- Email: [email protected]
Step gradients in 3-zone simulated moving bed chromatography
Available online at www.sciencedirect.com
Journal of Chromatography A, 1176 (2007) 69–78
Step gradients in 3-zone simulated moving bed chromatography Application to the purification of antibodies and bone morphogenetic protein-2夽 Lars Christian Keler a,∗ , Ludmila Gueorguieva b , Ursula Rinas c , Andreas Seidel-Morgenstern a,b b
a Max Planck Institute for Dynamics of Complex Technical Systems, Sandtorstr. 1, D-39106 Magdeburg, Germany Institute of Process Engineering, Otto von Guericke University, Universit¨atsplatz 2, D-39106 Magdeburg, Germany c Helmholtz Centre for Infection Research, Inhoffenstr. 7, D-38124 Braunschweig, Germany
Received 6 July 2007; received in revised form 16 October 2007; accepted 19 October 2007 Available online 1 November 2007
Abstract The simulated moving bed (SMB) technology is a proven tool for efficient separation of binary mixtures. However, relying on isocratic conditions limits the applicability of the classical SMB approach when considering the field of bioseparations. Here, the use of gradients opens up new possibilities. A gradient in a SMB process can be established by using different solvent strengths in the incoming feed and desorbent streams, resulting in two internal plateaus of elution strength. Thus, compared to the conventional process, the overall amount of solvent needed can be reduced, productivity can be increased and more concentrated product streams can be obtained. In this contribution, two case studies will be presented. At first, the separation of bovine IgG from lysozyme will be analyzed as a model system. Antibodies are a common target substance in bio-chromatography, as therapeutic monoclonal antibodies are among the most promising biopharmaceuticals. Using adsorption data obtained from single-column experiments, an appropriate SMB process was designed and implemented. The second target component is the active dimeric form of the bone morphogenetic protein-2 (BMP-2). This protein was isolated from a renaturation solution, which also contained its inactive monomeric form as well as other undefined proteins from the bacterial production strain. A 3-zone open-loop gradient-SMB approach was used successfully for both separations. © 2007 Elsevier B.V. All rights reserved. Keywords: Simulated moving bed; Gradients; Protein separation; BMP-2; Antibodies; Moving-column implementation; Equilibration; Open-loop
1. Introduction Chromatographic separation techniques are applied throughout the chemical, pharmaceutical and biotechnological industry for the purification of high-value products. The use of gradients to increase process performance has become state-of-the-art in discontinuous chromatography. In continuous chromatography, however, the majority of separations are still run isocratically. The most commonly used continuous process is simulated moving bed (SMB) chromatography. This concept was invented by UOP in the early 1960s and uses an intelligent movement 夽 Presented at the 20th International Symposium on Preparative and Process Chromatography, Baltimore, MD, USA, 3–6 June 2007. ∗ Corresponding author. Tel.: +49 391 6110 442; fax: +49 391 6110 403. E-mail address: [email protected] (L.C. Keler).
0021-9673/$ – see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.chroma.2007.10.087
of ports (moving-port implementation) or columns (movingcolumn implementation) to mimic a counter-current movement of liquid and solid phases [1]. The SMB process was found often to be superior to its discontinuous, isocratic counterpart in terms of achievable purity, productivity and reduced solvent consumption [2]. The idea of using gradients in a SMB process was described first by Nicoud et al. using a supercritical eluent [3]. In recent years, the more convenient use of two differently composed liquid mobile phases has been studied in some detail, e.g. [4–7]. As in batch chromatography, the use of gradients can further increase productivity and reduce solvent consumption. Additionally, separation problems exist that cannot be solved under isocratic conditions. Most of the contributions reported up to now deal with the separation of “small molecules” like cycloketones or nucleosides. In the emerging field of bioseparations, where undefined multicomponent mix-
70
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
tures are common and interactions are often more complex, only few gradient-SMB applications have been reported (e.g. [8,9]). In this context, an innovative continuous process which also uses gradient conditions was presented recently as an alternative to SMB chromatography [10]. Two case studies for continuous protein separation with a two-step gradient-SMB process are presented in this contribution. At first, the separation of a binary model mixture, composed of bovine immunoglobulin G (IgG) and lysozyme, using immobilized Heparin as weak cation exchanger is presented. The information gathered with this model system are then used as guidelines for implementing the purification of the active dimeric form of bone morphogenetic protein-2 (BMP-2) from its monomeric form as well as from numerous, unidentified other components. 2. Three zone two-step gradient-SMB technology All modifications suggested here originate in the classical isocratic closed-loop 4-zone SMB process (Fig. 1a, [1]). In this process, each of the four zones, framed by outlet and inlet ports, has to fulfill a specific task. The stream of the feed mixture to be separated, QF , enters the unit between zones II and III. In these so-called separation zones, the conditions selected have to ensure that the less adsorbing component A moves upstream with the mobile phase, where it is withdrawn at the raffinate outlet. At the same time, the more absorbing component B should be transported in counter-current direction by the periodic switching of either columns or ports, and likewise withdrawn at the extract port. Zones I and IV are regeneration zones; here, either the solid phase (zone I) or the mobile phase (zone IV) is freed from components A or B, respectively. This closed-loop approach reduces solvent consumption, but also introduces the risk of cross-contamination. If feed mixtures are to be processed which contain unknown or uncharacterized components, for example fermentation broths, one might inadvertently select conditions that do not prevent all substances from leaving zone IV. This leads to a loss of product in the raffinate outlet and a loss of purity in the extract outlet, which is especially troublesome if the target substance is an expensive, therapeutic compound. Thus, it can be beneficial to avoid closed-loop operation and use an open-loop configuration, leading to an increase in solvent consumption. However, this factor carries less weight when the mobile phase is inexpensive, as it is often the case in biochromatography. There, cheap and easy-to-handle water–salt solutions are most common. Also, disposal is less problematic when compared to the majority of processes for small molecules, where organic solvent are frequently used, renderring an open-loop configuration an attractive alternative. In an open-loop configuration, the recycling zone of the liquid phase (IV) can be removed, reducing the total number of zones to three (Fig. 1b). Additionally, in the process used for this study, a two-step gradient is realized by using different concentrations of a selectable modifier in desorbent and feed. This leads to a first plateau of solvent strength in zones I and II, which is equal to the solvent strength introduced at the desorbent port, and a second plateau in zone III, where the modifier level is determined by both the concentration of the
modifier in the feed stream and the first plateau concentration. Typically, in zones I and II high elution strength is set to better mobilize the more adsorbing component B. Similarly, the modifier concentration in the stream leaving zone II is reduced by the feed to reach lower elution strength in zone III, in order to promote adsorption of component B. Strategies for determining suitable operating conditions will be shortly discussed later in this contribution. 2.1. Realization of 3-zone SMB modes in a moving-column unit As described previously, the SMB principle can be realized using either a moving-port or a moving-column type of implementation [11]. Both are equivalent in theory. Practically, this holds for basic 4-zone operation, but if more sophisticated operating modes, for example the VariCol-process [12] or uneven numbers of zones, are considered, there are notable differences. The moving-port implementation uses multiple
Fig. 1. Schematic representations of a classical, closed-loop isocratic 4-zone SMB process [1] (a) and the open-loop gradient 3-zone SMB process (b) used in this study. In (b), the position of the first gradient plateau (high elution strength) is indicated by the crossed lines in the columns of zones I and II.
