Best Practices in Fermentation Bioprocess Development
We spoke with Stefan De Kok, PhD, Director of Fermentation at Zymergen, about his experience working cross-functionally across all aspects of biomanufacturing R&D.
Culture: Can you tell me more about your background in biomanufacturing R&D?
Stefan: I received most of my education at TU Delft where I did my bachelor’s, master’s, and PhD. The bachelor’s program, called Life Science & Technology, provided broad coverage from genetics and molecular biology all the way to fermentation and scale-up. The master’s program focused more on metabolic engineering, microbial physiology, and bioprocess design, and my PhD focused on yeast metabolic engineering and fermentation technology. Together, this served as a very good starting point to understand biomanufacturing as a whole.
My first industry experience was at Amyris, where we worked on developing improved methods for high-throughput DNA assembly - something completely new for me. I was also involved with their industrial farnesene production project and picked up a lot of real-world experience from it. I then joined Zymergen about five years ago to lead their DARPA Living Foundries: 1000 Molecules project. After about 2.5 years in that project, I switched to the fermentation team at Zymergen.
Culture: Could you tell me more about the DARPA project?
Stefan: DARPA, the Defense Advanced Research Projects Agency at the Department of Defense, invested in a research program dedicated to developing capabilities for biomanufacturing molecules and materials with unique performance that could not be made through conventional means. Hyaline, the high-performance film for electronics applications that Zymergen just launched, is an example of such a new material. Zymergen participated in the program, and the biology team worked on designing, building, testing, and analyzing new strains to produce these molecules as well as building out the infrastructure underlying the work.
Culture: What did that infrastructure build-out entail?
Stefan: Zymergen’s core business model up until that point had been focused on increasing the yield or productivity of existing strains used in industrial fermentation processes. These projects are on the upper part of the “strain performance S-curve” (pictured below). The DARPA project was our first foray into the bottom part of the S-curve: designing metabolic pathways, selecting heterologous enzymes, assembling and sequencing larger DNA constructs with multiple genes, inserting those into multiple hosts, verifying correct genomic insertion, testing molecule production in high-throughput microtiter plate assays, developing baseline fermentation processes, and scaling some of them up to pilot scale.
Culture: Can you walk me through this “strain performance S-curve?”
Stefan: It describes a curve with time or effort on the horizontal axis and strain performance (measured in titer, rate, yield, etc.) on the vertical axis, with three phases. The goal of the first phase is to get from “zero to milligrams,” to go from an idea to proof-of-concept molecule production. This is when you are designing and inserting metabolic pathways and developing initial methods for measuring production, and it takes some time to achieve that proof of concept. Then, once you have achieved initial production, you enter the second phase of “milligrams to kilograms,” where you are – among others – optimizing flux in the metabolic pathways, reducing byproducts, and mitigating toxicity. As there is usually a lot of low-hanging fruit here, rapid progress is often made in this phase.
After getting closer to theoretical perfection, you enter the third phase of “kilograms to commodity.” At this point progress typically slows down, since it takes more and more nuanced changes to make smaller and smaller improvements. With Zymergen’s high-throughput Design-Build-Test-Analyze-Learn platform and advanced data analysis, we try to steepen the initial parts of the curve to get closer to theoretical perfection sooner.
Culture: How do you assess the theoretical perfection of a bioprocess? Maybe I should back up even more: how do you start development for biomanufacturing a new molecule?
Stefan: It all starts with a business case and a techno-economic analysis. Before going into the lab and doing any experiments, it is important to make sure that there is a market for the molecule and to have an initial understanding of production economics and strain and process performance targets. At Zymergen, we routinely perform techno-economic analyses that leverage models of cellular metabolism, fermentation process parameters, and expected downstream recovery and purification steps. This analysis informs whether a product could ever be commercially viable: whether it can be made for less than its anticipated value. It can also help inform several fermentation parameters, for instance: what carbon source to use, how much aeration production will require, whether it is beneficial to run at low or neutral pH, and what type of microbe to use. The model also provides a framework for evaluating strain and process performance, which enables setting milestones along the R&D process. Early techno-economic modeling requires making many assumptions, but as strain and process development moves forward one can validate or revise those assumptions to refine the model further.
Culture: Can you walk me through the activities done in each of these phases? Let’s start with phase I, assuming you’ve built your techno-economic model and know the titer, productivity, and yield you need to hit to make a commercially viable product.
Stefan: Let me preface by emphatically stressing the importance of incorporating fermentation expertise from the very beginning. Beyond needing this knowledge to even build a viable techno-economic model in the first place, fermentation know-how is equally important to inform strain engineering approaches from the outset. For instance, understanding whether it is feasible to use antibiotics and (expensive) inducers in a large-scale fermenter can help inform – early on – what expression system to use and whether to use plasmids or integrate genes into the genome instead. ‘Keeping the end in mind’ from the get-go can thereby avoid rework along the way and accelerate overall timelines.
