Here we apply our approach to validation of animal cell culture process changes using qualified, scale-down bioreactors. As described in Part 1 (including Table 1, Figures 1 and 2, and References 1–23), the goal is to facilitate implementation for the benefit of both the patients and industry. “Qualification of Scale-Down Bioreactors: Validation of Process Changes in Commercial Production of Animal-Cell–Derived Products, Part 1 — Concept” appears on pages 38–45 of BioProcess International’s May 2014 issue.

Process changes often entail validation, but a common problem facing the industry is justifying the validation approach. Traditionally, process-development groups have performed validation at scale, in production, for the duration of a campaign. However, that entails dedicating production units of operation to the validation exercise and segregating commercial and validation materials from each other throughout the process. Taking a production unit out of the supply line (even temporarily, until validation is confirmed successful) lessens capacity output. That in turn can prohibit validation and impede implementation of beneficial process changes.

Product Focus: Recombinant proteins

Process Focus: Production

Who Should Read: Process development, analytical, regulatory affairs, and manufacturing

Keywords: Process optimization, cell culture, scale-up/scale-down, model systems, product quality/comparability

Level: Advanced

BAYER HEALTHCARE

Our proposed solution to this problem is performing validation outside the good manufacturing practice (GMP) area using qualified scale-down unit operations. We describe this qualification program in two-parts: Concept (Part 1) and Application (Part 2). Unit operations of the entire process (upstream/cell culture, downstream/purification, final-product manufacturing) would need to be qualified for the validation exercise. Our case study focuses on the bioreactor unit operation because, for many companies, cell culture is rate-limiting in production processes. This is as relevant to companies that use larger reactors in fed-batch mode over relatively short campaigns as it is to those using smaller reactors in perfusion mode over lengthy campaigns.

Part 1 discussed elements of the qualification program; here we describe its application. Our approach could reduce risk upon implementation because using qualified small-scale systems provides flexibility. That enables a larger number of validation runs to be performed, which is in line with current regulatory-agency expectations for a science-based approach to process validation.

Scale-Down of a Manufacturing- Scale Perfusion Bioreactor at a Bayer Commercial Manufacturing Facility: We applied scale-down system qualification principles discussed in
Part 1: design, performance, and product quality. To achieve proportions similar to those of our (over an order of magnitude larger) production-scale system, designers of the scale-down laboratory-scale perfusion bioreactor (including its cell retention device) made an effort to be faithful to physical parameters such as geometric aspect ratios, angles, port location, and so on. We tuned the operation parameters (e.g., stirring speed, volume-specific power inputs, liquid and gas fluxes to/from the bioreactor) to achieve comparable performance (e.g., in aeration, gas–liquid mass transfer, and mixing time) to that of the production system. Finally, we tested the two systems — scaled down and commercial scale —side by side in production cell culture campaigns.

Materials and Methods

Cell Scale-Up and Bioreactor Operation: We expanded a single vial from a research cell bank in roller-bottle culture using a proprietary medium formulation with glucose and glutamine as its main carbon sources. The bottles rolled in a 5% CO₂ incubator at an operating temperature of 36.5 °C. We used cells from that culture to inoculate a laboratory-scale perfusion bioreactor, which was subsequently used to inoculate a manufacturing-scale bioreactor.

Variable set-points and operating conditions were the same in both our laboratory and manufacturing-scale systems. Bioreactor temperature remained at 35.5 °C, dissolved oxygen (DO) concentration remained at 50% air saturation, and automatic additions of 6% Na₂CO₃ maintained pH at 6.8. Bioreactor cell density on inoculation was ∼1 × 10⁶ cells/mL; after a cell accumulation phase, the target steady-state concentration of 20 × 10⁶ cells/mL was reached. Steady-state bioreactor cell density was maintained by automatic discarding of cells from the bioreactor based on cell-density measurements.

Abbreviations

CFD: computational fluid dynamics

CGMP: current good manufacturing practice

CPP: critical process parameter

DO: dissolved oxygen

DoE: design of experiment

EDR: energy dissipation rates

EoF: edge of failure

FMEA: failure mode and effects analysis

ICH: International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use

IR: incident/investigation report

KPP: key process parameter

NOR: normal operation range

OFAT: one factor at a time

OUR: oxygen uptake rate

PAR: proven acceptable range

PQ: process qualification

PQA: product quality attribute

QA: quality assurance

QbD: quality by design

sGCR: specific glucose consumption

sLPR: specific lactate production

SOP: standard operating procedure

Bioreactor Monitoring: We sampled the bioreactor daily for cell density and viability analyses using a CEDEX system from Innovatis. Subsequently we centrifuged those samples using a Beckman Coulter instrument and analyzed the supernatants for nutrient and metabolite concentrations. Glucose, lactate, glutamine, and glutamate concentrations were determined using a YSI model 2700 analyzer from Yellow Springs Instruments. We measured pH and DO on-line with retractable electrodes from Mettler-Toledo Inc. after verifying their measurement through off-line analysis in a Rapidlab 248 blood-gas analyzer from Bayer HealthCare. We also used that same instrument to measure dissolved CO₂ concentrations.

