For companies developing cultured meat and other cultured ingredients, each step up in scale presents both technical and financial challenges. Predicting the performance of a cell expansion and differentiation process when climbing the process scale ladder from lab to pilot to commercial scale is an expensive and uncertain affair. At each stage, the investment in plant equipment grows exponentially, as does the volume of media required to run each development and optimisation culture.
Increasing the volume of the culture process can completely change the physics of key process behaviours such as fluidic mixing, gas transfer, damaging turbulent energy dissipation and laminar shear. The biology’s response to these hard to predict changes is at best non-linear and at worse, chaotic. All this adds up to an uncomfortable high financial risk, low-certainty situation that technology leaders and investors thoroughly dislike.
The challenge for bioprocessing engineering
This begs a key question for bioprocessing engineering: what can cultured protein ventures do to gain insight into the behaviour of their processes before placing a bet on CAPEX and development costs? The GFI has highlighted the ability to better computationally model behaviours of these processes at scale as a top priority for cellular agriculture and have funded some projects with the Cultured Meat Modelling Consortium to address the question.
We at CC also believe that sophisticated predictive models – digital twins – that couple biology, heat transfer, mass transfer and fluid dynamics can shed light on potential issues when scaling up. These biofluidic digital twins are capable of:
Rapidly gaining insights into process behaviour across a wide operational, geometric, and fluidic parameter space
Enabling ‘virtual’ sensors
Using a combination of the above to test control regimes, ultimately feeding into highly stable model-based control
Exploring process behaviours to inform investment decisions
Engineering biology is never straightforward. There is an enormous number of variables that influence the behaviour of a cultured meat process, ranging from media composition to aeration and agitation rates. Understanding the scaled process behaviour response across this vast design space far from economically viable, especially for expensive to change parameters like plant specifications.
Digital twins trained with real data from lower-cost lab experiments and thermofluidic physics can be used to explore the design space much more economically. As an example, we have trained a digital twin to give us insights into the local volumetric oxygen mass transfer rate (KLA) , damaging viscous dissipation energy, and corresponding local biomass growth rate of S. cerevisiae for a range of bioreactor operating conditions and geometries.
Below you can see a matrix of animations showing cell damaging zones around an impeller (left) and the local oxygen transfer rate (KLA, right) across a range of impeller tip speeds (top) and media viscosities (bottom).
More visibility, with no added cost
The control of a cultured protein bioprocess often depends on careful monitoring of process variables such as dissolved oxygen, glucose and lactate, via sensors. However, as processes scale up, the distribution of these variables widens in the reactor. A solution to this is to simply to increase the number of sensors, but his will have economic limits. Digital twins can unlock this trade-off via the creation of virtual sensors. Using the digital twin we developed in the previous example, we can infer the distribution of dissolved oxygen content, glucose and biomass growth rate throughout the bioreactor.
Maximising yield and identifying problems early
Cell culture bioprocesses are inherently complex and nonlinear, making it difficult to predict and control the behaviour of the system. Multivariate interactions, metabolism variability and spatial distributions all contribute to a very difficult control engineering problem.
Digital twins can improve upon typical feedback control regimes and alleviate these issues by acting like dynamic reduced order models of the system. This approach can dampen out the instabilities and slow response time associated with typical closed-loop feedback approaches, ultimately improving performance.
This approach would be possible with the digital twin we have built but would require testing and validation using a real bioprocess. An added benefit is that it enables the detection of unexpected conditions in the cell culture, for example elevated oxygen uptake rate due to contamination.
Delighting the consumer and reaching profitability
The team here at CC believes that these types of approaches can address some of the biggest challenges with delivering cultured protein products. We can see a future for industrial biotechnology where these approaches can be extrapolated to enhancing product taste and texture by also considering the expression of volatile metabolites, proteins and lipids as part of the digital twin. Doing so will accelerate cultured protein companies towards delightful consumer experiences within a viable and sustainable business model.
Ultimately the world needs to meet a significant challenge. At its heart is the commercial imperative to deliver a final product that meets not only the desired levels of quality and purity, but that also provides a company with a viable business model. Not only that, but food with bio-derived components must be put on consumers’ tables that is acceptable to their sensibilities and reflects the demand for sustainability. If you are looking for ways to revolutionise your approach to bioprocessing engineering and want to know more about how we can help, email Steve Thomas or James Westley. It’ll be great to hear from you.