Steadily increasing demand for more efficient and more affordable biomolecule-based therapies put a significant burden on biopharma companies to reduce the cost of R&D activities associated with introduction of a new drug to the market. Reducing the time required to develop a purification process would be one option to address the high cost issue. The reduction in time can be accomplished if more efficient methods/tools are available for process development work, including high-throughput techniques. This paper addresses the transitions from traditional column-based process development to a modern high-throughput approach utilizing microtiter filter plates filled with a well-defined volume of chromatography resin. The approach is based on implementing the well-known batch uptake principle into microtiter plate geometry. Two variants of the proposed approach, allowing for either qualitative or quantitative estimation of dynamic binding capacity as a function of residence time, are described. Examples of quantitative estimation of dynamic binding capacities of human polyclonal IgG on MabSelect SuRe and of qualitative estimation of dynamic binding capacity of amyloglucosidase on a prototype of Capto DEAE weak ion exchanger are given. The proposed high-throughput method for determination of dynamic binding capacity significantly reduces time and sample consumption as compared to a traditional method utilizing packed chromatography columns without sacrificing the accuracy of data obtained.
Steadily increasing demand for more efficient and more affordable biomolecule-based therapies put a significant burden on biopharma companies to reduce a cost of R&D activities associated with introduction of a new drug to the market (1). This cost can be reduced if a larger number of drug candidates are considered in a given time. From downstream purification perspective, a reduction in time required to develop a process for production of sufficient amount of material needed for clinical studies would be a step toward cutting development costs. The reduction in time can be accomplished if more efficient methods/tools are available for process development work, including high-throughput techniques.
High-throughput methods for characterizing different separation processes have been reported (2−6). Linden (7) has used microtiter plates filled with well-defined volumes of SP Sepharose XL to simultaneously measure single- and multicomponent adsorption isotherms under various binding conditions, including various pH and ionic strength levels. In a ground-breaking work entirely focused on utilizing high-throughput techniques for development of purification processes, the process development group at Wyeth have shown how microtiter filter plates filled with chromatography resins can be used for screening of elution conditions by generating hundreds of “virtual chromatograms” in less than an hour (8). In another example, Rege et al. (6) described a high-throughput development of a purification process of α-amylase from a cell culture broth based on screening of a wide variety of chromatography media and conditions.
However, none of these methods considered transient effects occurring during protein adsorption. All of them were based on incubation of sample with a given volume of chromatography resin for a predetermined amount of time varying from several hours (7) to minutes (9). Furthermore, in several cases multiple incubations of resin and sample were performed in order to ensure saturation of resins to a desired degree. Studies of transient protein uptakes investigated using microtiter plates were also reported, yet not all details of experimental procedures were given (3). Since under properly chosen conditions a well in a microtiter plate can mimic a stirred finite bath used in batch adsorption studies, uptake data obtained using microtiter plates can be used in concert with mathematical models to predict dynamics of a chromatography column. The use of batch uptake data in describing dynamics of a packed bed column in bioseparation has been shown to be very valuable (10−13). The method can be used for studying single- and multicomponent systems (14). Knowledge of the dynamic behavior of a chromatography column allows for predicting effects of process conditions on dynamic binding capacity.
In the typical development of a purification process, the primary capture step is one of first investigations to be performed. This investigation involves determination of dynamic binding capacity (DBC) as a function of process conditions, including the effect of residence time (τ) and feed concentration; efficient wash and elution protocols; resin life time studies; etc. Knowledge of DBC as a function of process conditions is a prerequisite to develop an economically sound capture step in a downstream purification process. A relationship between DBC, feed concentration, pH and conductivity of buffer used, and flow rate through the column will allow thorough optimization of the capture step. A typical experimental procedure for determining effect of τ on DBC is to perform experiments at different flow rates, varying from a maximum flow rate, which is often linked to the mechanical properties of a given resin, to a minimum flow rate that is typically specified by practical limitations related to hardware and/or time constraints. In general, these experiments are tedious and very often require large amounts of valuable sample. Consequently, scaled-down methods that could significantly reduce time and sample requirements in process development work are sought for.
This paper describes a basic methodology for using microtiter filter plates to investigate time dependent protein uptake onto a chromatography resin and how the information obtained can be used to estimate dynamic binding capacity data as a function of an apparent residence time in a chromatography column. Practical examples and theoretical considerations are provided.