Bioreactor Data Analysis: We used bioreactor data from days 45–95 (after inoculation) for comparison of both bioreactors at steady-state. Average comparisons were made using both the t-test and an equivalence test in which a practical threshold difference could be specified. We set that threshold difference to 10% for viable cell density (VCD) and perfusion rates, 15% for glucose and lactate consumption rates and specific productivity, and 25% for growth rate based on errors inherent in measurement and computation of those variables (24). For statistical analysis, we used JMP statistical software from SAS.

Results and Discussion

Design and System Performance: Shear stress is an important performance attribute we determined using a clay/polymer floccular system in floc-destruction experiments (25). Cumulative shear stress is measured in terms of the equivalent floc diameter (d
_ef, µm) as a function of volume-specific power input (P/V in W/m³). The lower the def value is, the higher is the cumulative shear stress (25). Experiments using both scale-down and production-scale bioreactors confirmed the equivalent floc
diameter to be proportional to (P/V) – 1/3 (data not shown).

The mechanism of oxygen gas–liquid mass transfer is another important performance attribute because it can significantly affect carbon dioxide removal and foam formation. Although macro-mixing time was not explicitly a scale-down criterion in this case, our hypothesis was that the geometry of the scale-down bioreactor aeration mechanism, stirrer, and sparger should be similar to their production-scale equivalents so that the mixing patterns would be as similar as possible. So we developed a detailed aeration model to determine the values for each scale-down bioreactor parameter (not shown).

The cell retention device is a crucial aspect of our perfusion system. So we needed a design that would facilitate similar conditions (the physical environment) to be experienced by cells at both scales. For example, scale-down bioreactors’ liquid-fill volumes in relation to the corresponding retention device (including its recirculation loops) were designed to emulate those of commercial-scale bioreactors.

Cell Culture Performance and Product Quality: Multiple variables were measured or computed based on daily sampling data, but here we compare six variables between the laboratory- and manufacturing-scale bioreactors. Bioreactor cell density and perfusion rate are perhaps the two most important controlled variables in a perfusion process, so comparing them provides information about the quality of process control in bioreactor operation. Cell-specific rates of glucose consumption and lactate production are important indicators of metabolic state, so comparing them provides evidence of potential scale impacts on cell metabolism. And finally, cell growth and protein productivity are key performance metrics in a perfusion system, so we also compared them for the laboratory- and manufacturing-scale bioreactors.

Bioreactor Cell Density and Perfusion Rate:
Figures 3a and 3b show profiles and averages, respectively, for bioreactor viable cell density (VCD). The set point for bioreactor VCD was 20 × 10⁶ cells/mL, and the averages for laboratory and manufacturing scale were 20.35 and 20.51 × 10⁶ cells/mL, respectively. Respective coefficients of variation at 5.1 and 7.2% (Table 2) suggest good cell-density control. Results from both the t-test and means equivalence test suggest no significant difference in the averages (Table 2).

Figure 3: Cell density (a, b) and normalized perfusion rate (c, d) plotted over the campaign days (a, c), with distribution analysis and averages (b, d) shown for the laboratory-scale (scale-down) and manufacturing-scale bioreactors in full and open circles, respectively

Table 2: Laboratory- (scale-down) and manufacturing-scale platforms are compared; average (coefficient of variation, %) values for each of the six variables shown in Figure 3-5, are given and noted for whether the difference is significant or not following t-test and equivalence test.

Another critical control variable for perfusion processes is perfusion rate. Figure 3c and 3d show time profiles and averages, respectively, for normalized perfusion rates. Averages for both reactors (0.99 and 0.97) were close to the target set point of 1.0 and characterized by low coefficients of variation (3.5 and 7.0%) reflecting well-controlled bioreactor operation. Results from both statistical comparison tests indicate no significant difference in the averages (Table 2).

Glucose and Lactate Metabolism:
Figure 4 shows time profiles and averages of cell-specific glucose consumption (sGCR) and lactate production (sLPR) rates. Both variables were relatively stable considering inherent measurement and computational errors. Averages of sGCR were the same (1.52 pmol/ cell/d) in both reactors, whereas a slightly higher sLPR average was seen in the laboratory-scale unit (1.93 pmol/cell/d) than at manufacturing scale (1.77 pmol/cell/d). Both the t and equivalence tests indicated no significant difference in sGCR between the laboratory and manufacturing-scale bioreactors (Table 2). Whereas the t-test (but not the equivalence test) indicated that the higher laboratory-scale sLPR value was significant, the t-test may not be the best way to compare that average because of inherent measurement and computation associated variability (24).

Figure 4: Cell-specific glucose consumption (a, b) and lactate production (c, d) plotted over the campaign days (a, c), with distribution analysis and averages (b, d) shown for the laboratory-scale (scale-down) and manufacturing-scale bioreactors in full and open circles, respectively.

