Ind. Eng. Chem. Res. 2006, 45, 8299-8309
8299
Combining Six-Sigma with Integrated Design and Control for Yield Enhancement in Bioprocessing Eyal Dassau, Israel Zadok, and Daniel R. Lewin* PSE Research Group, Department of Chemical Engineering, Technion IIT, Haifa, Israel
Improving the sigma level of a pharmaceutical process leads to increased overall efficiency and quality, which, in turn, reduces cycle times. These objectives can be achieved using an approach involving: (a) process analytical technologies (PAT), for data extraction; (b) Six-sigma methodology, which serves as the driving force for continuous improvement by identifying the root cause or causes of low process yield; (c) process modeling, which is based on system biology and first principles models; and (d) adVanced process control (APC) and statistical process control (SPC). This methodology is demonstrated on a simplified process for the production of penicillin, including a fermentor and the first of the downstream processing steps. We show that a combination of improved process control in the downstream processing section, as well as a modified substrate feeding profile in the fermentor, can together achieve a 40% reduction in batch time while, at the same time, significantly increasing throughput yield and decreasing impurities concentration. Evidently, this systematic approach can make a substantial impact in the pharmaceutical industry, through improved overall process yield, quality, and return on investment. 1. Introduction Active pharmaceutical ingredient (API) production can be divided into two major parts: reaction/fermentation, where the API is produced from the bio-system, and separation/purification, where product quantity and quality specifications are satisfied. Clearly, the purification process is a key step in the pharmaceutical industry, because an unpurified product cannot be marketed. In contrast, the overwhelming majority of activity published in the literature has focused on the study of reaction or fermentation, rather than focusing on the downstream processing (i.e., the product purification stages). Figure 1 demonstrates the extent to which research effort has been invested on upstream processing, as opposed to downstream processing, and the degree to which these issues have been studied in connection with the manufacture of penicillin. The emphasis and the majority of hits (papers) have been registered by work on upstream processing (by 3 orders of magnitude more than that on downstream processing). Although upstream processing (i.e., bioreactors) is clearly important, as will be shown in this study, downstream processing is, in some cases, even more important. It is reasonable to question whether the fine-tuning of the fermentation section of the bioprocess is justified, or whether more effort should be invested to improve operations in downstream processing, or better yet, to take a plantwide stance in the design and optimization of the process.1 Such a plantwide stance calls for a systematic method, such as Six-Sigma, that can serve both as a marker and as the driving force for continuous improvement by identifying the root cause or causes of low process yield, due to excessive variance in the desired specifications. This excessive variance could be the outcome of either a poorly designed process, or its control system, or a combination of the two. By improving the most significant drawbacks, one will generally improve the process controllability and resiliency leading to increased sigma levels.2 It is important * To whom all correspondence should be addressed. Tel.: 972-48292006. Fax: 972-4-8295672. E-mail:
[email protected]. URL: http://pse.technion.ac.il.
