ENGINEERING
Introduction
To achieve stability in our engineered biological systems, we relied on Design–Build–Test–Learn (DBTL) cycles as the foundation of our workflow. By using this iterative approach, we gradually redefined our ideas and moved towards effective solutions. Dry lab and wet lab efforts complement each other, creating strong DBTL cycles that combine modelling-based predictions with experimental validation. This approach ensured that our workflows were efficient, data-driven, integrated, and grounded in rational approaches. Although our project spans several modules for the development of our process and product, here we highlight two of the particularly strong DBTL cycles used for the production platform:
Consortium engineering – We aimed to investigate and implement the initial and operational conditions to support a robust co-culture of Komagataeibacter sucrofermentans and Saccharomyces cerevisiae. This subproject went through two DBTL iterations, combining modelling and experimental efforts to understand and optimise the balance between the two species.
Ethanol homeostasis circuit – We aimed to design a synthetic genetic circuit that stabilises the ethanol concentration in the culture medium to 1% (v/v), thereby maintaining optimal conditions for bacterial cellulose (BC) production in K. sucrofermentans1. This subproject progressed through two iterations of the DBTL cycle, each bringing us closer to a functional circuit.
Consortium engineering
Our project followed a structured DBTL approach to construct a stable yeast–bacteria co-culture for enhanced BC production, inducibility of the system, and functionalisation of the material. In two DBTL cycles, we simultaneously developed the model and experimental co-culture.
Characterising the monocultures and parameterising the model:
Our first iteration focused on establishing the baseline characteristics of potential toxicities, metabolic preferences, and kinetic parameters. These were also simulated by feeding the data acquired in the lab into a mathematical model for both strains individually, before introducing the added complexity of the co-culture interactions.
Constructing the co-culture:
Using insights from cycle 1, we extended the model to predict co-culture dynamics and guided the design of experimental inoculation strategies. We then compared the results observed in the lab to model simulations to test our hypotheses.
Cycle 1: Characterising the monocultures and parameterising the model
We built an ordinary differential equations-based model (ODE-based model) inspired by a model for the growth of Escherichia coli by Millard et al.2, to simulate the metabolic pathways of S. cerevisiae and K. sucrofermentans within a consortium. The Millard models were modified by updating the parameter values within a biologically relevant range.
The choice to design a consortium was inspired by Gilbert et al., 20213, who established a synthetic community (Syncom) of S. cerevisiae and K. rhaeticus as a way of incorporating programmable functionalities to BC using the toolbox of yeast. From our human practices (Dr. Amritpal Singh) , we determined that we could achieve higher BC yields in a consortium. K. sucrofermentans was shown to achieve higher yields of BC when 1% (v/v) ethanol is provided, which can be produced by S. cerevisiae1. Gilbert et al., 20213 used a method of codependence to design a stable consortium. This method was not applicable to our co-culture, so we designed a cross-feeding mechanism to impose codependence. The overview of the cross-feeding mechanism is described in (Figure 1). For the ethanol not to be self-consumed, a mitochondrial knockout was performed in the yeast.
We prepared the respiration deficient S. cerevisiae IMX1812 (rho- S. cerevisiae IMX1812) through three rounds of ethidium bromide (EtBr) plating (Figure 2).
We set up the monoculture growth kinetics experiments for both strains. K. sucrofermentans and rho- S. cerevisiae were grown in 96-well plates with three replicates in concentration series of ethanol, gluconate, acetate, and a combination of gluconate and acetate. These experiments were used to find the growth parameter values that are used in the model under different circumstances. In addition, the change in the growth of the respiration deficient strain compared to the wild-type (WT) was observed.
The colonies from the third round of EtBr plating were tested through replica plating on maltose and glycerol with two replicates (Figure 3).
We deemed the mitochondrial knockout experiment to be successful, as little to no colonies grew on glycerol.
We fit the OD vs. time data obtained from the growth experiments to a Richards growth function using a Python script. We integrated Michaelis-Menten parameters (\mu, K{_m}, V{_{max}}) into the model, and simulated the behaviour of each strain.
We fit the model to the monoculture growth curve data (Figure 4). After a successful fit, we identified the optimal values of the various metabolic parameters. Based on monoculture experiments on varying media, we also obtained parameters involved in the cross-feeding mechanism.
Cycle 2: Constructing the co-culture
Based on the simulations carried out for the co-culture, we built the model for the consortium. This model incorporated the cross-feeding mechanism, and additional fine-tuning was performed to account for an acetate sink. We did this because unrealistically high acetate production was observed within the simulation. Simultaneously, we designed a model that does not include this acetate sink, with the aim of comparing simulation results to experimental data.
