mobile phase composition of aqueous methanol was varied from 70% (v/v) to 100% (v/v) in increments of 5% (v/v). In addition was the flow rate raised from 0.9 to 1.0 ml/min and a different column was used, YMC ODS-AM 5 µm, 150 x 4.6 cm (Schermbeck/Weselerwald, Germany). 2.5.3. Results and discussion The average column hold up time, tu, for the column was found to be 1.67 minutes, which resulted in a total bed void fraction, ε , of 0.71. Assuming that the inter-particle void fraction, ε b, is 0.4, the intra-particle void fraction, ε p, can be found with help of the following relation [101] (1 )b p bε ε ε ε= + − (2.37)

The intra-particle void fraction is calculated to be 0.52. The uncertainty analysis of the experimental data is performed as presented elsewhere [102,103]. The confidence interval for all experimental results are found by Student t (95%, two-tailed) distribution and presented

29

Chapter 2

together with the main results in Table 2.3 for the solubility experiments and Table 2.4 for the distribution coefficient experiments. 2.5.3.1. Solubility The analytical HPLC results were transformed from EU to concentration (g/l). The solubility increases exponentially with increased methanol concentration. This was also observed by Michel [95]. The nystatin concentrations are calculated into mole fractions (Table 2.3). Table 2.3. Solubility of nystatin in aqueous methanol solvent at 298 K.

Solvent composition Solubility [Methanol mole fraction] [Methanol volume fraction] [g/l] [105 mol fraction]

1.00 1.00 10.28 45.13 0.95 0.98 6.30 26.89 0.89 0.95 5.65 23.34 0.80 0.90 4.10 15.99 0.70 0.84 2.01 7.37 0.64 0.80 1.85 6.48 0.57 0.75 1.69 5.63 0.51 0.70 1.35 4.32

Table 2.4. Results and uncertainties for the Kn value experiments at 298 K.

Methanol volume fraction,

oφ , [-] Retention time,

tu, [min] Distribution coefficient,

Kn, [-] ln Kn [-]

0.70 9.89 12.17± 0.57 2.50 0.80 3.37 2.50± 0.14 0.92 0.85 2.52 1.25± 0.35 0.22 0.90 2.16 0.71± 0.07 -0.35

2.5.3.2. Distribution coefficients

Measured chromatographic distribution coefficients are given in Table 2.4 as a function of the methanol volume fraction. Distribution coefficients are also calculated using Eq. (2.34), where experimental distribution coefficient at methanol volume fraction of 0.7 was used as . The relation between the experimentally determined distribution

coefficients and calculated distribution coefficients are shown in Fig. 2.3.

(0.7)nK

According to the simplified theory, the relation between the solvent composition of the mobile phase and the negative logarithm of the solubility is linear . If this is true, the logarithm of the distribution coefficients should also exhibit a linear dependence on the methanol volume fractions. As can be observed in Fig. 2.3, the calculated distribution coefficients and the experimentally determined distribution coefficients follow a similar, yet not identical linear dependence on the volume fraction methanol. The slopes of the calculated and experimental distribution coefficients versus volume fraction methanol differ by

30


approximately a factor of 1.7. This variation between experimental and modeled data might be due to the assumption that the activity coefficient of nystatin in the stationary gel phase does not change as a result of changes in mobile phase composition, i.e. ( )oG

nφγ = n

the drastic assumptions, the qualitative agreement of theory and experiments seems reasonable and offers a simple tool for process design.

(0.7)Gnγ . Give

Methanol volume fraction (-)

0.7 0.8 0.9 1.0 1.1

-ln (x

nsat ),

ln (K

n)

-2

0

2

4

6

8

10

12

Methanol volume fraction0.7 0.8 0.9 1.0

ln (K

n)

-2

0

2

4 CalculatedExperimental

Fig. 2.3. Distribution coeffi-cient and solubility of nystatin in aqueous-methanol solution. Rectangles represent the logarithm of the reciprocal solubility and circles represent the logarithm of distribution coefficient. Inset plot compares experimental ln Kn

with Eq. (2.34).

2.5.4. Conclusion of the case study

Solubility is important not only for crystallization processes, but also for the optimization of the chromatographic processes, since the solvent consumption is more or less a direct function of the solubility. To find the optimal solvent conditions is it necessary to have quantitative relations between both distribution constant or solubility and solvent composition. This is particularly true for processes with varying solvent composition such as gradient fixed and SMB chromatography processes.

Therefore and from a practical point of view, the linear relation between solubility and distribution coefficient and the methanol void fraction is an important result. Even though the quantitative difference of model parameters cannot be explained, the models are very useful to correlate the data and aid in process design. 2.6. Conclusion The thermodynamic basics describing the most important techniques for purification of biological macromolecules in downstream processing are given. It appears that several purification methods can be described with the help of Bmm, thereby allowing translation of experimental results over the boundaries of different purification unit operations. This will be a central aid in developing an integrated approach to bioseparation process design.

The applicability of the proposed framework and the translation between different unit operations was demonstrated for a medium sized biomolecules, nystatin, between crystallization and reversed-phase chromatographic separation.

31

Chapter 2

Acknowledgements These investigations were supported (in part) by the Netherlands Organization for Scientific Research (NWO) in the NWO-ACTS research program B-BASIC, which is directed by Prof. Luuk A.M. van der Wielen (Delft University of Technology). We acknowledge Alpharma, Inc. (Oslo, Norway) for the partial financial support and for the donation of nystatin, and T.B. Jensen for experiments. We also thank Dr. E.S.J. Rudolph (Delft University of Technology) for her valuable suggestions and comments. Symbols a Activity b Interaction parameter B McMillan-Mayer’s osmotic virial coefficient B Hill’s osmotic virial coefficient c Concentration in molar scale F Faraday constant K Distribution coefficient k ′ Chromatographic capacity factor or retention factor m Concentration in molal scale (mol/kg of solvent) M Molecular mass R Universal gas constant t Retention time T Temperature in Kelvin V Column volume x Concentration in mole fraction scale z Charge γ Activity coefficient ε Void fraction of column μ Chemical potential

μ° Standard state chemical potential π Osmotic pressure

0ρ Molar density 0ν Partial specific volume

φ Volume fraction

φ′ Phase ratio of the chromatography column Φ Electric potential Φ ’ Volumetric flow rate Subscripts 0 Zero ionic strength 1 Polymer 1 2 Polymer 2

32


ATPS Aqueous two-phase separation process b Inter-particle HIC Hydrophobic-interaction chromatography process m Biological macromolecule n Nystatin p Intra-particle s Solvent SEC Size-exclusion chromatography process u Unretained compound/tracer Superscripts (0) Ionic strength zero (0.7) Methanol volume fraction 0.7 (I) Ionic strength I (φ ) Volume fraction of methanol B Bottom phase G Chromatographic gel (stationary) phase L Liquid (solution) phase S Solid (crystal/precipitate) phase sat Saturation in terms of macromolecule concentration T Top phase References [1] J.A. Asenjo & B.A. Andrews. J. Mol. Recognit. 17 (2004) 236-247. [2] J.A. Asenjo & I. Patrick. In E.L.V. Harris & S. Angel (Eds.) Protein Purification Applications:

A Practical Approach, Oxford University Press, Oxford, 1990, p. 1-28. [3] B.L. Neal, D. Asthagiri, O.D. Velev, A.M. Lenhoff & E.W. Kaler. J. Cryst. Growth 196 (1999)

377-387. [4] A. George & W.W. Wilson. Acta Cryst. D50 (1994) 361-365. [5] R.A. Curtis, J.M. Prausnitz & H.W. Blanch. Biotechnol. Bioeng. 57 (1998) 11-21. [6] R.S. King, H.W. Blanch & J.M. Prausnitz. AIChE J. 34 (1988) 1585-1594. [7] J.G.S. Ho, A.P.J. Middelberg, P. Ramage & H.P. Kocher. Protein Sci. 12 (2003) 708-716. [8] P.-A. Albertsson. Partition of Cell Particles and Macromolecules, Wiley/Interscience, New

York, 1986. [9] L.A.M. van der Wielen & E.S.J. Rudolph. J. Chem. Technol. Biotechnol. 74 (1999) 275-283. [10] B. Guo, S. Kao, H. McDonald, A. Asanov, L.L. Combs & W.W. Wilson. J. Cryst. Growth 196

(1999) 424-433. [11] C. Haas, J. Drenth & W.W. Wilson. J. Phys. Chem. B 103 (1999) 2808-2811. [12] S. Ruppert, S.I. Sandler & A.M. Lenhoff. Biotechnol. Prog. 17 (2001) 182-187. [13] E.W. Leser & J.A. Asenjo. J. Chromatogr. 584 (1992) 43-57. [14] E.W. Leser & J.A. Asenjo. Ann. N. Y. Acad. Sci. 721 (1994) 337-347. [15] T.M. Przybycien, N.S. Pujar & L.M. Steele. Curr. Opin. Biotechnol. 15 (2004) 469-478. [16] J.A. Asenjo, J. Parrado & B.A. Andrews. Ann. N. Y. Acad. Sci. 646 (1991) 334-356. [17] J.A. Asenjo. In J.A. Asenjo (Ed.), Separation Processes in Biotechnology, Marcel Dekker, New

York, 1990, p. 3-16.

33

Chapter 2

[18] G.J. Prokopakis & J.A. Asenjo. In J.A. Asenjo (Ed.), Separation Processes in Biotechnology, Marcel Dekker, New York, 1990, p. 571-601.

[19] J.A. Asenjo. In D.L. Pyle (Ed.), Separations for Biotechnology II, Elsevier, London, 1990, p. 519-528.

[20] E. Watanabe, S. Tsoka & J.A. Asenjo. Ann. N. Y. Acad. Sci. 721 (1994) 348-364. [21] A.T. Andrews, I. Noble, S. Keeratatipibul & J.A. Asenjo. Biotechnol. Bioeng. 44 (1994) 29-37. [22] L.W. Nichol, R.J. Siezen & D.J. Winzor. Biophys. Chem. 9 (1978) 47-55. [23] K.E. Shearwin & D.J. Winzor. Eur. J. Biochem. 190 (1990) 523-529. [24] J. Bloustine, V. Berejnov & S. Fraden. Biophys. J. 85 (2003) 2619-2623. [25] W. Melander & C. Horvath. Arch. Biochem. Biophys. 183 (1977) 200-215. [26] W.R. Melander, D. Corradini & C. Horvath. J. Chromatogr. 317 (1984) 67-85. [27] W.R. Melander, Z. El Rassi & C. Horvath. J. Chromatogr. 469 (1989) 3-27. [28] H.P. Jennissen. Int. J. Biochomatogr. 5 (2000) 131-163. [29] A. Mahn, G. Zapata-Torres & J.A. Asenjo. J. Chromatogr. A 1066 (2005) 81-88. [30] A. Mahn, M.E. Lienqueo & J.A. Asenjo. J. Chromatogr. A 1043 (2004) 47-55. [31] J.L. Fausnaugh, L.A. Kennedy & F.E. Regnier. J. Chromatogr. A 317 (1984) 141-155. [32] D. Forciniti, C.K. Hall & M.R. Kula. Biotechnol. Bioeng. 38 (1991) 986-994. [33] W. Fan, U. Bakir & C.E. Glatz. Biotechnol. Bioeng. 59 (1998) 461-470. [34] J.R. Luther & C.E. Glatz. Biotechnol. Bioeng. 44 (1994) 147-153. [35] M.J. Sarmento, M.J. Pires, J.M.S. Cabral & M.R. Aires-Barros. J. Chromatogr. A 668 (1994)

117-120. [36] A. Olivera-Nappa, G. Lagomarsino, B.A. Andrews & J.A. Asenjo. J. Chromatogr. B 807

(2004) 81-86. [37] M.A. Eiteman & J.L. Gainer. Bioseparation 2 (1991) 31-41. [38] G. Tubio, B. Nerli & G. Pico. J. Chromatogr. B 799 (2004) 293-301. [39] B.A. Andrews, A.S. Schmidt & J.A. Asenjo. Biotechnol. Bioeng. 90 (2005) 380-390. [40] J.A. Asenjo, A.S. Schmidt, F. Hachem & B.A. Andrews. J. Chromatogr. A 668 (1994) 47-54. [41] T.C.V. Penna, E. Chiarini & A. Pessoa. Appl. Biochem. Biotechnol. 107 (2003) 481-492. [42] S. Jain, R. Singh & M.N. Gupta. J. Chromatogr. A 1035 (2004) 83-86. [43] I. Roy, A. Sharma & M.N. Gupta. Biochem. Biophys. Acta 1698 (2004) 107-110. [44] A. Sharma & M.N. Gupta. Carbohydrate Polym. 48 (2002) 391-395. [45] B.H. Zimm. J. Chem. Phys. 14 (1946) 164-179. [46] W.G. McMillan & J.E. Mayer. J. Chem. Phys. 13 (1945) 276-305. [47] H. Cabezas. J. Chromatogr. B 680 (1996) 3-30. [48] T.L. Hill. J. Chem. Phys. 30 (1959) 93-97. [49] H. Cabezas & J.P. O'Connell. Ind. Eng. Chem. Res. 32 (1993) 2892-2904. [50] P.R. Wills, W.D. Comper & D.J. Winzor. Arch. Biochem. Biophys. 300 (1993) 206-212. [51] C.A. Haynes, K. Tamura, H.R. Korfer, H.W. Blanch & J.M. Prausnitz. J. Chem. Phys. 96

(1992) 905-912. [52] H.M. Schaink & J.A.M. Smit. Phys. Chem. Chem. Phys. 2 (2000) 1537-1541. [53] P.E. Pjura, A.M. Lenhoff, S.A. Leonard & A.G. Gittis. J. Mol. Biol. 300 (2000) 235-239. [54] J. Behlke & O. Ristau. Biophys. Chem. 76 (1999) 13-23. [55] P.M. Tessier, A.M. Lenhoff & S.I. Sandler. Biophys. J. 82 (2002) 1620-1631. [56] C.A. Teske, H.W. Blanch & J.M. Prausnitz. J. Phys. Chem. B 108 (2004) 7437-7444. [57] T. Ahamed, M. Ottens, G.W.K. van Dedem & L.A.M. van der Wielen. J. Chromatogr. A 1089

(2005) 111-124.

34


[58] C.J. Coen, H.W. Blanch & J.M. Prausnitz. AIChE J. 41 (1995) 996-1004. [59] A.G. Ogston. Trans. Faraday Soc. 54 (1958) 1754-1757. [60] J.C. Bosma & J.A. Wesselingh. J. Chromatogr. B 743 (2000) 169-180. [61] M.E. Lienqueo, A. Mahn & J.A. Asenjo. J. Chromatogr. A 978 (2002) 71-79. [62] M.E. Lienqueo, A. Mahn, L. Vasquez & J.A. Asenjo. J. Chromatogr. A 1009 (2003) 189-196. [63] J.C. Salgado, I. Rapaport & J.A. Asenjo. J. Chromatogr. A 1075 (2005) 133-143. [64] A. Vailaya & C. Horvath. Ind. Eng. Chem. Res. 35 (1996) 2964-2981. [65] A. Staby & J. Mollerup. J. Chromatogr. A 734 (1996) 205-212. [66] W.J. Cheong & P.W. Carr. J. Chromatogr. 499 (1990) 373-393. [67] H. Walter, G. Johansson & D.E. Brooks. Anal. Biochem. 197 (1991) 1-18. [68] N.L. Abbott, D. Blankschtein & T.A. Hatton. Bioseparation 1 (1990) 191-225. [69] J. Jiang & J.M. Prausnitz. J. Phys. Chem. B 104 (2000) 7197-7205. [70] E. Edmond & A.G. Ogston. Biochem. J. 109 (1968) 569-576. [71] E. Edmond & A.G. Ogston. Biochem. J. 117 (1970) 85-89. [72] C.A. Haynes, R.A. Beynon, R.S. King, H.W. Blanch & J.M. Prausnitz. J. Phys. Chem. 93

(1989) 5612-5617. [73] C.A. Haynes, H.W. Blanch & J.M. Prausnitz. Fluid Phase Equilib. 53 (1989) 463-474. [74] M.R. Kula. In L.B. Wingard, E. Katchalski-Katzir & L. Goldstein (Eds.) Applied Biochemistry

and Bioengineering, Academic Press, New York, 1979, p. 71-95. [75] A. George, Y. Chiang, B. Guo, A. Arabshahi, Z. Cai & W.W. Wilson. Methods Enzymol. 276

(1997) 100-110. [76] D. Rosenbaum, P.C. Zamora & C.F. Zukoski. Phys. Rev. Lett. 76 (1996) 150-153. [77] M. Malfois, F. Bonnete, L. Belloni & A. Tardieu. J. Chem. Phys. 105 (1996) 3290-3300. [78] S.M. Agena, I.D.L. Bogle & F.L.P. Pessoa. Biotechnol. Bioeng. 55 (1997) 65-71. [79] S.M. Agena, M.L. Pusey & I.D.L. Bogle. Biotechnol. Bioeng. 64 (1999) 144-150. [80] J.M. Prausnitz. Pure Appl. Chem. 75 (2003) 859-873. [81] M.E. Lienqueo & J.A. Asenjo. Comput. Chem. Eng. 24 (2000) 2339-2350. [82] M.E. Lienqueo, J.C. Salgado & J.A. Asenjo. J. Chem. Technol. Biotechnol. 74 (1999) 293-299. [83] M.E. Lienqueo, E.W. Leser & J.A. Asenjo. Comput. Chem. Eng. 20 (1996) S189-S194. [84] E.W. Leser, M.E. Lienqueo & J.A. Asenjo. Ann. N. Y. Acad. Sci. 782 (1996) 441-455. [85] C.D. Garcia, D.J. Hadley, W.W. Wilson & C.S. Henry. Biotechnol. Prog. 19 (2003) 1006-1010. [86] L. Luo, T. Ahamed, M. Ottens, G.W.K. van Dedem & L.A.M. van der Wielen. In T. Ahamed,

M. Ottens & L.A.M. van der Wielen (Eds.) Program and Abstracts of the International Conference on Biopartitioning and Purification, Delft University of Technology, Delft, 2005, p. 68.