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
multiposition valves. Here, every valve can be operated at a different time, offering a high degree of flexibility. It has to be ensured, however, that when a simultaneous switching of all valves is desired, there are no delays or deviations. In the moving-column implementation, this aspect is taken care of by the design of the one single valve, with which all pumps and columns are directly connected. As these valves are typically designed with the conventional 4-zone operation in mind, the number of ports generally is a multiple of 4, often 12 or 16. This design implies that all columns can only be moved at the same time. For a 16-port valve, this complicates 3-zone operation (as 16 cannot be divided by three integrally). An example of a possible, but problematic realization of three zones in a 16-port multiposition valve, as supplied, for example, by Knauer (Berlin, Germany), is shown in Fig. 2. In Fig. 2a the situation at the beginning of the first cycle is given, with one column being at the beginning of each zone. The situation after 4 switching events, at the end of the first cycle, is shown in Fig. 2b. Still, one column can be found in each zone. But after one additional switch (Fig. 2c), there are now two columns in zone III and none in zone II. Over the course of the next cycles, this will occur in all three zones, which will inevitably lead to a decrease in purity. To avoid this problem, two possible alternative schemes are proposed. The first scheme (Fig. 3a) uses a disconnected (“silent”) zone IV, together with one additional column. During operation, only three columns participate actively, the fourth is used solely to generate the needed symmetry. The second scheme (Fig. 3b) utilizes the additional fourth zone to preequilibrate the column leaving the high-modifier level zone I with the lower modifier concentration selected for zone III. Again, four columns are used, but only three take part in separation. Depending on the separation problem and the cycle time, it is also possible to use zone IV for an integrated cleaning-inplace (CIP). For moving-port implementations, a similar CIP scheme was suggested in Refs. [13,14]. For the sake of clarity, it is necessary to have a closer look at the physical meaning of the cycle time tS when comparing moving-port and moving-column implementations. When performing a regular SMB separation using a moving-port apparatus, the switching of each multiposition valve occurs once at the end of each cycle. Consequently, tS is often also referred to as switching time. This term might be misleading when considering a moving-column implementation. There, each zone contains multiple valve ports, for example four as shown in Fig. 3. In order to realize the correct counter-current movement, the valve has to be operated four times within a cycle time, so the occurrence of a switching event is not equivalent to tS . In this publication, tS can be understood as the necessary residence time of a column in a zone, as this definition holds for both types of implementations. 2.2. Calculation of operating conditions based on equilibrium theory The selection of both internal and external flow rates, the corresponding cycle time and suitable modifier concentrations in the inlet streams requires a connection between these operat-
71
Fig. 2. Schematic representation of valve and column positions in an erroneous moving-column implementation of a 3-zone process using a 16-port valve. In (a–c) the positions of the three columns used are indicated by black dots at the beginning of the first cycle (a), after moving each column for four times, at the end of the first cycle (b), and at the beginning of the second cycle (c).
ing conditions and characteristic properties of the components to be separated. This link can be found in the distribution equilibria between the liquid phase concentration of a component k, ck , and the corresponding solid phase concentration, qk . For bioseparations, this relation is often described with the Langmuir isotherm. In the case of the separation problems discussed
72
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
A component k in zone z moves with the liquid phase, if mz is larger than Kk , and in counter-current direction, with the solid phase, when mz is smaller than Kk . The knowledge, which component has to be moved in which direction to achieve complete separation, can then be cast into two inequalities: KB < mI
(3)
KA < mII < mIII < KB
(4)
For gradient separations, the basic principle still holds, but due to the different modifier levels in zones I–II and III the inequalities have to be adjusted [4,6] B Mod mI > KI−II (cD )
(5)
A Mod B Mod KI−II (cD ) < mII < KI−II (cD )
(6)
A Mod B Mod KIII (cIII ) < mIII < KIII (cIII )
(7)
From Eqs. (5)–(7) it is quite obvious that a larger difference of k mII /mIII can be achieved with gradients, as KI−II will be smaller k than KIII , leading to an increase in productivity. Consequently, it is desirable to maximize the difference in modifier concentration between zones I–II and III [5]. In order to calculate operating conditions, the relation between Kk and cMod must be determined experimentally. Often, a simple equation as given in Ref. [5] can be used to describe the equilibrium functions: −p2k
Kk (cMod ) = (p1k cMod )
Fig. 3. Schematic representation of valve and column positions of two feasible moving-column implementation of a 3-zone process using a 16-port valve. In (a) is given the scheme of a 3-zone SMB process in a 16-port valve using a disconnected (silent) fourth zone. In (b) is given the scheme of a 3-zone SMB process in a 16-port valve using the fourth zone for pre-equilibration. In both (a) and (b), the positions of the columns used are indicated by black dots.
in this study, it was found that the adsorption behavior could be described reasonably well with a simple linear isotherm (Eq. (1)). For two components A and B, isotherms can be written as qk = K k c k
with k = A, B
(1)
with component A being less retained than B (KA < KB ). For isocratic conditions, it is straightforward to determine operating conditions using the well-known equilibrium theory. Key to this analysis is the introduction of dimensionless flow-rate ratios mz [15]. mz = QZ
(tS − VC ε) VC (1 − ε)
with
z = I, II, III
(2)
The direction of a component k in zone z depends on the relation between its adsorption constant KK , quantifying the strength of interaction with the stationary phase, and the flow-rate ratio mz .
(8)
After selecting one flow rate, usually QI or QF , it is possible to calculate the corresponding cycle time from Eq. (2). Afterwards, the same equation can be used to convert the chosen m-ratios to volumetric flow rates. This conversion, however, does not account for additional extra-column volumes. While this is fairly unproblematic for large columns, this does not hold for small columns as the ones used in this study. A convenient way to account for extra-column dead volume is given in Ref. [16]. The flow rates further have to fulfill the following mass balances: QI = QD
(9)
QII = QI − QE
(10)
QIII = QII + QF
(11)
When compared to the classical 4-zone process, there is a notable difference in Eq. (9). Due to the loss of zone IV, QI is now sustained only by the external desorbent stream. For the same reason, the stream leaving zone III, QIII , is completely considered as raffinate stream QR . Besides the mobile phase mass balance, the modifier mass balance also has to be fulfilled for gradient separations. It can be given as Mod Mod cI−II = cD Mod cIII =
Mod + Q cMod QII cD F F QII + QF
(12) (13)
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
From these equations, it is obvious that certain combinations of internal modifier concentrations are not realizable, because they would require negative values of the external concentrations. Since the lowest possible external modifier concentration is zero, Mod that still yields the lower boundary for a concentration cIII,Min feasible conditions, is given by: Mod = cIII,Min
Mod QII cDes QII + QF
(14)
This equation holds unless an additional dilution step is introduced. The aforementioned design strategy of maximizing the difference between the two internal modifier concentrations can introduce difficulties, especially when using long columns and a short cycle time. Thus, it has to be further ensured that for a given set of conditions, the cycle time fulfills tS ≥
VC ε QIII
and tS ≥
VC ε QI
(15)
Otherwise, the residence time of a column, for example in zone III, would not be long enough to equilibrate it completely with Mod . In the next cycle, when the correct modifier concentration cIII the column is then located in zone II, this will lead to short pulses of high-modifier concentration moving through zone III. Together with the pulses, the slow-moving component B can travel faster than desired, which may disturb separation. A similar situation can be found in zone I, but since QI is usually chosen to be as high as pressure drop limitations allow, the condition (Eq. (15)) will typically be fulfilled without additional attention. We will discuss possible strategies for ensuring complete column equilibration later. Thus, when choosing internal modifier concentrations, it has to be ensured that (A) the adsorption constants associated with the two gradient steps allow for complete separation (Eqs. (5)–(7)); (B) under the obtained operating conditions, which have to take into account further constraints (e.g. maximum pressure drop), it is feasible to reach the desired concentrations (Eqs. (14) and (15)) More detailed studies on the subject of specifying operating conditions for gradient-SMB processes can be found in Refs. [4,6]. 3. Experimental 3.1. Materials and equipment Model proteins lysozyme and IgG were purchased from Merck (Darmstadt, Germany) and Equitech Bio (Kerrville, TX, USA), respectively. For separation of this model mixture, buffers of varying sodium chloride concentrations and 2 mM NaH2 PO4 at pH 7.0 were used. Production of BMP-2 in recombinant E. coli (Rosetta (DE3), Novagen, Madison, WI, USA) in form of inclusion bodies and the subsequent procedures for solubilization and renaturation have been described before [17,18]. The
73
final renaturation mixture after buffer exchange containing renatured biologically active dimeric BMP-2, non-active monomeric BMP-2 and as other protein contaminants non-identified E. coli proteins in 4 M urea and 20 mM Tris–HCl (pH 8.0) was used as starting material in the chromatographic purification procedure. Prepacked HiTrap Heparin HP 1 mL columns (2.5 cm × 0.7 cm) were purchased from GE Healthcare (Uppsala, Sweden). Heparin acts as a weak cation exchanger for the separation of the model mixture and as an affinity ligand for BMP-2. For isotherm determination and analytical chromatography, one column was used. For SMB separations, two 1 mL columns per zone were connected and used as one 2 mL column, adding up to a total of four 2 mL columns. Isotherm experiments were carried out ¨ using an AKTA prime system (GE Healthcare). Continuous separations were performed using a Knauer CSEP 916 SMB unit (Knauer). Outlet UV profiles were measured using two WellChrom K-2600 UV detectors (Knauer). Additionally, two Amersham UPC-900 (GE Healthcare) were used for on-line measurements of conductivity and UV280 . 3.2. Determination of adsorption isotherms For lysozyme and IgG, adsorption constants were determined from pulse experiments. For each salt concentration (0.1; 0.2; 0.3; 0.4; 0.5 M NaCl), 100 L of 0.2 g L−1 protein in salt buffer were injected. Each injection was repeated at least three times. The retention time was calculated using the first statistical moment. The determination of isotherms for BMP-2 was published recently [8]. The isotherm data obtained in that study are summarized in Table 1. 4. Results of the experimental investigations 4.1. Isotherm measurements for IgG and lysozyme Adsorption constants of IgG and lysozyme were determined for salt concentrations between 0.1 and 0.5 M NaCl (Table 1). The relation between Kk and salt concentration is plotted in Fig. 4. The experimental data were fitted using Eq. (8). Over the range of salt concentrations investigated, lysozyme is better adsorbed than IgG and also exhibits a Table 1 Adsorption constants for IgG and lysozyme as a function of sodium chloride concentration cNaCl (mol L−1 )
KLysozyme
KIgG
0.10 0.18 0.20 0.22 0.25 0.30 0.40 0.45 0.50 0.60
–
1.31
– – 7.94 4.20 2.09 – 1.25 –
1.03 – – 0.71 0.69 – 0.52 –
*
KH,Monomer * – 75.52 – 22.20 – 11.08 3.38 – – –
Isotherm data for BMP-2 are taken from Ref. [8].
