High-throughput strain engineering platforms facilitate the introduction of a plethora of genome modifications targeted at strain performance improvements in a wide variety of host organisms. Zymergen works with clients to improve their specific production strains in processes to which they are already wedded. In these cases, we help mature production processes that are tuned to the client’s previous best strain; this, in turn, results in our dealing with fermentation of a wide strain diversity as we progress. We can also select microbial production hosts and develop scalable production processes in conjunction with strain optimization efforts to produce a variety of different products and materials in our internal projects. This involves developing fermentation processes from the ground up; we also need to work with a large strain diversity in these cases, though at a different stage of the fermentation characterization process. Thus, our ability to flexibly work with many different organisms at different project stages is essential to efficiently test strains for a multitude of different processes and products.
Once we have obtained our initial parent production strain, either from a client process or an internal product project, we apply both focused and genome-wide strain engineering strategies geared towards strain performance improvements to generate large quantities of engineered strains. Depending on the type of genome modification performed, the strains that emerge from our strain engineering pipeline may phenotypically differ significantly in many ways from its parent strain, including in growth rate, feed demands, and oxygen requirements. To accurately assess the optimal performance of strains from these diverse lineages and their respective optimal cultivation conditions, it is beneficial to develop fermentation protocols that can adapt to individual strain physiology.
With the large amount of strain diversity generated in Zymergen’s high-throughput build pipeline, it is important that we design fermentation processes with that in mind. When developing high-performing, bench-scale screening processes, we have two critical objectives: flexibility and robustness (defined as high process reproducibility and low statistical variability). For the former, the process must be flexible enough to accommodate wide ranges of phenotypic diversity and be able to react accordingly such that near-optimal strain performance is achieved. If a process is too rigid, in addition to harming our evaluation of a strain’s performance, an increase in variability may also be observed. Once we achieve a baseline process, we can apply fermentation characterization data to gain a deeper understanding of technical and biological parameters that are driving variability. By identifying these parameters, we are able to develop mitigation strategies to lower process variability. These, in turn, will directly impact our ability to identify improved strains while enabling the use of fewer fermentation runs to do so.
Many industrial fermentations are performed as fed-batch processes, in which a substrate-limited environment is maintained throughout the production phase to reduce overflow metabolism, drive metabolic flux towards product formation and reduce biomass accumulation. Typically, fed-batch fermentations start with a “main batch” phase, in which the cells metabolize the nutrients and carbon source provided in an initial growth medium, mainly to accumulate biomass.
Typically, microbial cells undergo exponential growth when the carbon source is in excess and no limitations are exhibited (such as oxygen or other nutrients required for growth). This requirement is usually met during a batch phase in an aerobic fermentation process. Under these conditions, cells ideally grow at their maximal growth rate and double at their respective minimal doubling time. Once this “batch” phase in fed-batch processes is completed, the carbon source is delivered via an external feed to maintain cell viability and production. However, growth rates may differ significantly from strain to strain, leading to different time points of substrate exhaustion. As such, a fixed time initiation of feeds may result in starvation for fast-growing strains or overfeeding for slow-growing strains. Exhaustion of the carbon source in the initial growth medium is typically accompanied by various physiological signals, such as a rise in dissolved oxygen (DO) and potentially a pH spike (Fig. 3).
With an increase in actively growing cells, the oxygen demand of the culture increases which, in turn, results in a decrease in the dissolved oxygen present in the culture broth, assuming that the oxygen transfer into the culture is relatively constant. Once the carbon source in the initial growth media is depleted, the lack of substrate to be metabolized causes a decrease in the culture’s oxygen demand, resulting in a sudden spike in dissolved oxygen. Additionally, many organisms undergo overflow metabolism at their maximal growth rate, resulting in the formation of short-chain organic acids that lower the pH of the culture. Once the cells have consumed the initial carbon supplied in the batch media, they might switch their metabolism to allow consumption of organic acids, both produced during overflow metabolism or supplied in the media, which in turn results in an increase in pH. Thus, both DO and pH spikes indicate the full consumption of the main carbon source and can be exploited as triggers to initiate feeds.