[87] K. Nakazato, T. Homma & T. Tomo. J. Synchrotron Rad. 11 (2004) 34-37. [88] K. Rege, M. Pepsin, B. Falcon, L. Steele & M. Heng. in 227th ACS National Meeting,

Anaheim, 2004, BIOT365. [89] P. Bastek, K. Molnar, J. Coffman & B. Kelley. in 227th ACS National Meeting, Anaheim,

2004, BIOT155. [90] J. Coffman. in 229th ACS National Meeting, San Diego, 2005, BIOT143. [91] T. Huang, N.S. Mosier & M.R. Ladisch. in 229th ACS National Meeting, San Diego, 2005,

BIOT63. [92] H. Kornmann, X. Lesaout, E. Vinter & G. Baer. in 229th ACS National Meeting, San Diego,

2005, BIOT379. [93] T.B. Jensen, H.A.H. Billiet & L.A.M. van der Wielen. In, UK, 2000, p. 1-11.

35

Chapter 2

36

[94] T.B. Jensen, T.G.P. Reijns, H.A.H. Billiet & L.A.M. van der Wielen. J. Chromatogr. A 873 (2000) 149-162.

[95] G.W. Michel. Nystatin Analytical Profiles of Drug Substances, Academic Press, New York, 1977.

[96] P.J. Shoenmakers, H.A.H. Billiet & L. De Galan. J. Chromatogr. 185 (1979) 179-195. [97] J. Milhaud, P. Gareil & R. Rosset. J. Chromatogr. 358 (1986) 284-287. [98] J. Adamovics & S. Unger. J. Liq. Chromatogr. 9 (1986) 141-155. [99] B. Sauer & R. Matusch. J. Chromatogr. A 672 (1994) 247-253. [100] M. Margosis & A. Aszalos. J. Pharm. Sci. 73 (1984) 835-838. [101] R.M. Nicoud & A. Seidel-Morgenstern. Isolat. Purif. 2 (1996) 165-200. [102] J. Kragten. Analyst 119 (1994) 2161-2165. [103] P.C. Meier & R.E. Zund. Statistical Methods in Analytical Chemistry, Wiley, New York, 2000.

3

Design of self-interaction chromatography as an analytical tool for predicting

protein phase behavior

Abstract Solution conditions under which proteins have a tendency to crystallize correspond to a slightly negative osmotic second virial coefficient (Bmm). A positive Bmm value guarantees no crystallization to occur. On the other hand, a Bmm value within the so called “crystallization slot” thermodynamically supports the crystallization processes but does not guarantee successful crystal growth. It is, however, a prerequisite for protein crystallization that the Bmm value must be in the slightly negative regime. Self-interaction chromatography (SIC) is designed in this work as an analytical tool for determining Bmm in a precise and reproducible way. The methodology was demonstrated in detail in terms of its theoretical basis, experimental methodology, troubleshooting, and data analysis for different protein samples and solution conditions. The inherent error limit of SIC is found to be comparatively less than other Bmm measurement techniques. The designed experimental approach was applied for mapping crystallization conditions of a model protein, i.e. lysozyme. Good agreement between the obtained lysozyme Bmm values and literature values confirms the accuracy of the approach.

Keywords: Self-interaction chromatography; Protein phase behavior; Systematic

screening; Predictive crystallization; Lysozyme ___________________________________________________________________________ Published as T. Ahamed, M. Ottens, G.W.K. van Dedem & L.A.M. van der Wielen. J. Chromatogr. A 1089 (2005): 111-124 and 1115 (2006): 272.

Chapter 3

3.1. Introduction Protein crystallization is one of the critical aspects of structural biology and pharmaceutical biotechnology. In addition, crystallization is one of the demanded techniques in biotechnology and pharmaceutical industries for downstream processing of proteins. Numerous independent variables, including solvent conditions (pH, ionic strength, salt type, temperature, crystallizing agent etc.) and initial protein concentration, are usually involved in nucleation and growth of protein crystals [1]. Most high-resolution protein structural information is obtained by X-ray diffraction, neutron crystallography or surface plasma resonance of protein crystals. The major obstacle in these processes is often to obtain a diffraction-quality crystal. Because of the involvement of several parameters and the lack of a systematic screening approach, optimum crystallization conditions are traditionally determined by empirical screening. Empirical screening provides neither any insight of crystallization thermodynamics nor any indication of how close the solution conditions were to the ones optimal for growing crystals. Consequently, the approach requires intensive screening of numerous solution conditions blindly and failure is often the case for many proteins, particularly membrane proteins and monoclonal antibodies. It is, therefore, highly desirable to develop high-throughput methods to determine conditions for protein crystallization in a rational manner, reducing the number of crystallization experiments, cost and time.

George and Wilson [2] observed that solution conditions under which proteins have a tendency to crystallize correspond to a slightly negative osmotic second virial coefficient (Bmm), resulting from weak attractive protein self-interactions. They correlated protein crystallization conditions with Bmm values in the form of so called “crystallization slot” [3]. Bmm is a thermodynamic parameter that reflects the magnitude and direction of deviations of a non-ideal solution from ideality. At the molecular level, Bmm characterizes pair-wise protein self-interactions including contributions from excluded volume, electrostatic interactions and short-range interactions [4]. According to the McMillan and Mayer [5] solution theory, Bmm is correlated to the potential of mean force, which describes all kinds of possible interactions between two protein molecules in a dilute solution. A negative value of Bmm indicates protein-protein attraction whereas a positive Bmm value indicates mutual repulsion. The thermodynamic insight regarding the macromolecular interactions involved in Bmm and why these interactions are related to protein crystallization were explained [4,6]. There is still doubt whether the crystallization slot guarantees successful crystal growth universally for all kinds of proteins. In this study, we review available literature in order to explore the relationship between Bmm and phase behavior of different proteins. We also discuss the question whether Bmm is the only thermodynamic parameter that governs the protein crystallization process.

The link between Bmm and protein crystallization conditions offers the hope that screening Bmm values may be useful for the predictive crystallization of proteins that are proven difficult to crystallize. In some cases, solution properties could also be pushed toward the crystallization slot. Nevertheless, there has been little use of Bmm for predictive

38

Design of self-interaction chromatography

crystallization, largely because of the difficulties in Bmm measurement. A considerable amount of modeling work has been done recently by calculating Bmm and/or predicting protein phase behavior on a theoretical basis [4,6-10]. However, no model can universally be applied for all kinds of proteins regardless of their molecular mass, size, shape and nature. Bmm is usually measured experimentally by colloidal characterization techniques, for instance static light [2-4,8,9,11-17], small-angle X-ray [18-20], laser-light [21,22] or neutron [9,23,24] scattering, membrane osmometry (MO) [25-27] and sedimentation equilibrium [28]. Unfortunately, all of these methods are too labor-intensive and expensive in terms of both protein and time to allow extensive screening. Moreover, Bmm measurement by scattering techniques becomes extremely difficult when the solubility of the protein is low (<5 mg/ml). Another obstacle of the Bmm aided protein crystallization approach is the inconsistency in Bmm values measured by different techniques and/or different researchers for exactly the same protein sample and the same solution conditions. In this paper we also discuss possible reasons for this inherent inaccuracy in different techniques. It is, however, essential to have an easy-to-perform Bmm measurement technique, which provides a precise result using a minimum amount of protein, time and effort.

Self-interaction chromatography (SIC) [29-35] and size-exclusion chromatography (SEC) [36] are two recently developed alternative methods of characterizing weak protein interactions that could potentially meet the requirements of being inexpensive in terms of both time and protein relative to other traditional techniques [37]. SIC measures the interaction of immobilized protein molecules in the stationary phase with free protein molecules in the mobile phase. The average retention of a protein pulse characterizes the protein-protein interaction and hence the value of Bmm. On the other hand, SEC characterizes the thermodynamic non-ideality of a protein solution as a function of protein concentration. Negative Bmm values correspond to decreased retention in a SEC column as a function of protein concentration, which consequently reflects the protein-protein attraction and vice versa. The advantages of SIC is that a single injection of dilute protein solution leads to a Bmm value, whereas several injections of dilute to concentrated protein solution are required for SEC, which obviously costs more protein and time. On the other hand, SIC requires immobilization of protein to the stationary phase, which may sometime cause structural and conformational change of the immobilized protein. Comparatively speaking, the required experimental time and protein consumption in SIC is at least 4-5 times less than that of SEC. In addition, SIC is better suited to miniaturize the process to the microchip level, which would provide easy and fast screening of crystallization conditions. The SIC principle and methodology can also be used to study interactions among unlike proteins [38].

A systematic crystallization approach was successfully performed to produce diffraction quality crystals of chymotrypsinogen [27] and OmpF porin [12] through Bmm mapping by static light scattering (SLS), and MO, respectively. The SIC approach has been applied so far for predictive crystallization of myoglobin [31] and ribonuclease A [32]. However, no study has been reported for Bmm mapping and predictive crystallization of proteins that are structurally complex and were previously proven difficult to crystallize by

39

Chapter 3

empirical screening. In the present work, SIC methodology was designed with respect to its theoretical basis, experimental methodology, troubleshooting, data analysis and applicability for different kinds of proteins. The accuracy and precision limit of the optimized methodology was compared by reproducing Bmm trends of the well-known protein, lysozyme. 3.2. Theory and design of SIC methodology

3.2.1. Inherent inaccuracy in Bmm measurement Lysozyme is the most widely studied protein in the field of Bmm. We plotted the reported Bmm values or trends among different literature sources for exactly the same or similar solution conditions (Fig. 3.1). The possible reason of these variations was suspected to be the source and purity of lysozyme, the employed measurement technique, minor deviations in the solution conditions or experimental errors. Hen egg lysozyme from different sources is different in terms of purity. For instance lysozyme from Sigma contains more contaminant proteins (ovalbumin and albumin) than those from Seikagaku and Boehringer-Mannheim [15,21,23,36]. We therefore discriminated between different sources of lysozyme and plotted the cases where lysozyme was purchased from Sigma and Bmm was measured at pH 4.5 and at 23-25 ˚C (Fig. 3.2). Fig. 3.2 still represents a variation of ±1.5×10-4 mol.ml.g-2 unit. We, therefore, conclude that because of the involvement of only weak protein interactions Bmm measurement techniques have significant inherent inaccuracy. Virial

Fig. 3.1. Bmm of lysozyme with varying NaCl concentration at pH 4.2-4.7 and temperature 20-25 ˚C. The corresponding experimental conditions, measurement technique, source of lysozyme and reference are shown in Table 7.1.

Fig. 3.2. Bmm of lysozyme purchased from Sigma with varying NaCl concentration at pH 4.5 and temperature 23-25 ˚C. The line represents the outcome of this work. The meaning of the other symbols as well as experimental conditions, measurement technique, source of lysozyme and reference are shown in Table 3.1.

40


coefficients are determined from chemical potential as the coefficients of a positive power series of protein concentration. The chemical potential of protein molecules in a given solvent is usually measured by osmotic pressure, light scattering or sedimentation behavior or

these measuring parameters uge in d m lu

tion ns and re c r Fig. 3.1

Source . (˚C) Buffer ent

retention times in chromatography. Minor inherent inaccuracy inleads to a h accuracy in the eter ined Bmm va es. Table 3.1. Solu conditio feren es fo and Fig. 3.2.

Data Sign Protein pH Temp Measurem Ref. Black diamond M Na-acetate Sigma 4.5 20 100 m SAXS [20] Ash diamond

S

ku

e e

ku ircle ku

circle

4.7 N/A 50 mM Na-acetate SEC [36]

Sigma 4.5 25 100 mM Na-acetate SAXS [20] White diamond Sigma 4.5 N/A N/A LALL [21]Black square Sigma 4.5 25 N/A LALLS [22]Ash square Seikaga 4.7 N/A 50 mM Na-acetate SLS [15]White square Sigma 4.5 25 Minimal citric acid SLS [9] Black triangl Sigma 4.5 25 Minimal citric acid SANS [9] Ash triangl Sigma 4.6 25 40 mM Na-acetate SLS [8] White triangle Seikaga 4.7 25 25 mM Na-acetate SLS [14] Black c Seikaga 4.2 25 100 mM Na-acetate SLS [11] Ash Seikagaku 4.6 25 50 mM Na-acetate SLS [13] White circle Sigma 4.5 23 5 mM Na-Acetate SIC [30] Cross Sigma 4.5 25 20 mM Na-acetate SIC [35]

ar Seikagaku stN/A: Not available; SAXS: Small-anglelight scattering; SANS: Small angle n

X-ray scattering; LALLS: Low-angle laser light scattering; SLS: Static eutron scattering; SIC: Self-interaction chromatography, SEC: Size-

clusio

3.2.1.1. Membrane osmometry For a dilute protein solution, the Bmm is defined in ter

ex n chromatography

ms of the osmotic pressure, π, by the osmotic virial expansion [5]

1m mm mRTc B c

Mπ ⎛ ⎞= + +⎜ ⎟

⎝ ⎠ (3.1)

where R is the universal gas con T is the temperature in Kelvin, cm is the concentration of protein and M is the molecular the protein. For the description of a dilute protein solution, three-body or higher order interactions are neglected from the virial expansion

equation (3.1). Thus, a plot of

stant, weight of

mRTcπ verses cm is linear for a sufficiently low range of cm

v with the slope equal to Bmm and the intercept equal to 1/M. Alternatively, Bmm can also be measured from the osmotic pressure data at a single point of protein concentration, if the molecular weight and aggregation state of the protein in that solvent condition are known.

The inherent inaccuracy of Bmm values measured by MO comes from the inaccuracy in the osmotic pressure measurement and from the partial aggregation of the protein. Membrane osmometers of different manufacturers have an inherent inaccuracy of 0.5-1% in osmotic pressure measurement. According to the information provided by different suppliers,

alues,

41

Chapter 3

Type 3300 Micro Osmometer of John Morris Scientific and Advanced 3250 osmometer of Advanced Instruments, Inc. have a precision limit of 0.5%. Semi-Micro Osmometer Type Dig. L of KNAUER has a measurement accuracy of <1%. In the work of Haynes and coworkers [25], the average mean square error of the measured osmotic pressure at different solvent concentration and at different protein concentration was 0.94%. This apparently small inaccuracy in osmotic pressure measurement dramatically affects the Bm value. For a simple case of lyso

m

zyme (M 14600 g/mol) at a concentration of 5 mg/ml, a Bmm value 0.00 10-4

osmotic pressure measurement would shift the Bmm value from 0.00

×mol.ml.g-2

corresponds to an osmotic pressure of 849 Pa at 25 ˚C. Only a 1% inaccuracy in ×10 to ±1.37 10 -4 × -4

mol.ml.g-2.

