OR WAIT null SECS
Case studies were run to test Process Analytical Technology applications for protein refolding, diafiltration, and cation exchange chromatography. It is shown that it is feasible to design control schemes that rely on measurement of product quality attributes and thereby enable real-time decisions.
Case studies were run to test Process Analytical Technology applications for protein refolding, diafiltration, and cation exchange chromatography. It is shown that it is feasible to design control schemes that rely on measurement of product quality attributes and thereby enable real-time decisions. Implementation of these schemes should result in consistent product quality and high operational efficiency. However, these advantages are balanced by requirements of a higher level of process understanding for designing these schemes, and increased operational complexity during implementation. Increased consistency in product quality is likely to increase variability in step recovery, a parameter that is commonly considered an indicator of process consistency.
The biopharmaceutical community is interested in using Process Analytical Technology (PAT) for continuous real-time quality assurance. PAT has the potential to improve operational control and compliance. The operational definition of PAT is as "a system for designing, analyzing, and controlling manufacturing through timely measurements (i.e., during processing) of critical quality and performance attributes of raw and in-process materials and processes, with the goal of ensuring final product quality."1
Several case studies demonstrating the utility of this approach in formulation unit operations such as granulation and compression have been recently published.2 Fewer studies have addressed bioprocessing unit operations mainly because they are difficult to control, and subject to the increased variability and relatively limited understanding of raw and feed materials.3,4
A desired goal of the PAT framework is to design and develop well-understood processes that will consistently ensure a predefined quality at the end of the manufacturing process.1 A process is generally considered well understood when (1) all critical sources of variability are identified and explained; (2) variability is managed by the process; and, (3) product quality attributes can be accurately and reliably predicted over the ranges of acceptance criteria established for materials used, process parameters, manufacturing, environmental, and other conditions.1
This article focuses on the feasibility of creating and implementing control schemes that enable real-time and product quality decisions for commonly used biotech unit operations, which include protein refolding, process chromatography, and ultrafiltration–diafiltration. This is the sixth article in the "Elements of Biopharmaceutical Production" series.5 We present below three case histories that use the PAT framework to improve process quality.
Protein refolding typically runs with a time-based recipe. The time is set based on the rate of refolding (as determined by process development studies) and operational constraints (manufacturing shifts). There are two consequences of performing refolds by fixed time. The refold step produces varying product quality as measured by percent purity or percent impurities (e.g., reduced and oxidized forms of product, misfolds, etc). The refold rate will vary from lot to lot due to variability in feed materials, and so refold time is often set conservatively. This makes production operating costs higher than optimum.
We ran a study to evaluate PAT methods for controlling a protein refolding step. In our test, frozen cell paste was resuspended using a Silverson L4R high-shear mixer (Silverson Machine, Chesham, Bucks, UK). A Model 100Y Microfluidizer (Microfluidics, Newton, MA) was used to perform cell lysis. The lysate and the subsequent wash suspension pellets were spun down using a Beckman J6-B bucket centrifuge with a JS 4.2 rotor (both from Global Medical Instrumentation, Ramsey, MN). The refold was performed in a Biostat MD 12 fermentor (Sartorius BBI Systems, Bethlehem, PA). Post-refold filtration was carried out using a CUNO 60 SP depth filter (CUNO, Meriden, CT).
Figure 1 illustrates the refold process. As time progresses the unfolded protein (peak B) is oxidized to produce the product (peak A). The peaks designated by numbers are other impurities that are also generated during the refold process.
Figure 1. Purity profile during refolding. Chromatograms show the progressive oxidation of B to A.
We wanted to design a PAT-based control strategy using on-line monitoring that would allow refolding operations to end at a time determined by product quality parameters (percent purity of the product or percent impurity). To generate data for such a strategy, samples were withdrawn at fixed time intervals, quenched, and transferred to the auto sampler for analysis using an Agilent 1100 Series high performance liquid chromatography (HPLC) system with a binary pump, micro vacuum degasser, thermostatted autosampler, thermostatted column compartment, and a multi-wavelength detector (Agilent Technologies, Palo Alto, CA).