Materials and Methods
Prototypes of PreDictor plates filled either with different volumes of MabSelect SuRe or with 2 μL of prototype of Capto DEAE were prepared in our lab.
Sodium phosphate monobasic, sodium phosphate dibasic, sodium chloride, bis-Tris, and Tris were obtained from Sigma (St. Louis, MO). Model proteins polyclonal Gammanorm human immunoglobulin, hIgG, and amyloglucosidase were obtained from Octapharma (Vienna, Austria) and Fluka (Switzerland), respectively.
After removal of storage solution from the wells of the prototype PreDictor plate by vacuum filtration, the plates containing different volumes of either MabSelect SuRe or a prototype of Capto DEAE weak ion exchanger were washed three times with 200 μL of respective equilibration buffer (for buffer compositions see later in the text). Transient protein adsorption experiments were carried out according to an experimental procedure that allows generation of a whole protein uptake curve using a single microtiter filter plate. The procedure uses data generated in one well to give one data point on the protein uptake curve. In order to generate a whole uptake curve using a single plate, protein solution is added to wells in reverse order with respect to the length of incubation time for which the liquid and stationary phases are in contact with each other. Thus, the protein sample to be incubated for the longest time is added first while the sample to be incubated for the shortest time is added as the last. At the end of the shortest incubation time a whole plate is filtered to separate phases by applying an external force field. A schematic of the experimental procedure described above is shown in Figure 1.
In all experiments stationary and liquid phases were separated by vacuum filtration at −300 mbar gauge for 5 s using Multi-Well plate vacuum manifold (Pall Life Sciences, MI). All filtrate fractions except the initial buffer washes were collected in Costar UV readable microtiter plates (Corning Incorporated, NY) and absorbance was read at 280 nm using SpectraMax 384 Plus microtiter plate reader (Molecular Devices, CA). The absorbance read was used to calculate protein concentration applying a standard curve prepared using protein stock solution. Binding capacities at different incubations times, expressed as mass of protein per unit volume of sedimented resin, were calculated from a mass balance equation.
In order to verify the microtiter plate results, dynamic binding capacities of hIgG on MabSelect SuRe and of amyloglucosidase on the prototype of Capto DEAE were obtained following a standard frontal analysis procedure (15). For these tests MabSelect SuRe was packed into HiTrap 1 mL columns and Capto DEAE was packed into Tricorn 5/100 columns. The columns were equilibrated with 5 CV of the equilibration buffer, loaded with a feed solution made of the protein dissolved in the equilibration buffer, washed with 5 CV volumes of the equilibration buffer, eluted with 3 CV of elution buffer, and regenerated with 5 CV of regeneration buffer. Between tests, the columns were stored in 20% EtOH. All loading steps were terminated when absorbance of eluent stream from the column was equal to 10% of absorbance of the feed solution. In the experiments with MabSelect SuRe, equilibration/wash and elution steps were performed using 20 mM phosphate buffer pH 7.4 with 150 mM NaCl and citrate buffer pH 3.0, respectively. In experiments with the prototype of Capto DEAE, equilibration and loading steps were done using either 30 mM Tris buffer (pH 8 and 9) or 30 mM bis-Tris buffer (pH 6 and 7) with different levels of NaCl, while the elution step was performed with 1.0 M NaCl in a loading buffer.
All steps but the loading steps were performed at flow rate of 2 mL/min. The loading steps were performed at different flow rates that were calculated from apparent residence times chosen and column volumes (the apparent residence time is defined as column volume divided by flow rate).
All experimental conditions tested were carried out in triplicates unless otherwise specified.
Results and Discussions
Theoretical Consideration. The method for determining of DBC described here is based on a well-known concept of batch uptake. The batch uptake method is based on contacting defined volumes of an adsorbent and buffer solution containing adsorbate in a closed system. The mass balance (eq 1) can be used to calculate adsorption capacities from changes in adsorbate concentration in the liquid phase:
where qi is the concentration of adsorbate i in the solid phase, Ci is the concentration of adsorbate i in the liquid phase, β is the phase ratio defined as the ratio of the volumes of liquid phase, Vliq, to solid phase, Vsolid, and t is the contact time. Subscript 0 indicates initial concentrations of adsorbate i in the system.