KH,Dimer * – 2167 – 1108 – – – 0.432 – 0.061
74
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
Fig. 4. Relation between modifier concentration and adsorption constant for IgG (circles) and lysozyme (diamonds). Data points fitted using Eq. (8): p1IgG = 5.688 L mol−1 ; p2IgG = 0.524; p1Lys = 1.994 L mol−1 ; p2Lys = 2.951.
stronger dependency on the modifier conditions. For sodium chloride concentrations smaller than 0.25 M, no elution profile was detected for lysozyme. The free parameters of Eq. (8) were estimated to be p1IgG = 5.688 L mol−1 ; p2IgG = 0.524; Lysozyme Lysozyme = 1.994 L mol−1 ; p2 = 2.951. p1 4.2. Continuous separation of lysozyme/IgG model mixture After determining adsorption properties for each component as a function of sodium chloride concentration, a 3-zone open-loop gradient-SMB process was realized using a silent zone-implementation (Fig. 3a). Salt concentration in zones I and II was chosen as 0.4 M NaCl, as this concentration led to a sufficient decrease of KLysozyme , allowing for an easier regeneration, without reducing the separation factor to unfavorable values. Following the principle of maximizing gradient difference, a concentration of 0.26 M NaCl was chosen in zone III, requiring a feed concentration of 0 M NaCl. The corresponding flow-rate ratios mII and mIII were selected to fulfill Eqs. (6) and (7). The same holds for mI and Eq. (5). Protein concentration in the feed was 0.5 g L−1 for both compounds. Operating conditions are summarized in Table 2. On-line measurements of UV280 at both outlets are shown in Fig. 5a. The positionTable 2 Operating conditions for 3-zone open-loop gradient separation of IgG and lysozyme Feed concentration Column volume Buffer Gradient KIgG KLysozyme mII /mIII Cycle time QD QF QR QE
0.5 g L−1 each 4 × 2 mL 2 mM NaH2 PO4 Zones I and II: 0.4 M NaCl; zone III: 0.26 M NaCl Zones I and II: 0.7; zone III: 0.82 Zones I and II: 2.1; zone III: 7.0 1.95/4.76 2.2 min 2 mL min−1 0.86 mL min−1 2.5 mL min−1 0.4 mL min−1
Fig. 5. On-line UV280 and conductivity measurements during IgG/lysozyme separation. UV280 profiles recorded at raffinate (dashed line) and extract outlet (solid line) are given in (a). Conductivity signal in zones I and II, corresponding Mod to cI−II (solid line), measured internally at the end of zone II, and the conductivity Mod signal at raffinate outlet after zone III, corresponding to cIII (dashed line), are given in (b).
ing of the gradients was monitored using conductivity detectors at the raffinate outlet and in-line at the end of zone II. The corresponding profiles are given in Fig. 5b. Separation performance was evaluated by an isocratic single-column analysis of collected fractions using 0.25 M NaCl buffer as mobile phase. 50 L of sample were injected in each run. Elution profiles for extract and raffinate samples, collected at the end of the separation, as well as for the feed mixture are shown in Fig. 6. From the recorded chromatograms, it can be deduced that both substances could be obtained pure. Due to gradient operation, it was possible to simultaneously purify and enrich the component eluting from the extract port, in this case lysozyme. Its concentration could be increased almost twofold from 0.5 to 0.9 g L−1 , while IgG was diluted to 0.2 g L−1 . Typically for SMB pro−1 cesses, productivity was high with 0.15 g min−1 stationary phase L −1 for IgG. Eluent for lysozyme and 0.19 g min−1 stationary phase L −1 consumption was found to be 4.2 L g for IgG and 5.6 L g−1 for lysozyme.
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
75
Table 3 Operating conditions for continuous separation of BMP-2 from renaturation mixture Feed concentration Column volume Buffer Gradient KMonomer KDimer mII /mIII Cycle time QD QF QR QE
Fig. 6. Evaluation of separation performance for IgG/lysozyme separation. Isocratic analysis of feed mixture (dotted line), raffinate (dashed line) and extract samples (solid line) were carried out in a HiTrap Heparin HP 1 mL column using 0.25 M NaCl.
The aforementioned downside to maximizing gradient difference can be seen quite evidently in the gradient profiles (Fig. 5b). The combination of a large difference between modifier concentration in zone II and feed, together with a short cycle time, caused very strong fluctuations in gradient conditions. A constant value could not be observed for a longer period. Interestingly, this did not seem to disturb separation. This probably can be ascribed to lysozyme being more adsorbed than IgG over the whole range of salt concentration used and the rather conservative choice of operating conditions.
Dimer: 0.045 g L−1 ; monomer: 0.06 g L−1 4 × 2 mL 4 M urea, 20 mM Tris pH 8.0 Zones I and II: 0.405 M NaCl; zone III: 0.198 M NaCl Zones I and II: 1.05; zone III: 56.81 Zones I and II: 194; zone III: 1818 24.3/300 59.0 min 0.7 mL min−1 1 mL min−1 1.04 mL min−1 0.66 mL min−1
ther complicated by the high urea content in the renaturation buffer. A previous measurement of concentrations showed that both dimer and monomer of BMP-2 are highly diluted in the feed mixture. Therefore, a simple linear isotherm was considered sufficient. UV and conductivity profiles of the separation after reaching cyclic steady state are shown in Fig. 7a and b. Samples collected during separation were analyzed using a two-
4.3. Continuous purification of dimeric BMP-2 from renaturation solution Continuous purification of the therapeutically active dimeric form of BMP-2 from a multicomponent renaturation broth has been studied theoretically in a previous work [8]. Using the isotherm data published in this article, and given here again in Table 1, a 3-zone open-loop gradient-SMB process was designed using an implementation with an equilibration zone (Fig. 3b). It should be remarked that even though Heparin Sepharose was used as stationary phase for both IgG/lysozyme and BMP-2 separation, the mode of interaction was considerably different. For most proteins, Heparin acts as a weak cation exchanger. For coagulation factors and certain growth factors, Heparin acts as an affinity ligand, leading to extremely high values of KDimer up to ∼2000 at 0.2 M NaCl [8]. In contrast to other affinity ligands, however, it is possible to elute bound BMP-2 dimer using a simple salt gradient, thereby greatly simplifying process design. Nevertheless, the high K values have a strong influence on operating conditions, most prominently in form of a very long cycle time tS (here 59 min). The selection of internal modifier concentrations, utilizing computer simulations, is also outlined in Ref. [8]. Again, operating conditions were chosen to fulfill Eqs. (5)–(7). They are summarized in Table 3. Separation is fur-
Fig. 7. On-line UV280 and conductivity measurements during BMP-2 separation. In (a), the UV280 profiles recorded at raffinate (dashed line) and extract outlet (solid line) are given. In (b) the conductivity profiles measured interMod nally at the end of zone II, corresponding to cI−II , and at the raffinate outlet, Mod corresponding to cIII , are given.