to note that understanding the process is vital in selecting appropriate output variables that can both reflect quality and can be easily controlled. For example, the penicillin extraction yield is strongly affected by the selected operating temperature and pH.3 1.1. Pharmaceutical Manufacturing. Pharmaceutical companies are highly motivated to find and introduce new therapeutic drugs to the marketplace, at the expense of vast quantities of time, effort, and capital. However, a company that succeeds in being first to market with a new drug “hits the jackpot.” An example is the case of Teva Pharmaceutical Industries, Ltd., with their drug Copaxone for multiple sclerosis, whose worldwide sales exceeded half a billion dollars in 2002.4 However, to lead, a company must be able to convert a chemical synthesis into an optimal, economic, robust, and reproducible process for the manufacture of a chemical of desired quality at the ultimate desired scale.5 Developing a new therapeutic compound involves numerous steps, starting from an idea for a new drug, which is a direct result of evaluating the needs and opportunities in a particular therapeutic area. For example, Copaxone is a multiple sclerosis drug therapy whose target site is the protein myelin that acts as protection for nerves. Although it is not exactly clear how Copaxone works, it is believed to block myelindamaging T-cells in the body by serving as a myelin decoy. The process of drug discovery can be divided into four main stages, as can be observed in Figure 2: (a) the discovery stage, which involves fundamental biological and/or medical research for a promising target site or a new therapeutic idea; (b) preclinical development, which is the chemical synthesis step; (c) clinical trials, which can take between 2 and 7 years and are divided into three phases (phase 1, in which the safety, tolerability, and dosage of the drug are tested on healthy volunteers; phase 2, where the safety, tolerability, and efficacy of the drug are tested on patients; and phase 3, in which the long-term safety, efficacy, and risk/benefits are tested on moderately and severely ill patients); and (d) the final stage, in which the company seeks final Food and Drug Administration (FDA) approval of the drug, and its subsequent market release. As can be seen from Figure 2, process developmentsthat is, the transformation of the chemical synthesis into a commercial
10.1021/ie051261q CCC: $33.50 © 2006 American Chemical Society Published on Web 05/16/2006
8300
Ind. Eng. Chem. Res., Vol. 45, No. 25, 2006
Figure 1. Research effort in bioprocessing.
Figure 2. Process development on the drug development critical path.9
processsis a key aspect in making new drugs. This is performed through the use of several stages of scale-up and process optimization. Good communication between the chemists, the development engineers, and the pilot plant engineers is crucial, so that problems during scaleup are resolved at the early development stages.5-8 A typical pharmaceutical process, illustrated in Figure 3, involves batch and semi-batch operations rather than continuous processing. Although these operation modes are inherently transient with typically nonlinear dynamics,10,11 they nonetheless enable flexible production of high-value added products in the pharmaceutical industry.12 These unit operations, or “unit procedures” as they are often referred to, are well-known in the classical chemical engineering literature. Among these operations are size reduction and classification, sterilization, mixing, filtration, evaporation and distillation, crystallization, solid-liquid extraction, drying, and bio-reactors (fermentors).13,14 The entire process of drug production is composed of batch and/or semibatch operations, which introduce problems in scheduling and linking these unit operations. 1.2. Design and Control. In the pharmaceutical industry, the term design and control is often used in reference to the development of the chemical synthesis associated with drug manufacture rather than to the overall process design and control as in the classical chemical industry (i.e., petrochemical and fine chemical production). The generic interpretation of design and control refers to the actual relationship between designing a process and the ability to control it. This interpretation can be used to improve plant yield and can be extended by
computer-aided process engineering techniques. Computer-aided process design (CAPD) and simulation have been successfully used in the classical chemical industry since the early 1960s to accelerate development and optimize the design and operation of processes and improve product yield. The batch nature of the pharmaceutical industry introduces difficulties in the development of such tools that need to consider both the dynamics and scheduling of the process. These tools can be used throughout the development stages as can be seen from Figure 2.5,15,16 In the example described here, simulations were performed using Matlab and Simulink to resolve these difficulties and to include advanced process control. 1.3. Penicillin. Penicillin was the first nonsulfa antibiotic substance, discovered (by accident) in September 1928 by Alexander Fleming. It is a metabolic byproduct of the Penicillium mold. Currently, natural penicillin is not used as an actual drug but as an intermediate in the production of two semisynthetic penicillins, which are shown in Figure 4, where the crude price of penicillin G is ∼10 U.S. dollars per kilogram. 2. Proposed Plantwide Improvement Method In response to the issues raised in the previous section, we propose a plantwide approach, combining five main components, as shown in Figure 5, to improve the overall yield in bioprocessing: process analytical technologies (PAT), Six-Sigma methodology, process modeling, advanced process control, and statistical process control. Process analytical technologies (PAT) is an interface between the process and the management system, providing for data
Ind. Eng. Chem. Res., Vol. 45, No. 25, 2006 8301
Figure 3. Typical pharmaceutical process.
Figure 4. Semi-synthetic penicillins.