The measured Bmm value is more reliable, if the measurement is done from the slope

of the plot of mRTc

π verses cm at multipoint protein concentration without any preliminary

knowledge of the molecular weight of the protein. However, an additional error may arise from he standard deviation of the slope. We have analyzed the data of Haynes and coworkers [25] and standard deviations of the slope corresponded to an error of ±0.6×10-4 mol.ml.g-2 in the calculated Bmm values. Haynes and coworkers [25] also ound an error m

t

f argin of ±0.4 10-4 mol.ml.g-2 only due to the mean square error in osmo easurement.

omes equal to ±1.0× tic pressure m

The overall error then bec ×10-4 mol.ml.g-2 ed that mm

. It is, therefore, concludB lues measured by MO have an inherent inaccuracy of ±1.0 va ×10-4 mol.ml.g-2.

3.2.1.2. Light scattering Macromolecular solutions scatter light due to the thermally induced fluctuation in local concentration. Using this property of protein solutions, light scattering techniques can be used to obtain the so called “static” parameters of a protein such as molecular weight, osmotic second virial coefficient, molecular dimensions and sometimes the radius of gyration. The ststic light scattering (SLS) or low-angle laser-light scattering (LALLS) experiment measures the average intensity of light scattered by a protein solution of defined concentration in excess of that scattered by the background solvent. Measurement of Bmm using this method relies on measuring the intensity of light scattered as a function of the protein concentration. Since protein molecules are usually much smaller than the wavelength of the incident light (<λ/20), their scattering intensity is independent of the scattering angle,

other words, within the Rayleigh limit. In this limit, thmm

in e Rayleigh ratio, Rθ, is by definition proportional to the scattered light intensity and is related to the M and B by the classic equation [5,39]

1 2mmm m

Kc B cR Mθ

= + (3.2)

where K is an optical or instrumental constant, which can be given by

22 2

04

4

A m

n dnKN dcπλ

⎛ ⎞= ⎜ ⎟

⎝ ⎠ (3.3)

42


where n0 is the refractive index of the solvent, m

dndc

is the refractive index increment for the

protein-solvent pair, NA is the Avogadro’s number and λ is the wavelength of the incident vertically polarized light in vacuum.

Equation (3.2) indicates that a plot of nKcRθ

verses cm allows the determination of M

and Bmm. The accuracy of the determined Bmm values however depends on the measurement

accuracy of Rθ, n0 and m

dndc

values as well as the linearity of the plot. Since light scattering in

the Rayleigh limit is isotropic, the values of Rθ in SLS are usually measured at 90˚. In order to determine the absolute R90 values for protein solutions, the SLS instrument is calibrated first using toluene or benzene. Toluene and benzene have an established R90 value of 1.406×10-5 cm-1 at 633 nm [8,15,17] and 3.86×10-5 cm-1

at 488 nm [4,9], respectively. The background scattering of the pure solvent alone is then subtracted first in order to measure the actual increment of scattering due to protein.

Solvents in Bmm determination system are usually salt/buffer solutions, whose

refractive index, n0, can be determined by a refractometer. The term m

dndc

is the change of the

solution’s refractive index with respect to a change in protein concentration, which can be measured using a differential interferometric refractometer. The difference in refractive index between the protein solution and the electrolyte solvent (Δn) is measured at the same wavelength at which light scattering is measured. To obtain the same chemical potential of salt and water in protein solutions as in the pure solvent, a tedious dialysis against the solvent

is rigorously required before measuring the Δn, which is often overlooked. The value of m

dndc

can then be measured by plotting Δn verses cm as

mm

dnn cdc

Δ = (3.4)

In order to study the sensitivity of the Bmm value on the measured Rθ value, we consider a simple case of light scattering intensity of lysozyme (M 14600 g/mol) at a concentration of 0.002 g/ml and at a wavelength of 633 nm. We also fix the values of n0 at 1.33 and at 1.81 ml/g, which eventually give an optical constant, K, of 2.37/ mdn dc ×10-7

mol.ml.cm-1.g-2. In this specific case, a Bmm value 0.00×10-4 mol.ml.g-2 corresponds to a Rθ value of 6.91×10-6 cm-1. Unfortunately, no information is available about the precision limit of the Rθ value measured by SLS or LALLS from the instrument suppliers. However, inconsistency in light scattering data is very common in practice. For this reason, the final data point is usually taken based on the average of at least 50 statistically consistent measurements. Disturbance from dust is a common source of error, which can be minimized by using the statistical dust rejection function and by setting a tight rejection ratio. We, therefore, assumed that the standard deviation of Rθ values would be about 0.05%.

43

Chapter 3

In addition to Rθ, small errors in the optical parameters, i.e. n0 and m

dndc

dramatically

affect the Bmm value because of their quadratic dependence. According to the information provided by different suppliers, most interferometric refractometers have a measurement

inaccuracy of ~1%. In practice, a 0.5% inaccuracy is quite normal in the eventual m

dndc

value.

For example, Rosenbaum and coworkers [13] found the value of lysozyme to be

1.181 ml.g-1 with an error bar of ±0.005 ml.g-1 (0.42%) at 633 nm. We, therefore,

approximate the standard deviation of measured n0 and

/ mdn dc

m

dndc

values to be 0.5%, which gives

an error of approximately 2% in the K value using equation (3.3).

A realistic error margin in Bmm as measured by light scattering techniques is obtained as follows. If Bmm is calculated from a single data point of lysozyme concentration considering a 0.05% error in Rθ values and a 2% error in K value, the overall error in Bmm value is as high as ±2.58×10-4 mol.ml.g-2. The error margin decreases with an increasing number of data points at different protein concentrations. For the case of 5 data points, the standard deviation of the slope corresponds to an error of about ±2.0×10-4 mol.ml.g-2 in the determined Bmm value. It is, therefore, quite realistic that a Bmm value measured by the light scattering technique may have an inherent inaccuracy of ±2.0×10-4 mol.ml.g-2. The accuracy limit or error bar is not presented in most published works. However, the extent of the error margin and the experimental reproducibility are obvious from the experimental approach of Curtis and coworkers [22]. They found an error in the M determination on the order of 400 g.mol-1 for lysozyme and 1200 g.mol-1 for ovalbumin. If we translate this error in terms of the Bmm value, the error could be as large as ±4.0×10-4 mol.ml.g-2. A similar error is also reported in other work [15]. Rosenbaum and coworkers [13] also mentioned an estimated uncertainty for the lysozyme Bmm measurement by their SLS experiment to be ±18 nm3, which is equivalent to ±0.5×10-4 mol.ml.g-2. 3.2.1.3. Size-exclusion chromatography Thermodynamic non-ideality of a solute leads to a concentration dependent partition coefficient in a non-interacting column [40]. Using this non-ideal behavior of a protein in a typical SEC column, a theoretical framework was established to determine Bmm by frontal-exclusion chromatography as [36,41]

0

ln 2 (1 )Dmm i D

K B MC KK

⎛ ⎞= −⎜ ⎟

⎝ ⎠ (3.5)

where Ci is the plateau value of the protein concentration in the mobile phase, K0 is the partition coefficient of protein at the limit of infinite dilution and KD is the local protein distribution coefficient, which is described as

( )( )

0

0

rsD

m t

V VCKC V V

−≡ =

− (3.6)

44


where Cs and Cm are the local protein concentrations in the pore/intraparticle volume (stationary phase) and interparticle volume (mobile phase), respectively, Vr is the average retention volume of the mobile protein, V0 is the extra-particle or interstitial volume and Vt is the total mobile phase volume. If ln(KD) is plotted as a function of , the slope of

the plotted line becomes equal to 2

(1 )iC K− D

mmB M . The Bmm determination procedure is then to inject

different concentrations of protein sample and measure KD from the retention volumes. Therefore, the accuracy of a determined Bmm value depends on the run-to-run deviation of retention volume for a fixed protein concentration and the linearity of ln(KD) verses

plot. (1 )iC K− D

Frontal-exclusion chromatography was, however, rarely used as a Bmm measurement technique because it requires a huge amount of protein and long experimental times to reach the plateau region. In contrast, a pulse SEC technique was recently developed [36], where the plateau value of the mobile phase protein concentration, Ci, was replaced by average protein concentration of the mobile phase in the pulse, iC⟨ ⟩

0

ln 2 (1 )Dmm i D

K B M C KK

⎛ ⎞= ⟨ ⟩ −⎜ ⎟

⎝ ⎠ (3.7)

For the error estimation in SEC, we consider a realistic case with V0 and Vt values of 6.07 and 10.77 ml, respectively. In such a SEC column, the Vr values vary from 9.85 to 9.88 ml for values from 0.70 to 2.95 mg/ml, respectively [36]. The random run-to-run

difference in retention volumes in this system is usually within the range of 0.1%. According to error propagation statistics, an error of only ±0.1% in the retention volumes corresponds to an error of ±0.0023 in the KD values and ±0.00285 in the ln(KD) values. Considering the fact of ±0.00285 errors in the ln(KD) values, our analysis shows huge error bars in every data point in the direction of Y axis in the

iC⟨ ⟩

iC⟨ ⟩ (1-KD) verses ln(KD) plot (Fig. 3.3). The standard

error of the slope due to Y-axis error bar was analyzed as described in the ref. [42]. Since every data point shows a huge error bar, the standard error of the slope is as large as 0.0083 ml/mg, which gives an error of about ±3.0×10-4 mol.ml.g-2 in the eventual Bmm value. Therefore, Bmm values measured by either pulse or frontal elution SEC system would have an inherent inaccuracy of ±3.0×10-4 mol.ml.g-2. The error limit could even be higher in a pulse SEC system, because the mobile phase protein concentration in a pulse system changes during transport down the column whose measurement difficulty could be an extra source of error. Since it is quite difficult to measure the accurate average protein concentration of the mobile phase in the pulse, , a reasonably realistic value of iC⟨ ⟩ iC⟨ ⟩ can be determined as the

total amount of protein in the pulse divided by the volume of the column accessible to the protein. Bloustine and co-workers [36] showed that the maximum concentration of protein in the eluted mobile phase, Cmax, can also be used instead of iC⟨ ⟩ . In such a situation an error of

few percentages in the values is quite optimistic. In practice, Cmax values differ from iC⟨ ⟩

45

Chapter 3

respective values by about 5% [36]. However, this error was neglected in this analysis

of error margin of of ±3.0×10-4 mol.ml.g-2. iC⟨ ⟩

Fig. 3.3. iC⟨ ⟩ (1-KD) versus ln(KD) plot

considering error margin of ±0.1% in the retention volumes. Solid line represents the trend line according to least-square fitting and dotted lines represent standard error of the trend line according to ref. [42]. This plot was generated using some data of ref. [36].

3.2.2. Determination of Bmm by SIC

3.2.2.1. Calculation of Bmm from SIC retention data Retention of a protein sample in a SEC column is typically characterized by the distribution coefficient, KSEC, which is given by

( ) ( )( )

0

0

r rSEC

i t

V V V VK

V V V− −

= =−

0 (3.8)

where Vr, V0 and Vt terms are described in equation (3.6). The term (Vt – V0) was assumed as Vi, which denotes the intra-particle pore volume. The KSEC term in equation (3.8) may vary from 0 to 1 for fully excluded large molecules to fully included small molecules, respectively.

SIC is essentially a quantitative affinity chromatography system, which estimates weak interactions of mobile protein molecules with immobilized protein molecules. Because of the weak and bi-directional (both attractive and repulsive) nature of protein self-interactions, the SIC system cannot be characterized in terms of association or dissociation constants. In the case of low protein load in the mobile phase, which is essentially the condition for SIC, the slope of the linear region of the adsorption isotherm is however related to the potential of mean force between protein molecules. The distribution coefficient in such a quantitative affinity chromatography system, Kaff, can be described as [38,43]

,r affaff

m

VqKc m

Δ= = (3.9)

where q is the amount of protein adsorbed per volume of resin, m is the amount of resin in terms of volume and the term denotes the changes in retention volume due to protein

self-interactions only. The distribution coefficient, Kaff, in equation (3.9) is a measure of protein-protein interaction. Positive values of Kaff correspond to higher retention of the mobile protein molecules due to attractive interaction with stationary molecules. Similarly,

,r affVΔ

46


negative values of Kaff may also arise from repulsive interaction between stationary and mobile protein molecules. In thermodynamic terms, Kaff can also be defined as [30,43]

( 1)

i

GkT

iVaff

i

e dK

V

Δ−

−=∫ V

(3.10)

where ΔG is the free energy change of bringing a protein molecule from the interstitial volume into the pore volume so that it interacts with a single immobilized protein molecule and k is the Boltzman constant. In a typical SIC system, we measure an overall partition coefficient, Koverall, which is contributed by both size-exclusion and weak-protein interactions. The overall partition coefficient can therefore be represented as overall SEC aff SECK K K K= + (3.11)

Only the Kaff term in equation (3.10) is related to protein-protein interaction, and eventually to Bmm. Since the retention volume in SEC column also varies with the injected protein concentration due to protein-protein interactions [36], the KSEC term as defined in equation (3.8) is also related to protein-protein interaction. However, the injected protein concentration in SIC is usually very dilute and in the linear region of adsorption isotherm. Changes of retention volume in a SEC column as a function of the injected protein concentration is, however, very little and within the range of run-to-run errors of retention volume. Therefore, any relation of KSEC to the protein-protein interactions is neglected. ΔG in equation (3.10) is essentially equivalent to the potential of mean force, W, which is described as the anisotropic interaction free energy required to bring two infinitely spaced solute molecules into a defined separation distance, r, averaged over all possible orientations of the solute molecules. The free energy change, ΔG, is also a function of intermolecular separation distance (r) and possible angular positions/orientations of both immobilized (Ω1) and mobile (Ω2) interacting molecules. However, 1 2( , , )G rΔ Ω Ω may not be equal to in the

sense that one mobile protein molecule may simultaneously interact with more than one immobilized molecule or among mobile molecules themselves or with the chromatography resin. In addition, the immobilized protein molecules may lose their rotational freedom and may not be accessible from different angular positions or orientations. Assuming that the experimental SIC system does not encounter these uncertainties,

1 2( , , )W r Ω Ω

1 2( , , )G rΔ Ω Ω becomes

equal to . 1 2( , , )W r Ω Ω

Considering all statistically possible orientations for both immobilized and free protein molecules, the distribution coefficient due to protein-protein interaction (Kaff) can be expressed as discussed by Tessier and co-workers [30].

1 2

1 2

( , , )2

2 101

W rkT

aff SECi

N e r dr dK K

V

Ω Ω−∞

Ω Ω

⎛ ⎞d− Ω Ω⎜ ⎟

⎝ ⎠=∫ ∫ ∫

(3.12)

where N is the total number of immobilized protein molecules accessible for mobile protein molecules. The lower limit of the separation integral in Eq. (3.12) was taken as zero rather

47

Chapter 3

than the center-to-center distance upon intermolecular contact, because the Kaff term has to be obtained from chromatography retention data. The nature of protein-protein interactions in the SIC column is similar to that in a real protein solution and the chromatographic retention data represents all sorts of interactions including the excluded volume contribution.

The osmotic second virial coefficient (Bmm) refers to the interaction between two protein molecules in solution and is rigorously related to the two-body potentials of mean force between protein molecules in solution [5,44]

1 1

1 2

( , , )2

201

W rkT

mm 1B e r dr dΩ Ω

−∞

Ω Ω

⎛ ⎞= − − Ω Ω⎜ ⎟

⎝ ⎠∫ ∫ ∫ d (3.13)

Comparing equation (3.11), (3.12), and (3.13)

( )SEC overall imm

K KB

N−

=V

(3.14)

Equation (3.14) can be used to calculate Bmm by the SIC methodology. KSEC can be determined using equation (3.8) for a protein-free column with the same resin material, where no protein-protein interaction can take place and the retention volume of the mobile protein is only determined by size-exclusion. In such a protein-free column, the value of Kaff is zero. The overall distribution coefficient, Koverall, can also be determined using equation (3.8) for an immobilized-protein column, where the retention of the mobile protein is guided by both size-exclusion and protein-protein interactions. Vi in equation (3.14) is equal to (Vt – V0) for the protein-immobilized column. N can be represented as the total amount of immobilized protein in gram. The unit of Bmm obtained in this approach is ml/g which has to be divided by the molecular weight (M) of the protein in order to obtain the usual unit of Bmm mol.ml.g-2. Therefore

( )SEC overall imm

K KB

NM−

=V

(3.15)

Our approach to the calculation of Bmm is comparable to that derived in previous studies [30,35] and essentially the same as that of Teske and co-workers [35] for the case of identically packed immobilized-protein and protein-free columns. However, the immobilized-protein column in our work does not necessarily have to be the same as the protein-free column in terms of column volume and packing integrity. In addition, our approach justifies the use of a protein-free column and does not require the retention volume in the theta condition (when mobile phase proteins have no net interactions with the immobilized proteins), which is quite impractical to determine accurately [34].