76
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
Fig. 8. Evaluation of separation performance for BMP-2 separation. Samples of feed mixture (dotted line), raffinate (dashed line) and extract (solid line) were separated in a HiTrap Heparin HP 1 mL column equilibrated at 0.18 M NaCl using two-step gradient elution (0.3 M NaCl; 0.6 M NaCl).
step gradient, starting at 0.18 M NaCl, and using 0.3 and 0.6 M NaCl as second and third step. A comparison of the elution profiles for feed, raffinate and extract samples (Fig. 8) shows that even though an unknown amount of proteins besides monoand dimeric forms of BMP-2 was present, the dimeric form of BMP-2 (second peak) could be simultaneously isolated and enriched. The purity of the BMP-2 fraction was also verified by SDS–PAGE analysis (results not shown). Based on a comparison of peak areas, its concentration could be increased almost two-fold. A more detailed analysis, focusing on BMP-2 specific effects occurring during separation, will be presented soon [19]. This example shows that when dealing with an unknown number of less adsorbing components, choosing an open-loop SMB configuration really becomes beneficial. If the target component is the most adsorbed substance, and if it is possible to characterize the adsorption behavior of the component closest in terms of adsorptivity, then such a process is feasible regardless of other proteins present. 4.4. Equilibration zone vs. silent zone implementation The case studies presented can be used to illustrate possible advantages or disadvantages of the two types of 3-zone implementations in a moving-column unit studied. It is obvious that the cycle time plays a major role in evaluating possible options, since tS is the time frame available during each cycle. Let us consider the first example, the separation of IgG and lysozyme. A silent-zone implementation in combination with a short residence time of each column in zone III led to a fluctuating salt gradient (Fig. 5b), but did not circumvent separation. This may be different if the components change elution order at a certain salt level, and the boundary is violated due to the aforementioned fluctuations. However, this will not be a common case as most gradient conditions would be selected to avoid
a scenario, where one of the step concentrations is close to such a intersection of isotherm curves. A far more common source of difficulties will be low safety margins when converting the inequalities (5–7) to equalities, whether they were chosen arbitrarily or brought about by the actual separation task. As shown in numerous studies on SMB chromatography, there are optimal sets of operating conditions, which will give maximum productivity for a given purity constraint. However, these conditions are not robust. Even small deviations in an internal volumetric flow can cause a contamination of one or both outlets, and since adsorption constants are often highly dependent on the salt concentration present, a situation like in Fig. 5b might lead to a violation of the inequalities as given by Eqs. (5)–(7). Generally speaking, separation can potentially be disturbed by three different possibilities. The first has already been discussed briefly in Section 2.2, with the residence time of a column in zone III being too short to ensure complete equilibration. The second and third possibilities can arise despite a sufficiently long equilibration time. Due to a large difference in modifier concentration, flow rate or insufficient mixing volumes, it might occur Mod is not yet reached homogethat the resulting concentration cIII nously upon entering the first column of zone III, resulting in local concentration differences. This might lead to an accelerated movement of proteins in small bands of increased salt strength. These differences, however, will mostly be small and the effects not pronounced. The third possible reason for a disturbed separation is more troublesome, because it represents an intrinsic problem of continuous gradient operation. After a column is transported either by port or column switching from high-salt zone I into low-salt zone III, it is still equilibrated with the high-salt buffer. After a minimum time of t0 (the dead time of the column), this buffer is replaced by the low-salt buffer. Two aspects have to be noted, however. Due to dispersion effects, the time necessary to reach the desired salt concentration will be larger than t0 . Secondly, t0 for the salt buffer can differ from t0 for the proteins involved. This is caused by differences in pore accessibility, and thereby in porosity, between the small salt ions and the large proteins. Consequently, even though the highsalt concentration is replaced within a short time, the resulting peak of high elution strength can drag along some of the more adsorbing components in direction of the raffinate outlet. Such a situation was observed in one of the first experiments in the BMP-2 case study (Fig. 9). In the last third of a cycle, a pronounced spike in salt concentration was found at the end of zone III. Consequently, a small amount of BMP-2 dimer was found in raffinate samples. The introduction of an equilibration zone (Fig. 3b), as described in Section 4.3, eliminated the spikes and led to pure outlet streams. The impact using this Mod and type of implementation depends on the difference in cIII Mod cI−II . A large difference can lead to the aforementioned problems sooner than a small difference, so the beneficial effect of using pre-equilibration will be more prominent in the first case. Thus, using a design strategy that emphasizes a maximized gradient difference has to be treated with a certain amount of caution, if an equilibration step cannot be used. For some combinations of column dimensions, backpressure limitations and cycle times, it might not be possible to equilibrate the col-
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
77
enhances internal gradient stability and process performance. If required, both sanitization and equilibration can be integrated in the described continuous chromatographic separation process. Nomenclature
Fig. 9. On-line conductivity profile of a BMP-2 separation in silent zone implementation.
umn completely within the given time frame. It should be noted, though, that even a partial pre-equilibration will be better than no pre-equilibration. A logical next step is to combine equilibration with an integrated sanitization step. Due to the long cycle time in the second case study, an implementation of the standard sanitization protocol of 30 min at 0.5 M NaOH, followed by an equilibration step, can be easily realized. For the first case study, pre-equilibrating the columns before entering zone III will lead to far more stable gradient conditions. An integrated CIP step, however, is hardly possible for the operating conditions selected in this example. It has to be evaluated case by case, if a “fast” SMB process, which has to be shut down periodically to perform CIP, is more productive than a “slower” SMB process with an integrated CIP.
c K m p q Q tS t0 V
fluid phase concentration (g L−1 ) adsorption constant flow-rate ratio free parameters in Eq. (8) absorbed phase concentration (g L−1 ) fluid phase flow rate (mL min−1 ) cycle time (min) column dead time (min) volume (mL)
Greek letters ε porosity Roman letters I, II,III zone number Subscripts and superscripts A,B components D desorbent E extract F feed k component index Mod modifier R raffinate z zone Acknowledgements
5. Conclusions The separation of both a model mixture of IgG/lysozyme and the purification of dimeric BMP-2 from a multicomponent mixture using two different implementations of a 3-zone open-loop gradient-SMB process in a 16-port valve was studied experimentally. First, isotherm data for both model proteins were determined as functions of sodium chloride concentration. It was found that equilibrium could be described reasonably well with linear isotherms. Based on this data, a SMB process using a disconnected zone IV was successfully designed and realized for the model system. Despite strong fluctuations of internal salt gradients, both proteins could be obtained pure. The use of gradients led to an almost two-fold increase of lysozyme concentration in the extract stream. Employing isotherm data obtained in previous studies, a continuous process using an equilibration-zone implementation was successfully designed and realized for BMP-2 separation. Dimeric BMP-2 could be simultaneously purified and enriched. The influence of column pre-equilibration on continuous gradient separation processes was discussed and illustrated on the basis of the case studies presented. It was found that pre-equilibration
The support of Deutsche Forschungsgemeinschaft/SFB 578, Fonds der Chemischen Industrie and “Bayer-Schering AG” is gratefully acknowledged. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
D.B. Broughton, C.G. Gerhold, US Patent 2,985,589 (1961). R.M. Nicoud, Pharm. Technol. Eur. 11 (1999) 28. R.M. Nicoud, M. Perrut, G. Hotier, US Patent 5,422,007 (1995). S. Abel, M. Mazzotti, M. Morbidelli, J. Chromatogr. A 944 (2002) 23. D. Antos, A. Seidel-Morgenstern, J. Chromatogr. A 944 (2002) 77. D. Antos, A. Seidel-Morgenstern, Chem. Eng. Sci. 56 (2001) 6667. T.B. Jensen, T.G.P. Reijns, H.A.H. Billiet, L.A.M. van der Wielen, J. Chromatogr. A 873 (2000) 149. L. Gueorguieva, L.F. Vallejo, U. Rinas, A. Seidel-Morgenstern, J. Chromatogr. A 1135 (2006) 142. K. Wekenborg, A. Susanto, S.S. Frederiksen, H. Schmidt-Traub, Chem. Ing. Tech. 76 (2004) 815. G. Strohlein, L. Aumann, M. Mazzotti, M. Morbidelli, J. Chromatogr. A 1126 (2006) 338. C.Y. Chin, N.-H.L. Wang, Sep. Purif. Rev. 33 (2004) 77.
78
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
[12] O. Ludemann-Hombourger, R.M. Nicoud, M. Bailly, Sep. Sci. Technol. 35 (2000) 1829. [13] S. Abel, M.U. B¨abler, C. Arpagaus, M. Mazzotti, J. Stadler, J. Chromatogr. A 1043 (2004) 201. [14] G. Paredes, S. Makart, J. Stadler, M. Mazzotti, Chem. Eng. Technol. 28 (2005) 1335. [15] M. Mazzotti, G. Storti, M. Morbidelli, J. Chromatogr. A 769 (1997) 3.
[16] C. Migliorini, M. Mazzotti, M. Morbidelli, AIChE J. 45 (1999) 1411. [17] L.F. Vallejo, M. Brokelmann, S. Marten, S. Trappe, J. Cabrera-Crespo, A. Hoffmann, G. Gross, H.A. Weich, U. Rinas, J. Biotechnol. 94 (2002) 185. [18] L.F. Vallejo, U. Rinas, Biotechnol. Bioeng. 85 (2004) 601. [19] L. Gueorguieva, H. Baars-Hibbe, U. Rinas, A. Seidel-Morgenstern, J. Chromatogr., in preparation.