Figure 5. Role of process systems engineering in the manufacture of biopharmaceuticals.
extraction, local control, and failure diagnosis, and involving instrumentation as simple as a temperature indicator to morecomplex instrumentation, such as near infrared (NIR) analysis or in-line high-performance liquid chromatography (HPLC). The ability of this interface to perform automated data acquisition and data-transfer in real time is vital to enable process improvements. Six-Sigma methodology (6σ)17 serves as the driving force for continuous improvement, and assists in identifying the root cause or causes of low process yield.
Process modeling is based on system’s biology and first principles models, which provides a basis for model-based control. AdVanced process control (APC) serves as a high-level control system tier that can coordinate the operation of cascaded, lower-level controllers, to provide optimized regulatory performance. Statistical process control (SPC) serves as a safety net for the APC, and as a preliminary stage in the 6σ methodology. The key to improvements is the process model. As indicated in Figure 5, this must be validated using plant data. Process modeling is an essential step toward process improvement; a reliable model can help in improving a process by exploring different solutions that can later be confirmed using experiments on the real plant. Thus, if the verification step implicit in our approach exposes significant disagreements between model prediction and plant response, this calls for improvements in the quality of the model. The five building blocks of the proposed approach allow the investigation of the relationships between process variables and product quality and, most importantly, allow the degrees of freedom in the process to be manipulated to achieve the desired level of quality. Figure 6 is a sketch showing the relationship between cost and compliance, with the production cost varying in a convex fashion, as a function of compliance. If the level of compliance is reduced below the “sweet spot,” the operating costs will increase,
8302
Ind. Eng. Chem. Res., Vol. 45, No. 25, 2006
Figure 7. Schematic of simplified penicillin process.
Figure 6. Cost-compliance plane.
because of the waste of raw materials and poor productivity. Conversely, increasing the compliance level above a threshold value will increase costs, because of excessive quality-control methods. Thus, increasing the sigma level allows the operating point in the cost-compliance plane to be shifted in such a way that the process compliance is ensured, while, at the same time, increasing the profit margin. This is an important outcome, because it permits continued compliant production, even when subjected to drifts. 2.1. Six Sigma. In Six-Sigma methodology, an iterative fivestep procedure is followed to progressively improve product quality. The five steps are define, measure, analyze, improVe, and control; these can be referenced by the acronym, DMAIC. Define. First, a clear statement is made defining the intended improvement. At this stage, the main focus is on customer concerns, which are used to define critical-to-quality (CTQ) and/ or critical-to-productivity (CTP) output variables. Measure: The CTQ variables are monitored to check their compliance with the upper and lower control limits (UCLs and LCLs, respectively). The data for the critical quality variables are analyzed and used to compute the number of defects per million opportunities (DPMO):
DPMO ) 106(1 -
UCL f(x) dx) ∫LCL
(1)
where f(x) is the probability of the quality being at a value of x ∈ (x,x + dx), UCL and LCL are the upper and lower control limits, respectively. As described in Rath and Strong,17 the sigma level of a process is inversely proportional to its DPMO. Analyze. When the sigma level is below its target, steps are taken to increase it, starting by defining the most significant causes for the excessive variability. This is assisted by a systematic analysis of the sequence of steps in the manufacturing process, and the interactions between them. Using this analysis, the common root cause of the variance is identified. Improve. Having identified the common root cause of variance, it is eliminated or attenuated by redesign of the manufacturing process or by employing process control. Seider et al.2 presented several examples in which process redesign can improve the controllability and resiliency of a process, and hence, reduce the variance in controlled output variables. Alternatively, feedback control can be installed, which transfers product variability to manipulated variables such as the manipulated neutralizing stream in a pH control system. Control. After implementing steps to reduce the variance in the CTQ/CTP variables, this is evaluated and maintained. Thus, the DMAIC procedure is repeated in cycles to continuously improve process quality. Note that achieving Six-Sigma performance is rarely the goal, and seldom achieved. Note that
Figure 8. Schematic of the fermentor and its control system.