The development of dynamic feeding protocols can present a means to account for differences in strain physiologies and help reveal conditions necessary for optimal strain performance by ensuring optimal carbon availability. Dynamic feeding protocols can include different aspects of the fed-batch process. On the one hand, they may include the automated start of the feed phase triggered by a biological signal that indicates that the cells have fully consumed the carbon source provided in the initial growth medium. On the other hand, these protocols allow dynamic control of the feed rate to account for different carbon and nutritional requirements during the fed-batch phase. The optimal rate at which the external feed is delivered to the culture is strain- and process-dependent. Deviation from the optimal feed rate to either a rate that is too low (underfeeding) or too high (overfeeding) can result in suboptimal strain performance that is not reflective of the true strain performance under optimal conditions.
An example of the potentially detrimental effect of screening strains under suboptimal feeding strategies is shown in Fig. 4. We ran several strains in a fixed feed process that was optimized for a production strain. The detection of residual glucose at the end of the run, as well as significantly increased byproduct formation for strains A, B, and C compared to the production strain, indicate that the selected feed rate was too high to reveal the true potential of these strains. Specifically, the production of byproducts can reduce yield and productivity, as certain byproducts can have an inhibitory effect on the critical genetic pathway or on growth. In contrast, underfeeding can reduce productivity and yield due to the relatively fixed carbon demands of biomass production and maintenance. In addition, suboptimal feeding strategies can result in increased variability in the performance of a single strain, as the feed rate cannot adjust based on subtle differences between replicates.
We can use the same physiological cues that can be exploited to initiate the fed-batch phase during the feeding phase to determine if the selected feed rate is suitable for strains that exhibit different growth or metabolic rates. There are various methodologies of feeding based on these biological signals, in which the pH or DO is controlled at a setpoint by adjusting the feed supply (pH stat or DO stat); a biological starvation signal, for example, a spike in DO or pH, may trigger a short feed pulse. Once the substrate supplied during this feed bolus is depleted, another biological starvation signal will trigger the next feed pulse. Other feeding strategies may be more complex. When implemented correctly, these dynamic feeding protocols can adapt to varying strain physiologies over the entire duration of the fermentation and thus consistently assess the optimal performance of each strain.
Figure 5 shows an example of a dynamic feeding strategy that we implemented for an E. coli fermentation process, adapted from a strategy published by Akesson et al. [24]. To maximize product yield and productivity, the substrate feed rate must be optimized throughout the fermentation process to minimize byproduct production, mainly acetate in the case of E. coli. Acetate formation is a result of overflow metabolism and occurs as a result of overfeeding in aerobic cultures. A characteristic feature of overflow metabolism in E. coli is that the specific oxygen uptake rate reaches a maximum just prior to switching to overflow metabolism [24].
We used our knowledge of E. coli physiology to implement a probing strategy that adapts feed rates based on substrate demand throughout the fermentation (Fig. 5). The probing method tests the system periodically as it seeks to maintain the substrate feed rate, and thus the specific substrate uptake rate, just below the maximum oxygen uptake rate to prevent overflow metabolism. By periodically adjusting the feed rate up or down, and analysing the response of the dissolved oxygen concentration in the culture, we can determine if we need to decrease, increase, or maintain the current feed rate to reach an optimal substrate supply. We optimized several aspects of this strategy, including cycle length, pulse size, pulse duration, and magnitude of the feed rate change.
It should be noted that dynamic feeding protocols should take into account any given constraints of the process at scale, such as maximum feasible feed rates and oxygen transfer rates, to avoid screening strains in conditions that are not reflective of the production process at scale. Ultimately, we strive for optimal performance within the fundamental boundaries of the commercial-scale production environment to ensure that strain performance is preserved through scale-up. Once an improved strain has been identified using a dynamic feed protocol, it may be necessary to translate back to a fixed feed protocol that can be used at production scale and validate strain performance in a fixed feeding regimen.
In conclusion, employing dynamic feed initiation triggers and feeding protocols enables us to reveal the optimal performance of each strain and reduce variability. We translate the dynamic feeding profiles into fixed processes that can be run at a larger scale. This strategy allows us to efficiently assess optimal strain performance using dynamic screening processes and reliably scale our processes to production scale.