Using the predictions made by the co-culture model, we aimed to investigate the effect of the initial balance of the microbes (inoculation concentrations) on the final cellulose yield of the co-culture.
We inoculated pre-cultures of K. sucrofermentans and rho- S. cerevisiae 48 hours and around 30 hours prior to the pre-cultures, and until the OD reached 2.5 and 0.01 respectively. The media required for the monoculture controls and co-cultures, namely, YPD (glucose), YPM (maltose), and YPD+M (glucose and maltose) were prepared. The bacterium was diluted 1/50 while the yeast was diluted 1/100 in the media.
After four days, we analysed co-culture results. Sugar consumption and the concentrations of cross-feeding metabolites were measured with HPLC. To do this, duplicate samples were centrifuged at 6000 g for 12 minutes. The supernatant was frozen and later used for HPLC. In addition, a small amount of the co-culture was plated on YPD+M to assess if both strains had survived.
We confirmed that both strains survived by comparing the sample plated from the co-culture to the monoculture plates of the strains (Figure 5). The phenotypically distinct characteristics in the (potential) co-culture plate matched the observed characteristics from the monoculture plates, so the organisms both survived.
We observed differing results between the model simulation and the laboratory experiments, mainly in the levels of acetate production. The model simulates an overproduction of acetate, which was not observed in the lab, thus to modify the model we incorporated an acetate sink. The model’s simulation with the acetate sink qualitatively matched the results seen in the laboratory.
From the HPLC results of the K. sucrofermentans
monoculture, we noticed that there was a carry-over of maltose
(0.5 g/L, around 5.5% of the amount in the co-culture) (Figure 6A). No maltose was
added to the original monoculture sample, so was thought to be
left from the previous injection. Moreover, low production levels
of ethanol and acetate, with a relatively higher proportion of
ethanol, were observed. We brainstormed two potential hypotheses
to explain the results:
(1) There could have been yeast contamination at a very low
concentration within the HPLC samples, leading to the production
of ethanol.
(2) The Komagataeibacter produced ethanol, which is not
described for its metabolism.
Of the two, we consider (1) to be more likely, since our model’s prediction was able to show a production of ethanol (Figure 7B) at a higher concentration than that of acetate (Figure 7B), when accounting for S. cerevisiae as a contaminant (Figure 7B&D), and there was around 0.5 g/L maltose in the medium. This qualitatively matched the HPLC results (Figure 6A), suggesting that we likely did have a contamination.
Ethanol homeostasis circuit
A central challenge in developing a stable yeast–bacteria consortium for BC production is controlling the ethanol concentration. S. cerevisiae naturally produces ethanol at levels exceeding the desired threshold of 1% (v/v), which risks destabilising the system and compromising BC production by K. sucrofermentans1. To overcome this, we applied a structured DBTL approach to iteratively design, model, and test an ethanol homeostasis circuit. Across two DBTL cycles, we progressively advanced both our wet lab and dry lab work:
Model design and validation: We began by designing an ODE model for S. cerevisiae metabolism and ethanol production and validating whether the model described S. cerevisiae under normal circumstances. When this was confirmed, the model was extended to describe the ethanol inducible circuit.
Build genetic circuit based on model predictions: After validation of the model, it was used to guide design choices for the wet-lab circuit design and help with the experimental implementation of the genetic circuit.
Cycle 1: Model design and validation
The synthetic circuit was conceptually developed to allow for the inhibition of ethanol production when threshold concentrations are reached. The synthetic circuit consists of a promoter that senses ethanol concentrations. This promoter regulates the expression of pTEV+. pTEV+ is a protease that cleaves a specific amino acid sequence. We used this for the targeted degradation of PGK1 or GPM1, key enzymes in ethanol production (Figure 8). When pTEV+ is expressed, this now causes degradation of PGK1 and therefore a decrease in ethanol production. When the ethanol concentration drops as a result of this, the signal transduction also ceases, and the inhibition of ethanol production stops.
The circuit was modelled with ordinary differential equations (ODEs). The model was based on Michaelis-Menten equations and the fermentation pathway in S. cerevisiae. Literature values for growth rates, promoter activity, and degradation rates were used as initial parameter estimates. The null-model behaved plausibly in silico, but the model predicted slower S. cerevisiae growth than expected. This suggested that either the parameter estimates were inaccurate or that the model was oversimplifying growth behaviour. Experimental data were needed to confirm which explanation was correct.