An extra hard sphere contribution term (BHS) is disappeared in this derivation because lower limit of the separation integral in equation (3.12) was taken as zero. The BHS contribution to Bmm is always positive and roughly equal to 6.7 times the molecular volume of the protein [45]. Considering the lysozyme molecule as a hard sphere of 3.11 nm, the excluded volume contribution is 4 times the molecular volume, which is equal to 1.84×10-4 mol.ml.g-2. The BHS term of the larger protein is even smaller in the unit of mol.ml.g-2.

48


3.2.2.2. Selection of stationary phase For the purpose of SIC, chromatograph particles with wide pores are desirable in order to minimize the mass transfer limitation and to ensure that immobilized protein molecules do not block the pore space for the mobile molecule to pass and interact outside the pore surface. In addition, the packed particles should not have any interaction with mobile phase protein. In this work, we used N-hyroxysuccinimide (NHS) activated Sepharose FF which consists of a 14-atom (6-aminohexanoic acid) spacer arm between the ligated protein molecules and the surface of the particle. This long spacer arm gives the immobilized protein molecule more flexibility to interact with the mobile phase protein from different angular positions and orientations. The mean particle size, pore diameter and porosity of Sepharose FF are 90 µm (provided by Amersham Biosciences), 50 nm [46] and 0.63 [46], respectively. The NHS groups react with N-terminal amino groups of the peptide chains and with the ε-amino groups of lysine residues [47], which provides random orientation of the immobilized protein molecules on the particle surface. In addition, the NHS-activated support shows very fast and complete binding of protein with comparatively minimal protein leakage during storage and chromatography [48]. The only minor disadvantage of NHS-Sepharose is that it may produce some anionic groups on its surface by hydrolysis of NHS, when the coupling reaction is done at high pH and/or high temperature. However, this problem can be minimized and the coupling reaction rate can be controlled in order to obtain the desired coupling concentration by conducting the reaction at pH 6.0 and at a temperature of 4 ºC [47]. 3.2.2.3. Optimization of the extent of immobilization

Tessier and co-workers [30,31] found that the retention of protein depends on the injected concentration at higher surface coverage (~33%), while the effect of injected concentration on retention volume is negligible at a surface coverage of 17-18% [30,35]. At very high surface coverage, a free protein molecule may have the opportunity to interact simultaneously with multiple immobilized molecules, which results in an injection concentration-dependent retention behavior. In addition, the higher surface coverage may block some particle pores, which hence become inaccessible to free proteins. We, therefore, controlled the immobilization process to work within a surface coverage of about 15%. The immobilization concentration of 20 mg lysozyme/ml of packed column corresponds to 15% surface coverage for NHS-activated Sepharose (Appendix). The incubation time, temperature, pH and protein concentration of the immobilization reaction mixture are the parameters for controlling the immobilization reaction. Only the incubation time was varied in this work and it was found that 12 hours of incubation was sufficient to obtain optimum coupling. 3.2.2.4. Optimization of injection sample

While a l5% surface coverage is used to avoid multi-body interaction, the protein concentration in the mobile phase has also to be low enough to avoid interaction among mobile phase proteins themselves. Since the range of protein self-interactions is very short [4] and the injected protein concentration in the SIC experiment is typically quite dilute, this is

49

Chapter 3

not a prominent matter of concern. However, in order to make sure that we are working in the linear region of the adsorption isotherm, the total amount of protein in an injection pulse must be much lower than the amount of immobilized protein in the stationary phase. In that case, the retention of the pulse should not vary with little fluctuations in injection concentration. Garcia and co-workers [33] found no significant difference in the peak position of lysozyme over an injection concentration range of 2-9 mg/ml. Teske and co-workers [35] also obtained retention times independent of the mobile phase protein concentration provided that this concentration is less than 0.25 mg/ml. On the other hand, Tessier and co-workers [30] observed that the retention time of lysozyme increases when the injection concentration goes below 5 mg/ml. Considering these variations in previous studies, we have studied the optimum range of injected concentrations and volumes in our system. Using an injection volume of 50 µl and concentrations of 1-5 mg/ml for a 1.2-1.4 ml column produced sharp Gaussian peaks with a fairly high detection limit for lysozyme. The peak position was independent of the injected protein concentration in the tested range (1-5 mg/ml). 3.2.2.5. Determination of K, KSEC and N K and KSEC can be determined from the immobilized-protein column and the protein-free column, respectively, using equation (3.8). It is therefore important to inject a pulse of non-interactive fully included and of fully excluded molecules in both columns. Acetone was selected as the fully included molecule because of its small size, delectability in UV at 280 nm and non-interactive nature. The optimum concentration of acetone was found to be 2% (v/v). On the other hand, the interstitial volume was determined by injecting a fully excluded large molecule, i.e. blue dextran (M ~2 mDa). Blue dextran has some interaction with the Sepharose particles and probably with the immobilized protein molecules at low ionic strength. Therefore, the blue dextran pulse was eluted in the presence of 1.0 M NaCl in order to eliminate any interactions with the particle surface and the immobilized proteins.

N is the number of immobilized protein molecules accessible for mobile phase protein. The number of protein molecules immobilized per volume of settled particles can be determined first, then multiplied by the amount of settled particles used to pack the column. Considering the fact that the entire pore spaces in Sepharose FF are not wide enough for the protein molecules [46], it was nevertheless assumed that all the immobilized protein molecules are accessible to mobile molecules. Since only 15% surface coverage is applied, the remaining 85% of the pore surface is still free. Therefore, the immobilized protein molecules entered into the pore spaces are not expected to restrict the entry of mobile molecules. Such a restriction can only be expected in the case of high surface coverage or monolayer coverage. Therefore, the value of N is equal to the total number of protein molecules immobilized onto the stationary phase.

50


3.3. Materials and methods

3.3.1. Materials

Lysozyme from chicken egg white (3x crystallized, dialyzed and lyophilized) was bought from Sigma-Aldrich Co. NHS-activated SepharoseTM 4 Fast Flow was purchased from Amersham Biosciences.

Acetic acid, sodium chloride, hydrochloric acid (36-38%), sodium hydrogen carbonates, potassium hydrogenphosphate anhydrous, potassium dihydrogenphosphate, acetone, and sodium hydroxide were bought from J.T. Baker. Sodium hydrogen phosphate dihydrate and sodium dihydrogen phosphate dodecahydrate were bought from Merck. Ethanolamine, tween 80, blue dextran, and magnesium bromide hexahydrate were bought from Aldrich. Bicinchoninic acid (BCA) protein assay reagents were bought from Pierce.

A TricornTM 5/50 column and TricornTM 5 adapter unit was bought from Amersham Biosciences. Chromatography experiments were done in a Pharmacia FPLC system, which was controlled by Unicorn version 2.0. Ultracentrifugation experiments were done by a Beckman L-70 ultracentrifuge with a Ti-60 rotor type. Normal centrifugation was done in a Beckman GP centrifuge. All spectrophotometric analyses were done in a Pharmacia spectra UV/Visible spectrophotometer. 3.3.2. Immobilization of protein

An amide linkage is formed between the amino groups of the protein and the NHS activated group of Sepharose in the pH range of 6-9. The coupling reaction is very fast and almost uncontrollable at higher pH and temperature [47]. In addition, NHS groups are hydrolyzed rapidly at higher pH to give free COO- groups, which makes the particle a weak cation exchanger [48]. In order to avoid this undesirable side reaction, the coupling reaction was done at pH 6.0 and at 4 ˚C. The protein solution was prepared first at a concentration of 5 mg/ml in the coupling buffer (0.1 M sodium phosphate, 0.5 M NaCl, pH 6.0). Isopropanol suspended particles were washed 5 times with ice cold 1 mM HCl by centrifugation. Three milliliters of washed particles were incubated with 10 ml of 1 mM HCl for 15 minutes at 4 °C for swelling. HCl was removed from the settled particles and immediately replaced by 10 ml of pre-prepared protein solution. The coupling reaction was allowed to proceed at 4 ˚C with gentle shaking. The desired surface coverage of protein was obtained by manipulating the incubation time. The coupled particles were then washed a few times with ice cold coupling buffer to remove unbound proteins and released NHS. The immobilized particles were incubated again with 10 ml of blocking buffer (1 M ethanolamine, 0.5 M NaCl, 0.1 M Na-phosphate, pH 6.0) for 12 hours at 4 °C in order to block any remaining reactive groups. The protein-free particles were prepared in the same way without doing the coupling reaction.

Since NHS, released during the coupling reaction, has a very high UV absorption at 280 nm, it was not possible to determine the amount of immobilized protein from the amount of protein in the wash out solutions. The density of protein immobilized in the particle was therefore determined by a standard BCA technique [49], applied to the solid phase [50].

51

Chapter 3

3.3.3. Chromatography and data analysis Tricorn 5/50 (5 x 50 mm) columns were packed with both immobilized-protein and protein-free particles at a flow rate of 3 ml/min for at least 10 min. The flow rate was subsequently reduced to 1 ml/min for confirming the stability of the bed. The integrity of the packed column was characterized by height equivalent to a theoretical plate analysis, and peak shape and symmetry of a pulse of a small molecule, for instance acetone and/or high salt. The chromatography procedure was accomplished as described by Tessier and co-workers [30] in an automated Pharmacia FPLC system controlled by Unicorn version 2.0. The injection sample was prepared at a concentration of 1-2 mg/ml, unless mentioned otherwise. The column was equilibrated with the appropriate solution until the UV, pH and conductivity base lines became completely straight before every injection. Retention volumes were automatically determined by Unicorn as the peak position. The column was stored at 4 °C in 10 mM sodium phosphate (pH 7.0), when not in use. Each column was used for a period of maximum 4 weeks. 3.4. Results and discussion

3.4.1. Inherent inaccuracy of SIC

A theoretical framework for determination of Bmm by SIC is presented in section 3.2.2. According to equation (3.15), the uncertainty in Bmm may come from errors in the estimation of pulse retention volumes in the columns and in the determination of immobilized protein concentration on the stationary phase. In our experimental set-up, the reproducibility of retention volume was within the limit of ±0.01 ml and the maximum inaccuracy in determining the immobilized protein concentration on the gel particle was ±20%. These amounts of error in V and N in equation (3.15) yield an overall error of maximum ±1.0×10-4 mol.ml.g-2 in the calculated Bmm value. A comparative overall error analysis of the different techniques is shown in Fig. 3.4.

Fig. 3.4. Inherent inaccuracy limits of different Bmm measurement techniques.

52


3.4.2. How efficient is the SIC technology compared to other techniques? The efficiency of a Bmm measurement technique is determined by the amount of protein and experimental time required to measure one Bmm value. Although a fairly good estimation can be made regarding the amount of protein required to determine one Bmm value, the estimation becomes complicated regarding the time needed to obtain one Bmm. This is because each method has its own long preparation time, data acquisition difficulties and extraneous complications. In reality, in all of these troubleshooting consume most of the time compared to the real data acquisition experimentation.

In MO, a micro-osmometer may have sample volume as low as 20 µl. Five different concentrations are usually required to compare the / mRTcπ verses cm plot. We assume

concentration levels of 2, 4, 6, 8 and 10 mg/ml. Therefore, the minimum amount of protein required to measure one Bmm value is 0.6 mg. Regarding time requirement, once the installation of the osmometry equipment is completed, the duration of each osmotic pressure measurement is about 15 min [25]. MO, however, suffers from some practical problems, for instance fouling and adsorption. Troubleshooting of these difficulties require unusually long time, which are case oriented and difficult to estimate.

In light scattering, 1 ml of sample volume is usually required to place in the sample cell. If we consider 5 data points with the same protein concentration as estimated for MO, total amount of protein required to measure one Bmm value is 6 mg. However, light scattering measurement is not usually run-to-run consistent and require a large number of replications in order to validate one data point. Sample preparation and the ability of the light scattering equipment to measure scattering intensity rapidly over a range of protein concentrations are great challenges. If we assume that all experiments run perfectly and provide acceptable data, 15 min is usually enough to measure the light scattering intensity and the refractive index of a sample.

Five different concentrations of protein are usually injected in a SEC column to obtain a linear relationship in the ln(KD) verses (1 )iC KD− plot. Each protein concentration has to be

far apart from the next ones in order to provide better resolution and consequently higher accuracy. For a 20 µl pulse injection with protein concentrations of 10, 20, 30, 40 and 50 mg/ml, the total protein requirement is 3.0 mg. The elution time of one pulse can be assumed to be 25 min. However column preparation, characterization, and equilibration take much longer time than the peak elution time.

If the experiment is done in a frontal-exclusion system in a 1 ml volume column, the injected sample volume has to be at least 1 ml in order to reach the plateau stage. In a frontal elution system, protein concentrations of 2, 4, 6, 8 and 10 mg/ml would be enough to provide good resolution. The amount of protein required is then 30 mg in order to obtain one Bmm value. The elution time will also be at least twice as long as for a pulse system.

Measurement of Bmm by SIC requires at least 2-pulse injections, one on an immobilized-protein column and another on a protein-free column. Total experimental time

53

Chapter 3

for each pulse elution is not more than 25 min for our current set-up. Typically each pulse contains 50 µl of a protein solution at a concentration of 2 mg/ml. Therefore, two pulses contain a total of 0.2 mg of protein. In addition, optimization of the injection concentration and flow rate requires several pulses at different protein concentration and different flow rate level, which eventually costs more time and protein. A comparatively large amount of protein (about 30 mg) is required to prepare an immobilized-protein column, which can be used for a month for the determination of hundreds of Bmm values. We assume that an immobilized-protein column is used to determine 160 Bmm values (20 days × 8 measurements per day) throughout a month. Therefore, an average of minimum 0.45 mg of protein is required to determine a Bmm value by the SIC technique. However the SIC method can easily be miniaturized to microchip level, thereby lowering the analysis time and the protein requirement by orders of magnitude. 3.4.3. Mapping of lysozyme Bmm profile Fig. 3.2 shows that the accuracy of Bmm data obtained by SIC at pH 4.5 is quite good in comparison to other techniques. However, we have extended our work on lysozyme in order to investigate the reproducibility of the SIC technique in different gel materials, protein immobilization strategies and solution conditions.

The stationary phase and the protein immobilization strategy used in this work were quite different from those in previous studies [30,35]. We have used narrow pore-size particles where protein immobilization took place via a spacer arm. The result obtained in our work shows that the SIC system is able to reproduce Bmm data irrespective of the type of stationary phase and protein immobilization strategy (Fig. 3.5). We have also found for our system that a column packed with immobilized lysozyme can be used as long as microorganisms do not degrade it. Microbial degradation deteriorates the packing integrity of the column and the column can no longer produce sharp Gaussian peaks. However, the lifetime of SIC columns can be different depending upon the stability the protein immobilized on it.

Fig. 3.5. Bmm of lysozyme measured by SIC using different gel materials at pH 4.5 and at a temperature of 23-25 ˚C. The line represents Sepharose from Amersham Biosciences (this work), the circles represent Toyopearl from Tosoh Bioscience [30] and crosses represent cross-linked agarose from Sigma [35].

54


The experimental approach was further extended to calculate Bmm values in conditions available from the literature and in some unknown conditions. It was found earlier that Bmm trends of lysozyme are quite ideal, decrease smoothly with pH and ionic strength [11], but increase proportionally with temperature [3,20]. The next mapping of Bmm was done at pH 7.6 in the presence of 10 mM Na-phosphate buffer. There are remarkable variations in published Bmm values for this condition (Fig. 3.6). The data obtained in this work fell below SLS data available in the literature. A notable feature of Fig. 3.6 is that the Bmm values obtained by SIC at higher NaCl concentrations (>0.5 M) are well below than that of SLS. The reason for this behavior can be explained as simultaneous interaction of a mobile-phase molecule with two or more immobilized molecules [35]. The overall trend of Bmm as a function of NaCl concentration is obviously due to electrostatic interaction. Since the pI of lysozyme is quite high (11.2), electrostatic repulsion is more prominent during self-interaction at low pH and low salt. Lysozyme is almost changeless at pH 9, where short-range attractions play the vital role.