Following changes in adsorbate concentration in the liquid with time allows for characterizing kinetic and thermodynamic effects governing the adsorption process. Since under most practical conditions the same effects are responsible for dynamic behavior of the chromatography column, information obtained in the batch system can be used to predict what would happen in a column.
Batch uptake experiments are typically performed in agitated vessels/contactors, where under sufficient mixing conditions the whole stationary phase is in contact with the bulk liquid of the same concentration. Close to equilibrium, when changes in adsorbate concentration in the two phases are infinitesimally small, such a single contactor could, to a first approximation or loosely speaking, represent a single theoretical plate in a chromatography column. The theoretical/hypothetical plate is defined as a place where a complete equilibrium between stationary and mobile phases takes place and, furthermore, a continuous exchange of mobile phase occurs, i.e., the mobile phase leaving the plate is in equilibrium with the stationary phase present in the plate (16). Yet a certain characteristic time is necessary to allow the system to reach equilibrium. The magnitude of this characteristic time is a result of external and internal mass transfer steps and ligand−ligate interactions, before the phases would be in equilibrium. A well mixed vessel would provide an experimental system to investigate these characteristic steps and to determine when the system has reached a state close to equilibrium. From this perspective, a single well in a microtiter plate could represent a single theoretical plate, but only if conditions are chosen such that the system is allowed to reach equilibrium. These conditions will depend on many factors such as protein amount, available surface for adsorption, and its accessibility and contact time. The first two factors are directly related to the volumes of stationary and mobile phases in a single well.
From thermodynamic and kinetic perspectives, most protein adsorption processes can be characterized using batch uptake principle. The rate of change in protein concentration provides kinetic data, while the concentrations after long incubations times will provide thermodynamic data that can be described by an adsorption isotherm.
Because both the rates and the equilibrium concentrations will affect dynamic binding capacity, knowing how they are affected by different operating conditions, such as pH, buffer types, additives, etc., would allow estimation of DBC at various residence times at different process conditions. Depending on the type of data obtained from batch uptake experiments, this estimation can be either qualitative or quantitative in nature. The qualitative method focuses only on capturing trends in effects of different process conditions on the dynamic binding capacities at short and/or long residence times. This method relies on a very limited data set and, in principle, requires experimental data collected at only two contact times, one relatively short and one relatively long. In contrast, the quantitative estimation requires more advanced data sets but will provide explicit and detailed relation between the DBC and the residence time in a column. In the center of this method is a mathematical model that captures kinetic and thermodynamic effects measured in a batch system (10, 11) and is subsequently used to describe the same effects in column geometry. In principle, the model need not describe mass transfer mechanisms responsible for protein adsorption in detail, but it must capture effects of process variables on a characteristic time constant for the adsorption process studied. The extensive data set should contain data of such quality that a precise estimation of all parameters in the model chosen could be accomplished. However, care must be taken not to blindly extrapolate validity of the model by checking if all model assumptions are fulfilled. If necessary, the model can be supplemented with data based on engineering correlations developed for batch systems and for packed columns, such as correlations for film mass transfer and for axial dispersion coefficients. A brief summary of the qualitative and quantitative methods is shown in Figure 2.
Obviously, the qualitative method relies on an assumption that the adsorption process studied is not convection-dependent. In case of protein adsorption on a typical porous chromatography resin this is a true assumption because the rate-limiting step is typically related to intraparticle mass transfer (e.g., pore diffusion) which is practically flow-independent. However, in case of large macromolecules such as DNA and viruses the film mass transfer is the rate-limiting step and the characteristic time constant for this step depends on hydrodynamic conditions occurring in a given system. Consequently, when applying the qualitative method for estimating DBC as a function of process conditions and apparent residence time in a column, care must be taken to ascertain, if possible, that convection dependent mass transfer steps are negligible, or at least quantifiable. The latter applies also to the case when the quantitative method is used.