Beyond these specific examples, it is important to embrace a more dynamic view of the interplay between process development and strain screening. Too often they are seen as linear or sequential, where process development only begins once a certain titer milestone is reached through strain engineering. Developing a biomanufacturing process is iterative and non-sequential; an optimal, scalable process requires the best strain in the best process, where the two are interdependent and therefore need to be developed in tandem.
Culture: That makes sense to me when you’re in “phase II,” but why would fermentation process considerations be so critical when you’re building and testing strains initially?
Stefan: Having a vision of the ultimate fermentation process helps to develop a high-throughput strain screen that picks up the right strains. Mimicking fermentation conditions in strain screening protocols in 96-well plates will ensure that you select strains that perform well in fermentation conditions, since “you get what you screen for.” Neglecting to keep the fermentation process in mind when developing a high-throughput strain screen might result in promoting strains that won’t perform well in bioreactors and missing those that might have really been the best.
Culture: What do you mean by a vision of the fermentation process? How would you know that before you’ve even done any work on a strain?
Stefan: This is why having someone with fermentation expertise on the team early on is beneficial, because they will know from experience some approximation of how the fermentation process will look like. From an understanding of the organism and product in question, as well as the optimal conditions from the techno-economic model, one can predict at least a ballpark range of where the process parameters will ultimately land. For instance, if the envisioned fermentation process is an aerobic, fed-batch process at pH 3 and 30 ℃ that uses particular medium components, one should be trying to mimic that in microtiter plate assays from the start.
It can also be useful to do some light-touch work in bioreactors with a wild-type strain to understand its growth characteristics such as pH optimum, growth rate, biomass yield, nutrient requirements, and metabolic response to excess substrate or oxygen limitation. Based on that information, you can then start to develop a baseline feeding scheme and nutrient composition for your media, and couple that information back to the 96-well plate screen workflow development. When working with E. coli or S. cerevisiae where much of this is already known, this exercise can be very straightforward. However, when working with a more exotic microbe, then understanding metabolic and physiological parameters early on is a critical activity.
Finally, product toxicity should also be assessed very early on - even before you begin engineering a strain. How do wild-type versions of candidate host organisms tolerate high concentrations of the product? This can be assessed in plates or flasks and then confirmed in bioreactors. At Zymergen, we often evaluate multiple hosts for product tolerance in order to select a good host.
All of this work is part of “phase 0:” ensuring selection of a host that is tolerant to the product, that the host can be grown in the envisioned fermentation process in a bioreactor, and that those conditions can be reflected in a microtiter plate assay. This work can avoid going down the wrong path and having to do a lot of re-work down the road by switching host and process conditions.
Culture: Once those first strains are made, do you test them with microtiter plates or in bioreactors?
Stefan: At Zymergen, we usually build more strains then we can ever test in bioreactors, even with all of Culture’s reactors at our disposal. Accordingly, we always test our strains first in microtiter plates and then promote the most promising ones to bioreactors. Typically, more than 90% of our strains will never make it to bioreactors so we need to make sure that our 96-well plates are predictive of bioreactor performance.
At Zymergen, we focus heavily on developing and optimizing our plate workflows, and this is not an easy task, as plates are not bioreactors, and plates will never and can never be bioreactors. For instance, plates are limited in the oxygen transfer rate, by an inability to control the pH, and by not having a controlled substrate or nutrient feeding. On the other hand, the plate screen also doesn’t have to be perfectly predictable; it “just” needs to be a good enough model to pick up improved strains. It is therefore critical (especially when the fermentation process is being refined) to continuously monitor the plate-to-bioreactor correlation and improve the plate screen if its predictive power is drifting.
Culture: When would you begin additional process development work in bioreactors?
Stefan: As mentioned before, we often already do some early process development work using a wild-type strain to get the basic physiological parameters down. Then, once we generate a strain with some decent production level, we can start optimizing the fermentation process by changing the feed regimen, nutrient availability, pH, temperature, etc. Starting process development work is not very useful when you’re only producing micrograms or milligrams per liter, as the pathway flux is still very limiting at that point. But you also don’t want to start process development too late in case strain engineering improves strain performance rapidly.
These initial fermentation experiments often start with univariate testing of process parameters to identify the most sensitive parameters and appropriate ranges. Then you would conduct multivariate Design-of-experiment (DoE) studies to find optimal combinations of those parameters. The results of these bioreactor experiments should then be fed back into the strain engineering work: if there is an intermediate or a byproduct identified that accumulates during fermentation runs, the strain engineers can then work to optimize the pathway accordingly. As such, the fermentation data from bioreactors helps prioritize the gene edits to make.
Culture: How many different bioreactor runs are typically done as part of this initial process development work?
Stefan: It depends on how well characterized the microbe and metabolic pathways in question are, but it usually involves testing some 10s of conditions and using 2-3 replicates for each condition with a few different strains, which would add up to maybe 100 bioreactor runs in total.
Culture: Do you always recommend running replicates? How do you know how many replicates to run?