It is known that Bmm of lysozyme in MgBr2 shows a minimum at 0.3 M MgBr2 at pH ~7.6 [11]. This phenomenon was confirmed by the SIC method [30]. The phenomenon was further confirmed in this work and the minimum was found at 0.4 M instead of 0.3 M MgBr2

(Fig. 3.7). However, the Bmm of lysozyme does not change significantly with the MgBr2 concentration. Tessier and co-workers [30] explained this phenomenon as an increase in repulsion at higher MgBr2 concentrations due to binding of the divalent Mg2+ to the acidic residues of lysozyme. In order to correctly determine the reason of this behavior, Bmm was also measured in MgCl2.

Fig. 3.7. Bmm trend of lysozyme as a function of MgBr2 concentration at pH 7.6. Triangle: pH 7.6, 10 mM Na-phosphate, SIC [this work]; Diamond: pH 7.8, 5 mM bis-tris, SIC [26]; Rectangle: pH 7.8, 20 mM HEPES, 23 ˚C, SLS [11].

Fig. 3.6. Bmm trend of lysozyme at pH 7.6. Black diamond: pH 7.5, 25 ˚C, SLS [9]; White diamond: pH 7.4, 25 ˚C, SLS [8]; White rectangle with line: pH 7.6, 10 mM Na-Phosphate, SIC [this work]; Black triangle with dashed line: pH 7.0, 5 mM bis-tris, SIC [30]; White triangle with dashed line: pH 7.6, 20 mM Na-Phosphate, SIC [35].

55

Chapter 3

The Bmm trend was found to be steadily decreasing with MgCl2 at pH 4.5 (Fig. 3.8). However, there is no literature data available for this condition except one point. The available data point [13] is lower than that found in this work, probably because of lower ionic strength of the buffer. The Bmm trend obtained in MgCl2 is comparable with that in NaCl for pH 4.5. The trend line in MgCl2 is slightly lower than that in NaCl. The reason is the presence of more electrolytes at equal molarity in MgCl2 than in NaCl because of the divalency of magnesium. The Bmm trend of lysozyme was also determined in MgCl2 at pH 7.6 (10 mM Na-phosphate) and we found that the Bmm does not change much with MgCl2 concentration (Fig. 3.9). Instead of a minimum, a maximum was found at 0.2-0.3 M MgCl2. The trend was further going down with increasing MgCl2 concentration. Since the effect of MgCl2 and MgBr2 on the Bmm trend is not very large at pH 7.6, it is hard to determine a minimum or maximum point at a particular ionic strength. It is therefore clear that the Bmm trend of lysozyme in MgCl2 is similar to that of NaCl but differs between MgCl2 and MgBr2. A likely explanation why trends in MgCl2 are not similar to MgBr2 is the chloride binding affinity of lysozyme. Lysozyme does not exhibit salting in behavior with NaCl due to predominant electrostatic screening of the positively charged protein and/or by adsorption of chloride ions by the protein [51]. Lack of this phenomenon in presence of bromide salt produces a downward peak in the Bmm trend.

Fig. 3.8. Bmm trend of lysozyme as a function of MgCl2 concentration at pH 4.5. Line: pH 4.5, 10 mM Na-acetate, SIC [this work]; Rectangle: pH 4.6, 50 mM Na-acetate, 25 ˚C, SLS [13].

Fig. 3.9. Bmm trend of lysozyme as a function of MgCl2 concentration at pH 7.6 (10 mM Na-Phosphate).

3.4.4. Crystallization slot George and Wilson [2] determined Bmm values of nine different proteins by SLS at their crystallization conditions and found them to be within a narrow range (between -0.8×10-4 to -8.4×10-4 mol.ml.g-2), regardless of the size and nature of the proteins. The pattern of Bmm for non-crystallization solvent conditions was not studied in detail. It was, however, observed for lysozyme that solution conditions corresponding to positive and highly

56


negative Bmm values promote no phase separation and amorphous precipitation, respectively. After further investigation of a few other proteins, they defined the so called “crystallization slot” as the range of Bmm values between -1×10-4 to -8×10-4 mol.ml.g-2 [3]. This Bmm based crystallization slot was used thereafter for predictive crystallization of lysozyme [9], chymotrypsinogen [9,27], ribonuclease A [32], myoglobin [31] and OmpF porin [12]. Bmm clearly has a predictive value for the conditions of protein crystallization. The question one could ask is whether this range of Bmm values -1×10-4 to -8×10-4 mol.ml.g-2 applies to all kinds of proteins, regardless of their size, shape, charge, hydrophobicity and surface roughness. In order to explore the versatility and applicability of Bmm as a predictor of protein phase behavior, crystallization conditions of known proteins were mapped from available literature in terms of Bmm. Bmm values of all of these proteins (Fig. 3.10) at their crystallization conditions fell fairly within the range of -1×10-4 to -8×10-4 mol.ml.g-2. Fig. 3.10 confirms that it is important to have a Bmm value within the crystallization slot for crystallization of any protein. However, does a Bmm value within the crystallization slot guarantee successful production of protein crystals? An extended Bmm mapping was therefore done for proteins, of which the conditions of crystallization, amorphous precipitation and no phase separation were available from the literature (Fig. 3.11). Fig. 3.11 supports Fig. 3.10 in the sense that crystals do not grow in conditions at which the Bmm value is either positive or largely negative. Interestingly however, crystals were not obtained in a number of cases where the Bmm values were within the crystallization slot.

Fig. 3.10. Bmm map of different proteins at their crystallization conditions. Solid rectangle: canavalin [2,3]; open rectangles: concanavalin A [2,3]; solid diamonds: bovine serum albumin [2,3]; open diamond: ovostatin [2,3]; solid circle: α-chymotrypsin [2,3]; open circle: satellite tobacco mosaic virus [2,3], solid triangle: ovalbumin [2,3]; open triangle: α-lactalbumin [3]; plus: β-lactoglobulin A [3]; cross: β-lactoglobulin B [3]; star: pepsin [3]; shaded diamond: thaumatin [3]; shaded rectangles: OmpF porin [12].

We have also conducted ultracentrifugal crystallization experiments with lysozyme in six solution conditions, out of which three conditions correspond to crystallization, one corresponds to amorphous precipitation and the remaining two correspond to no phase change (Table 3.2). According to the previously described models [52,53], for an initial protein concentration of 5 mg/ml and rotational speed of 45,000 rpm, 8 hrs of untracentrifugation was enough to produce crystals in our system (Beckman Ti 60 rotor). After finishing the ultracentrifugation, about 80% of the supernatant was removed gently with a pipette. The remaining solution and pellet were examined visually for the presence of crystals or precipitate. It was unexpectedly found that no phase separation occurred in two samples at pH 7.6 where Bmm values were –4.3×10-4 mol.ml.g-2 and –8.8×10-4 mol.ml.g-2. In

57

Chapter 3

our experiment pH 7.6 was buffered using K-phosphate, which seems to be unfavorable for growing lysozyme crystals. It was also previously found that phosphate and sulfate ions are comparatively less effective for crystallization of lysozyme [13,54].

Fig. 3.11. Phase behavior of proteins as a function of Bmm. Solid, shaded and white symbols denote crystal, precipitate and no change, respectively. Dashed lines denote the upper and lower boundaries of the crystallization slot. (a) Chymotrypsinogen: rectangular symbols are from ref. [9] and diamond symbol are from ref. [27]. (b) Ribonuclease A: rectangular symbols are from ref. [2,3] and diamond symbols are from ref. [32]. (c) Myoglobin: all symbols are from ref. [31]. (d) Lysozyme: rectangular symbols are from ref. [2], diamond symbols are from ref. [9] and triangular symbols are from ref. [3]. It is, therefore, fair to conclude that Bmm values within the range of 0 to -10-3 mol.ml.g-

2 are thermodynamically favorable for protein crystallization but do not guarantee successful crystal growth. On the other hand, protein crystallization is difficult or impossible at a condition where the Bmm value is positive. Successful crystal growth may depend on several other parameters, for instance solubility and the effect of specific ions. Several authors investigated whether any direct relationship exist between protein solubility and Bmm [3,11,13,16,20,23,55]. Their outcome suggests that a simple correlation may exist, but the

58


relationship is not strong enough to design crystallization experiments. Bmm may not sufficiently account for all interactions that are reflected in solubility, especially protein-salt interactions [21]. In addition, the crystallization process is significantly affected by the effect of specific ions. Bmm of lysozyme decreases with increasing chloride ionic strength. However, the presence of phosphate and sulfate as buffering salts is not favorable for lysozyme crystallization even though the Bmm value is driven into the crystallization slot by extra chloride. Indeed, the solubility of lysozyme is also very high in the presence of phosphate and sulfate ions [55]. Similarly, the solubility of lysozyme is the lowest in buffers containing Na+ salts compared to other cations at equal ionic strength [56]. Therefore, in addition to Bmm, the successful design of crystallization experiments may require solubility data and the knowledge of the effect of specific ions on that protein. However, Bmm is the preliminary guide for systematic screening of protein crystallization conditions in the sense that it must be in the slightly negative regime for crystallization is likely to occur. Table 3.2. Ultracentrifugal crystallization of lysozyme from Bmm aided prediction.

Sample No.

Solution condition Bmm (10-4 mol.ml.g-2)

After ultracentrifugation

A pH 4.5 (0.01 M Na-acetate), 20 ˚C > 30.0 None B pH 4.5 (0.01 M Na-acetate), 0.51 M NaCl, 20 ˚C ~ -2.0 Crystal C pH 4.5 (0.01 M Na-acetate), 0.86 M NaCl, 20 ˚C ~ -4.0 Crystal D pH 7.6 (0.01 M K-phosphate), 20 ˚C ~ 0.0 None E pH 7.6 (0.01 M K-phosphate), 0.17 M NaCl, 20 ˚C ~ - 4.3 None F pH 7.6 (0.01 M K-phosphate), 0.86 M NaCl, 20 ˚C ~ - 8.8 None A molecular or thermodynamic understanding why a particular range of Bmm values promotes crystallization was described in the literature [2-4,6,16,57]. Here we recall that negative values of Bmm indicate that attractive forces between protein molecules are dominant and protein-solvent interactions are less favored than those between protein molecules. A positive value for Bmm does not completely exclude the possibility of crystallization, but typically requires impractically high concentration of protein in order to bring about any kind of phase separation and the probability of obtaining acceptable crystals is very low. For the negative regime of the Bmm map, Wilson [58] discriminated between craggs and praggs. Craggs are highly structured microcrystalline aggregates formed at slightly negative Bmm. On the other hand, praggs are non-specific aggregates formed at highly negative Bmm, which usually leads to amorphous structures. 3.5. Conclusion A theoretical framework was established to correlate self-interaction chromatography retention data with Bmm value. The approach requires retention data from an immobilized-protein column as well as from a protein-free column for the determination of a Bmm value. However, the protein-free column does not necessarily have to be the same as the immobilized-protein column in terms of column volume and packing integrity. Details of the

59

Chapter 3

chromatography methodology, troubleshooting and data analysis approaches were designed. The reproducibility and accuracy limit of the Bmm data by the SIC methodology was discussed in comparison to other traditional techniques, and SIC was shown to performs in a superior way. The SIC methodology can even be improved further by miniaturization to microchip level. The designed methodology was applied for Bmm mapping of a model protein, i.e. lysozyme. It was also found that a Bmm value within the crystallization slot is an essential prerequisite of crystallization, but does not guarantee successful crystal growth. In addition to protein-protein interaction, protein solubility and the effect of specific ions also play a vital role for successful crystallization of protein, by mechanism that is not completely understood.

Appendix: Calculation of Surface Coverage The accessible surface area per volume of packed column, also called phase ratio, of a typical chromatography media decreases with increasing mobile phase particle size. The circumradius of the lysozyme molecule is 1.56 nm. An estimate of the phase ratio of Sepharose FF for lysozyme can be obtained from the data of DePhillips and Lenhoff [46]. Although the material used in this project was neither SP nor CM Sepharose, an approximation can be made for NHS-Sephasore using this data. Interpolating the data in ref [46] the phase ratio for lysozyme is approximately 42.5 m2/ml. In order to obtain 15% surface coverage, the required immobilization concentration is 20 mg of lysozyme/ml of settled particle for lysozyme. Symbols Bmm Osmotic second virial coefficient Ci Plateau value of protein concentration in the mobile phase cm Concentration of protein (in the mobile phase) Cs Local protein concentration in the pore/intraparticle volume (stationary phase) <Ci> Protein concentration in the pulse K Optical or instrumental constant of light scattering equipment k Boltzman constant K0 Protein distribution coefficient at the limit of infinite dilution Kaff Distribution coefficient in quantitative affinity chromatography KD Local protein distribution coefficient in size-exclusion chromatography Koverall Overall distribution coefficient due to size-exclusion and self-interaction KSEC Distribution coefficient in size-exclusion chromatography M Molecular mass m Amount of resin N Number of immobilized protein molecule NA Avogadro’s number n Refractive index of solution n0 Refractive index of solvent q Amount of protein adsorbed per volume of resin R Universal gas constant r Intermolecular separation distance Rθ Rayleigh ratio/angle T Temperature in Kelvin

60


V0 Extra-particle or interstitial volume of column Vi Intra-particle (pore) volume of column Vr Retention volume of protein Vt Total mobile phase volume of a column W Potential of mean force π Osmotic pressure λ Wavelength of light Δ Change or increment in another parameter ΔG Gibbs free energy change of bringing a protein molecule from the interstitial to pore volume ΔVr,aff Changes in retention volume due to protein self-interaction only Ω1 Angular position/orientation of immobilized molecule Ω2 Angular position/orientation of mobile molecule References [1] A. McPherson. Methods Enzymol. 114 (1985) 112-120. [2] A. George & W.W. Wilson. Acta Cryst. D50 (1994) 361-365. [3] A. George, Y. Chiang, B. Guo, A. Arabshahi, Z. Cai & W.W. Wilson. Methods Enzymol. 276

(1997) 100-110. [4] B.L. Neal, D. Asthagiri, O.D. Velev, A.M. Lenhoff & E.W. Kaler. J. Cryst. Growth 196 (1999)

377-387. [5] W.G. McMillan & J.E. Mayer. J. Chem. Phys. 13 (1945) 276-305. [6] B.L. Neal, D. Asthagiri & A.M. Lenhoff. Biophys. J. 75 (1998) 2469-2477. [7] D. Asthagiri, B.L. Neal & A.M. Lenhoff. Biophys. Chem. 78 (1999) 219-231. [8] D.F. Rosenbaum & C.F. Zukoski. J. Cryst. Growth 169 (1996) 752-758. [9] O.D. Velev, E.W. Kaler & A.M. Lenhoff. Biophys. J. 75 (1998) 2682-2697. [10] P. Prinsen & T. Odijk. J. Chem. Phys. 121 (2004) 6525-6537. [11] B. Guo, S. Kao, H. McDonald, A. Asanov, L.L. Combs & W.W. Wilson. J. Cryst. Growth 196

(1999) 424-433. [12] C. Hitscherich, J. Kaplan, M. Allaman, J. Wiencek & P.J. Loll. Protein Sci. 9 (2000) 1559-

1566. [13] D.F. Rosenbaum, A. Kulkarni, S. Ramakrishnan & C.F. Zukoski. J. Chem. Phys. 111 (1999)

9882-9890. [14] R. Piazza & M. Pierno. J. Phys.: Condens. Matter 12 (2000) A443-A449. [15] M. Muschol & F. Rosenberger. J. Chem. Phys. 103 (1995) 10424-10432. [16] D. Rosenbaum, P.C. Zamora & C.F. Zukoski. Phys. Rev. Lett. 76 (1996) 150-153. [17] W.W. Wilson. J. Struct. Biol. 142 (2003) 56-65. [18] H.V. Porchel & G. Damaschun. Stud. Biophys. 62 (1977) 69. [19] A. Ducruix, J.P. Guilloteau, M. Ries-Kautt & A. Tardieu. J. Cryst. Growth 168 (1996) 28-39. [20] F. Bonnete, S. Finet & A. Tardieu. J. Cryst. Growth 196 (1999) 403-414. [21] R.A. Curtis, J.M. Prausnitz & H.W. Blanch. Biotechnol. Bioeng. 57 (1998) 11-21. [22] R.A. Curtis, J. Ulrich, A. Montaser, J.M. Prausnitz & H.W. Blanch. Biotechnol. Bioeng. 79

(2002) 367-380. [23] C. Gripon, L. Legrand, I. Rosenman, O. Vidal, M.C. Robert & F. Boue. J. Cryst. Growth 178

(1997) 575-584. [24] V. Receveur, D. Durand, M. Desmadril & P. Calmettes. FEBS Lett. 426 (1998) 57-61.