Quantitative Analysis. Microtiter filter plates prefilled with MabSelect SuRe were used to study adsorption of hIgG in a batch system. Effects of mixing intensity, phase ratio, and initial protein concentrations on the rates of adsorption were considered. The effect of mixing intensity is shown in Figure 3. Agitating PreDictor plates at an orbital speed of 500 rpm has not improved mass uptake rates as compared to the case of an agitating speed of 150 rpm (data not shown). In both cases uptake curves were fairly shallow, indicating that the rate-limiting step was the diffusion of hIgG from bulk liquid to the surface of chromatographic particles. In contrast to these results an agitation speed of 1100 rpm resulted in a significantly different uptake curve (Figure 3). The same results were obtained irrespective of the initial protein concentrations. Further increase in agitation intensity did not affect uptake rates. Similar results were reported by Bensch et al. (3) and Kramarczyk (9), who showed that minimum agitation intensity should be above 1300 rpm. However, one should bear in mind that the minimum agitation intensity will depend on the system studied including well diameter, liquid volume, liquid viscosity and density, particle density, and orbital amplitude of the shaking table used (data not shown).
The effect of phase ratio and initial protein concentration on the rate of hIgG uptake onto MabSelect SuRe is shown in Figure 4. The results obtained clearly show expected trends, i.e., the smaller the phase ratio the lower is the concentration at the end of batch uptake experiments and the higher are the initial uptake rates, dC/dt.The data also show how these rates change with initial protein concentration. Because from the quantitative method perspective it is of paramount importance to work with data that allow simultaneous quantitative estimation of both kinetic and thermodynamic effects responsible for the adsorption process, only data obtained at phase ratios of 40 and 20 were deemed to be of sufficient quality. These data series allowed meaningful estimation of initial uptake rates with the sampling times chosen in this work. From experience, choosing a phase ratio and initial protein concentration that would result in 80% decrease in bulk protein concentration will provide data of sufficient quality to allow simple models derived for highly favorable isotherms to be fitted to the data.
In this work a model describing the so-called shrinking core mass transfer mechanism was used. The model was fitted to the experimental data at the two lowest phase ratios. The least-squares estimates of the model parameters are given in Table 1.
Table Table 1.. Least-Square Estimates of Parameters in Shrinking Core Model Used To Describe Batch Uptake Data Shown in Figure 4
a Q is the equilibrium capacity at the initial feed concentration Cini.
51.43 ± 0.06
2.75 ± 0.08
48.84 ± 0.46
2.63 ± 0.08
19.30 ± 0.80
14.92 ± 3.05
9.48 ± 0.03
23.47 ± 4.60
As shown in Figure 4 the agreement between model predictions and experimental data is very good. Data series obtained with the highest phase ratio studied was not considered for parameter estimations because validity of the model is not warranted under these conditions. At the highest phase ratio, an almost instantaneous adsorption of all the protein present in solution occurs on the surface of chromatographic particles. This in turn leads to a decrease in bulk protein concentration to a very low level, the level at which binding capacity will be much lower than a capacity obtained at the excess of bulk protein when all binding sites are saturated.
It should be emphasized at this point that while the intraparticle concentration profiles calculated using the shrinking core model do resemble concentration profiles characteristic for a mass transfer mechanism based on pore diffusion and a favorable adsorption isotherm, the shrinking core model does not describe any realistic mass transfer mechanism (17) and has been used in this work because of a compromise between model simplicity, number of data points obtained, and screening character of the experiments.
Model parameters estimated from batch uptake data (Table 1) were used in concert with eq 2 to predict the dynamic binding capacities as a function of residence time in a chromatographic column. This equation was derived on the basis of a model describing a case when the external mass transfer was negligible and the intraparticle mass transfer was approximated by the shrinking core behavior (18):
where DBC10% is the dynamic binding capacity at 10% breakthrough, c0 is feed concentration, Qmax and K are the maximum adsorption capacity and dissociation constant in Langmuir isotherm, respectively, and N is the number of transfer units defined by eq 3:
where De is the effective pore diffusivity, Rp is the particle radius, ϵ is the bed porosity, and τapp is the apparent residence time in a column defined as ratio between the column length and superficial velocity of a mobile phase.
Values of parameters in the Langmuir adsorption isotherm used in eq 2 were estimated from the batch uptake data (Figure 4) obtained after 60 min of incubation in all batch uptake experiments performed. The least-square estimates found for Qmax and K were 51.43 ± 3.58 g/Lbed and 0.072 ± 0.02 g/L, respectively. While these data could not be considered as a true equilibrium isotherm, they provided good approximation of the equilibrium state because the rates of uptake after 60 min of incubation were fairly low, indicating that the system was approaching equilibrium.
