QbD and PAT in Upstream and Downstream Processing

To gain perspective on the implementation of quality by design (QbD) and process analytical technology (PAT) in biopharmaceutical processing, BioPharm International spoke with Clinton Weber, associate director of bioprocess sciences, CMC Biologics Organization; Henrik Johanning, director QAtor; Nathan L. McKnight, PhD, principal engineer, late stage cell culture, BioProcess Development, Genentech; Anurag Rathore, professor, Indian Institute of Technology (IIT) Delhi; Frederic Girard, CEO, Spinnovation Biologics; Thomas J. Vanden Boom, PhD, vice-president, global biologics research, development and manufacturing operations, Hospira.

Upstream Processing

BioPharm: In implementing QbD, what would you identify as the critical quality attributes (CQAs) in a typical upstream bioprocess using cell-culture?

Vanden Boom (Hospira): Generally, upstream critical quality attributes for a cell-culture manufacturing process are limited to the adventitious agent and bioburden testing of the cell-culture harvest material. This also holds for biosimilar products.

McKnight (Genentech): CQAs are defined for the product, not identified as part of upstream or downstream portions of the manufacturing process. There are key performance indicators (KPIs) that are defined for upstream steps, including culture productivity (i.e., titer), cell growth, and viability. While KPIs may be correlated with CQA results, KPIs are not themselves product quality attributes. However, particular CQAs may be generated or modified during specific upstream or downstream steps. Protein glycosylation, for example, is generally determined during cell culture and minimally altered in downstream unit operations (at least for uncharged glycosylation species). Using this definition of CQAs, those CQAs observed to be potentially impacted during cell-culture steps include product attributes contained within the protein glycan distribution (e.g., afucosylated glycan), charge-variant distribution (e.g., glycated, deamidated forms), and to a lesser extent, molecular-size distribution (e.g., dimer or aggregate forms).

It should be noted that knowledge of an association between cell-culture steps with certain product quality attributes is not a result of implementing a QbD approach but, rather, a result of knowledge gained through the basic scientific and engineering endeavors that should be elements of developing a bioprocess.

Rathore (IIT Delhi): In the past, most CQAs could not be measured directly in the fermenter broth due to the interference from the numerous components present in the broth. As a result of major advancements in analytical science, direct measurements of CQAs are performed in bioprocessing today. These would be product-quality related parameters, including host-cell impurities (e.g., host-cell proteins, DNA), process-related (e.g., Protein A leachate), and product related (e.g., aggregate, basic variants, acidic variants, and glycoxylation pattern). Some of these CQAs, such as glycoxylation pattern for monoclonal antibody (mAb) products, are primarily impacted by the upstream process and are particularly important to monitor during process development.

Weber (CMC Biologics): The goal of a QbD approach is to develop additional knowledge of the impact of upstream process unit operation performance on the final purified product quality. The most likely or desired outcome is to develop a quantifiable correlation between upstream process outputs, such as cell viability or viable cell density (VCD), and product attributes, such as glycosylation. This approach can lead in the determination of CQAs for the cell production bioreactor unit operation. Upstream outputs classified as CQAs have proven to be controversial, because the bioreactor is so far upstream from the final product. However, if a strong correlation can be established between VCD/viability and other CQAs or final product specifications, this can be an appropriate approach.

Girard (Spinnovation): The underlying concept for QbD of an upstream bioprocess is that the desired quality of the biological or biopharmaceutical product is assured every time. CQAs vary for each cell line depending on the nature of the bioprocess, with typical critical qualities such as metabolites and contaminants. CQAs usually include properties that affect product quality and eventually overall performance of the bioprocess. CQAs are typically also release tests, although they don’t have to be as there are no real release tests as such in upstream processing.

The variability and complexities associated with the upstream biological process make QbD a complex process, one that relies on defining operation specific critical process parameters (CPPs). CPPs are those likely to impact on the quality of a product or intermediate. For biological products, process control can be difficult to define and implement. O2 pressure, catalyst concentration, and pH are examples of critical parameters. It is important to note that mAbs are currently the leading area of biopharmaceutical research. One of the key parameters to monitor in the implementation of QbD in mAb production is the glycosylation process during formulation. Glycosylation is one of the overriding contributors to mAb heterogeneity and has significant implications for the function of the antibody in vivo and immunogenicity. This means that glycosylation has been isolated as a critical parameter to follow during mAb manufacture. QbD for mAb development with specific glycosylation patterns enables researchers to optimize manufacturing and clinical efficiency.