Journal of Chromatography A, 1176 (2007) 69–78
Step gradients in 3-zone simulated moving bed chromatography Application to the purification of antibodies and bone morphogenetic protein-2夽 Lars Christian Keler a,∗ , Ludmila Gueorguieva b , Ursula Rinas c , Andreas Seidel-Morgenstern a,b b
a Max Planck Institute for Dynamics of Complex Technical Systems, Sandtorstr. 1, D-39106 Magdeburg, Germany Institute of Process Engineering, Otto von Guericke University, Universit¨atsplatz 2, D-39106 Magdeburg, Germany c Helmholtz Centre for Infection Research, Inhoffenstr. 7, D-38124 Braunschweig, Germany
Received 6 July 2007; received in revised form 16 October 2007; accepted 19 October 2007 Available online 1 November 2007
Abstract The simulated moving bed (SMB) technology is a proven tool for efficient separation of binary mixtures. However, relying on isocratic conditions limits the applicability of the classical SMB approach when considering the field of bioseparations. Here, the use of gradients opens up new possibilities. A gradient in a SMB process can be established by using different solvent strengths in the incoming feed and desorbent streams, resulting in two internal plateaus of elution strength. Thus, compared to the conventional process, the overall amount of solvent needed can be reduced, productivity can be increased and more concentrated product streams can be obtained. In this contribution, two case studies will be presented. At first, the separation of bovine IgG from lysozyme will be analyzed as a model system. Antibodies are a common target substance in bio-chromatography, as therapeutic monoclonal antibodies are among the most promising biopharmaceuticals. Using adsorption data obtained from single-column experiments, an appropriate SMB process was designed and implemented. The second target component is the active dimeric form of the bone morphogenetic protein-2 (BMP-2). This protein was isolated from a renaturation solution, which also contained its inactive monomeric form as well as other undefined proteins from the bacterial production strain. A 3-zone open-loop gradient-SMB approach was used successfully for both separations. © 2007 Elsevier B.V. All rights reserved. Keywords: Simulated moving bed; Gradients; Protein separation; BMP-2; Antibodies; Moving-column implementation; Equilibration; Open-loop
1. Introduction Chromatographic separation techniques are applied throughout the chemical, pharmaceutical and biotechnological industry for the purification of high-value products. The use of gradients to increase process performance has become state-of-the-art in discontinuous chromatography. In continuous chromatography, however, the majority of separations are still run isocratically. The most commonly used continuous process is simulated moving bed (SMB) chromatography. This concept was invented by UOP in the early 1960s and uses an intelligent movement 夽 Presented at the 20th International Symposium on Preparative and Process Chromatography, Baltimore, MD, USA, 3–6 June 2007. ∗ Corresponding author. Tel.: +49 391 6110 442; fax: +49 391 6110 403. E-mail address: [email protected] (L.C. Keler).
0021-9673/$ – see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.chroma.2007.10.087
of ports (moving-port implementation) or columns (movingcolumn implementation) to mimic a counter-current movement of liquid and solid phases [1]. The SMB process was found often to be superior to its discontinuous, isocratic counterpart in terms of achievable purity, productivity and reduced solvent consumption [2]. The idea of using gradients in a SMB process was described first by Nicoud et al. using a supercritical eluent [3]. In recent years, the more convenient use of two differently composed liquid mobile phases has been studied in some detail, e.g. [4–7]. As in batch chromatography, the use of gradients can further increase productivity and reduce solvent consumption. Additionally, separation problems exist that cannot be solved under isocratic conditions. Most of the contributions reported up to now deal with the separation of “small molecules” like cycloketones or nucleosides. In the emerging field of bioseparations, where undefined multicomponent mix-
70
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
tures are common and interactions are often more complex, only few gradient-SMB applications have been reported (e.g. [8,9]). In this context, an innovative continuous process which also uses gradient conditions was presented recently as an alternative to SMB chromatography [10]. Two case studies for continuous protein separation with a two-step gradient-SMB process are presented in this contribution. At first, the separation of a binary model mixture, composed of bovine immunoglobulin G (IgG) and lysozyme, using immobilized Heparin as weak cation exchanger is presented. The information gathered with this model system are then used as guidelines for implementing the purification of the active dimeric form of bone morphogenetic protein-2 (BMP-2) from its monomeric form as well as from numerous, unidentified other components. 2. Three zone two-step gradient-SMB technology All modifications suggested here originate in the classical isocratic closed-loop 4-zone SMB process (Fig. 1a, [1]). In this process, each of the four zones, framed by outlet and inlet ports, has to fulfill a specific task. The stream of the feed mixture to be separated, QF , enters the unit between zones II and III. In these so-called separation zones, the conditions selected have to ensure that the less adsorbing component A moves upstream with the mobile phase, where it is withdrawn at the raffinate outlet. At the same time, the more absorbing component B should be transported in counter-current direction by the periodic switching of either columns or ports, and likewise withdrawn at the extract port. Zones I and IV are regeneration zones; here, either the solid phase (zone I) or the mobile phase (zone IV) is freed from components A or B, respectively. This closed-loop approach reduces solvent consumption, but also introduces the risk of cross-contamination. If feed mixtures are to be processed which contain unknown or uncharacterized components, for example fermentation broths, one might inadvertently select conditions that do not prevent all substances from leaving zone IV. This leads to a loss of product in the raffinate outlet and a loss of purity in the extract outlet, which is especially troublesome if the target substance is an expensive, therapeutic compound. Thus, it can be beneficial to avoid closed-loop operation and use an open-loop configuration, leading to an increase in solvent consumption. However, this factor carries less weight when the mobile phase is inexpensive, as it is often the case in biochromatography. There, cheap and easy-to-handle water–salt solutions are most common. Also, disposal is less problematic when compared to the majority of processes for small molecules, where organic solvent are frequently used, renderring an open-loop configuration an attractive alternative. In an open-loop configuration, the recycling zone of the liquid phase (IV) can be removed, reducing the total number of zones to three (Fig. 1b). Additionally, in the process used for this study, a two-step gradient is realized by using different concentrations of a selectable modifier in desorbent and feed. This leads to a first plateau of solvent strength in zones I and II, which is equal to the solvent strength introduced at the desorbent port, and a second plateau in zone III, where the modifier level is determined by both the concentration of the
modifier in the feed stream and the first plateau concentration. Typically, in zones I and II high elution strength is set to better mobilize the more adsorbing component B. Similarly, the modifier concentration in the stream leaving zone II is reduced by the feed to reach lower elution strength in zone III, in order to promote adsorption of component B. Strategies for determining suitable operating conditions will be shortly discussed later in this contribution. 2.1. Realization of 3-zone SMB modes in a moving-column unit As described previously, the SMB principle can be realized using either a moving-port or a moving-column type of implementation [11]. Both are equivalent in theory. Practically, this holds for basic 4-zone operation, but if more sophisticated operating modes, for example the VariCol-process [12] or uneven numbers of zones, are considered, there are notable differences. The moving-port implementation uses multiple
Fig. 1. Schematic representations of a classical, closed-loop isocratic 4-zone SMB process [1] (a) and the open-loop gradient 3-zone SMB process (b) used in this study. In (b), the position of the first gradient plateau (high elution strength) is indicated by the crossed lines in the columns of zones I and II.