the term control has a different interpretation than that used by control engineers. Initially, the DMAIC procedure is applied to define the basecase conditions. Cycles of the procedure then are implemented to improve the process iteratively. 3. Demonstrative Example The proposed methodology is demonstrated on a simplified process for the production of penicillin, considering only the fermentation and the first downstream processing step, as shown in Figure 7. 3.1. Modeling. The models of the fermentor and extractors and their control systems are reported in Appendices A, B, and C, and were implemented in Matlab and Simulink. The calibration of the models involved nonlinear regression to fit key model parameters, to minimize the error between model predictions and data reported in the literature. See Appendices A and B for more details. 3.1.1. Fermentor. The penicillin fermentation stage is simulated in the development environment based on the model described in Appendix A. This is combined with a control system, as shown in Figure 8, in which the coolant flow rate is manipulated using a proportional-integral (PI) controller to regulate the fermentor temperature, and the pH is regulated by manipulating the flow rate of acid and base to the fermentor using a PI controller with a nonlinear gain, intended to approximately “invert” the titration curve (see Appendix C). The product recipe calls for a makeup stream of substrate to be introduced when the substrate concentration reaches a threshold value. The oxygen flow rate is assumed to be constant. The simulation results, which are presented in Figure 9, are in very good agreement with the published results of Bajpai et al.18 and Birol et al.,19 where the set points were 25 °C for temperature and 5 for pH, and the substrate threshold value was 0.3 g/L. A simulated time of 422 h is needed to attain the maximum concentration level of penicillin, which was 1.5 g/L. The makeup stream of substrate is introduced after 48 h. 3.1.2. Primary Recovery System: Extractor and Reextractor. Although Reschke and Schuegerl20-22 presented a
Ind. Eng. Chem. Res., Vol. 45, No. 25, 2006 8303
Figure 9. Schematic of the fermentation trajectories (base case).
model describing the reactive extraction of penicillin, it lacks the degrees of freedom (DOFs) to enable the control of pH and flows. A formulation based on a two-film model allows these desired DOFs to be included, as described in Appendix B. Key model parameters were fitted to concentration/time measured data by Reschke and Schuegerl with good agreement, as can be seen in Figure 10. 4. Results The penicillin production process, as described previously, is implemented, and the DMAIC procedure is applied to define the base-case conditions, as summarized in Table 1. Subsequently, cycles of the procedure are implemented to improve the process iteratively, noting that improvements at each cycle are implemented in the unit exhibiting the highest DPMO value. 4.1. Cycle I. The first unit operation that is selected for improvement is the reactive extractor, which exhibits in the basecase, the highest DPMO value of 462 456. Figure 11 shows that the degree of extraction reaches 73% after 5 h and that the pH value is clearly not at its set point of 5. Moreover, the value of Cx, which represents the degradation products of penicillin, is increasing constantly during the batch. The total throughput yield (TY) of the sequence is 63% and the production time is 432 h. Evidently, the root cause of such poor performance is the absence of a pH control system in the reactive extractor, leading to an increase in the amount of impurities, which has a negative affect on the downstream units. At this point, the process is improved by installing a control system to maintain the pH at its set point of 5, which, as shown in Figure 11, not only regulates the pH as required, but also reduces the amount of impurities by 74%, increasing the TY of the unit from 73% to 89%, leading to an increase in the overall TY value from 63% to 77%. 4.2. Cycle II. Having implemented pH control to improve the extractor operation, the DMAIC procedure is repeated to further improve the process. We note that the improvement implemented in Cycle I increases the quality of the feed to the reactive re-extractor and thus reduces the DPMO for that unit from 31 264 to 13 378 before having made any additional changes. Moreover, noting that the fermentation part dominates the overall production time (see Figure 9), its reduction would provide a means to increasing the overall productivity of the process. One way of reducing the fermentation time is by changing the glucose concentration in the fermentor at which
Figure 10. Extractor calibration: (a) reactive extractor and (b) reactive re-extractor. Table 1. Summary of Control Limits, DPMO, and Throughput Yield for the Base Case LCL fermentor pH temperature production time reactive extractor TY ) 73% pH Cx () production time TY ) 86% pH Cx production time
UCL
DPMO
4.9 22 422 h
5.1 28
45 445 465
4.8 6.75 × 10-5 mol/L 5h
5.2
462 456
7 4.2 × 10-5 mol/L 5h
9
31 264
total production time total throughout yield, TY
432 h 63%
additional substrate is added (i.e., the threshold value) from 0.3 g/L to >15 g/L. Doing so reduces the fermentation time for a maximum penicillin concentration of 1.5 g/L from 422 h to 258 h, as shown in Figure 12. This reduced production time is achieved at a price of pH and temperature distributions with a higher variance than in the base case, with DPMO levels of 49 628 and 15 625, These DPMO values are higher than the
8304
Ind. Eng. Chem. Res., Vol. 45, No. 25, 2006
Figure 11. Comparison between trajectories in the extractor before and after cycle I.
