Designing in Quality: Approaches to Defining the Design Space for a Monoclonal Antibody Process

May 1, 2010

BioPharm International

Volume 23, Issue 5

How to use risk assessment strategies to integrate operations.

ABSTRACT

A systematic Quality by Design (QbD) strategy was used to develop and characterize a monoclonal antibody production process. A risk assessment approach incorporating design of experiments (DOE), especially multivariate analyses, was used to define and prioritize laboratory-scale experiments and made it possible to focus on high-risk process parameters and study the interactions of those parameters to define the design space. Scale-up/scale-down strategies, such as the effective characterization of a scale-down bioreactor model, ensured the applicability of small-scale studies, and secondary risk assessment approaches were used to ensure that various unit operations were properly integrated in the development of the design space.

Quality by Design (QbD) is a scientific, systematic, risk-based approach applied throughout a product's life cycle to ensure safe, effective products.1,2,3 In applying QbD approaches to biopharmaceutical processes that produce complex biomolecules, a systematic approach to process understanding is essential. A key goal of process understanding studies is to establish the functional relationship between the process parameters and quality attributes, including parameter interactions; therefore, process understanding invariably uses multivariate experimental strategies.

(PFIZER INC.)

The design space, which is an output of the process understanding studies, provides a definition of process input variables and their ranges to ensure consistent quality for large-scale commercial manufacture. The continuous development of the knowledge space that ultimately makes it possible to determine the design space begins at product conceptualization, evolves in pace with product commercialization, and is ongoing throughout the product lifecycle.

Pfizer has developed a systematic approach to implementing QbD principles for process design for small molecules that encompasses process understanding, process control, and continuous improvement.4 This article explores the strategies for and challenges involved in developing a thorough process understanding and defining the design space for a monoclonal antibody manufacturing process. We briefly summarize Pfizer's strategies for systematic risk assessment to define and prioritize laboratory-scale experiments, scale-up/scale-down strategies to ensure the applicability of the small-scale studies, and secondary risk assessment approaches that integrate various unit operations.

PFIZER'S APPROACH TO QBD FOR PROCESS DESIGN

The QbD process design is achieved through Pfizer's Right First Time (RFT) approach (Figure 1), which consists of achieving process understanding, process control, and continuous improvement using a life-cycle approach based on the International Conference on Harmonization (ICH) Q8, Q9, and Q10 guidelines.2,5,6 This article focuses on the first part, the development of process understanding that is used to develop the design space for a process. Commercial manufacturing processes are operated in a "control space," which is an area within the design space. A control strategy is applied to ensure that the process operation stays within the designated control space.

Figure 1

The QbD process design starts with an intensive characterization of the product through a large array of biochemical and biophysical analyses at normal and stressed conditions and through careful analysis of clinical and nonclinical data. This characterization provides the basis to define the criticality of product quality attributes (QAs), according to knowledge of safety and efficacy of the product. QA criticality is used to prioritize experiments for process understanding and is incorporated into the risk-assessment tools that are used.

Risk Assessment for Prioritization

A typical monoclonal antibody (MAb) manufacturing process involves >20 distinct unit operations with >200 process parameters, and more than 50 different raw materials, making the complexity level significantly higher than that of a small-molecule drug. Therefore, a clear and efficient strategy is required to identify high-risk process parameters for process characterization.

Figure 2

A multidisciplinary team consisting of representatives from the quality, process development, regulatory, manufacturing, and analytical groups actively participates in the risk-assessment process, using data and knowledge from various sources, including previous development, platform process knowledge, manufacturing data from relevant bioprocesses, and literature information. The inputs for risk assessments are summarized in Figure 2. The output from the risk-assessment exercise is captured in a "process understanding plan."

Figure 3

The process for risk assessment leading to experiment prioritization is illustrated in Figure 3. Based on equipment and operation similarity, the process was segmented into various focus areas with defined boundaries. For each focus area, the relevant quality attributes are ranked according to a predefined scale (1 to 10). Then, the effect of each process parameter on every relevant quality attribute is assessed using a predefined scale (1 to 10). Based on this ranking, a cumulative score is calculated for each parameter. This score represents the relative importance of the parameter for the focus area and is used to prioritize experiments.

Table 1. "Cause/Effect" matrix (abbreviated) for an upstream focus area. The scale is 1 to 10, with 10 being the most significant.

Tables 1 and 2 provide two abbreviated examples of a "cause and effect matrix" for an upstream cell culture and a downstream purification process. The examples include the rankings for quality attributes as well as the rankings of the impact of process parameters on quality attributes. The parameters are ranked based on the calculated scores.

Table 2. "Cause/Effect" matrix (abbreviated) for a downstream focus area. The scale is 1 to 10, with 10 being the most significant.