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
multiposition valves. Here, every valve can be operated at a different time, offering a high degree of flexibility. It has to be ensured, however, that when a simultaneous switching of all valves is desired, there are no delays or deviations. In the moving-column implementation, this aspect is taken care of by the design of the one single valve, with which all pumps and columns are directly connected. As these valves are typically designed with the conventional 4-zone operation in mind, the number of ports generally is a multiple of 4, often 12 or 16. This design implies that all columns can only be moved at the same time. For a 16-port valve, this complicates 3-zone operation (as 16 cannot be divided by three integrally). An example of a possible, but problematic realization of three zones in a 16-port multiposition valve, as supplied, for example, by Knauer (Berlin, Germany), is shown in Fig. 2. In Fig. 2a the situation at the beginning of the first cycle is given, with one column being at the beginning of each zone. The situation after 4 switching events, at the end of the first cycle, is shown in Fig. 2b. Still, one column can be found in each zone. But after one additional switch (Fig. 2c), there are now two columns in zone III and none in zone II. Over the course of the next cycles, this will occur in all three zones, which will inevitably lead to a decrease in purity. To avoid this problem, two possible alternative schemes are proposed. The first scheme (Fig. 3a) uses a disconnected (“silent”) zone IV, together with one additional column. During operation, only three columns participate actively, the fourth is used solely to generate the needed symmetry. The second scheme (Fig. 3b) utilizes the additional fourth zone to preequilibrate the column leaving the high-modifier level zone I with the lower modifier concentration selected for zone III. Again, four columns are used, but only three take part in separation. Depending on the separation problem and the cycle time, it is also possible to use zone IV for an integrated cleaning-inplace (CIP). For moving-port implementations, a similar CIP scheme was suggested in Refs. [13,14]. For the sake of clarity, it is necessary to have a closer look at the physical meaning of the cycle time tS when comparing moving-port and moving-column implementations. When performing a regular SMB separation using a moving-port apparatus, the switching of each multiposition valve occurs once at the end of each cycle. Consequently, tS is often also referred to as switching time. This term might be misleading when considering a moving-column implementation. There, each zone contains multiple valve ports, for example four as shown in Fig. 3. In order to realize the correct counter-current movement, the valve has to be operated four times within a cycle time, so the occurrence of a switching event is not equivalent to tS . In this publication, tS can be understood as the necessary residence time of a column in a zone, as this definition holds for both types of implementations. 2.2. Calculation of operating conditions based on equilibrium theory The selection of both internal and external flow rates, the corresponding cycle time and suitable modifier concentrations in the inlet streams requires a connection between these operat-
71
Fig. 2. Schematic representation of valve and column positions in an erroneous moving-column implementation of a 3-zone process using a 16-port valve. In (a–c) the positions of the three columns used are indicated by black dots at the beginning of the first cycle (a), after moving each column for four times, at the end of the first cycle (b), and at the beginning of the second cycle (c).
ing conditions and characteristic properties of the components to be separated. This link can be found in the distribution equilibria between the liquid phase concentration of a component k, ck , and the corresponding solid phase concentration, qk . For bioseparations, this relation is often described with the Langmuir isotherm. In the case of the separation problems discussed
72
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
A component k in zone z moves with the liquid phase, if mz is larger than Kk , and in counter-current direction, with the solid phase, when mz is smaller than Kk . The knowledge, which component has to be moved in which direction to achieve complete separation, can then be cast into two inequalities: KB < mI
(3)
KA < mII < mIII < KB
(4)
For gradient separations, the basic principle still holds, but due to the different modifier levels in zones I–II and III the inequalities have to be adjusted [4,6] B Mod mI > KI−II (cD )
(5)
A Mod B Mod KI−II (cD ) < mII < KI−II (cD )
(6)
A Mod B Mod KIII (cIII ) < mIII < KIII (cIII )
(7)
From Eqs. (5)–(7) it is quite obvious that a larger difference of k mII /mIII can be achieved with gradients, as KI−II will be smaller k than KIII , leading to an increase in productivity. Consequently, it is desirable to maximize the difference in modifier concentration between zones I–II and III [5]. In order to calculate operating conditions, the relation between Kk and cMod must be determined experimentally. Often, a simple equation as given in Ref. [5] can be used to describe the equilibrium functions: −p2k
Kk (cMod ) = (p1k cMod )
Fig. 3. Schematic representation of valve and column positions of two feasible moving-column implementation of a 3-zone process using a 16-port valve. In (a) is given the scheme of a 3-zone SMB process in a 16-port valve using a disconnected (silent) fourth zone. In (b) is given the scheme of a 3-zone SMB process in a 16-port valve using the fourth zone for pre-equilibration. In both (a) and (b), the positions of the columns used are indicated by black dots.
in this study, it was found that the adsorption behavior could be described reasonably well with a simple linear isotherm (Eq. (1)). For two components A and B, isotherms can be written as qk = K k c k
with k = A, B
(1)
with component A being less retained than B (KA < KB ). For isocratic conditions, it is straightforward to determine operating conditions using the well-known equilibrium theory. Key to this analysis is the introduction of dimensionless flow-rate ratios mz [15]. mz = QZ
(tS − VC ε) VC (1 − ε)
with
z = I, II, III
(2)
The direction of a component k in zone z depends on the relation between its adsorption constant KK , quantifying the strength of interaction with the stationary phase, and the flow-rate ratio mz .
(8)
After selecting one flow rate, usually QI or QF , it is possible to calculate the corresponding cycle time from Eq. (2). Afterwards, the same equation can be used to convert the chosen m-ratios to volumetric flow rates. This conversion, however, does not account for additional extra-column volumes. While this is fairly unproblematic for large columns, this does not hold for small columns as the ones used in this study. A convenient way to account for extra-column dead volume is given in Ref. [16]. The flow rates further have to fulfill the following mass balances: QI = QD
(9)
QII = QI − QE
(10)
QIII = QII + QF
(11)
When compared to the classical 4-zone process, there is a notable difference in Eq. (9). Due to the loss of zone IV, QI is now sustained only by the external desorbent stream. For the same reason, the stream leaving zone III, QIII , is completely considered as raffinate stream QR . Besides the mobile phase mass balance, the modifier mass balance also has to be fulfilled for gradient separations. It can be given as Mod Mod cI−II = cD Mod cIII =
Mod + Q cMod QII cD F F QII + QF
(12) (13)
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
From these equations, it is obvious that certain combinations of internal modifier concentrations are not realizable, because they would require negative values of the external concentrations. Since the lowest possible external modifier concentration is zero, Mod that still yields the lower boundary for a concentration cIII,Min feasible conditions, is given by: Mod = cIII,Min
Mod QII cDes QII + QF
(14)
This equation holds unless an additional dilution step is introduced. The aforementioned design strategy of maximizing the difference between the two internal modifier concentrations can introduce difficulties, especially when using long columns and a short cycle time. Thus, it has to be further ensured that for a given set of conditions, the cycle time fulfills tS ≥
VC ε QIII
and tS ≥
VC ε QI
(15)
Otherwise, the residence time of a column, for example in zone III, would not be long enough to equilibrate it completely with Mod . In the next cycle, when the correct modifier concentration cIII the column is then located in zone II, this will lead to short pulses of high-modifier concentration moving through zone III. Together with the pulses, the slow-moving component B can travel faster than desired, which may disturb separation. A similar situation can be found in zone I, but since QI is usually chosen to be as high as pressure drop limitations allow, the condition (Eq. (15)) will typically be fulfilled without additional attention. We will discuss possible strategies for ensuring complete column equilibration later. Thus, when choosing internal modifier concentrations, it has to be ensured that (A) the adsorption constants associated with the two gradient steps allow for complete separation (Eqs. (5)–(7)); (B) under the obtained operating conditions, which have to take into account further constraints (e.g. maximum pressure drop), it is feasible to reach the desired concentrations (Eqs. (14) and (15)) More detailed studies on the subject of specifying operating conditions for gradient-SMB processes can be found in Refs. [4,6]. 3. Experimental 3.1. Materials and equipment Model proteins lysozyme and IgG were purchased from Merck (Darmstadt, Germany) and Equitech Bio (Kerrville, TX, USA), respectively. For separation of this model mixture, buffers of varying sodium chloride concentrations and 2 mM NaH2 PO4 at pH 7.0 were used. Production of BMP-2 in recombinant E. coli (Rosetta (DE3), Novagen, Madison, WI, USA) in form of inclusion bodies and the subsequent procedures for solubilization and renaturation have been described before [17,18]. The
73
final renaturation mixture after buffer exchange containing renatured biologically active dimeric BMP-2, non-active monomeric BMP-2 and as other protein contaminants non-identified E. coli proteins in 4 M urea and 20 mM Tris–HCl (pH 8.0) was used as starting material in the chromatographic purification procedure. Prepacked HiTrap Heparin HP 1 mL columns (2.5 cm × 0.7 cm) were purchased from GE Healthcare (Uppsala, Sweden). Heparin acts as a weak cation exchanger for the separation of the model mixture and as an affinity ligand for BMP-2. For isotherm determination and analytical chromatography, one column was used. For SMB separations, two 1 mL columns per zone were connected and used as one 2 mL column, adding up to a total of four 2 mL columns. Isotherm experiments were carried out ¨ using an AKTA prime system (GE Healthcare). Continuous separations were performed using a Knauer CSEP 916 SMB unit (Knauer). Outlet UV profiles were measured using two WellChrom K-2600 UV detectors (Knauer). Additionally, two Amersham UPC-900 (GE Healthcare) were used for on-line measurements of conductivity and UV280 . 3.2. Determination of adsorption isotherms For lysozyme and IgG, adsorption constants were determined from pulse experiments. For each salt concentration (0.1; 0.2; 0.3; 0.4; 0.5 M NaCl), 100 L of 0.2 g L−1 protein in salt buffer were injected. Each injection was repeated at least three times. The retention time was calculated using the first statistical moment. The determination of isotherms for BMP-2 was published recently [8]. The isotherm data obtained in that study are summarized in Table 1. 4. Results of the experimental investigations 4.1. Isotherm measurements for IgG and lysozyme Adsorption constants of IgG and lysozyme were determined for salt concentrations between 0.1 and 0.5 M NaCl (Table 1). The relation between Kk and salt concentration is plotted in Fig. 4. The experimental data were fitted using Eq. (8). Over the range of salt concentrations investigated, lysozyme is better adsorbed than IgG and also exhibits a Table 1 Adsorption constants for IgG and lysozyme as a function of sodium chloride concentration cNaCl (mol L−1 )
KLysozyme
KIgG
0.10 0.18 0.20 0.22 0.25 0.30 0.40 0.45 0.50 0.60
–
1.31
– – 7.94 4.20 2.09 – 1.25 –
1.03 – – 0.71 0.69 – 0.52 –
*
KH,Monomer * – 75.52 – 22.20 – 11.08 3.38 – – –
Isotherm data for BMP-2 are taken from Ref. [8].