feeding profiles in the fermentor, as well as improvements in the downstream processing section, can achieve a 40% reduction in batch time, accompanied by a 17% increase in throughput yield, as well in a 33% reduction of impurities, as summarized in Table 2. It has been demonstrated that these improvements result from adopting a plantwide stance in operations, because each improvement has its price tag, and budget and time constraints usually limit the total number of improvements that one can perform. Thus, rather than focus entirely on the upstream process, additional attention is needed in the downstream section(s). Evidently, this systematic approach can make a substantial impact in the pharmaceutical industry, through improved overall process yield, quality, and return on investment. Acknowledgment Figure 12. Fermentation trajectories after cycle II.
base case, but the increased variance must be weighed against the resulting reduction in batch time of 40%. The modifications reduce the total production time to only 268 h, instead of the 432 h required previously, without any decrease in the total TY value. 4.3. Cycle III. Once again, the DMAIC procedure is repeated, this time on the re-extractor unit. For the base case without pH control, the degree of extraction reaches 86%, as seen in Figure 13. Note, however, that the penicillin degradation is rather high. This situation is improved by introducing a pH controller in this unit also, with the most important outcome being a decrease of 33% in the concentration of impurities in this unit. The downside is a slight decrease in the degree of extraction, from 86% to 83%, reducing the total TY value to 74%. This improvement should be evaluated with respect to what is more critical to the management: a 33% decrease in impurity level is achieved at the cost of a 3% decrease in yield. 5. Conclusions We have shown that our approach, which involves a combination of improved process control and modified substrate
On a personal note, D.R.L. recognizes that his shift in research interests from pure control to integrated design and control emerged largely as a consequence of his collaborations over the years with Warren Seider. This began with the joint courses on nonlinear analysis that they taught in the early 1990s, culminating in their collaboration on the textbook Product and Process Design Principles,2 with Bob Seader. The work reported in this paper contains many components of that ongoing joint activity. E.D. was introduced to the work of Warren Seider on design and control in 1998 while working on developing the Multimedia Guide for the Core Curriculum23 as a system manager and programmer. This work planted the seed for his interest in this field and in improving chemical engineering education. Appendix A. Fermentor Model The following penicillin fermentation model is based on the publications of Birol et al.19 and Bajpai et al.,18 which were shown to give good agreement with the experimental results of Pirt and Righelato.24 The parameter values, which are summarized in Table 3, are the same as those in Birol, apart those marked with a single asterisk (*), which were adjusted to fit
Ind. Eng. Chem. Res., Vol. 45, No. 25, 2006 8305
Figure 13. Re-extractor trajectories before and after cycle III. Table 2. Record of Improvements Using the Proposed PSE Procedure base-case fermentor DPMO - pH DPMO - temperature reactive extractor DPMO - pH Cx reactive re-extractor DPMO - pH Cx throughput yield, TY production time
cycle 1 45 445 465
49 628 15 625
49 628 15 625
462 456 6.75 × 10-5 mol/L
