Harvey Branton, senior translational consultant at cell and gene therapy CDMO eXmoor Pharma, shares invaluable insights into early process development, exploring common misconceptions that could impact long-term manufacturing success for ATMP developers.

eXmoor
Recent scientific advances are driving an increase in the number of advanced therapeutic medical products (ATMPs) coming to market, and keeping costs under control is essential to support the democratisation of these life-changing therapies. Traditionally, biomanufacturing relies on process optimisation and scale-up to drive down costs. Unfortunately, these approaches do not translate to an autologous therapy where each batch is personalised and made from a patient’s own cells. Understanding how to adapt to the challenges of manufacturing thousands of batches a year is critical to support future commercialisation.
Processes designed to facilitate parallel manufacture will maximise batch throughput and support efficient utilisation of manufacturing resources, helping to reduce individual batch costs. Robust processes will always be the cornerstone of any bioprocess, ensuring consistent performance. But some in the autologous cell therapy space focus on yield as a key cost driver, which may have less impact on cost than overall process design. Developers often assume that it will be straightforward to increase the number of batches manufactured, but issues arise if they fail to think through the significant impact of early process choices on operational efficiency, especially where parallel processing is required.
Early consideration of the commercial manufacturing strategy will ensure that initial process decisions do not detrimentally impact the future commercialisation potential of a life-saving therapy. Two key areas need to be considered: how the process can be designed to facilitate efficient manufacturing, and how innovative technology such as machine learning (ML)/AI and automation can be leveraged to reduce labour demands and simplify the manufacturing approach.
Process design
Driven by the considerable number of batches required for a commercial product, processes must be designed to enable parallel manufacture, helping to dilute manufacturing costs through efficient batch scheduling. It must be noted that facilities running constantly have three times the capacity of a facility used for 8 hours a day. Night shifts are often used for monitoring, preparation, and set up activities, but this approach underutilises potential manufacturing capacity – especially where the economics of the entire manufacturing operation are dominated by amortising the capital costs of the facility and other overheads.
A detailed manufacturing schedule can improve throughput, ideally identical every day. This can reduce the need for complex handover between shifts and limit the risk of errors, but it may not be possible where starting materials (such as fresh apheresis) are not steadily available. Crucially, batch timings must be consistent, avoiding the need for rescheduling. This ensures shift staff know exactly what needs to be delivered each day and patients treatments are always available against agreed timelines.
Processes should be designed to complement the shift structure, maximising the value of the additional capacity. These are built around different unit operations, designed to be completed within a working shift, negating the requirement to hand over partly completed operations to the next shift. Optimal scheduling will always be unique to every product. But general approaches – such as starting batches during the day and finishing batches overnight, so they are ready for a review next morning – can help manufacturing approaches to feel “similar.”
Automation, ML/AI and data
The move to manufacturing thousands of batches of the same products will generate large data sets, opening the opportunity to use ML to support more dynamic approaches to monitoring and control. Cell therapy processes often run for 10 to 15 days, with intense activity required on the first and the last day of the process, but with most of the time used to bulk cells post-selection and reprogramming.
The high number of identical batches to be manufactured will drive the move towards routine automation. Companies are offering off-the-shelf, scalable cGMP solutions that are helping to make automation easily achievable and cost-effective. Invariably, automation comes with compromises, as the most efficient automation strategy may not easily align with the more manual approaches used during initial stages of process development. Automation strategy and partnerships should be considered early in the design process to avoid the costly impact of needing to later adapt the process design to suit automated manufacturing approaches.
AI models could be used to automatically monitor batches during cultivation and make decisions around how to adapt the control strategy to achieve defined outcomes. Model-based control can predict future process performance based on historical data, then make informed decisions about how to best adapt process control to maximise performance. It is difficult to accelerate a cell-based process but in principle, processes could be run under suboptimal conditions to slow them down and then speed them up if required.
This approach would enable process timings to be fixed and support efficient batch scheduling. Such approaches are already being used in the health care sector to monitor patient status, using a platform that combines sensors and AI-based algorithms to help inform clinical decisions.
Another area where AI is delivering game-changing advances is around image analysis. Sophisticated models can evaluate images and rapidly provide accurate insights into cell growth and morphology, which can be used to support advanced control strategies.
The data generated – and the process insights hidden within – have significant added value. Defining who owns the data will be important, especially where the process, automation and control strategies are all supplied by different parties.
The vast amounts of data, and the computing power required to update models and make them available across multiple manufacturing sites, will increase the use of cloud-based solutions. Providers will need to accelerate the requirement to clearly define the data ownership rights. The use of trust law frameworks to create data trusts will help with complex governance, conditional consent, and stewardship, helping to clarify IP positions and open new value streams. Companies need to be aware of these challenges, but it is likely that off-the-shelf solutions will be available soon.
Conclusion
Successful commercialisation of a cell therapy is not straightforward, and as the number of therapies rises, the pressure on cost will increase. Manufacturing technology is advancing fast but adopting innovative technology early can be risky. Early process choices have a significant impact on the long-term cost of manufacture and therefore need to be carefully considered. A diverse range of skills sets covering quality, technical, engineering, automation computing and legal will be needed to make projects successful.
It is worth highlighting that cost-of-goods modelling is an invaluable guide which can be used to support process planning and decision-making by evaluating the impact of early choices. Running different scenarios in a model is a cost-effective approach to quantify the impact of different choices, providing a data-driven route to support whole-process design and decision-making and help pull together the multi-disciplinary teams required to deliver a successful project.