KH,Dimer * – 2167 – 1108 – – – 0.432 – 0.061
74
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
Fig. 4. Relation between modifier concentration and adsorption constant for IgG (circles) and lysozyme (diamonds). Data points fitted using Eq. (8): p1IgG = 5.688 L mol−1 ; p2IgG = 0.524; p1Lys = 1.994 L mol−1 ; p2Lys = 2.951.
stronger dependency on the modifier conditions. For sodium chloride concentrations smaller than 0.25 M, no elution profile was detected for lysozyme. The free parameters of Eq. (8) were estimated to be p1IgG = 5.688 L mol−1 ; p2IgG = 0.524; Lysozyme Lysozyme = 1.994 L mol−1 ; p2 = 2.951. p1 4.2. Continuous separation of lysozyme/IgG model mixture After determining adsorption properties for each component as a function of sodium chloride concentration, a 3-zone open-loop gradient-SMB process was realized using a silent zone-implementation (Fig. 3a). Salt concentration in zones I and II was chosen as 0.4 M NaCl, as this concentration led to a sufficient decrease of KLysozyme , allowing for an easier regeneration, without reducing the separation factor to unfavorable values. Following the principle of maximizing gradient difference, a concentration of 0.26 M NaCl was chosen in zone III, requiring a feed concentration of 0 M NaCl. The corresponding flow-rate ratios mII and mIII were selected to fulfill Eqs. (6) and (7). The same holds for mI and Eq. (5). Protein concentration in the feed was 0.5 g L−1 for both compounds. Operating conditions are summarized in Table 2. On-line measurements of UV280 at both outlets are shown in Fig. 5a. The positionTable 2 Operating conditions for 3-zone open-loop gradient separation of IgG and lysozyme Feed concentration Column volume Buffer Gradient KIgG KLysozyme mII /mIII Cycle time QD QF QR QE
0.5 g L−1 each 4 × 2 mL 2 mM NaH2 PO4 Zones I and II: 0.4 M NaCl; zone III: 0.26 M NaCl Zones I and II: 0.7; zone III: 0.82 Zones I and II: 2.1; zone III: 7.0 1.95/4.76 2.2 min 2 mL min−1 0.86 mL min−1 2.5 mL min−1 0.4 mL min−1
Fig. 5. On-line UV280 and conductivity measurements during IgG/lysozyme separation. UV280 profiles recorded at raffinate (dashed line) and extract outlet (solid line) are given in (a). Conductivity signal in zones I and II, corresponding Mod to cI−II (solid line), measured internally at the end of zone II, and the conductivity Mod signal at raffinate outlet after zone III, corresponding to cIII (dashed line), are given in (b).
ing of the gradients was monitored using conductivity detectors at the raffinate outlet and in-line at the end of zone II. The corresponding profiles are given in Fig. 5b. Separation performance was evaluated by an isocratic single-column analysis of collected fractions using 0.25 M NaCl buffer as mobile phase. 50 L of sample were injected in each run. Elution profiles for extract and raffinate samples, collected at the end of the separation, as well as for the feed mixture are shown in Fig. 6. From the recorded chromatograms, it can be deduced that both substances could be obtained pure. Due to gradient operation, it was possible to simultaneously purify and enrich the component eluting from the extract port, in this case lysozyme. Its concentration could be increased almost twofold from 0.5 to 0.9 g L−1 , while IgG was diluted to 0.2 g L−1 . Typically for SMB pro−1 cesses, productivity was high with 0.15 g min−1 stationary phase L −1 for IgG. Eluent for lysozyme and 0.19 g min−1 stationary phase L −1 consumption was found to be 4.2 L g for IgG and 5.6 L g−1 for lysozyme.
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
75
Table 3 Operating conditions for continuous separation of BMP-2 from renaturation mixture Feed concentration Column volume Buffer Gradient KMonomer KDimer mII /mIII Cycle time QD QF QR QE
Fig. 6. Evaluation of separation performance for IgG/lysozyme separation. Isocratic analysis of feed mixture (dotted line), raffinate (dashed line) and extract samples (solid line) were carried out in a HiTrap Heparin HP 1 mL column using 0.25 M NaCl.
The aforementioned downside to maximizing gradient difference can be seen quite evidently in the gradient profiles (Fig. 5b). The combination of a large difference between modifier concentration in zone II and feed, together with a short cycle time, caused very strong fluctuations in gradient conditions. A constant value could not be observed for a longer period. Interestingly, this did not seem to disturb separation. This probably can be ascribed to lysozyme being more adsorbed than IgG over the whole range of salt concentration used and the rather conservative choice of operating conditions.
Dimer: 0.045 g L−1 ; monomer: 0.06 g L−1 4 × 2 mL 4 M urea, 20 mM Tris pH 8.0 Zones I and II: 0.405 M NaCl; zone III: 0.198 M NaCl Zones I and II: 1.05; zone III: 56.81 Zones I and II: 194; zone III: 1818 24.3/300 59.0 min 0.7 mL min−1 1 mL min−1 1.04 mL min−1 0.66 mL min−1
ther complicated by the high urea content in the renaturation buffer. A previous measurement of concentrations showed that both dimer and monomer of BMP-2 are highly diluted in the feed mixture. Therefore, a simple linear isotherm was considered sufficient. UV and conductivity profiles of the separation after reaching cyclic steady state are shown in Fig. 7a and b. Samples collected during separation were analyzed using a two-
4.3. Continuous purification of dimeric BMP-2 from renaturation solution Continuous purification of the therapeutically active dimeric form of BMP-2 from a multicomponent renaturation broth has been studied theoretically in a previous work [8]. Using the isotherm data published in this article, and given here again in Table 1, a 3-zone open-loop gradient-SMB process was designed using an implementation with an equilibration zone (Fig. 3b). It should be remarked that even though Heparin Sepharose was used as stationary phase for both IgG/lysozyme and BMP-2 separation, the mode of interaction was considerably different. For most proteins, Heparin acts as a weak cation exchanger. For coagulation factors and certain growth factors, Heparin acts as an affinity ligand, leading to extremely high values of KDimer up to ∼2000 at 0.2 M NaCl [8]. In contrast to other affinity ligands, however, it is possible to elute bound BMP-2 dimer using a simple salt gradient, thereby greatly simplifying process design. Nevertheless, the high K values have a strong influence on operating conditions, most prominently in form of a very long cycle time tS (here 59 min). The selection of internal modifier concentrations, utilizing computer simulations, is also outlined in Ref. [8]. Again, operating conditions were chosen to fulfill Eqs. (5)–(7). They are summarized in Table 3. Separation is fur-
Fig. 7. On-line UV280 and conductivity measurements during BMP-2 separation. In (a), the UV280 profiles recorded at raffinate (dashed line) and extract outlet (solid line) are given. In (b) the conductivity profiles measured interMod nally at the end of zone II, corresponding to cI−II , and at the raffinate outlet, Mod corresponding to cIII , are given.
