by Brandon Downey
Associate Director, R&D Biologics
An overriding mission at Lonza, and for many CDMOs, is to consistently create effective, safe, and high-quality drug substances and products. The structural attributes of those substances are among the key components that determine the safety, efficacy, and quality of those products. The process used to produce those products can profoundly impact those structural attributes. Therefore, systems must be put in place to ensure consistent product quality attributes across the entire product life cycle.
To achieve this high level of consistency, developers and manufacturers need a system that makes quality and safety central to the development and manufacturing process. One such system is the Quality by Design (QbD) approach. QbD is a holistic and proactive approach to defining and controlling process impacts on product quality. The QbD approach includes:
- Characterizing a design space, which identifies process inputs that have a significant risk of impacting product quality.
- Quantifying the potential impact on quality for specified variation of process inputs
- Implementing a process control strategy to minimize the risk of process variations occurring or mitigate the impact on product quality should they occur.
Prospective Process Control
A key part of implementing processes incorporating a QbD approach is the control strategy. The goal of the process control strategy in a QbD process is to maintain consistent product quality using a well-characterized design space. However, there are two approaches to implementing this control strategy –a retrospective approach and a prospective approach (see Figure 1).
Figure 1: Schematic diagram of retrospective and prospective control strategies.
By far, the most common QbD control strategy implemented in the industry today is a retrospective approach, in which consistency of product quality is achieved by minimizing process disturbances and tightly controlling process inputs on a recipe. Product quality is only measured after the completion of the process in this approach. The QbD design space framework helps identify key process inputs that need to be tightly controlled and potential disturbances which must be mitigated to maintain product quality. This approach is simple, proven, and effective, but critical issues can arise when unanticipated process disturbances are encountered with this strategy.
In the prospective approach, also referred to here as product attribute control, consistency of product quality is achieved by monitoring product quality during the process and actively manipulating specified process control handles to maintain product quality attributes within specified limits. In prospective control, QbD design space knowledge is used to determine which process control handles to manipulate and how to manipulate them to achieve the desired product quality profile. This control approach has some implementation barriers but maximizes the ability to mitigate unanticipated process disturbances.
Drivers for Prospective Control
A Prospective Control Approach Can Maximize Successful Batches
As mentioned above, current control strategies for biopharmaceutical production processes involve tightly controlling process inputs and raw materials, thereby minimizing the risk of producing product attributes outside of specifications. This retrospective control approach can be very effective, but it does not provide assurances against unforeseen disturbances to the process. Since product quality is not actively controlled in this approach, quality variations are not corrected during the process, and that can result in a failed batch – a catastrophic result for the CDMO and client. Adding product attribute control to the QbD toolbox, where product quality is actively monitored and controlled to a defined setpoint, can ensure batch success by catching quality variations during the process and mitigating them in real time before the batch becomes unusable.
The Rise of Complex Proteins Calls for a More Flexible Approach to Manufacturing
Complex proteins, such as bispecific antibodies, contain more diverse structures and mechanisms of action than “vanilla” mAbs. This structural diversity can result, for example, in difficulty expressing in CHO cells, potentially requiring more flexibility in the scale and format of the cell culture process to meet material demands. Additionally, it can be more difficult early in a development program to forecast the market uptake and potential off-label usage for a novel medicine compared to more established molecules and indications. That means the processes for making these molecules need to be more flexible and modular to handle potential unforeseen changes in market demand over the product's lifetime.
To address the potential diversity in scale and process format for manufacturing complex proteins while still delivering consistency of product quality, we believe a modular manufacturing approach is needed. In a modular format, individual unit operations are designed to be interchangeable as needed by the required scale and unique process considerations for that molecule. Incorporating prospective control strategies into process unit operations can deliver consistent product quality in this flexible manufacturing paradigm by adapting for potential process disturbances that can be difficult to anticipate when prior process experience is limited.
Addressing Barriers to Prospective Control Strategies:
Measuring Product Quality Attributes in Real-Time
A foundational requirement for implementing prospective control in bioprocesses is the ability to measure product quality attributes in time to allow the process to be adjusted to impact quality attributes. Current methods are designed for retrospective product quality analysis and are usually not fast enough or automated to be used in a prospective control strategy. The challenge is to adapt existing methods (or identify alternative methods) for measuring quality attributes in bioprocesses to be fast enough to enable prospective control and completely automate the method and resulting data such that a computer can use the output to manipulate the process appropriately. Many product quality attributes are determined in the production bioreactor step. Therefore, Lonza invests in systems to measure product quality attributes in near real-time in the bioreactor. An example system Lonza is developing to enable prospective control of glycan structures for protein therapeutics is shown in Figure 2.