61

Chapter 3

62

[25] C.A. Haynes, K. Tamura, H.R. Korfer, H.W. Blanch & J.M. Prausnitz. J. Chem. Phys. 96 (1992) 905-912.

[26] H.M. Schaink & J.A.M. Smit. Phys. Chem. Chem. Phys. 2 (2000) 1537-1541. [27] P.E. Pjura, A.M. Lenhoff, S.A. Leonard & A.G. Gittis. J. Mol. Biol. 300 (2000) 235-239. [28] J. Behlke & O. Ristau. Biophys. Chem. 76 (1999) 13-23. [29] S.Y. Patro & T.M. Przybycien. Biotechnol. Bioeng. 52 (1996) 193-203. [30] P.M. Tessier, A.M. Lenhoff & S.I. Sandler. Biophys. J. 82 (2002) 1620-1631. [31] P.M. Tessier, S.D. Vandrey, B.W. Berger, R. Pazhianur, S.I. Sandler & A.M. Lenhoff. Acta

Cryst. D58 (2002) 1531-1535. [32] P.M. Tessier, H.R. Johnson, R. Pazhianur, B.W. Berger, J.L. Prentice, B.J. Bahnson, S.I.

Sandler & A.M. Lenhoff. Proteins: Struct. Funct. Gen. 50 (2003) 303-311. [33] C.D. Garcia, S.C. Holman, C.S. Henry & W.W. Wilson. Biotechnol. Prog. 19 (2003) 575-579. [34] C.D. Garcia, D.J. Hadley, W.W. Wilson & C.S. Henry. Biotechnol. Prog. 19 (2003) 1006-1010. [35] C.A. Teske, H.W. Blanch & J.M. Prausnitz. J. Phys. Chem. B 108 (2004) 7437-7444. [36] J. Bloustine, V. Berejnov & S. Fraden. Biophys. J. 85 (2003) 2619-2623. [37] P.M. Tessier & A.M. Lenhoff. Curr. Opin. Biotechnol. 14 (2003) 512-516. [38] C.A. Teske, H.W. Blanch & J.M. Prausnitz. Fluid Phase Equilib. 219 (2004) 139-148. [39] B. Zimm, H. J. Chem. Phys. 16 (1948) 1093-1099. [40] P.J. Wyatt. Anal. Chem. Acta 272 (1993) 1-40. [41] L.W. Nichol, R.J. Siezen & D.J. Winzor. Biophys. Chem. 9 (1978) 47-55. [42] J.R. Taylor. An Introduction to Error Analysis: The Study of Uncertanities in Physical

Measurements, University Science Books, Mill Valley, 1982. [43] J. Staahlberg, B. Joensson & C. Horvath. Anal. Chem. 63 (1991) 1867-1874. [44] B.H. Zimm. J. Chem. Phys. 14 (1946) 164-179. [45] B.L. Neal & A.M. Lenhoff. AIChE J. 41 (1995) 1010-1014. [46] P. DePhillips & A.M. Lenhoff. J. Chromatogr. A 883 (2000) 39-54. [47] P. Cuatrecasas & I. Parikh. Biochemistry 11 (1972) 2291-2299. [48] A.P.G. van Sommeren, P.A.G.M. Machielsen & T.C.J. Gribnau. J. Chromatogr. A 639 (1993)

23-31. [49] P.K. Smith, R.I. Krohn, G.T. Hermanson, A.K. Mallia, F.H. Gartner, M.D. Provenzano, E.K.

Fujimoto, N.M. Goeke, B.J. Olson & D.C. Klenk. Anal. Biochem. 150 (1985) 76-85. [50] A.L. Plant, L. Locascio-Brown, W. Haller & R.A. Durst. Appl. Biochem. Biotechnol. 30 (1991)

83-98. [51] P. Retailleau, M. Ries-Kautt & A. Ducruix. Biophys. J. 73 (1997) 2156-2163. [52] V.V. Barynin & V.R. Melik-Adamyan. Sov. Phys. Crystallogr. 27 (1982) 588-591. [53] A.M. Lenhoff, P.E. Pjura, J.G. Dilmore & T.S. Godlewski. J. Cryst. Growth 180 (1997) 113-

126. [54] M.M. Ries-Kautt & A.F. Ducruix. J. Biol. Chem. 264 (1989) 745-748. [55] S. Ruppert, S.I. Sandler & A.M. Lenhoff. Biotechnol. Prog. 17 (2001) 182-187. [56] M.L. Broide, T.M. Tominc & M.D. Saxowsky. Phys. Rev. E 53 (1996) 6325-6335. [57] M. Malfois, F. Bonnete, L. Belloni & A. Tardieu. J. Chem. Phys. 105 (1996) 3290-3300. [58] W.W. Wilson. Methods 1 (1990) 110-117.

4

Phase behavior of an intact monoclonal antibody

Abstract Understanding protein phase behavior is important for purification, storage and stable formulation of protein drugs in the biopharmaceutical industry. Glycoproteins, such as monoclonal antibodies (MAbs) are the most abundant biopharmaceuticals and probably the most difficult to crystallize among water-soluble proteins. This study explores the possibility of correlating osmotic second virial coefficient (Bmm) with the phase behavior of an intact MAb, which has so far proven impossible to crystallize. The phase diagram of the MAb is presented as a function of the concentration of different classes of precipitants, i.e. NaCl, (NH4)2SO4 and polyethylene glycol. All these precipitants show a similar behavior of decreasing solubility with increasing precipitant concentration. Bmm values were also measured as a function of the concentration of the different precipitants by self-interaction chromatography and correlated with the phase diagrams. Correlating phase diagrams with Bmm data provides useful information not only for a fundamental understanding of the phase behavior of MAbs, but also in understanding the reason why certain proteins are extremely difficult to crystallize. The scaling of the phase diagram in Bmm units also supports the existence of a universal phase diagram of a complex glycoprotein when it is recast in a protein interaction parameter.

Keywords: Precipitation; Crystallization; Phase diagram; Solubility; Osmotic second virial coefficient; Self-interaction chromatography ___________________________________________________________________________ Published as T. Ahamed, B.N.A. Esteban, M. Ottens, G.W.K. van Dedem, L.A.M. van der Wielen, M.A.T. Bisschops, A. Lee, C. Pham & J. Thömmes. Biophys. J. 93 (2007) 610-619.

Chapter 4

4.1. Introduction A protein solution remains homogenous only up to a certain protein concentration. Once this solubility limit is exceeded, a new state or phase appears as a result of different mechanisms such as crystallization, precipitation, gelation, aggregation or liquid-liquid phase separation. These phase transformations in a protein solution are generally defined as ‘phase behavior’. Understanding protein phase behavior is important for a variety of reasons. From a medical point of view, protein phase transition is the cause of many diseases, such as cataracts [1], Sickle-cell diseases [2], and neurodegenerative or amyloidogenic diseases [3,4]. The controlled release of certain protein drugs, such as insulin, depends on their particular state [5]. From a biological perspective, the microcompartmentation of the cell cytoplasm is thought to be driven partially by protein phase separation [6].

Protein crystallization, on the other hand, is an important tool of structural biology. Most high-resolution protein structural information is obtained by X-ray diffraction, neutron crystallography or surface plasmon resonance of protein crystals. Protein crystallization is also instrumental in elucidating protein function, mode of action, reaction mechanism and so on. In addition, precipitation and crystallization are important unit operations in the purification of industrial proteins and are receiving increasing attention in the industrial separation of therapeutic proteins. Salt-induced precipitation is often the first step in protein purification from a fermentation broth, or from plant and animal extracts, while crystallization may be the last. Solid (crystal or precipitate) forms of a protein are also convenient for storage and transportation. Furthermore, the stability of a biologically active protein is well maintained in the crystal form.

The quality of crystals may not be an important issue when crystallization is applied for purification and/or storage of protein. It is rather important to obtain crystals in bulk quantity without loosing protein functionality. However, structural elucidation by X-ray crystallography requires diffraction quality crystals. Growing such crystals has always been the major barrier to the crystallographic determination of protein structure, in particular in the case of integral membrane proteins [7] and glycoproteins [8]. Membrane proteins are generally considered to be the most difficult to crystallize, mainly due to their amphiphilic character, which implies the use of detergent for their solubilization and crystallization [9]. Among water-soluble proteins, glycoproteins are considered the most difficult to crystallize [10].

Monoclonal antibodies (MAbs) are flexible macromolecules that can assume a wide range of conformations as a consequence of their intrinsic domain mobility and segmental flexibility. MAbs have recently been recognized as enormously efficacious therapeutics, which can be applied to treat numerous life threatening diseases, including cancer and immune diseases. MAbs are also well established as specific serologic reagents for a number of immunoassays and diagnostics for the detection of a wide variety of antigens thanks to their unlimited availability [11]. Unfortunately, MAbs are extremely difficult to crystallize as an intact molecule, probably because of their structural complexity and variability. Most of the published MAb crystallization experiments were restricted to either Fab-antigen

64

Phase behavior of monoclonal antibody

complexes or MAb fragments or MAbs without a hinge region. Successful crystallization of intact MAbs was, nevertheless, reported in several occasions [12-16]. However, successful crystallization of one MAb does obviously not imply that other MAbs will equally readily crystallize under the same solution conditions.

Crystallization or more generally transformation into different phases of a protein occurs due to changes in solution conditions. There are numerous solution variables that can influence protein phase behavior [17] and a variety of phases may form that are difficult to distinguish. The changes of state of a protein as a function of these solution variables are generally known as a ‘protein phase diagram’. The fundamental relationship between a particular solution condition and a particular state of protein is poorly understood. However, evidence has been accumulating that protein interactions play a governing role in determining the structure of the phase diagram. The most tangible result so far is that the phase behavior of a protein solution is correlated with a protein interaction parameter named ‘osmotic second virial coefficient’ (Bmm). Bmm, by definition, is a thermodynamic parameter that reflects the magnitude and direction of deviations of a protein solution from ideality. At the molecular level, Bmm characterizes pair-wise protein self-interactions including contributions from excluded volume, electrostatic interactions and short-range interactions [18]. According to statistical thermodynamics, Bmm is correlated to the potential of mean force, which describes all known interactions between two protein molecules in a dilute solution [19]. A negative value of Bmm indicates protein-protein attraction, whereas a positive Bmm value indicates mutual repulsion.

Solution conditions under which a protein is likely to crystallize correspond to a certain range of slightly negative Bmm values, known as the ‘crystallization slot’ [20,21]. If a Bmm value is more negative than the crystallization slot, disordered precipitation is the phase most likely to develop. However, conducting crystallization experiments under conditions that correspond to the crystallization slot does not guarantee a successful crystallization. The predictive value of the crystallization slot can be improved by studying specific ion effects and the phase diagram. On the other hand, conducting experiments under conditions corresponding with positive Bmm values is sure to prevent phase separation to occur. This correlation is also exploited in pharmaceutical industries for screening stable conditions, at which Bmm values are largely positive, e.g. for liquid formulations of protein drugs. A detailed review on Bmm values of different proteins and their corresponding phases was published earlier [22,23]. The thermodynamic insight regarding the macromolecular interactions involved in Bmm [18,24] and why these interactions are related to protein phase behavior were explained elsewhere [25-28].

Beside the crystallization slot it was established that Bmm is a critical parameter in controlling or accelerating protein aggregation, folding and stability [29-31]. Recent studies show that Bmm is an important thermodynamic parameter not only in predicting protein phase behavior, but also in understanding and designing a molecular approach to different bioseparation processes [32-36]. The reason why Bmm was not applied in bio-separations earlier is because of the difficulty to experimentally determine Bmm. The recent development

65

Chapter 4

of self-interaction chromatography (SIC) allows more accurate and rapid measurement of Bmm using a minimal amount of protein [22,23,37-45].

This study explores the correlation between Bmm and phase behavior of an intact monoclonal antibody, designated in this paper as IDEC-152, which has a molecular mass of ~144 kDa. The IDEC-152 MAb has not been possible to crystallize as of now, even when commercial crystallization kits particularly designed for intact MAbs were employed. The unsuccessful crystallization efforts on IDEC-152 prompted a study of its phase behavior with respect to Bmm values. In this paper, we present phase diagrams of IDEC-152 as a function of different classes of precipitants. Bmm values were also measured for the same conditions by SIC and correlated with the phase diagrams. As a result a single MAb phase diagram displays solubility, Bmm and the optimal crystallization region. This phase diagram is useful not only to develop a fundamental understanding of the phase behavior of MAbs, but also in understanding the reason why certain proteins are extremely difficult to crystallize. To the best of our knowledge, this is the first work to present experimental data on phase behavior of an intact MAb. In addition, the scaling of the phase diagram in Bmm units provides useful information on a structurally complex glycoprotein demonstrating the universality of the phase diagram of many proteins when it is recast in a protein interaction parameter. 4.2. Protein phase diagram A phase diagram shows the state of a material as a function of all of the relevant variables of the system. The simplest form of a protein phase diagram usually displays the state of a protein as a function of protein concentration and another parameter, i.e. the precipitant concentration, with all other variables held constant. This simple phase diagram of different proteins under different solution conditions is quantitatively quite different, although their basic shape is similar. Many broad classes of proteins display a single universal phase diagram, if the phase diagram is recast on a protein interaction parameter [25-27,33,46,47], instead of correlating a single parameter with a solution condition. This is not surprising because a protein interaction parameter, such as Bmm, reflects all the solution parameters into a single dimension. Once Bmm values of a protein are known, the approximate shape and position of the phase diagram would be known.

Rosenbaum and co-workers [25-27] pioneered the development of a generalized protein phase diagram based on its similarity to those of colloids immersed in polymer solutions. They showed that proteins and other globular macromolecules with a size greater than 1 nm do not display gas-liquid phase transitions [25], rather display a broad region of metastable liquid-liquid co-existence along with the region of the liquid-crystal solubility line. In other words, the protein phase diagram exhibits only a liquid-solid equilibrium in three distinct phases: a dilute liquid phase (analogous to the vapor phase), a dense liquid phase (analogous to the liquid phase), and a crystal phase (analogous to the solid phase). The liquid-liquid co-existence region in a protein phase diagram is further separated into two parts (Fig. 4.1), a metastable liquid-liquid region named binodal and an unstable liquid-liquid region named spinodal [48-50]. If a protein solution is quenched into the spinodal of this

66


liquid-liquid phase transition, the solution will spontaneously separate into two metastable phases corresponding the binodal, i.e. one light phase depleted in protein and the other dense phase concentrated in protein. In the regions between the binodal and the spinodal curves, the solution is metastable with respect to liquid-liquid phase separation, i.e. liquid-liquid phase separation occurs rather slowly in these regions. At the critical point, the two phases become identical and liquid-liquid phase separation is not possible beyond this point.

Fig. 4.1. Schematic representation of generalized protein phase diagram.

The location of the George and Wilson’s crystallization slot [20,21] in the universal protein phase diagram is around the metastable liquid-liquid immiscibility region [33,47-49]. The precise location of the crystallization is below the liquid-liquid critical point [27,47] as well as around the critical point [51-54], where nucleation occurs by two different mechanisms. Because of the presence of a nearby metastable liquid-liquid two-phase region, critical density fluctuations are strongly enhanced around the liquid-liquid critical point, which lower the free energy barrier to the formation of critically sized nuclei [52-54]. Therefore, nucleation occurs around the critical point spontaneously. Below the critical point, small liquid-like droplets with a density corresponding to the dense branch of the liquid-liquid binodal may lead to the enhancement of the nucleation just outside the low density branch of the liquid-liquid binodal [48,55]. As a consequence of the high concentration in the droplets, a large fraction of the protein molecules forms aggregates. Crystalline nuclei are then formed from the aggregates inside the droplets. Once the aggregate grows beyond a critical size (about a few hundred molecules), it can convert into a stable crystalline nucleus [56]. Each nucleus is covered with a thin liquid film with a high protein concentration. This thin film lowers the surface energy of the crystal. Protein molecules, diffusing from the dilute solution to the crystal are first incorporated into the surface film. The molecules in this liquid surface film are quite mobile and have ample time to find the proper orientation for incorporation in the crystal [50]. If protein solution conditions fall within the spinodal, the formation of large droplets of high density favors over the formation of crystalline nuclei. The rapid phase separation in this region leaves little time for establishing the proper order

67

Chapter 4

and steric orientations of the protein molecules required for crystallization. The desolubilization and self-association rate is faster in this region than the rate at which molecules achieve the proper orientations that would favor crystallization. Below George and Wilson’s crystallization slot, binodal and spinodal are so wide that the light branches of both binodal and spinodal correspond to impractically low concentration of protein to induce crystallization. Therefore, this region favors amorphous precipitation rather than crystallization.