Johanning (QAtor): The QbD concept works its way back so to speak from the patient to product to process and ultimately to the facility. A risk assessment will outline the risk profile in and between each area.

The starting point for the risk assessment is R&D, which upstream has determined the CQAs on the product. CQAs are often product specifications, including eventual GMP requirements (if GMP is used as part of the biopharmaceutical process, which is often the case in multinational pharma companies in order to ensure a fast-track initiative from R&D to license to operate and market). The risk assessment includes a review and assessment of the products CQAs when manufactured on specific equipment.

BioPharm: In implementing QbD, what would you identify as the critical process parameters (CPPs) in a typical upstream bioprocess using cell culture?

McKnight (Genentech): Among cell-culture parameters, culture pH is typically the most difficult to control relative to its impact on cell-culture performance and product quality. This is generally the case for both mAbs and other products.

Johanning (QAtor): Common CQAs in an upstream bioprocess (i.e., bio formulation process) using cell-culture include sterility and bioactivity.

Rathore (IIT Delhi): Typical CPPs for a fermentation step would be pH, sparge rate, agitation rate, and temperature. These are typically easy to control. Issues arise when one goes to volumes greater than 1000 L and when it increasingly becomes more difficult to strip off the CO2 generated by the cells and in ensuring uniform supply of O2 and other critical nutrients to the cells. Another set of challenges comes from raw materials that are complex and not well characterized (such as yeastolates) as these can result in significant variation in the CQAs from lot to lot. For mAbs, besides the afore-mentioned factors, concentration of critical nutrients has been known to affect the glycoxylation pattern of the product, thereby impacting product efficacy.
Vanden Boom (Hospira): Parameters such as temperature, pH, osmolality, and dissolved oxygen have the potential to impact the CQAs of mAbs and other mammalian cell-culture derived products. However, with current bioreactor engineering controls, these parameters may be tightly and confidently controlled within the design space of the manufacturing process permitting these parameters to be downgraded from a CPP to a key process parameter (or other lower parameter designation used by different drug sponsors).

Weber (CMC Biologics): For an upstream process, the process of expanding the cells is the primary purpose until the culture goes into the production bioreactor. Because the majority of product produced is in the production bioreactor, the expansion process is considered to have minimal impact on the final product. Additionally, most products have shown the ability to recover from suboptimal conditions during expansion without serious product quality impact. Therefore, for a typical upstream process, CPPs are identified at the production bioreactor stage. CPPs for the production bioreactor may include seed density, temperature, and process duration. Initial seed density can impact the overall growth profile and viability. Temperature is critical in maintaining viability and may impact the quality of product being produced by the cells. Process duration will impact the viability of the culture at harvest, which can be tied to product quality.

Among the most difficult parameters to control in an upstream process is CO2 concentration. The difficulty of controlling this parameter depends on the complexity of the aeration control strategy and availability of dissolved CO2 probes. Though there can be typically broad ranges of acceptable CO2 during production, very high CO2 concentrations can impact the product quality. A balance needs to be maintained between lowering CO2 and maintaining pH at a set point. Furthermore, these conclusions seem to be supportive of most cell-culture processes, not just mAb production.

BioPharm: What measurement tools are typically used to measure CPPs and the resultant process inputs and outputs, including PAT tools, in an upstream process?

McKnight (Genentech): CPPs are a subset of the environmental and batch-recipe settings (e.g., timing of feeds, culture duration) used to perform the process. Temperature and pH, which are commonly CPPs, are measured using on-line probes (a technology that is officially PAT, but far preceding the ‘PAT initiative’). Timing of events and generation of basal and feed media are controlled with traditional process controls and standard operating procedures. Online cell-density measurement is an area of active development as is online nutrient measurement technology to enable advanced feeding or timing strategies.