76
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
Fig. 8. Evaluation of separation performance for BMP-2 separation. Samples of feed mixture (dotted line), raffinate (dashed line) and extract (solid line) were separated in a HiTrap Heparin HP 1 mL column equilibrated at 0.18 M NaCl using two-step gradient elution (0.3 M NaCl; 0.6 M NaCl).
step gradient, starting at 0.18 M NaCl, and using 0.3 and 0.6 M NaCl as second and third step. A comparison of the elution profiles for feed, raffinate and extract samples (Fig. 8) shows that even though an unknown amount of proteins besides monoand dimeric forms of BMP-2 was present, the dimeric form of BMP-2 (second peak) could be simultaneously isolated and enriched. The purity of the BMP-2 fraction was also verified by SDS–PAGE analysis (results not shown). Based on a comparison of peak areas, its concentration could be increased almost two-fold. A more detailed analysis, focusing on BMP-2 specific effects occurring during separation, will be presented soon [19]. This example shows that when dealing with an unknown number of less adsorbing components, choosing an open-loop SMB configuration really becomes beneficial. If the target component is the most adsorbed substance, and if it is possible to characterize the adsorption behavior of the component closest in terms of adsorptivity, then such a process is feasible regardless of other proteins present. 4.4. Equilibration zone vs. silent zone implementation The case studies presented can be used to illustrate possible advantages or disadvantages of the two types of 3-zone implementations in a moving-column unit studied. It is obvious that the cycle time plays a major role in evaluating possible options, since tS is the time frame available during each cycle. Let us consider the first example, the separation of IgG and lysozyme. A silent-zone implementation in combination with a short residence time of each column in zone III led to a fluctuating salt gradient (Fig. 5b), but did not circumvent separation. This may be different if the components change elution order at a certain salt level, and the boundary is violated due to the aforementioned fluctuations. However, this will not be a common case as most gradient conditions would be selected to avoid
a scenario, where one of the step concentrations is close to such a intersection of isotherm curves. A far more common source of difficulties will be low safety margins when converting the inequalities (5–7) to equalities, whether they were chosen arbitrarily or brought about by the actual separation task. As shown in numerous studies on SMB chromatography, there are optimal sets of operating conditions, which will give maximum productivity for a given purity constraint. However, these conditions are not robust. Even small deviations in an internal volumetric flow can cause a contamination of one or both outlets, and since adsorption constants are often highly dependent on the salt concentration present, a situation like in Fig. 5b might lead to a violation of the inequalities as given by Eqs. (5)–(7). Generally speaking, separation can potentially be disturbed by three different possibilities. The first has already been discussed briefly in Section 2.2, with the residence time of a column in zone III being too short to ensure complete equilibration. The second and third possibilities can arise despite a sufficiently long equilibration time. Due to a large difference in modifier concentration, flow rate or insufficient mixing volumes, it might occur Mod is not yet reached homogethat the resulting concentration cIII nously upon entering the first column of zone III, resulting in local concentration differences. This might lead to an accelerated movement of proteins in small bands of increased salt strength. These differences, however, will mostly be small and the effects not pronounced. The third possible reason for a disturbed separation is more troublesome, because it represents an intrinsic problem of continuous gradient operation. After a column is transported either by port or column switching from high-salt zone I into low-salt zone III, it is still equilibrated with the high-salt buffer. After a minimum time of t0 (the dead time of the column), this buffer is replaced by the low-salt buffer. Two aspects have to be noted, however. Due to dispersion effects, the time necessary to reach the desired salt concentration will be larger than t0 . Secondly, t0 for the salt buffer can differ from t0 for the proteins involved. This is caused by differences in pore accessibility, and thereby in porosity, between the small salt ions and the large proteins. Consequently, even though the highsalt concentration is replaced within a short time, the resulting peak of high elution strength can drag along some of the more adsorbing components in direction of the raffinate outlet. Such a situation was observed in one of the first experiments in the BMP-2 case study (Fig. 9). In the last third of a cycle, a pronounced spike in salt concentration was found at the end of zone III. Consequently, a small amount of BMP-2 dimer was found in raffinate samples. The introduction of an equilibration zone (Fig. 3b), as described in Section 4.3, eliminated the spikes and led to pure outlet streams. The impact using this Mod and type of implementation depends on the difference in cIII Mod cI−II . A large difference can lead to the aforementioned problems sooner than a small difference, so the beneficial effect of using pre-equilibration will be more prominent in the first case. Thus, using a design strategy that emphasizes a maximized gradient difference has to be treated with a certain amount of caution, if an equilibration step cannot be used. For some combinations of column dimensions, backpressure limitations and cycle times, it might not be possible to equilibrate the col-
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
77
enhances internal gradient stability and process performance. If required, both sanitization and equilibration can be integrated in the described continuous chromatographic separation process. Nomenclature
Fig. 9. On-line conductivity profile of a BMP-2 separation in silent zone implementation.
umn completely within the given time frame. It should be noted, though, that even a partial pre-equilibration will be better than no pre-equilibration. A logical next step is to combine equilibration with an integrated sanitization step. Due to the long cycle time in the second case study, an implementation of the standard sanitization protocol of 30 min at 0.5 M NaOH, followed by an equilibration step, can be easily realized. For the first case study, pre-equilibrating the columns before entering zone III will lead to far more stable gradient conditions. An integrated CIP step, however, is hardly possible for the operating conditions selected in this example. It has to be evaluated case by case, if a “fast” SMB process, which has to be shut down periodically to perform CIP, is more productive than a “slower” SMB process with an integrated CIP.
c K m p q Q tS t0 V
fluid phase concentration (g L−1 ) adsorption constant flow-rate ratio free parameters in Eq. (8) absorbed phase concentration (g L−1 ) fluid phase flow rate (mL min−1 ) cycle time (min) column dead time (min) volume (mL)
Greek letters ε porosity Roman letters I, II,III zone number Subscripts and superscripts A,B components D desorbent E extract F feed k component index Mod modifier R raffinate z zone Acknowledgements
5. Conclusions The separation of both a model mixture of IgG/lysozyme and the purification of dimeric BMP-2 from a multicomponent mixture using two different implementations of a 3-zone open-loop gradient-SMB process in a 16-port valve was studied experimentally. First, isotherm data for both model proteins were determined as functions of sodium chloride concentration. It was found that equilibrium could be described reasonably well with linear isotherms. Based on this data, a SMB process using a disconnected zone IV was successfully designed and realized for the model system. Despite strong fluctuations of internal salt gradients, both proteins could be obtained pure. The use of gradients led to an almost two-fold increase of lysozyme concentration in the extract stream. Employing isotherm data obtained in previous studies, a continuous process using an equilibration-zone implementation was successfully designed and realized for BMP-2 separation. Dimeric BMP-2 could be simultaneously purified and enriched. The influence of column pre-equilibration on continuous gradient separation processes was discussed and illustrated on the basis of the case studies presented. It was found that pre-equilibration
The support of Deutsche Forschungsgemeinschaft/SFB 578, Fonds der Chemischen Industrie and “Bayer-Schering AG” is gratefully acknowledged. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
D.B. Broughton, C.G. Gerhold, US Patent 2,985,589 (1961). R.M. Nicoud, Pharm. Technol. Eur. 11 (1999) 28. R.M. Nicoud, M. Perrut, G. Hotier, US Patent 5,422,007 (1995). S. Abel, M. Mazzotti, M. Morbidelli, J. Chromatogr. A 944 (2002) 23. D. Antos, A. Seidel-Morgenstern, J. Chromatogr. A 944 (2002) 77. D. Antos, A. Seidel-Morgenstern, Chem. Eng. Sci. 56 (2001) 6667. T.B. Jensen, T.G.P. Reijns, H.A.H. Billiet, L.A.M. van der Wielen, J. Chromatogr. A 873 (2000) 149. L. Gueorguieva, L.F. Vallejo, U. Rinas, A. Seidel-Morgenstern, J. Chromatogr. A 1135 (2006) 142. K. Wekenborg, A. Susanto, S.S. Frederiksen, H. Schmidt-Traub, Chem. Ing. Tech. 76 (2004) 815. G. Strohlein, L. Aumann, M. Mazzotti, M. Morbidelli, J. Chromatogr. A 1126 (2006) 338. C.Y. Chin, N.-H.L. Wang, Sep. Purif. Rev. 33 (2004) 77.
78
L.C. Keβler et al. / J. Chromatogr. A 1176 (2007) 69–78
[12] O. Ludemann-Hombourger, R.M. Nicoud, M. Bailly, Sep. Sci. Technol. 35 (2000) 1829. [13] S. Abel, M.U. B¨abler, C. Arpagaus, M. Mazzotti, J. Stadler, J. Chromatogr. A 1043 (2004) 201. [14] G. Paredes, S. Makart, J. Stadler, M. Mazzotti, Chem. Eng. Technol. 28 (2005) 1335. [15] M. Mazzotti, G. Storti, M. Morbidelli, J. Chromatogr. A 769 (1997) 3.
[16] C. Migliorini, M. Mazzotti, M. Morbidelli, AIChE J. 45 (1999) 1411. [17] L.F. Vallejo, M. Brokelmann, S. Marten, S. Trappe, J. Cabrera-Crespo, A. Hoffmann, G. Gross, H.A. Weich, U. Rinas, J. Biotechnol. 94 (2002) 185. [18] L.F. Vallejo, U. Rinas, Biotechnol. Bioeng. 85 (2004) 601. [19] L. Gueorguieva, H. Baars-Hibbe, U. Rinas, A. Seidel-Morgenstern, J. Chromatogr., in preparation.