Before ethanol homeostasis circuit simulations were run to test the functioning of the synthetic circuit, parameter estimates first had to be validated. To this end, experiments were performed to assess S. cerevisiae growth, glucose consumption, and ethanol production without the synthetic circuit. To obtain these data, a rho- strain was developed. This strain lacked respiratory capacity and had energy and biomass production coupled to the production of ethanol, which matched the model.
The rho- strain was grown for 24 hours in glucose concentrations of 2% (w/v), 5% (w/v) and 10% (w/v), all in triplicate. Every hour for the initial 8 hours, and then after 22 hours, a measurement was taken to determine OD600, the glucose percentage (w/v), and the ethanol percentage (v/v). OD600 was measured with a spectrophotometer, and ethanol and glucose concentrations were determined with HPLC.
We used the experimental S. cerevisiae data to fine-tune and validate the model, such that it could be used to make predictions to help with the experimental implementation of the genetic circuit. Initially, the experimental data showed faster ethanol accumulation than the model. The model parameters were successfully optimised so that the null-model matched the experimental data (Figure 9). We used the validated model to make predictions about whether 1) the genetic circuit indeed reduces the S. cerevisiae ethanol production rate, 2) which version (PGK1 or GMP1) of the genetic circuit to implement in the lab, and 3) which parameters to engineer in the lab.
Cycle 2: Build a genetic circuit based on model predictions
From the model, we learned that the synthetic circuit can indeed be used to reduce the S. cerevisiae ethanol production rate. According to the model, it did not matter whether PGK1 or GPM1 is targeted for degradation in the actuator domain of the synthetic circuit. For the wet-lab we therefore decided to focus solely on the development of the circuit with PGK1. Finally, the model helped identify parameters that were important for the functioning of the synthetic circuit (Figure 10). All parameters that were related to promoter activity showed non-uniform distributions, which meant that their activity is important. Therefore, different promoters with a range of dynamic ranges were designed.
The build of the synthetic circuit in the wet-lab was split up into two parts: the signal transducer and actuator domain were designed separately from the promoter. For the actuator, genome engineering with CRISPR/Cas9 was used to fuse an inactive bidirectional degron to PGK1. This inactive degron is activated by the signal transducer pTEV+. Plasmids were designed to express pTEV+ under the Z_{3}EVpr, which is \beta-estradiol inducible. With the signal transducer and actuator domain, the effect of the circuit can be tested upon activation by a promoter with a known dynamic range. For the ethanol-inducible promoter, we developed a modular cloning strategy that allowed ethanol-responsive Upstream Activation Sequence (UASe) repeats, isolation sequences, and core promoter elements to be easily combined in a single Golden Gate cloning step. Using this approach, we constructed 12 promoter variants by pairing four different core elements with one, two, or three UASe repeats.
Due to time constraints, this testing step has not yet been completed, but we aim to perform it shortly. By splitting up the circuit into a sensor domain and a domain containing the signal transducer and actuator, we are able to explore the effect of different levels of circuit activity on the production of ethanol and cell growth. Varying amounts of \beta-estradiol will be added to the growth medium. During a 24 hour growth experiment, samples will be collected to measure OD600, glucose concentration, and ethanol concentration. Similarly, ethanol promoter activity can be tested in a range of ethanol concentrations to find the promoter dynamic ranges. These measurements are taken continuously in a plate reader experiment, since promoter activity can be measured with fluorescence.
Once results about the circuit activity have been obtained, they can be coupled back to the model. The results of the signal transduce and effector domain can be used to further validate the model. When varying the \beta-estradiol concentrations, resulting in different levels of Z_3EV promoter activity, this changes the extent to which the ethanol homeostasis circuit is active. We expect this to result in different S. cerevisiae behaviour. Information about the cell growth and ethanol production for different activity levels of the ethanol homeostasis circuit can be compared to the model prediction. If the model predicts similar behaviour as the experimental results, we can further specify model parameters. This can then be used to guide improvements for future designs of the experimental genetic circuit. In addition, the experimental promoter dynamic range assays help determine the true parameter values of the ethanol-inducible promoter available in the lab. Simulating the model with the experimentally identified promoter-related parameters will help identifying promoters that help stabilise the ethanol concentration at 1% (v/v). This will provide concrete suggestions for the final version of the ethanol homeostasis circuit in which the actuator domain is combined with an ethanol-inducible promoter.