George and Wilson’s narrow range of Bmm values provides necessary but insufficient information for successfully predicting crystallization of different proteins. The universal protein phase diagram in the dimension of Bmm gives further insight into the possibility of crystallizing a protein that is difficult to crystallize, such as IDEC-152. For instance, nucleation can occur more easily when there is a larger distance between the liquid-liquid critical point and the liquid-solid solubility line [26]. However, the phase behavior of certain proteins may not precisely be mapped using the universal phase diagram because of the presence of long-ranged ranged interactions and/or highly anisotropic nature of protein interactions occurred due to non-sphericity and surface patchiness. For example, the existence of a protein solution well between the critical point and the solubility line might also be proven unsuccessful to crystallize because of the non-complementary shape [24]. Certain proteins may have a strong mutual attraction in few specific orientations, which leads to an overall slightly negative Bmm values but are not compatible with any solid lattice formation. If all the bonds necessary for crystallization cannot form in the crystalline phase but only in the liquid phase, then the protein will not crystallize. Rather, it will exist as a condensed liquid state, analogous to the precipitate phase [51,57]. The presence of moderately long-ranged interactions, which are generally neglected, would shift the liquid-liquid critical point to the gelation regime [58-60].

4.3. Materials and methods 4.3.1. Materials

IDEC-152 MAb was provided by Biogen-Idec Inc., San Diego, California. N-hydroxysuccinimide (NHS)-activated SepharoseTM 4 Fast Flow, a TricornTM 5/50 column and a TricornTM 5 adapter unit were purchased from GE Healthcare Uppsala, Sweden.

Acetic acid, sodium chloride, hydrochloric acid (36-38 %), acetone, and sodium hydroxide were bought from Mallinckrodt Baker, Deventer, The Netherlands. Sodium hydrogenphosphate dihydrate, sodium dihydrogenphosphate dodecahydrate, ammonium sulfate, and sodium acetate trihydrate were bought from Merck, Darmstadt, Germany. Ethanolamine, blue dextran, and polyethylene glycol (PEG) 400 were bought from Sigma-Aldrich Chemie, Zwijndrecht, The Netherlands. Bicinchoninic acid (BCA) protein assay reagents were bought from Perbio Science, Etten-Leur, The Netherlands.

Dialysis equipment (Spectra/Por® Float-A-Lyzer® with Biotech Cellulose Ester Membranes, 100 kDa nominal molecular weight cut-off (NMWCO), 10 ml) was bought from Spectrum Europe, Breda, The Netherlands. A Centriprep® centrifugal filter unit (15 ml, 3

68


kDa NMWCO) was bought from Millipore, Amsterdam, The Netherlands. Chromatography experiments were done in a Pharmacia FPLC system controlled by Unicorn version 2.0 software. All spectrophotometric analysis was done in an Agilent 8453 UV-Visible Pharma System. 4.3.2. MAb sample preparation The IDEC-152 MAb was prepared for SIC and precipitation experiments by dialysis using a 100 kDa NMWCO membrane at 4˚C for at least 24 hours. Dialysis results in an approximately 2-fold dilution of the MAb solution. The dialyzed MAb solution was further diluted or concentrated according to the requirement. 4.3.3. Self-interaction chromatography Two different columns, one with immobilized MAb Sepharose particles and the other MAb-free Sepharose particles, were packed for SIC. The MAb-free column was prepared simply by blocking the NHS-activated groups of the Sepharose particles with ethanolamine. The immobilized MAb column was prepared by immobilizing MAb on NHS-activated sepharose particles. The details of the column preparation processes were described earlier [22,23]. The concentration of immobilized MAb per volume of gel particles was determined by the BCA technique [61] applied to the solid phase [22,23,62]. The integrity of the packed column was characterized by analyzing its height equivalent to a theoretical plate, peak shape and symmetry of a small molecule, such as acetone and NaCl. When not in use, the columns were stored in 10 mM sodium phosphate (pH 7.0) at 4°C. Each column was used for a period of maximally 4 weeks.

The chromatography procedure was accomplished as described previously [37] in an automated Pharmacia FPLC system controlled by Unicorn version 2.0 software. Before every injection, the column was equilibrated until the UV, pH and conductivity base lines became stable. The retention data were used for calculating the Bmm values according to Ahamed et al. [22,23]. 4.3.4. Determination of phase diagram The phase diagram of the IDEC-152 MAb was determined as a function of different precipitant concentrations at a constant temperature of 30°C. The precipitants used in this study were NaCl, (NH4)2SO4 and PEG-400. Since the MAb was only available in liquid form and no crystallization of this MAb was possible, its solubility or phase diagram was measured from precipitation experiments. The dialyzed MAb solution was concentrated to 150 mg/ml at pH 7.6 (10 mM Na-phosphate). Then a 3.0 M (NH4)2SO4 solution (pH 7.6, 10 mM Na-phosphate) was added drop-by-drop on an analytical balance to 1 ml of a 150 mg/ml MAb solution until phase separation was visually observed. The minimum (NH4)2SO4 concentration at which the MAb solubility was approximately 100 mg/ml was considered as the starting point. On the other hand, the minimum (NH4)2SO4 concentration at which the addition of a single drop of a 1 mg/ml MAb solution in 1 ml of an (NH4)2SO4 solution caused

69

Chapter 4

phase separation was regarded as the end point. Several points were chosen between the start and the end point. At every point, the MAb solutions were diluted to a number of different concentrations and incubated for 48 hours in order to reach a stable supernatant concentration. The solutions were then investigated at 600 nm in a spectrophotometer in order to confirm the existence of phase separation. Finally the MAb concentration in the supernatant was measured. The concentration of MAb in the supernatant phase was treated as its apparent solubility. On the other hand, the maximum MAb concentration at which no phase separation was observed was identified as a point on the precipitation line. In order to determine the precipitate concentration, the solutions were allowed to settle under gravity and supernatants were removed gently. 4.3.5. ding to Haas and Drent lume soluti

Modeling of protein phase diagram The phase diagram of the MAb was generated in Bmm scale accor

h [48-50]. According their model, the Gibbs free energy per unit vo of a proteinon can be expressed as [48]

( )

2

21( ) ln

1c

G g kT kTmλ λ

φ φ φ φφ φφ φ

⎡ ⎤2 36 4φ⎧ ⎫⎛ ⎞ − +⎪ ⎪⎛ ⎞⎢ ⎥= + − ⎨ ⎬⎜ ⎟ ⎜ ⎟Ω ⎝ ⎠ −⎢ ⎥⎝ ⎠ ⎪ ⎪⎩ ⎭⎣ ⎦ (4.1)

In Eq. (4.1), φ is the volume fraction of protein, Ω is the volume a protein molecule, cφ is

the protein volume fraction in the crystal and /m ω= Ω , where ω is the molar volume of water divided by Avogadro’s number. The parameter for protein-protein interaction in solution, gλ , was calculated as ( )4B ρ22cg kTλ φ= M − , where M is the molecular mass of

protein and ρ is the protein density. In the calculations, the volume of an intact MAb molecule, , was assumed as 166.5 nm3, considering the fact that the unit cell volume (2 molecules per unit cell) of the MAb crystals were 900 nm3 [14], in which the MAb volume

on,

Ω

cφ , was 0.37 [63]. This resulted the value of ρ equal to 1.44 g/cm3, considering the

ular mass of the IDEC-152 MAb, M, of 144 kDa.

fracti

molec

The compositions of the two coexisting liquid phases in the binodal ( αφ and βφ ) were

lated from the following equations

calcu ( ) ( ) G GG Gβ α

λ λλ β λ α β α

φ φ

φ φ φφ φ

⎛ ∂ ⎞ ⎞− = −⎜ ⎟ ⎟∂ ∂⎝ ⎠ ⎠

φ ⎛ ∂⎜⎝

(4.2)

G G

β α

λ λ

φ φφ φ⎛ ∂ ⎞ ⎛ ∂ ⎞

=⎜ ⎟ ⎜ ⎟∂ ∂⎝ ⎠ ⎝ ⎠ (4.3)

Similarly, the compositions of the phases in the spinodal ( *αφ and *βφ ) were calculated

from the conditions ( and )2 2

*/ 0G

αλ φ

φ∂ ∂ = ( )2 2

*/G

βλ φ

φ 0∂ ∂ = . For the calculation of the

solubility, the Gibbs free energy per unit volume of protein crystal was expressed as /c c cG gφ= Ω , where . Then the solubility line of the crystal was calculated from

the condition

/cg g fλ=

70

Phase behavior of monoclonal antibody ( )c cGG G λ

λ φ φφ

∂− = −

∂ (4.4)

4.4. Results and discussion

4.4.1. Self-interaction chromatography

4.4.1.1. Optimization of the SIC methodology In order to comply with the theory and assumptions of SIC, 15-20% surface coverage of the gel particles was found to be the optimum for avoiding both multi-body interactions and injection concentration-dependent retention behavior [22,23,37,42]. An immobilization protein concentration of 9.4 mg of IDEC-152 per milliliter of packed column corresponds to 15% surface coverage for NHS-activated Sepharose. The incubation time, temperature, pH and protein concentration of the immobilization reaction mixture are the parameters by which the immobilization reaction can be controlled. In this work it was found that 12 hours of incubation at pH 6.0 and 4°C was sufficient to obtain optimum coupling.

Obtaining injected protein concentration independent retention behavior is another important requirement in SIC. Our set up of a 1.2-1.4 ml column, an injection volume of 50 µl and a MAb concentration of 1 to 5 mg/ml produced sharp and detectable peaks. The shape of the peaks showed tailing, due to the mass transfer limitation of large MAb molecules through the gel pores. The shape of all peaks was the same regardless of column type, injection protein concentration or solution condition. Peak retention was found not to vary with injection concentration within a range of 1-2.5 mg/ml (Fig. 4.2). Therefore, the concentration of MAb in the injection sample was always kept between 1-2 mg/ml and the retention volume was determined from the peak maximum.

Fig. 4.2. Effect of the injected protein concentration on the retention of the MAb. The experiment was conducted at pH 4.2 (100 mM Na-acetate) in the MAb-immobilized column.

4.4.1.2. Bmm profile of the monoclonal antibody Bmm values of IDEC-152 MAb were mapped as a function of NaCl concentration at different pH values. Most of the Bmm values were found to be in the positive regime up to a

71

Chapter 4

NaCl concentration of 1.0 M at pH 4.5 to 9.5 (Fig. 4.3). Very few points were found to be on the negative side, out of which none was negative enough to be in the George and Wilson’s crystallization slot. Since all of the SIC experiments were performed in a single column, the margin and direction of inherent error [22,23] in the Bmm data must be the same. This run-to-run error free data suggested a downward peak in Bmm at pH 6.5 and 7.6, in which Bmm values were minimal at 0.17 M NaCl. The Bmm values were slightly negative in these conditions. Overall Bmm mapping in NaCl showed that Bmm does not depend much on pH at higher ionic strength. This simple Bmm mapping in NaCl shows the reason why it was impossible to crystallize IDEC-152 by empirical screening.

In contrast to NaCl, (NH4)2SO4 and PEG are two well-known precipitants of proteins. Instead of a crystalline solid phase, amorphous precipitates are often observed in presence of (NH4)2SO4 or PEG. Bmm values of MAb as a function of the concentration of these precipitants would also help to gain a better understanding of protein phase behavior in the presence of these precipitants. Both (NH4)2SO4 and PEG-400 showed a similar trends, i.e. an unchanged Bmm value up to 0.6 M (NH4)2SO4 and 15% (v/v) PEG-400 (Fig. 4.4). Above these points, the Bmm value decreases dramatically to reach the negative regime of the Bmm scale. However, it was impossible to run SIC experiments at higher PEG-400 concentrations. At PEG-400 concentration of >20% (v/v), the solution is too viscous to pump through the chromatography column. On the other hand, the solubility of the MAb was too low to obtain a well distinguishable peak at an (NH4)2SO4 concentration of >1.0 M.

Fig. 4.4. Bmm profile of IDEC-152 in (NH4)2SO4 and PEG-400 at pH 7.6.

Fig. 4.3. Bmm profile of IDEC-152 in NaCl at different pHs.

72


4.4.2. Phase behavior of the monoclonal antibody

4.4.2.1. Precipitates of the monoclonal antibody The solubility of a protein is usually measured by dissolving crystals in a protein-free solution until the concentration of the protein in the liquid phase reaches a constant equilibrium value. Alternatively it is also possible to start with a supersaturated solution, in which the solution reaches equilibrium through the growth of crystals. It typically takes days to months to reach equilibrium [64]. Both methods were impossible to implement for IDEC-152, since it has not been possible to crystallize the MAb yet. Although precipitates are generally considered as non-equilibrium phases, the residual protein concentration in the supernatant is widely referred to as the solubility. In this work, solubility was measured by allowing the MAb solutions to precipitate.

When the solution conditions and the initial MAb concentrations were met for the phase separation, precipitates are formed immediately. In some experimental conditions, the immediately appearing precipitates dissolved during 48 hrs of incubation with shaking. The conditions at which the precipitates did not dissolve showed two distinct phases, a clear solution phase on the top and a white precipitate phase on the bottom. Under light microscope, the precipitates were characterized as fibril like opaque. The obtained precipitates were reversible, and they could be driven to re-dissolve by addition of solvent. The MAb concentration in the supernatant was found to be independent of the initial protein concentration within the limits of the experimental error of 10%, consistent with observations with lysozyme solutions [65,66]. In contrast to this observation, the apparent solubility of α-chymotrypsin, bovine serum albumin and bovine liver catalase in precipitation experiments have been reported to be a functions of the initial protein concentration [65,67]. The MAb concentration in the supernatant also showed a smooth decrease with increasing precipitant concentrations. However, if the initial MAb concentrations were too high with respect to the solubility, a sticky gel type of solid phase was formed. No clear solution could be recovered from the top in such circumstances. The characteristics of IDEC-152 precipitates obtained in NaCl, (NH4)2SO4 or PEG-400 were the same. All of these observations during the precipitation of IDEC-152 MAb suggest that there is indeed an equilibrium phase separation, despite the fact that the precipitates are kinetically trapped non-equilibrium phases. However, the solubility obtained from the supernatant concentration may not be equal to the real equilibrium solubility, therefore it is referred to as ‘apparent solubility’ in this paper. Indeed, one would expect a lower supernatant concentration to exist in equilibrium with a crystalline solid phase as it was observed for lysozyme [66,68,69]. 4.4.2.2. Phase diagrams of monoclonal antibody

The apparent solubility and precipitation behavior of IDEC-152 was studied as a function of the concentration of three precipitants, NaCl, (NH4)2SO4 and PEG-400, at a constant temperature of 30°C. The apparent solubility was found to decrease smoothly with increasing precipitant concentration for all three precipitants, in accordance with Cohn [70]. A series of dilutions was made with varying MAb concentrations around the apparent

73

Chapter 4

solubility line. Interestingly, no phase separation was observed up to a certain concentration above the apparent solubility line (Fig. 4.5). The minimum MAb concentration, at which precipitation was observed, was designated as the precipitation line in the phase diagram. The MAb solution was completely transparent between the solubility and precipitation lines. It was, however, not possible to differentiate nucleation and growth region in the supersaturated area [71], because of unsuccessful crystal growth. Fig. 4.5 shows a phase diagram of the IDEC-152 MAb as a function of NaCl concentration. The solubility of the MAb was extremely high up to 2.4 M NaCl. A solid phase was observed at a minimum NaCl concentration of 2.5 M. At 4.5 M NaCl, the solubility of the MAb was extremely low (below 1 mg/ml). Phase diagrams were obtained in similar fashion for (NH4)2SO4 (Fig. 4.6) and PEG-400 (Fig. 4.7).

Fig. 4.5. Phase behavior of IDEC-152 MAb in NaCl at 30°C and pH 7.6.

Fig. 4.7. Phase behavior of IDEC-152 MAb in PEG-400 at 30°C and pH 7.6.