Figure 2 presents profiles of the product (percent purity), reduced form of product (percent Impurity 2), and other impurities (percent Impurities 1, 3, and 4) over the refold time. The data suggest that it is feasible to implement a PAT-based control scheme that allows for ending refold based on product quality data. This would ensure consistency in product quality of the refold end pool, and improve operational efficiency by keeping refold time to completion (10 h may suffice versus 16 h presently used).
Figure 2. Purity and impurity profiles as generated by on-line HPLC
This PAT-based control scheme does pose several problems. First, operations are simpler with time-based recipes. In the proposed scheme, the refold time will vary from run to run so it can generate the process stream of consistent product quality. The likely result is a need for more dynamic planning and scheduling of manufacturing steps. Second, this scheme would require that the manufacturing operators be trained to interpret and act upon the data from the HPLC. Third, since the decision to end the refold is now based on HPLC data, it is critical that the analytical methods have a high degree of robustness. It may be necessary to have redundancy (e.g., duplicate analysis) built into the control scheme to ensure accuracy of the data.
Implementation of a PAT-based control scheme for a protein-refolding unit operation is feasible and will offer several benefits. This will be particularly attractive for cases when one or more impurities are created in the refold process and excess refold time is likely to result in higher levels of one of them.
The diafiltration step of the UF–DF process is often specified for a fixed number of diavolumes. The number is based on process development studies and, as for refold time, the number is often set conservatively, resulting in excess buffer use and process time. In this study, we wanted to design a PAT-based control strategy which would allow us to end the UF–DF step when the diafiltration process is complete, as signaled by a product quality criterion. This would result in consistent product quality at the end of the step and would also minimize the buffer usage and operation time.
UF–DF was performed with Millipore 0.1m2 Pellicon 2, 5kDa regenerated cellulose membrane (Millipore, Billerica, MA) operated by the Millipore Proflux M12 system. Samples were withdrawn at various times during the diafiltration step.
Figure 3. Concentration profiles of various excipients during diafiltration. The pH level is measured as a surrogate for the concentrations of Specie 1. All other species are not as limiting.
Figure 3 presents concentration profiles of the various species through the diafiltration process. Transport across the UF–DF membrane is governed by the following expression:
where Yp is the concentration of a species in the permeate, Yo is the concentration in the feed solution, S is the sieving coefficient, and DV is the number of diavolumes. The data presented in Figure 3 were analyzed using Equation (1) to calculate the sieving coefficient for the various species and the results are listed in Table 1.
Table 1. Determination of the sieving coefficient by fitting Equation (1) to the data presented in Figure 3.
The data suggest that Specie 1 has the smallest sieving coefficient, i.e., it is the slowest specie to transfer across the membrane and hence, it was chosen as an indicator of the completion of the diafiltration process. Another interesting observation that can be made from Figure 3 is that the pH profile shows a good correlation with the Specie 1 concentration profile. The pH increases as Specie 1 concentration drops with both stabilizing at approximately five diavolumes.
This relationship suggests the possibility of using pH, which is easy to measure on-line, as an indirect indicator for completion of the diafiltration process. This possibility was further examined by performing experiments at different Specie 1 concentrations. The data are presented in Figure 4. The data support the above-mentioned correlation with both the pH and Specie 1 profiles reaching the plateau earlier when the Specie 1 concentration is low, and later when the Specie 1 concentration is high. Thus, for this application, pH could be used for a PAT-based control study to trigger the end of diafiltration. We can stop diafiltration when pH reaches 8 at 5 diavolumes instead of at 8 diavolumes as presently specified. In Figure 5 we show a normalized fit for diavolumes as a function of concentration of Specie 1 in the feedstock. This is a version of Equation 1, and the fit is excellent (R2 = 0.98). If the feed concentration of Specie 1 is known, an operator can use this graph to estimate the number of diavolumes required to complete the diafiltration process.