Dynamic binding capacities calculated using eq 2 were adjusted by a factor accounting for resin compression when the resin is packed into a column. In the microtiter plate method, volume of the resin is based on a sedimented volume, whereas resin volume in the column is defined by column dimension and packing procedure. For noncompressible resins the difference between these volumes should be relatively low, whereas for chromatography resins made of a compressible backbone, such as agarose, the difference can be on the order of up to 20% (19). The correction factor of 1.22 used in this work was determined through dry weight measurements of MabSelect SuRe found in wells of microtiter plates and in HiTrap columns. One should bear in mind that the correction factor accounting for the difference between resin volumes is not constant and depends on resin type and the method of preparation of microtiter plates filled with the resin (data not shown).
Comparison of the dynamic binding capacities predicted on the basis of the batch uptake data and respective DBCs measured in packed columns for different feed concentrations and residence times is shown in form of a parity plot in Figure 5. A very good agreement between the two ways of obtaining DBC proves that the proposed quantitative method is a very attractive alternative for process development work, especially considering that savings in time and amount of sample required were in order of 10 and 50 times, respectively.
Qualitative Analysis. The other example of using prototype of PreDictor plates for estimating dynamic binding capacity is focused on the qualitative approach. The example is based on screening the effect of pH and ionic strength on binding capacity of amyloglucosidase onto a prototype of Capto DEAE ion-exchange resin. The results obtained using prototypes of PreDictor plates and using columns are compared in Figure 6. In general, the effect of pH (Figure 6A) and of ionic strength (Figure 6B), showed exactly the same trends for both formats used in the screening experiment. Furthermore, not only the trends but also capacity levels were in remarkable agreement between the two formats used. Obviously, the reason for a larger number of data points obtained with plates was related to the plate format itself, which allows parallel screening of either a wider experimental space or a more detailed investigation of a narrower space. One of the reasons for the good agreement observed, besides a correctly chosen resin volume, was a choice of contact times used in the microtiter plate experiments. The contact times chosen were based on the concept of similarity between the contact time and the column loading time, despite the fact that the apparent residence time can be defined in the same way for the column and the plate geometries as a time an inert solute would spend in an empty column/well. The column loading time, i.e., the time in which the chromatography column is exposed to feed solution, depends on operating conditions, such as flow rate, pH and ionic strength, and feed concentration, and thus on dynamic binding capacity. For instance, for a case when feed concentration is either 1 or 4 g/L and resin dynamic binding capacity is 100 g/L, the length of the loading step expressed in terms of number of column volumes loaded onto the column will be either 100 or 25, respectively. Now, if the apparent residence time is, for instance, 2 min, then the time at which resin in the beginning of the column will be in contact with the feed containing adsorbate will be either 200 or 50 min, respectively. If the residence will be 0.5 min then the contact time will be either 50 or 12.5 min. Consequently, in order to estimate dynamic binding capacities using microtiter plates, incubations times similar to the contact times anticipated when operating a chromatography column will be needed in order to saturate the resin used to a level, which will be similar to the one observed in greater parts of a packed column. However, it should be noted that the experimental approach based on the similar contact times may not work if the mass transfer mechanism behind protein adsorption is dominated by convection, such as in cases when the external film mass transfer or an intraparticle convection were the rate-limiting steps. In such cases, because of different hydrodynamic conditions occurring in a single well and in a column under a similar contact times, the dynamic binding capacities estimated using the qualitative version of the HTS method and the capacities measured in a chromatography column could be different. Furthermore, a magnitude of this difference could depend on column bed height, because for the same residence times in columns of different bed heights, the dynamic binding capacities and therefore loading times will not be the same. In all other cases, when the dynamic binding capacities for a given residence time were the same regardless of columns bed heights, the screening approach using microtiter plates based on similarity of the contact time in batch format and sample loading time in the column would be valid as the contact time does not depend on the column size but on dynamic binding capacity.