Scale-Down Considerations

The applicability of the process understanding obtained from small-scale experiments depends on the validity of the small-scale model.7 Before executing laboratory-scale experiments, a scale-down strategy must be established for all relevant unit operations in the process. As an example, the scale-down strategy for the production bioreactor has been provided, which is one of the most challenging unit operations in terms of establishing a scale-down model.

The strategy encompasses two approaches: one for scale-independent parameters and the other for scale-dependent parameters. For scale-independent parameters, such as temperature, pH, dissolved oxygen, seeding density, and nutrient feed rate, the small-scale model operates at the same set points and uses the same or comparable online or offline control strategies as the commercial-scale bioreactors. This ensures similarity between small-scale and commercial-scale operations. For scale-dependent parameters, such as agitation and gas sparging, scale-up effects should be minimized between the small-scale model and commercial-scale bioreactors. This is achieved through scale-up/scale-down studies that facilitate the determination of appropriate operating conditions.

Of all scale-dependent parameters, four high-risk parameters—agitation, pressure, surface:volume ratio (S:V), and air sparging—were identified through a risk analysis. Their major scale-up effects and the characterization strategy are illustrated in Figure 4. The five major scale-up effects were identified based on process development and scale-up experience. In the multidimensional space, the farther away from the central point, the more pronounced the scale-up effect. For each scale-up effect, an acceptable range was established, within which comparable process performance was observed using corresponding scale-down models. The combined acceptable ranges formed the acceptable space.

Figure 4

When both the commercial-scale and small-scale bioreactors are operated within the same acceptable space, significant scale-up or scale-down effects can be avoided. As a result, the design space established for scale-independent parameters in the laboratory-scale model can be applied to the commercial-scale operation.

Figure 5

As an example, the establishment of the acceptable range for hydrodynamic stress for an NS0 process is summarized below. Two scale-down models were used: a 3-L small-scale bioreactor with a wide range of agitation rates, and a recirculation "torture chamber" model in which cells circulated between a high-shear microfluidic device and a 2-L small-scale bioreactor.8 Both models demonstrated that the NS0 cell line used in the MAb process can withstand an intensive energy dissipation rate (shear) without significant effects on cell growth, productivity (Figures 5 and 6), or product quality (not shown). Though slightly higher in the torture chamber model, cell growth and product titer were within regular batch-to-batch variation.

Figure 6

Establishing the Design Space

An experimental strategy was developed based on several factors, including parameter risk score, operational considerations, and overall process understanding. For example, although chromatography column lifetime may be an important parameter, it is infeasible to study it in a multivariate study along with other parameters because of complexity and cost considerations.

Design of Experiments (DOE) and multivariate analyses were our preferred methods to study parameters with potential interactions, but other approaches, such as one factor at a time (OFAT) and challenge studies, also were used. One DOE study for the production bioreactor focus area is summarized below as an example of the experimental strategy and the definition of the design space.

Seven parameters with potential interactions in the fed-batch production process were studied in one set of DOE studies:

  • seeding cell density

  • cell density at nutrient feed initiation

  • nutrient feed supplementation rate

  • process duration

  • bioreactor temperature

  • bioreactor pH

  • bioreactor dissolved oxygen.

It would have been ideal to study all seven parameters together in a single design, but it was not practical because of resource (bioreactor) limitations. Therefore, we used a two-round DOE approach. The first round was a screening study that included only the first four parameters in the list. The parameters of significant impact were then combined with the remaining three parameters in the second round DOE study.

First Round DOE. A two-level (high and low) full factorial design including center points was used. The study showed that antibody acidic species level was the most sensitive attribute. The two significant parameters that affect the acidic species level were seeding cell density and culture duration, as shown in Figure 7. These two parameters influence acidic species levels cumulatively, without interaction.

Figure 7

Second Round DOE. Seeding cell density and culture duration were incorporated into the second round DOE study with culture pH, temperature, and DO. A central composite design was used to study four of these five parameters. Culture duration was not included in the central composite design. Instead, it was studied by analyzing samples on different days of all batches in the second round DOE study. The second round DOE study revealed again that antibody acidic species level was the most sensitive quality attribute. As indicated in Figure 8 below, three parameters—temperature, seeding density and culture duration—affect acidic species level cumulatively.

Figure 8

Establishing a Control Strategy for Acidic Species

The two-round DOE study indicated that acidic species level was not sensitive to pH, DO, nutrient feed rate, or cell density at the initiation of feeding within the studied range (knowledge space). Therefore, the control strategy focused on the three significant parameters: temperature, seeding cell density, and process duration. The effect of these three parameters is cumulative. If the three parameters were investigated separately in single-variable studies, a design space similar to the one illustrated in Figure 9a would be established. However, the multivariate study showed that a small portion of this space would result in unacceptable product. We evaluated two options to define the design space, a structured one (Figure 9b) and a truncated one (Figure 9c). Although the structured space is easier to implement, the operation ranges are narrow, especially for temperature (35.5–37.0 °C). For this limitation, we chose the truncated space, which allowed us to expand operating ranges without compromising product quality. Specifically, we removed the combination of high seeding density (>4 x 105 viable cells/mL) and long process duration (>11.5 days).