Figure 4. Relationship between pH and Specie 1 concentration during diafiltration. The two centerpoint runs are for replication purposes and start at a concentration of 50 mg/mL. The 2x concentration starts at 85 mg/mL and the 0.5x starts at 20 mg/mL. The initial pH is consistent at 6.5.
Similar to the case study for protein refolding, this PAT-based control scheme also has several problems with respect to implementation in a manufacturing environment. While the pH-based control scheme is simple to implement, the empirical model presented in Figure 5 would require well-trained manufacturing operators. Also, pH probes can be quite unreliable. Therefore, if the process control strategy is pH-based, then it may be necessary to have redundancy (e.g, duplicate probes) built into the control scheme to ensure accuracy of the data.
Figure 5. Relationship between number of diavolumes required and Specie 1 concentration
Implementation of a PAT-based control scheme for a DF step is feasible and will offer several benefits. This is particularly attractive when an expensive diafiltration buffer is used, or when the duration of diafiltration needs to be minimized due to product stability concerns.
Chromatography steps in biotech processes often use pooling criteria that are based on UV absorbance (e.g., absorbance at 280 nm). The key advantage is the simplicity of implementing UV-based pooling criteria in a manufacturing environment. For bind and elute applications, when the whole peak is to be collected from baseline to baseline, absorbance-based pooling criteria offer a simple solution that ensures that all the product is collected. This works because, in most applications, the protein concentration is linearly related to the absorbance signal.
Figure 6. JMP analysis of data from chromatography experiments. Term refers to the variables that were reviewed in the analysis, with Intercept as the mean value. Scaled estimate is the value for the effect of each parameter. Error is the absolute error for each parameter. The t-ratio is the ratio of the parameter estimated to its standard error. A t-ratio greater than 2 in absolute value is a common rule of thumb for judging significance. Prob. > |t| is the probability of getting, by chance alone, a t-ratio greater (in absolute value) than the computed value. Given a true hypothesis, a value below 0.05 is often interpreted as evidence that the parameter is significantly different from zero.
While simpler to implement operationally, pooling by absorbance has several limitations in applications where a high-resolution separation is being performed, and part of a peak is being collected to pool the product and pool out the impurities. Absorbance methods are not able to differentiate between product and other proteins or other species that have similar absorbance profile. As a result, since impurity levels vary from lot to lot due to variations in feed purity and column operating conditions, pool purity also varies from lot to lot. Further, pooling criteria are often set conservatively in order to deliver a pool of sufficient product quality, resulting in the loss of acceptable product for some lots.
In this case study, we wanted to design a PAT-based control strategy that would ensure that the column pools had the desired purity. Cation-exchange chromatography was carried out in an XK 16 column packed with Sepharose Fast Flow resin (GE Healthcare, Piscataway, NJ). Chromatog-raphy runs were carried out on an AKTA Explorer 100 (GE Healthcare). Samples from the column were analyzed by reversed phase HPLC (RP-HPLC) for product purity and percent impurity. A design of experiments study consisting of 17 experiments was conducted with purity of the load material, start point of pooling (start collect), and end point of pooling (stop collect) as the variables. Pool purity and step yield were monitored for each experiment.
Figure 7. Testing the JMP model for cation exchange chromatography against the actual data. The central point is at 96.4%.
The results were analyzed using JMP software (SAS Institute, Cary, NC) and are presented in Figure 6. It is seen that load purity and stop collect have significant impact on pool purity, while start collect does not. We tested the JMP model by calculating the pool purity for all experiments. Figure 7 is a plot of measured purity vs. calculated purity. The correlation is good (R2 = 0.87), supporting feasibility of this scheme.
Table 2. Estimations of start collect and stop collect from the JMP model for a given load purity and to get a targeted pool purity (mAU = milli absorbance unit).
In production, this JMP model could be used to estimate start collect and stop collect for a known load purity to yield a targeted pool purity. Four cases of this calculation are presented in Table 2. It is seen that for feed material with different purities, targeted pool purity can be achieved by changing the stop collect.