An additional aspect of the high-throughput capabilities of using the PreDictor plates, emphasizing the power of this type of screening, is shown in Figure 7. The figure represents capacity maps obtained at different pH and ionic strength at three incubation times, 2, 60, and 1200 min. The maps were generated using a plotting software (SigmaPlot v. 10, Systat Software, Inc., NY, US) directly from raw data, i.e., no modeling or fitting was involved. This way of data representation provides a quick method for analyzing large and sometimes complex data sets. In this particular case, the capacity maps show how the location of optimum conditions for maximum binding capacity varies with the contact time. While the pH optimum was the same for all incubation times at a pH of 7.0, the optimum salt concentration varied from 70 mM at 2 min (Figure 7A) through 40 mM at 60 min (Figure 7B) to 20 mM at 20 h (Figure 7C). This is an important result from a process development perspective because it will provide guidance for planning of optimization experiments that need to be performed using columns of a larger size. The shape of capacity maps shown in Figure 7 is characteristic for ion exchange resins that are characterized by opposite effect of changes in ionic strength and pH on intraparticle mass transfer rates and on equilibrium capacities. For these resins, optimum pH and ionic strength will depend on residence time. For resins where the effect of changes in ionic strength and pH on intraparticle mass transfer rates is small, the shape of the contours in the capacity maps would show a monotonic change in capacity with ionic strength and pH and conditions for obtaining maximum binding capacity would not depend on residence time.
It should be emphasized here that all experiments performed with the prototypes of PreDictor plates filled with the prototype Capto DEAE resin were done with 2 μL of the resin per well. Use of a larger volume of the resin would require much a larger volume of sample or higher sample concentration. The former would cause difficulties in sustaining sufficient mixing in a single well of the microtiter plate, invalidating the requirement for the same mass transfer regime to occur in a single well of the microtiter plate and in a column during adsorption process, whereas the latter would significantly reduce the lower sample requirement advantage associated with use of plates over columns.
Determination of dynamic binding capacity as a function of residence time requires tedious and expensive experiments using packed columns. A high-throughput method for screening a vast experimental space using microtiter filter plates filled with a well defined and carefully chosen volume of chromatography resin allows for estimating the effect of residence time on DBC. The estimation can be performed on either a qualitative or quantitative basis. In either case, the correct choice of experimental conditions, especially the incubation time and phase ratio, is a prerequisite to generate data of sufficient quality for dynamic binding capacity estimation purposes. The incubation/contact time should be on the same order of magnitude as the loading time in a chromatographic step. The incubation time and the apparent residence time in a column are two completely different concepts and care should be taken to avoid their misuse. At the same time, if the phase ratio in the batch system is not correctly chosen, the resulting semi-equilibrium saturating capacity will represent an artificial situation that only happens in a column in front of the mass transfer zone. This is especially important if the qualitative screening method is considered, because these capacities are completely non-representative of the average situation occurring in chromatography columns during the loading step.
The high-throughput method described here reduces column experiments to a necessary minimum. These experiments need only be performed at the most promising conditions, thus significantly reducing time and sample necessary for optimization purposes. In the two examples discussed above the reduction in sample size and experimental time were significant. For the qualitative studies, all data were generated 10 times faster and the amount of sample was reduced 80 times as compared to column experiments. In the case of quantitative analysis, savings in time and sample consumptions were in order of 10 and 50 times, respectively.
While fairly general, the method, especially in its qualitative form, may not work for all separation systems. In the case of very low protein concentrations and/or large biomolecules adsorbing on typical chromatography resins, the effect of external film mass transfer resistance on the rate of protein uptake in a filter microtiter plate cannot be neglected. In such cases, one may not be able to ascertain sufficient mixing within a single well that could mimic hydrodynamic conditions occurring in a chromatography column at short residence times. However, the quantitative analysis still can be performed if proper correlations for estimating external film mass transfer coefficients in agitated vessels and in chromatography columns are used in concert with the thermodynamic data generated using the plate method.
The high-throughput method should be used in day to day process development when screening a large experimental space of process variables to identify their effects on dynamic binding capacity. The best conditions identified should then dominate the subsequent studies/experiments, which need not be carried out using microtiter plates, but in fact they should be performed using packed columns, preferably in an automated chromatography system.
Capto, HiTrap, MabSelect SuRe, PreDictor, Sepharose, Tricorn, ÄKTAexplorer are trademarks of GE Healthcare companies. Parts of the material included in the paper were previously presented as a poster at SBE's Conference on Accelerating Biopharmaceutical Development, San Diego, CA, 2007 (20).