Figure 2: System for measuring glycan abundance in near real-time from bioreactor processes.
The system in Figure 2 has been implemented in R&D at the laboratory scale and can yield glycan abundance measurements every 4 hours, with results available in 1.5 hours. These analysis times are short enough to allow meaningful control actions in cell culture processes.1 The entire process is also automated, from sampling the bioreactor, sample preparation and detection by intact mass spectrometry through data analysis and reporting of results. Completely automated measurement of product quality attributes is needed to implement a prospective control strategy.
Characterizing a Design Space to Enable Product Attribute Control
To implement prospective QbD and product attribute control consistently and reliably, we must be able to characterize a design space that relates process inputs to product quality. The challenge for implementation is to achieve the needed characterization of the design space within the timeline and resource constraints of modern drug development.
A prospective control design space has to define the time-dependent impact of process inputs on product quality, as process inputs must be manipulated during the process to maintain product quality attributes during the process. This requires a sufficient volume of process data, and that data must effectively characterize the design space in a time-dependent fashion.
To effectively characterize design spaces for prospective control, Lonza R&D is leveraging accumulated process knowledge from datasets collected across development programs and collecting more of the data needed to characterize design spaces for prospective product quality control.
First, we are building the systems to enable us to learn from past experiences and apply those learnings to future molecules. Data generated over the course of developing and manufacturing molecules can be stored, accessed, and used to predict likely process input ranges that will impact product quality attributes. The data captured in these systems is the type that is needed to build machine learning models that can augment the traditional QbD toolkit for prospective control. These systems include:
- An automated data capture pipeline to capture, store and make the data created during development programs transparent.
- A library of control handles for common quality attributes that can be used to control product quality in the process.
- High throughput data generation tools to augment data arising from manufacturing and development programs.
Second, we are building automated methodologies to efficiently generate the additional experimental data needed to augment existing data streams. We have borrowed a proven approach commonly used in some other manufacturing industries called system identification to augment our datasets. This approach allows us to gather data in 60 days, which would usually take a year or more to collect. This approach intentionally manipulates process inputs in a prescribed manner and measures the process in real-time using the automated sampling system in our lab. Continuous measurement allows us to observe what is happening to the product in real time throughout the process. This gives Lonza access to a treasure trove of data that relates product quality attributes to process parameters.
The Future of QbD at Lonza:
We believe augmenting existing QbD retrospective control strategies with prospective control will increase the consistency of product quality while allowing for increased process flexibility that doesn’t compromise product quality. We have already demonstrated elements of prospective control in the lab. Lonza R&D is investing in the technologies needed to lower the technical barriers for prospective product quality control implementation, including systems to collect the data needed to define such control strategies and the automation and process analytical technologies to measure product quality during the process and actively control it. This approach will shorten time and risk for product development and commercialization while allowing process flexibility to support product quality over the entire product lifecycle as molecular formats continue to evolve in the future.
References:
1. Downey B, Schmitt J, Beller J, et al. A system identification approach for developing model predictive controllers of antibody quality attributes in cell culture processes. Biotechnol Prog. 2017;33(6):1647-1661. doi:10.1002/btpr.2537
Additional Information and Disclaimer
Lonza Group Ltd has its headquarters in Basel, Switzerland, and is listed on the SIX Swiss Exchange. It has a secondary listing on the Singapore Exchange Securities Trading Limited (“SGX-ST”). Lonza Group Ltd is not subject to the SGX-ST’s continuing listing requirements but remains subject to Rules 217 and 751 of the SGX-ST Listing Manual.
Certain matters discussed in these articles may constitute forward-looking statements. These statements are based on current expectations and estimates of Lonza Group Ltd, although Lonza Group Ltd can give no assurance that these expectations and estimates will be achieved. Investors are cautioned that all forward-looking statements involve risks and uncertainty and are qualified in their entirety. The actual results may differ materially in the future from the forward-looking statements included in these articles due to various factors. Furthermore, except as otherwise required by law, Lonza Group Ltd disclaims any intention or obligation to update the statements contained in these articles.