The proposed scheme requires that the manufacturing operators be trained to interpret and act upon the data from the HPLC. Further, since the pooling decision to end the refold is based on HPLC data, it is critical that the analytical methods have the required robustness. It may be necessary to have redundancy (e.g., duplicate analysis) built into the control scheme to ensure accuracy of the data. Since the aim is to achieve consistent pool purity by shifting stop collect criteria, recovery across the chromatographic step is expected to vary from lot to lot. It is seen in Table 3 that when load purity drops from 86% to 78% and the target for pool purity is 94%, column recovery decreases from 99% to 95%.
Table 3. Estimations of recovery from the JMP model for a given load purity and to get a targeted pool purity of 94% (mAU = milli absorbance unit).
Implementation of a PAT-based control scheme for a process chromatography step is feasible and will offer several benefits. This would be particularly attractive when a high-performance separation is being accomplished. Then a purity-based pooling criterion will allow for a dynamic control scheme that can yield consistent pool purity.
For the three unit operations that were examined in this article, we have shown that it is feasible to design control schemes that rely on measuring product quality attributes. We project gains in consistency and operational efficiencies. However, these schemes require highly trained operators to deal with complexity. There is likely to be increased variability in step recovery, a parameter that is commonly considered as an indicator of process consistency. It is evident that adoption of control schemes such as these presented here will require changes in our approaches toward process and analytical development, manufacturing, quality assurance, and regulatory filings.
The authors would like to thank Duane Bonham and Duncan Low of Amgen for useful discussions.
Anurag S. Rathore, PhD, is associate director of process development at Amgen Inc., 30W-2-A; One Amgen Center Drive, Thousand Oaks, CA 91320, 805.447.4491, fax 805.499.5008, firstname.lastname@example.org
Ashutosh Sharma and David Chilin are process development engineers at Amgen Inc.
Anurag Rathore is a member of the Editorial Advisory Board.
1. US Department of Health and Human Services, Food and Drug Administration (FDA), Center for Drug Evaluation and Research (CDER), Center for Veterinary Medicine (CVM), Office of Regulatory Affairs (ORA). PAT guidance for industry—A framework for innovative pharmaceutical development, manufacturing and quality assurance. 2004 September; p. 7.
2. Scott B, and Wilcock A. Process analytical technology in the pharmaceutical industry: A toolkit for continuous improvement. PDA Journal of Pharmaceutical Science and Technology 2006; 60(1): 17–53.
3. Larson TM, and Lam H. Process analytical technology in biopharmaceutical production: Past successes and future challenges. The Journal of Process Analytical Technology 2004; 1(1):20–22.
4. Larson TM, Davis J, Lam H, Cacia J. Use of process data to assess chromatographic performance in production-scale protein purification columns, Biotechnol. Prog. 2003; (19):485–492.
5. Past articles in the "Elements of Biopharmaceutical Production" series includes:
5A. Rathore AS, Nofer JF, Arling ER, Sofer G, Watler P, and O'Leary R. Process validation: How much and when. BioPharm 2002 October; 15(10):18–28.
5B. Rathore AS, Levine H, Latham P, Curling J, and Kaltenbrunner O. Costing issues in production of biopharmaceuticals. BioPharm 2004 February; 17(2):46–55.
5C. Rathore AS, Wang A, Menon M, Riske F, Campbell J, Goodrich E, and Martin J. Optimization, scale-up and validation Issues in filtration of biopharmaceuticals – Part I, BioPharm 2004 August; 17(8):50–58. Part II, BioPharm 2004 September; 17(9):42–50.
5D. Rathore AS, Krishnan R, Tozer S, Rausch S, and Seely J. Optimization, guidelines and examples for scale-down of biopharmaceutical unit operations, – Part I. BioPharm 2005 March; 18(3):60–68. Part II. BioPharm 2005 April; 18(4):58-64.
5E. Moscariello J, Lightfoot E, and Rathore AS. Efficiency measurements for chromatography columns. BioPharm 2005 August; 19(8):58–64.