Fig. 4.6. Phase behavior of IDEC-152 MAb in (NH4)2SO4 at 30°C and pH 7.6

74


4.4.2.3. MAb in the framework of universal phase diagram In this work, phase diagrams of IDEC-152 MAb were obtained as a function of precipitant concentration at constant temperature of 30 °C. One would expect protein phase behavior as a function of a precipitant concentration to be similar to that of inverse of temperature. However, measurement of the liquid-liquid co-existence curve, as a function of precipitant concentration, is extremely difficult by the cloud-point method [72,73], because of the difficulty of slowly increasing or decreasing the precipitant concentration while keeping all other parameters constant. In addition, the source of the turbidity in IDEC-152 solutions was mostly due to the formation of precipitates, unlike the liquid droplets as observed in a previous study [73]. However, Lenhoff and co-workers [66] showed that the supernatant curve of lysozyme obtained from precipitation and cloud-point measurements is consistent with the low-density branch of the metastable liquid-liquid co-existence curve. On the other hand, the formation of the dense branch of the liquid-liquid binodal is perturbed by the appearance of flocks or aggregates. Therefore, gels or precipitates represent a different, but kinetically trapped, structure for the dense liquid phase. The physical appearance of the lysozyme precipitate described in ref [66] was similar to that observed this study. In both cases, the precipitates were opaque and settled at the bottom of the tube. If shaken, it caused the solution turbid. The liquid phase concentration was independent of initial protein load conforming the pseudo-equilibrium. Furthermore, the initial higher load of protein caused the formation of gel in both cases. Therefore, precipitates of MAb can be interpreted as the dense branch of the liquid-liquid binodal. In that case, the dense phase in the liquid-liquid phase separation must be similar to the dense liquid phase described by Prausnitz and co-workers [65,74,75], where precipitates were named as dense liquid phases and protein was treated as the partitioning solute between the liquid and precipitate phases.

Fig. 4.8. P hase diagram of IDEC-152 MAb showing liquid-liquid coexistence.

A generic format of phase diagrams was made with above assumptions. The concentration of MAb in both the supernatant phase (apparent solubility) and the precipitate (dense) phase was plotted against the reverse scale of the (NH4)2SO4 concentration (Fig. 4.8).

75

Chapter 4

The supernatant and precipitate MAb concentrations show an apparent equilibrium curve, which represents the liquid-liquid co-existence curve. Since the crystal-liquid equilibrium solubility curve could not be determined experimentally, the position of the liquid-liquid co-existence curve with respect to the solubility curve cannot be shown directly. However, the position of the liquid-liquid co-existence curve can be visualized by plotting it against Bmm. Such a phase diagram was generated in Bmm scale according to the model described by Haas and Drenth [48-50]. The calculated phase diagram (Fig. 4.9) shows that the critical point is located at a Bmm value of –0.32 × 10-4 mol.ml/g2, which corresponds to slightly negative Bmm values of George and Wilson’s crystallization slot [20,21]. Therefore, the calculated phase diagram of MAb agrees with the fact that the precise location of the nucleation is around the critical point. The crystal-liquid solubility line (dashed line in Fig. 4.9) equals to the light branch of the binodal for an f value of >0.3217. It is, therefore, more likely that the phase diagram of MAb exhibits a triple point, rather than an isolated crystal-liquid solubility line.

Fig. 4.9. Phase diagram of IDEC-152 MAb in the format of generic protein phase diagram. Calculations were made according to Haas and Drenth [48]. Assumptions: = 1.665 × 10-19 cm3;

Ω

cφ = 0.37; M = 144000 g/mol;

ρ = 1.4362 g/cm3; f = 0.3217. Existence of binodal and spinodal within the crystal phase is not realistic. However, it is shown to visualize their approximate location.

The experimentally determined liquid-liquid co-existence curve in (NH4)2SO4 concentration scale was transformed into the Bmm scale by interpolation, since the Bmm values were known as a function of (NH4)2SO4 concentration. The experimental data points of liquid-liquid co-existence are shown in Fig. 4.9. Although the light branch of the data points matched well with the spinodal, the dense branch showed much lower concentration than theoretical expectation. This is because precipitates are considered as the dense phase and the precise determination of the precipitate volume and concentration is difficult. However, above observations suggest that the MAb, a complex glycoprotein, certainly supports the so-called universal format of phase diagram.

The question arises why crystallization of IDEC-152 MAb was impossible when the phase diagram supports a generic format. It is obvious from the phase diagram (Fig. 4.9) that

76


spontaneous classical homogeneous nucleation just above the critical point is not possible because of two reasons. Firstly, the critical point corresponds to a protein volume fraction of 0.131 or a concentration of 188 mg/ml. Conducting crystallization experiment above such a high protein concentration is impractical and inapplicable. Secondly, there is insufficient or no space between the liquid–liquid critical point and the solubility line. The only possible mechanism remaining for the nucleation of the MAb is the liquid-liquid phase separation. Therefore, crystallization of MAb is only expected between the light branch of the binodal and spinodal, which might be a very narrow range. A possible reason of unsuccessful crystallization is that even if an experiment is designed between the light branch of the binodal and spinodal, the crystallization process itself might be too slow for crystals to form within a practical time frame. A third reason of unsuccessful crystallization could be the shape of the MAb molecule. A slightly negative Bmm value around 0.8 M (NH4)2SO4 could be due to strong attractions in few specific orientations, which are not favorable to solid lattice formation. 4.5. Conclusion

The work presented here shows a phase behavior study of a complex glycoprotein. Like most well studied proteins, phase behavior of IDEC-152 MAb shows a behavior of decreasing solubility with increasing precipitant concentration according to Cohn [70]. Rescaling of the phase diagram in Bmm units shows that spontaneous classical homogeneous nucleation of MAb crystals is not possible just above the liquid-liquid critical point, because of insufficient or no space between the critical point and the solubility line. Nucleation of IDEC-152 MAb could only be possible by liquid-liquid phase separation in a narrow window. However, the idea of a universal protein phase diagram was supported for this large complex glycoprotein. Further study is required on uncommon and structurally complex proteins in order to understand protein phase behavior in a generalized way. This study further concludes that the crystallization of proteins in (NH4)2SO4 is rather difficult because both solubility and Bmm decreases drastically above a certain (NH4)2SO4 concentration and leaves an extremely narrow window of crystallization. References [1] A. Pande, J. Pande, N. Asherie, A. Lomakin, O. Ogun, J. King & G.B. Benedek. Proc. Nat.

Acad. Sci. USA 98 (2001) 6116-6120. [2] W.A. Eaton & J. Hofrichter. Adv. Protein Chem. 40 (1990) 63-279. [3] D.J. Selkoe. J. Neuropathol. Exp. Neurol. 53 (1994) 427-428. [4] E.H. Koo, P.T. Lansbury & J.W. Kelly. Proc. Nat. Acad. Sci. USA 96 (1999) 9989-9990. [5] M.L. Brader, M. Sukumar, A.H. Pekar, D.S. McClellan, R.E. Chance, D.B. Flora, A.L. Cox, L.

Irwin & S.R. Myers. Nat. Biotechnol. 20 (2002) 800-804. [6] H. Walter & D.E. Brooks. FEBS Lett. 361 (1995) 135-139. [7] C. Ostermeier & H. Michel. Curr. Opin. Struct. Biol. 7 (1997) 697-701. [8] E.A. Stura, G.R. Nemerow & I.A. Wilson. J. Cryst. Growth 122 (1992) 273-285. [9] J.-L. Rigaud, M. Chami, O. Lambert, D. Levy & J.-L. Ranck. Biochem. Biophys. Acta 1508

(2000) 112-128.

77

Chapter 4

[10] L.J. Harris, E. Skaletsky & A. McPherson. J. Mol. Biol. 275 (1998) 861-872. [11] W.J. Payne, D.L. Marshall, R.K. Shockley & W.J. Martin. Clin. Microbiol. Rev. 1 (1988) 313-

329. [12] S. Larson, J. Day, A. Greenwood, E. Skaletsky & A. McPherson. J. Mol. Biol. 222 (1991) 17-

19. [13] E.A. Stura, A.C. Satterthwait, J.C. Calvo & R.S. Stefanko. Acta Cryst. D50 (1994) 556-562. [14] L.J. Harris, E. Skaletsky & A. McPherson. Proteins: Struct. Funct. Gen. 23 (1995) 285-289. [15] Y.G. Kuznetsov, A.J. Malkin & A. McPherson. J. Cryst. Growth 232 (2001) 30-39. [16] E.O. Saphire, P.W.H.I. Parren, C.F. Barbas, D.R. Burtona & I.A. Wilsona. Acta Cryst. D57

(2001) 168-171. [17] A. McPherson. Methods Enzymol. 114 (1985) 112-120. [18] B.L. Neal, D. Asthagiri, O.D. Velev, A.M. Lenhoff & E.W. Kaler. J. Cryst. Growth 196 (1999)

377-387. [19] W.G. McMillan & J.E. Mayer. J. Chem. Phys. 13 (1945) 276-305. [20] A. George & W.W. Wilson. Acta Cryst. D50 (1994) 361-365. [21] A. George, Y. Chiang, B. Guo, A. Arabshahi, Z. Cai & W.W. Wilson. Methods Enzymol. 276

(1997) 100-110. [22] T. Ahamed, M. Ottens, G.W.K. van Dedem & L.A.M. van der Wielen. J. Chromatogr. A 1089

(2005) 111-124. [23] T. Ahamed, M. Ottens, G.W.K. van Dedem & L.A.M. van der Wielen. J. Chromatogr. A 1115

(2006) 272. [24] B.L. Neal, D. Asthagiri & A.M. Lenhoff. Biophys. J. 75 (1998) 2469-2477. [25] D. Rosenbaum, P.C. Zamora & C.F. Zukoski. Phys. Rev. Lett. 76 (1996) 150-153. [26] D.F. Rosenbaum, A. Kulkarni, S. Ramakrishnan & C.F. Zukoski. J. Chem. Phys. 111 (1999)

9882-9890. [27] D.F. Rosenbaum & C.F. Zukoski. J. Cryst. Growth 169 (1996) 752-758. [28] M. Farnum & C. Zukoski. Biophys. J. 76 (1999) 2716-2726. [29] J.G.S. Ho, A.P.J. Middelberg, P. Ramage & H.P. Kocher. Protein Sci. 12 (2003) 708-716. [30] E.Y. Chi, S. Krishnan, T.W. Randolph & J.F. Carpenter. Pharm. Sci. 20 (2003) 1325-1536. [31] J.J. Valente, R.W. Payne, M.C. Manning, W.W. Wilson & C.S. Henry. Curr. Pharm.

Biotechnol. 6 (2005) 427-436. [32] T. Ahamed, M. Ottens, B.K. Nfor, G.W.K. van Dedem & L.A.M. van der Wielen. Fluid Phase

Equilib. 241 (2006) 268-282. [33] R.A. Curtis & L. Lue. Chem. Eng. Sci. 61 (2006) 907-923. [34] H.W. Blanch, J.M. Prausnitz, R.A. Curtis & D. Bratko. Fluid Phase Equilib. 194-197 (2002)

31-41. [35] L.A.M. van der Wielen & E.S.J. Rudolph. J. Chem. Technol. Biotechnol. 74 (1999) 275-283. [36] T.M. Przybycien. Curr. Opin. Biotechnol. 9 (1998) 164-170. [37] P.M. Tessier, A.M. Lenhoff & S.I. Sandler. Biophys. J. 82 (2002) 1620-1631. [38] P.M. Tessier, S.D. Vandrey, B.W. Berger, R. Pazhianur, S.I. Sandler & A.M. Lenhoff. Acta

Cryst. D58 (2002) 1531-1535. [39] P.M. Tessier, H.R. Johnson, R. Pazhianur, B.W. Berger, J.L. Prentice, B.J. Bahnson, S.I.

Sandler & A.M. Lenhoff. Proteins: Struct. Funct. Gen. 50 (2003) 303-311. [40] C.D. Garcia, S.C. Holman, C.S. Henry & W.W. Wilson. Biotechnol. Prog. 19 (2003) 575-579. [41] C.D. Garcia, D.J. Hadley, W.W. Wilson & C.S. Henry. Biotechnol. Prog. 19 (2003) 1006-1010. [42] C.A. Teske, H.W. Blanch & J.M. Prausnitz. J. Phys. Chem. B 108 (2004) 7437-7444.

78


79

[43] B.W. Berger, C.J. Blamey, U.P. Naik, B.J. Bahnson & A.M. Lenhoff. Cryst. Growth Des. 5 (2005) 1499-1507.

[44] B.W. Berger, C.M. Gendron, C.R. Robinson, E.W. Kaler & A.M. Lenhoff. Acta Cryst. D61 (2005) 724-730.

[45] J.J. Valente, K.S. Verma, M.C. Manning, W.W. Wilson & C.S. Henry. Biophys. J. 89 (2005) 4211-4218.

[46] R.A. Curtis, R.S. Pophale & M.W. Deem. Fluid Phase Equilib. 241 (2006) 354-367. [47] G.A. Vliegenthart & H.N.W. Lekkerkerker. J. Chem. Phys. 112 (2000) 5364-5369. [48] C. Haas & J. Drenth. J. Phys. Chem. B 102 (1998) 4226-4232. [49] C. Haas & J. Drenth. J. Cryst. Growth 196 (1999) 388-394. [50] C. Haas & J. Drent. J. Phys. Chem. B 104 (2000) 368-377. [51] R.A. Curtis, H.W. Blanch & J.M. Prausnitz. J. Phys. Chem. B 105 (2001) 2445-2452. [52] P.R. ten Wolde & D. Frenkel. Science 277 (1997) 1975-1978. [53] P.R. ten Wolde & D. Frenkel. Theor. Chem. Acc. 101 (1999) 205-208. [54] V. Talanquer & D.W. Oxtoby. J. Chem. Phys. 109 (1998) 223-227. [55] O. Galkin & P.G. Vekilov. Proc. Nat. Acad. Sci. USA 97 (2000) 6277-6281. [56] A. Lomakin, N. Asherie & G.B. Benedek. Proc. Nat. Acad. Sci. USA 100 (2003) 10254-10257. [57] R.P. Sear. J. Chem. Phys. 111 (1999) 4800-4806. [58] M.G. Noro, N. Kern & D. Frenkel. Europhys. Lett. 43 (1999) 332-338. [59] M. Muschol & F. Rosenberger. J. Chem. Phys. 107 (1997) 1953-1962. [60] G. Foffi, G.D. McCullagh, A. Lawlor, E. Zaccarelli, K.A. Dawson, F. Sciortino, P. Tartaglia, D.

Pini & G. Stell. Phys. Rev. E 65 (2002) 031407. [61] P.K. Smith, R.I. Krohn, G.T. Hermanson, A.K. Mallia, F.H. Gartner, M.D. Provenzano, E.K.

Fujimoto, N.M. Goeke, B.J. Olson & D.C. Klenk. Anal. Biochem. 150 (1985) 76-85. [62] A.L. Plant, L. Locascio-Brown, W. Haller & R.A. Durst. Appl. Biochem. Biotechnol. 30 (1991)

83-98. [63] L.J. Harris, S.B. Larson, K.W. Hasel & A. McPherson. Biochemistry 36 (1997) 1581-1597. [64] M. Akata. Phase Transit. 45 (1993) 205-219. [65] Y.-C. Shih, J.M. Prausnitz & H.W. Blanch. Biotechnol. Bioeng. 40 (1992) 1155-1164. [66] Y.-C. Cheng, R.F. Lobo, S.I. Sandler & A.M. Lenhoff. Biotechnol. Bioeng. 94 (2006) 177-188. [67] H.V. Iyer & T.M. Przybycien. AIChE J. 40 (1994) 349-360. [68] S.B. Howard, P.J. Twigg, J.K. Baird & E.J. Meehan. J. Cryst. Growth 90 (1988) 94-104. [69] J.J. Moretti, S.I. Sandler & A.M. Lenhoff. Biotechnol. Bioeng. 70 (2000) 498-506. [70] E.J. Cohn. Physiol. Rev. 5 (1925) 349-437. [71] N.E. Chayen. Prog. Biophys. Mol. Biol. 88 (2005) 329-337. [72] C. Liu, N. Asherie, A. Lomakin, J. Pande, O. Ogun & G.B. Benedek. Proc. Nat. Acad. Sci. USA

93 (1996) 377-382. [73] N. Asherie. Methods 34 (2004) 266-272. [74] C.J. Coen, H.W. Blanch & J.M. Prausnitz. AIChE J. 41 (1995) 996-1004. [75] C.J. Coen, J.M. Prausnitz & H.W. Blanch. Biotechnol. Bioeng. 53 (1997) 567-574.

ahamed_ (13.2 MB) - TU Delft Institutional Repository

Documents

Transcript of ahamed_ (13.2 MB) - TU Delft Institutional Repository