Weber (CMC Biologics): Assuming this question is referring to the monitoring and control of CPPs and not the tools used to actually establish CPPs, the majority of upstream CPPs are monitored online and offline. Online instruments such as CO2 probes are verified against offline instruments to ensure that the conditions within the bioreactor have not caused the probes to ‘drift.’ Adjustments are made to the online instruments if drift is detected. Other outputs such as cell count and viability are measured strictly offline on a routine basis. PAT tools can be effective, though not necessary, to monitor parameters such as temperature and pH, but are not necessarily value-added for culture health outputs such as viable cell density, viability, or doubling time.

Rathore (IIT Delhi): For upstream process, tools that researchers have used include:

• Surface plasmon resonance (to assess product concentration and affinity)
• High-performance liquid chromatography (HPLC) (to assess product concentration and structure)
• Capillary electrophoresis (to assess product concentration and structure)
• Dielectric spectroscopy (to determine biomass)
• In-situ microscopy (to characterize cell population)
• Flow cytometry (to characterize cell population)
• Metal oxide field effect transistor (to sense biological contaminants)
• Infrared spectroscopy (to detect media components)
• In-situ 2D fluorometry (to detect media components and metabolic end products)
• Raman spectroscopy (to detect media components and metabolic end products)
• UV spectroscopy (to measure homogenate components)
• Mass spectroscopy (to detect metabolic end products)
• HPLC (to detect media components and metabolic end products).

Not all of these are amenable for online applications, but together they capture various attributes of upstream processing.

BioPharm: Given the inherent variability in biologics manufacturing, how does QbD improve process understanding and control? What are the limitations of QbD in upstream bioprocessing?

McKnight (Genentech): The inherent variability in biologics manufacturing, and the general inability to define mechanistic equations (i.e., mechanistic process models versus empirical process models), limits the ability to precisely predict the outcome of specific runs at manufacturing scale. Application of multivariate, statistically designed experiments, however, is still valuable for identifying CPPs, defining parameter acceptable ranges, and understanding the variability that may be expected from the manufacturing-scale process. Biological variability likely limits the ability to control even the best understood process solely through control of process parameters—the need for some degree of product testing will be necessary to control for inherent variability.

Rathore (IIT Delhi): Implementation of QbD necessitates creation of information relating the process to the product and the product to the clinic. It is this understanding that lays the foundation for appropriate process control. A major limitation that I see with respect to implementation of QbD in upstream processing is the complexity of the sample medium due to the presence of the large variety of process related, host-cell related, and product-related impurities. Another limitation is the fact that the fermentation process is so complex. With the aforementioned, tools it is easy to monitor different process and product attributes. However, many different alterations in operating conditions and raw-material attributes can lead to similar changes in the process outputs and hence monitoring is merely the first and simpler step. The difficulty comes in diagnosing the root cause of variability and effectively dealing with it in real time.

Girard (Spinnovation): Precisely because QbD is a scientific, risk-based, and proactive approach to biologics development, one can use it to define the ideal characteristics of a product to achieve CQAs relating directly to its clinical performance. On the basis of this information, product formulation and processes are designed within a specific framework to ensure the product meets these attributes. Variability within this framework can be monitored allowing scrutiny of the process to assure consistent product quality. However, it is important to consider the CQAs in a matrix since one knows that a biological system has the capability to compensate or adjust its metabolic pathway.

Vanden Boom (Hospira): The enhanced design-space knowledge derived from the systematic risk assessments and design of experiment (DOE) work completed in association with a QbD approach offers the potential to significantly improve the level of process control for mammalian cell culture-derived products. A key factor to realize the full benefit of QbD is the establishment of robust small-scale model(s) of the upstream manufacturing process. This may be more challenging for certain products resulting in limitations to fully using QbD for upstream bioprocessing steps for these products. In the case of biosimilar products, the bioanalytical characterization of the originator product provides another useful input to determine the significance of product and process variability.