Stefan: It is certainly best practice at Zymergen. How many replicates to run depends on three factors: the level of noise observed in your process, the difference that you need to detect, and the confidence level required. At the beginning, you are probably looking to double titer so a higher noise level can be tolerable and fewer replicates can be sufficient - either duplicate or triplicate. On the “upper part” of the S-curve when looking to squeeze out for instance 5-10% in strain performance, a low noise level with – sometimes - a higher number of replicates is typically required. When Zymergen ships an improved strain to a client, we want to be very confident that the strain is truly improved over the previous strain and so we run quite a few replicates of both strains across different weeks to confirm performance.
Culture: Okay, so to recap, we’re now at the point where we are using a microtiter plate model for strain screening and then confirming the best strains’ performance by running them in a baseline bioreactor process. The bioreactor data not only informs refinements to the microtiter plate model, but also helps to prioritize gene edits. When would you start extensive process development and optimization?
Stefan: It depends on the project milestones and factoring in end-goals in terms of timing and budget. Initially, you may only need to scale to a pilot tank to generate some material for regulatory or customer testing or to give the downstream team more broth for their work. If that’s the case, knowing how much material is needed informs how much process development work is required to hit those titers. When working towards scaling to commercial production, much more process development work is performed to improve the production economics and understand how the process responds to disruptions that can happen at larger scale.
Culture: How many bioreactor runs are we talking about in each of those stages?
Stefan: It depends on the strain and product, but we’re easily talking about hundreds of runs on the “lower part” of the S-curve to get to the point to produce kilograms of product. When shifting to the “upper part” of the curve and looking to approach the theoretical maximum, it may require hundreds to thousands of bioreactor-runs for process development as well as validating the performance of strains picked up by the high-throughput screen. At that point, it is appropriate to start doing larger DoEs, optimizing process conditions such as feed profile, media composition, oxygen availability, and pH to reduce production cost and ensure the process performs reproducibly.
Culture: What is involved in the work of scaling-up a process? I often hear that this is very difficult work that is prone to failure, and the way to mitigate risk is to incorporate scale-up considerations early in process development.
Stefan: Scale-up is not trivial because industrial fermenters differ from bench-scale reactors in several key ways. However, keeping those differences in mind can help to avoid several scale-up challenges.
First, the oxygen transfer rate (OTR) achievable in large reactors is typically lower than in bench-scale reactors. Therefore, when developing a bench-scale fermentation process, it is important to cap substrate feed rates and oxygen uptake rates (OUR) to industrially relevant ranges to ensure process scalability.
Second, large reactors are more heterogeneous than bench-scale reactors. There are pressure and gas gradients and mixing times are longer. At bench-scale, you can mimic some of these factors and study how the organisms respond to those fluctuations. Ideally, an organism is chosen or developed that is not sensitive to fluctuations and shows a similar performance in heterogeneous environments.
Third, the seed train to inoculate a large fermenter usually consists of several additional steps beyond the bench-scale process, resulting in additional doublings from glycerol stock to production fermenter. This can lead to significant differences in performance between bench and large-scale fermenters, especially if the strain has some genetic instability. An unstable strain might perform very well at bench-scale after ~30 generations, but not perform as well in a large reactor after ~50 generations due to the longer seed train. To compare apples to apples, you should mimic the large-scale seed train at bench-scale, for instance by having multiple seed steps before inoculating a bench-scale reactor. This helps to dissect whether performance differences at different scales are due to true scale effects or simply the effect of number of generations.
Fourth, make sure to use scalable ingredients. At bench-scale it is usually affordable to use high-quality ingredients and expensive antibiotics or inducers. At a larger scale, however, it can be challenging to source enough of those ingredients and they might not be affordable. Again, to compare apples to apples, it is best to use the same ingredients in bench-scale runs as in your scale-up system.
Keeping the end in mind helps to develop a scalable fermentation process from the beginning. As large-scale bioreactor runs are expensive, bench-scale fermentations should be used to de-risk at least the main factors before going to larger bioreactors. In the early stages of a project, the exact specifications of the production reactor may not be fully understood yet, but an experienced fermentation scale-up engineer can give ranges and ensure the bench-scale process is in the right ballpark.
Lastly, it is important to collaborate early and frequently with the down-stream processing (DSP) team. Producing a molecule is – in a sense – just the beginning of a biomanufacturing process; purifying the molecule out of the broth and formulating it into a product are just as important. A close collaboration between strain engineering, fermentation, and purification can help identify technical opportunities to make each other’s job easier!
Culture: This has been really helpful to get an end-to-end view of how bioprocess development works, particularly in understanding the relationship between strain screening and process development. My final question - with all the bioreactors you have in-house at Zymergen, how and when does working with Culture fit into your development projects?
Stefan: Zymergen has a certain bandwidth in terms of bioreactors, but the demand for experiments fluctuates over time. Culture has been a great resource for when we have a surge in demand and can free-up space internally by offloading some of the bioreactor work.
Outsourcing work always involves some uncertainty and risk, but Culture overcame that quickly by demonstrating their capabilities, both in terms of their fully-equipped bioreactors (with scales on all feed bottles, continuous off-gas measurements, and a web-interface that makes it easy to collaborate) as well as their experienced, professional, and readily available team.