Growth and Specific Productivity:
Figure 5 shows cell-growth rate and normalized protein productivity data, with associated averages and statistical comparison given in Table 2. Of parameters we evaluated (Figure 3–Figure 5), growth rate had the highest variability. This we expected was primarily due to computation-associated errors (24). The growth rate average in the laboratory-scale bioreactor was slightly higher than that of the manufacturing-scale bioreactor (0.80 versus 0.68 d^–1), a difference that was significant based on the t-test but not the equivalence test (Table 2). Identical specific protein productivity averages were seen for both bioreactor scales, with no statistically significant differences (Figure 5d; Table 2).

Figure 5: Growth rate (a, b) and normalized productivity (c, d) plotted over the campaign days (a, c), with distribution analysis and averages (b, d) shown for the laboratory-scale (scale-down) and manufacturing-scale bioreactors in full and open circles, respectively

Overall, the data presented in Figure 3–Figure 5 and Table 2 suggest consistency in the operation of the laboratory- and manufacturing-scale bioreactors (based on cell-density and perfusion-rate data), resulting in very similar metabolic, growth, and protein-productivity profiles. We also found a high degree of similarity in product-quality attributes from the scale-down (laboratory scale) and manufacturing-scale bioreactors (data not shown).

From a biological perspective, if a homogeneous population of production cells are subjected to highly consistent culture conditions (medium and environmental inputs), their metabolism including specific productivity and quality attributes of expressed protein should remain consistent. Optimization and robustness studies to define process inputs would help in achieving desired product attributes/ specifications and similar responses from process input changes — no matter what the specific bioreactor scale might be, as long as the cellular environment responds to the process change similarly.

The type and scope of experimental work — design space, choice of variables, ranges, specific excursions, and so on — needed to qualify a small-scale bioreactor are ideally performed by anticipating certain plausible changes that it would be used to validate. Our results suggest that qualified scale-down systems are a justifiable approach to validation of process changes — and we have provided a road map to that end. Nevertheless, a risk assessment would be prudent regarding specific change to be validated in certain cases.

A risk assessment exercise would help you determine whether an available data package (demonstrating scale representativeness) and process knowledge collectively provide an appropriate degree of confidence for performing validation using a qualified scale-down system (or whether resorting to the production scale is warranted). The FDA’s new validation guidance moved away from the familiar requirement of three consecutive validation runs (6). Using qualified small-scale systems provides added flexibility needed for increasing the number of validation runs (beyond three) based on process capability and statistical methodologies (26) to ultimately reduce the risk of implementing a change.

A Valid Approach

Applying a qualified scale-down bioreactor to certain validation studies would greatly facilitate implementation of process improvements such as drug safety, process robustness, and production cost — providing mutual benefits to both the industry and public health (2). We focused on the specific requirements and approach to qualifying a scale-down bioreactor for validating process changes to be implemented in a commercial manufacturing process. Furthermore, we justified this approach by demonstrating considerations in design through performance of a scaled-down, mammalian cell-culture perfusion bioreactor in a case study comparing it with a commercial-scale system. Tailoring the qualification approach to anticipated process changes would further enhance the utility of a scale-down bioreactor and facilitate a successful outcome during validation.

This concept is not new to the biotechnology industry: qualifying a scale-down system located outside the good manufacturing practice (GMP) production facility for the purpose of validation toward direct implementation at scale. It has been endorsed by regulatory bodies and used extensively, for example, to qualify downstream (protein purification) systems for applications such as validating viral clearance (5,7) and to justify using good laboratory practice (GLP) compliant pilot plants for validation of changes to a recovery process (8). Using qualified scale-down systems provides the flexibility of performing as many validation runs as it scientifically makes sense to perform (rather than just three by rote) in line with current regulatory expectations (6). Therefore, validation using qualified scale-down systems could in fact reduce risks associated with change implementation (26). The qualified scale-down bioreactor (and associated process systems) would enable validation of process changes that might not be feasible to validate otherwise (at scale in production). Although we focus here on validating changes to a licensed product/process, a qualified scale-down bioreactor may also be used in performing design-of-experiment (DoE) runs in support of quality by design (QbD) and design space work, the results of which could be included in regulatory submissions to health authorities (27). Ultimately it could facilitate implementation of important process changes for the benefit of patients and the biopharmaceutical industry alike.

Author Details

Corresponding author Yuval Shimoni is principal engineer, and Venkatesh Srinivasan is director of manufacturing sciences at Bayer HealthCare LLC, 800 Dwight Way, PO Box 1986, Berkeley, CA, 94710; 1-510-705-5775; yuval.shimoni@bayer.com. Marc Jenne is at Bayer Technology Services in Leverkusen, Germany. Chetan Goudar is currently director of cell science and technology at Amgen Inc., 1 Amgen Center Drive, Thousand Oaks, CA 91320. Shimoni presented parts of this work at Cell Culture Engineering XIII in Scottsdale, AZ (22–27 April 2012).

Acknowledgments

We thank Hugh Graham, Soumitra Ghosh, Juri Seletzky, and Peter Kramer for critical reading of our manuscript and are grateful to Volker Moehrle and Helmut Brod for helpful discussions. Edward Long and Ricardo Ibarra are acknowledged for help with the commercial-scale experiments.

References, Part 2