Figure 9

An Integrated Approach to Defining the Design Space

One of the key challenges in defining a design space based on results from these multifaceted studies is that they are not amenable to the comparatively simple "contour plots" that typically are used when defining the design space. For example, in this simplified approach, the functional relationship (obtained through an empirical statistical model) between the process parameters and quality attributes is used to predict the expected range of quality attributes for a given level of process parameters. The parameter ranges that provide acceptable levels of all the quality attributes (the common areas) are then described as the design space for the process parameters for that step.

Although this approach is acceptable for simple focus areas that are essentially independent, it is of limited use when applied to more complex focus areas (as is often the case in bioprocesses). MAb manufacturing processes invariably involve process parameters that are outputs or attributes of prior unit operations. As an example, we can look at the acetate ion and tris ion concentrations that are process parameters for the anion exchange chromatography (AEX) and cation exchange chromatography (CEX) steps, which are the unit operations that typically follow the low pH inactivation step in a MAb process. The AEX and CEX chromatography steps are affected by changes in the inactivation pH, which is an input parameter for the low pH inactivation step. Figure 10 shows the effect of the change in inactivation pH on acetate and tris ion concentrations. To identify the design space that is acceptable for the entire process, the effect of the parameters affecting performance in subsequent steps must be understood.

Figure 10

Therefore, in addition to using outputs (quality attributes) from one focus area as inputs (parameters) in other focus areas, we also used challenge studies to evaluate the interrelationships between the several focus areas, with the ultimate goal of ensuring product quality and process performance. An example of an experimental design for a challenge study is provided in Figure 11. The results from these studies are then used to further refine the design space. For example, if the CEX step is able to clear high levels of impurities generated under the suboptimal conditions, that provides flexibility for defining the design space for the previous steps.

Figure 11

Another aspect of process integration involves evaluating risk by combining the understanding of the effect of parameters on quality attributes with operational considerations (e.g., facility, equipment, procedures, controls). In this approach, the severity score of the failures modes and effects analysis (FMEA) is derived from the process understanding studies and combined with the occurrence and detectability scores to provide an overall risk assessment, which can then be used to assess parameter criticality. This approach makes it possible to establish the design space and the associated control strategies while considering the process as the whole, thus enabling Quality by Design.

CONCLUSIONS

Pfizer has developed a systematic strategy for implementing Quality by Design (QbD) principles for process understanding and design. This strategy was used to develop and characterize a monoclonal antibody production process. Combined with design of experiments (DOE), especially multivariate analyses, this strategy allowed us to focus on high-risk process parameters and study these parameters as integral components of the process. The effective characterization of a scale-down bioreactor model also aided in this process. This QbD process design enriches process understanding, boosts manufacturing process robustness to consistently deliver safe and efficacious product, spurs technological innovation and new approaches to process design and validation, and helps minimize and mitigate risk.

Amit Banerjee, PhD, is a research fellow at Pfizer Global Research & Development, St. Louis, MO, 636.247.5516, amit.banerjee@pfizer.com. NingNing Ma and Natarajan Ramasubramanyan also contributed to this article.

REFERENCES

1. US Food and Drug Administration. Guidance for industry. Quality system approach to pharmaceutical CGMP regulations. Rockville, MD: 2006.

2. International Conference on Harmonization (ICH). Q8(R1), Pharmaceutical development. Geneva, Switzerland; 2007.

3. US FDA. Pharmaceutical CGMPs for the 21st Century—a risk-based approach. Final report. Rockville, MD; 2004.

4. Ende D, Bronk KS, Mustakis J, O'Conor G, Santa Maria CL, Nosal R, Watson TJ. API Quality by Design. Example from the torcetrapib manufacturing process. J Pharm Innov. 2007;2(3–4):71–86.

5. ICH. Q9, Quality risk management. Geneva, Switzerland; 2006.

6. ICH. Q10, Pharmaceutical quality systems. Geneva, Switzerland; 2008.

7. Godavarti R, Petrone J, Robinson J, Wright R, Kelly BD. Scale down models for purification processes. Process. In: A. S. Rathore and Gail S. Sofer, editors, Validation in manufacturing of biopharmaceuticals. Boca Raton, FL: CRC Press, Taylor and Francis Group; 2005. p. 69–142, .

8. Godoy-Silva R, Chalmers JJ, Casnocha SA, Bass LA, Ma N. Physiological responses of CHO cells to repetitive hydrodynamic stress. Biotechnol bioeng. 2009;103(6):1103–17.