Over 2200 cell therapies are in clinical trials at time of writing . It seems prudent to assume that some will succeed, and not unduly optimistic to hope that some will move from experimental third line treatments for specific cancers to more common second line treatment. Some may even emerge as treatments for currently incurable and common autoimmune diseases such as type 1 diabetes.
When this happens, cell therapy manufacturing processes that currently benefit 1000s of patients per annum could need to be deployed for 100,000s of patients. If so, any weaknesses in the cell therapy manufacturing equipment used by the industry will be painfully exposed – and challenge the currently preferred scale-out approach of the industry. At scale, even 1% failure rates are unacceptably costly in terms of time and, especially in autologous cell therapy products, lives.
What this crystal ball gazing shows is that while the industry is remarkably successful at the boutique scale, where human operators are integral to even the most automated of solutions, the reliability of current methods of cell therapy manufacture will need to increase, or be de-risked – or both – to upwards of 99.9% reliability. No implementation will be perfect, but we can come close. What approaches can be used to provably achieve this?
Improving reliability is a hugely multivariate problem with no simple answer. Reliability can be increased by exercises in improving quality - ensuring the process or product is what you expect – but this does not ensure complete reliability.
This is because reliability accounts for any failure, not just manufacturing defects. Human error during product use, chemical exposure or a software crash can all cause failures. We therefore recommend an interdisciplinary, exhaustive approach that considers improving reliability from concept to use. This includes reliability by design, testing, manufacture, and online process control.
Improving reliability through design
Assuring reliability through design is best illustrated by way of an example. Reliability by design requires a thorough understanding of the risks present in the manufacture, product, and process, then mitigating these risks as much as possible. An example of all of these can be found in the consumables that C> therapy manufacturing requires for automation.
These complex consumables have grown out of necessity, joining relatively standard, standalone components such as pinch valves or fluid bags into an automatable network of inputs and outputs with a web of medical-grade tubing.
We have previously talked about how introducing an increased number of parts can significantly increase failure rate. Every additional weld to form the consumable adds a new failure mode, which could strike at any point from manufacture to use, and a single failure is enough to render the consumable useless. If patient material is lost due to failure, the patient would lose their potentially life-saving therapy.
An interesting approach to this problem would be to use multi-functional parts. These can be designed to provide the same fluidic routing as the complex consumable with both significantly reduced part count and a less error-prone manufacturing method. Even better, less moving parts and easier installation into the therapy processing equipment means reliability is increased in multiple areas.
At this point, the consumable reliability is likely to be affected by the supply chain and unit operation performance more than induced failure modes. As such, we will require the integration of biology with engineering, requiring clever design to keep the process closed while automated with non-invasive, online process analytics to enable the control required.
Assuring reliability through testing
Any changes to the workflow, product or process must somehow be proven to be improvements – not all plans survive first contact with reality, after all. Here, one might investigate the ease of use, the effects of degradation, or use exhaustive testing to identify other one-in-a-million gremlins that will otherwise beset therapy manufacture. This is alongside integration testing to ensure biological variability is controlled as much as possible.
There is, however, no way to test every single possible change exhaustively. It may be that some failure modes are one-in-a-million, while others are one-in-a-thousand. A predictive method that can indicate the most likely failure modes can guide these tests and design iterations to mitigate or eliminate these failures. This is a more efficient use of time and money.
For example, testing the distribution of a tolerance stack of a valve assembly can indicate if there is a statistical likelihood that some valves may not be the right size, causing the whole therapy manufacture process to fail. We know what that means for the patient.
In silico testing is making rapid progress and may also be appropriate. A computational study of the dynamics of fluid transfer between compartments, and the resultant effects on cell health, could give confidence in the design of the consumable. With in silico testing, problems can be flagged and solutions identified faster than if physical parts need to be created and procured for every design iteration. This does not eliminate the need for physical prototypes or statistical testing, but instead supplements the design process. Eventually, a digital twin of the whole therapy manufacture may be possible, but this requires significant data collection, and therefore time.
Quality control and Industry 4.0
Another crucial step in assuring the reliability of cell therapy manufacturing equipment, especially when products are on the open market, would be some form of quality control. Traditionally, random sampling of manufactured parts at various times and stages of the process may have been used, but consumables may require terminal QC, just prior to use, to flag issues that would otherwise prevent safe usage. This is often the case for the consumables mentioned earlier, where a leak detection step will determine if the consumable has failed prior to the precious patient-harvested cells entering the system.
Capabilities are, however, growing in the process monitoring space – described as “Industry 4.0”. Through the clever use of sensors and machine learning, manufacturing processes can now have a real-time stream of data regarding their operating conditions and product. This has the potential to eliminate the need for random sampling QC as each automated process is performing 100% QC. This method has already entered traditional manufacturing systems and is slowly being adopted by pharmaceutical manufacture . This data collection during manufacture can easily identify any remaining defects in the manufacturing process not identified by earlier work, allowing for their mitigation – a tenet of Six Sigma processes for continuous quality improvement. It should be noted that biological systems do add another layer of complexity to the system, but this is not an insurmountable challenge.
The right direction
With time and effort, these approaches can be used to prove a product or process has upwards of 99.9% reliability. It does, however, require real products and therapies to be made using methods that have potentially been moved from lab to line with little thought for scale. It may not even remove the most error-prone and costly part of the system – the human operator.
Human operators were absolutely key to getting cell and gene therapy manufacturing to the point it is at today, and there will always be humans in the manufacturing loop. However, if the therapies are to become affordable, a new scale of near-autonomous therapy manufacture must be reached. This has been the path to scale for manufacturing in a range of industries. It also both supplements and advances the work on increasing reliability the industry needs. If this is the future for cell and gene therapies, maybe a provably reliable automated factory may be the next crucial step in bringing these therapies to the masses.
Elliot Morley is a Year in Industry student at TTP. He studies Mechanical Engineering with Biomechanics at the University of Sheffield.