Weber (CMC Biologics): A QbD approach can lead to an early focus and understanding of what influences the CQAs. While scale-down models combined with screening and design space DOEs can be used to understand cell expansion and the cell-production bioreactor processes, full purification can be time consuming and costly for full characterization of the upstream process. In addition, as the downstream process is characterized, optimizations/changes in the downstream process can lead to the need to repeat upstream characterization efforts. Hence, the sooner a correlation between the main upstream outputs (such as viability) and primary product quality attributes (such as glycosylation) can be established, bench-scale work can be minimized.

Downstream Processing

BioPharm: In implementing QbD, what would you identify as the CQAs in a typical downstream bioprocess in which the product was produced using cell culture?

Weber (CMC Biologics): This is probably the most difficult question to answer. We have found the CQAs for downstream are very product dependent. Glycosylation and sialylation are definitely two outputs that require product and product quality understanding and are the most common across different cell-culture processes. After that, variation in the product profile begins to manifest itself too much, preventing universal CQAs to be established.

McKnight (Genentech): Some CQAs are generated in, or substantially changed by, certain downstream unit operations. Size variants or charge variants, for example, may be generated during hold times, and product quality attributes such as host-cell proteins and size variants are typically reduced through the purification process and controlled to an acceptable level by consistent purification process performance.

Rathore (IIT Delhi): The downstream process attributes would be similar to those mentioned earlier for upstream processing. These include process-related impurities (e.g., antifoam, additives added during the processing, Protein A leachate), host-cell impurities (e.g., host-cell proteins), and product-related impurities (e.g., aggregate, basic variants, acidic variants, glycoxylation pattern). The big difference is that the ease of measurement of these attributes improves significantly in downstream processing as the samples are relatively cleaner.

BioPharm: In implementing QbD, what would you identify as the CPPs in a typical downstream bioprocess in which the product was produced using cell culture? What are the most difficult parameters to control and why?

McKnight (Genentech): Chromato-graphy unit operations are typically most sensitive to the pH and ionic strength of wash and elution conditions. Control of these critical buffers is managed through batch preparation and release prior to use based on acceptable pH and conductivity ranges. Process capability is sufficient to reproducibly prepare buffers within narrow pH and conductivity ranges, such that only a subset of the most sensitive buffers are typically determined to be CPPs.

Rathore (IIT Delhi): Typical CPPs in downstream processing would include parameters such as pH, conductivity, temperature, and gradient for chromatography steps; temperature, agitation rate, and sparge rate for refolding steps; and pH and hold time for the viral inactivation step. The challenge in downstream processing mainly comes from the fact that the steps are relatively short in duration. For example, a typical elution in chromatography column may be 30–60 minutes. Our ability to measure and then take action, therefore, is quite challenged if the assay is not a real-time assay (such as HPLC).

Johanning (QAtor): CPPs in a typical downstream process involving cell culture include holding time, pH control, temperature control, and UV control. UV control/cutting is the most difficult parameter to control because UV cutting techniques/equipment are still an area for further development. They relatively often challenge batch release (out of specification).

Vanden Boom (Hospira): Originator and biosimilar products have similar downstream CPPs. For chromatography steps, these may include protein load, pH (load or elution depending on step), temperature, flow rate, and conductivity for ion-exchange steps. Viral inactivation steps may have temperature, pH, or detergent concentration as a CPP depending on the modality of inactivation. Pressure and filter volume represent CPPs for viral filtration steps. Again, if engineering controls permit tight control of these parameters, some may be downgraded to a lower parameter designation.

Weber (CMC Biologics): Using a QbD approach, downstream CPPs would be proposed only through using risk assessments of the possible impact to pre-identified CQAs. Downstream CPPs will vary greatly depending on the nature of the molecule, the purification strategy, and the order and timing of unit operations. Nevertheless, there are certain potential CPPs common to many downstream processes. The following list of operations/parameters is potentially responsible for affecting product quality/CQAs, particularly toward the end of a purification process.

Not all of these would necessarily end up as CPPs:

• Viral inactivation: pH of inactivation, time of inactivation, and concentration of inactivation solution
• Viral filtration: filter load density (mg/mL membrane) and filtration pressure
• Chromatography operations: load density, pH and/or conductivity of buffers, residence time, volumes, and eluent concentration
• Filtration operations: load density, transmembrane pressure, crossflow rate, and diafiltration diavolumes.

The most difficult parameters to control are those with narrow allowed ranges. Using QbD, well-designed DOEs are intended to grant maximum flexibility in defining allowed ranges.

BioPharm: What tools are typically used to measure CPPs and the resultant process inputs and outputs in a downstream process, including PAT tools?

McKnight (Genentech): CPPs for chromatography operations frequently include pH and conductivity of the most sensitive wash or elution buffers. Control of these critical buffers is managed through batch preparation and release prior to use based on acceptable pH and conductivity ranges. This practice predates the PAT initiative, but serves the intended purpose of PAT.

Rathore (IIT Delhi): The same aforementioned tools would be applicable here as well as we are measuring the same attributes. As mentioned previously, the difference lies in the fact that the samples are much cleaner and hence easier to analyze. However, the time for decision is significantly less, thus making PAT implementation more of a challenge.

Weber (CMC Biologics): Put simply, the right downstream equipment for chromatography and filtration should have built-in monitoring of all potential CPP parameters and should cover a wide range of operational values. This can either transmit out for a PAT approach, or can allow for well-designed process monitoring.

QbD and PAT In Bioprocessing

BioPharm: What are some recent advances in PAT or other analytical tools that better enable product characterization and the increased process understanding desired in a QbD paradigm? How is QbD applied to equipment design and implementation in biologic API manufacturing?

Girard (Spinnovation): FDA identified the use of new analytical methods, such as nuclear magnetic resonance (NMR), to monitor and control processes as important in increasing manufacturing quality through QbD. NMR could change the face of bioprocess development and monitoring because it provides access to component identity, plus quantitative data, rapidly and easily from a single analysis. NMR profiling can provide full visibility of the presence and concentration of feed components, contaminants, and metabolites. The technique is capable of providing access to accurate concentration data for media components. NMR-based methods can provide rapid and accurate quantitative monitoring of more than 50 media components, contaminants, and metabolites within a culture at any stage in the process to meet QbD requirements.
The use of NMR, combined with statistical approaches, provides rapid solutions to performance  inconsistencies in simple and complex raw materials within upstream processes of virtually any bioprocess, offering the ability to characterize chemically complex media. NMR techniques have the potential to contribute significantly to an understanding of process-critical parameters, helping to reduce performance variability and minimize the risk of process failure in large-scale biopharma production.

Johanning (QAtor): Before (traditional) installation, test and qualification of equipment a QbD approach involves a design qualification, where the user requirement specification (URS), which includes the basic CQAs from the master and manufacturing production files, is reviewed and qualified against the proposed equipment design. This, of course, involves specialists from R&D, manufacturing, and quality assurance. The design qualification and review is challenging the equipment design with special attention on CQAs such as contamination with germ and fibrillation.

Vanden Boom (Hospira): In the case of biosimilar products, the bioanalytical characterization of originator products over the approved shelf life provides a powerful input for use of a QbD paradigm in process development. Biosimilar developers also have the opportunity to employ modern manufacturing technologies, including PAT tools, to enhance process control. The enhanced monitoring of potential CPPs on current buffer dilution and chromatography skids represent one example of improved equipment design contributing to the improved process control of biopharmaceutical manufacturing processes.

Weber (CMC Biologics): PAT implies an online monitoring system that can determine if a process is trending negatively in ‘real time.’ The essentials of PAT have been in the industry nearly from the beginning: data historian. Although PAT allows for maximal design space control and understanding, whether one has real-time information and automated response capability is not essential to the QbD paradigm. However, having a well-understood and robust historian is. As long as processing trends and capabilities are assessed on a per-run basis, there is an opportunity to monitor and adjust to ensure that the process performs optimally within the design space. Many unit operations within cell-culture manufacturing processes have control capabilities built into the equipment; as long as the assessment is performed, the equipment can repeatedly perform within the required space. Couple that with strong operator training and having real-time process adjustments is already inherent in a process without PAT. Generating dynamic and static control charts on a per-run basis of the process allows for trend analysis, and adjustments can be made should a negative trend be detected. This can all be accomplished without a formal PAT system.