Engineering Success

Introduction

Scroll down to learn about how we used the engineering cycle to:

1. Engineering the Problem
2. Goal 1: Validate Device Functionality
3. Goal 2: Understand Device Implementation
4. Goal 3: Evaluate Device Utility

Engineering the Problem

Introduction

Responsible engineering requires a comprehensive understanding of the problem being addressed and how it fits into broader goals. Our team decided to tackle climate change and decided that addressing wastewater treatments would be the modality in which a biological device would serve the most utility. Wastewater expenditures in the U.S. total nearly one-trillion dollars, mostly due their extensive energy demand. We started by looking into biological systems that could serve to decrease this energy demand.

Electricity and Microbes

Our first thought was Microbial Fuel-Cells (MFCs). MFCs produce electricity by catalyzing the release of electrons from energy rich bonds of organic substrates under anoxic conditions [1]. Under non-anoxic conditions the efficiency of MFCs is severely degraded, resulting in a system that is not commercially viable [1]. To synthetically produce anoxic conditions in MFCs, high concentrations of L-cysteine are used [2]. In large scale reactors this becomes a limiting expense and creates excessive start-up costs [3].

To address this shortfall, we proposed a biological device able to supplement the medium of MFCs with 100 mg/L of L-cysteine before undergoing cell lysis. Cell lysis was deemed necessary to not upset the microbial balance in MFCs and ensure maximum efficiency. A literature review revealed E. coli K-12 strains as the most engineered for biological L-cysteine production, and corresponding genetic circuitry was designed.

Engaging with MFC experts made us realize that there were fundamental issues with MFCs, namely that theoretical yields were not high enough for market viability. For that reason, stakeholders like Dr. Daniel Bond (read more about our conversation with him on our IHP page) encouraged us to look towards Microbial-Electrolysis Cells (MECs). MECs take organic material and produce H₂, which can then be transformed into energy.

However, MECs face two major challenges. Firstly, methanogens often overtake these systems resulting in more methane production than hydrogen production. Because methane is a greenhouse gas, and hydrogen production is suppressed, this impacts the commercial viability of the system. Secondly, MECs only work at scale when fed with acetates. This severely limits its practicality.

The Road to Dark Fermentation

We learned from Professor Sovik Das about how he uses Dark Fermentation (DF) with MECs to increase hydrogen yields. DF is a light-independent process in which anaerobic bacteria convert glucose into H₂ and acetate [4]. DF commonly utilizes organic waste feedstocks and can be waste-to energy (WTE) streams. This allows a combination DF-MEC system that can both increase hydrogen yields and make the system more commercially viable. However, as Professor Das discovered, this did not tackle the methanogens.

To deal with methanogens and other hydrogen-consuming bacteria (HCB) during DF, Professor Das uses a variety of feedstock pre-treatments (pH, chemical, temperature, loading shock, etc.). Even still he faces difficulties with propagating hydrogen-producing bacteria (HPB). Literature reviews revealed that oxidation due to high oxidation-reduction potential (ORP) environments inhibit hydrogenase activity and HPB propagation [5].

Our first device construct was built to address excess oxidation, and with a detailed literature review we revised our original construct. Our new device, the Cysteinator, is a biological device that can supplement dark fermentation with L-cysteine up to a concentration of 300 mg/L and then undergo cell lysis. The revised concentration reflects the optimal oxidation-reduction potential, and cell lysis remains essential as to not have our device become a metabolic competitor to HPBs.

Defining What the Cysteinator Will Do

While we defined what the Cysteinator does, we also thought that it would be necessary to define how it would be implemented into DF process flows. Figure 1 shows the four parts of the DF process flow, and how they interact. As the Cysteinator is aimed to help propagate HPBs and increase hydrogenase activity, we found that its implementation best fits in Stage 1 (feedstock to inoculum conversion).

Figure 1

Stage 1 is where the greatest barrier to market viability for DF and its conjugate systems lies and is where the Cysteinator would be the most impactful. Feedstocks in Stage 1 range from different wastewaters to solid wastes and its resultant inoculums are affected by bioreactor type, climate, and specific process flows among many other factors. Increasing inoculum quality means that the Cysteinator must 1) be integrated into the Stage 1 Process Flow (see Figure 2) and 2) be modular to adapt to the diversity of these processes.

Figure 2

Stakeholder feedback led us to realize that we needed to prevent horizontal gene transfer between our device and native bacteria. Our modified process flow (see Figure 3) does this in two ways 1) we introduce the Cysteinator after pretreatment to minimize exposure to different bacterial strains, and 2) we advocate that our modified process flow includes a heat treatment step after Stage 3 (DF) to prevent any accidental release of the device and bacteria that may be carrying our engineered circuitry.

Figure 3

Bringing it Together

In summary our project helps make H₂ energy a commercially viable option. By sustainably enhancing the efficiency of DF-originated H₂ production, our device will help accelerate the global transition to sustainable energy. By supplementing DF feedstocks with precisely 300 mg/L of L-cysteine, the Cysteinator will be the key to making DF-based WTE streams a viable future energy stream.

Goal 1: Validate Device Functionality

The objective of this first goal is to build and test a modular device that can supplement DF feedstocks with 300 mg/L of L-cysteine. Along with the device we determined that a comprehensive model able to simulate device behavior would be beneficial for the implementation of the device. The below sections detail the Design-Build-Test-Learn cycles of our device and its accompanying model.

Design (Wet lab)

We decided to segment the functionality of the Cysteinator into two plasmid components: (1) the L-cysteine overproduction plasmid and (2) an L-cysteine regulated kill-switch. This segmentation allows us to more effectively build circuitry and fine-tune L-cysteine overproduction and export. Modularity was built into every aspect of this project, synthesizing both parts with flanking restriction enzyme sites around every gene segment, allowing for ease of swapping between mutants in future designs, to optimize function in different bioreactor systems.

Our goal was to design a bio-based device that could induce a large change in L-cysteine concentration in a medium yet simple enough to leave cell fitness relatively unaffected. It was shown that bioreactor systems functioned optimally within a range of 300-500 mg/L of added L-cysteine, and that exceeding this range would work against the efficiency of the system [6]. Thus, our overexpression plasmid is designed to produce up to 300 mg/L of L-cysteine.

Because of this constraint, we also wanted our device to stop functioning before it adds too much L-cysteine to the medium. Thus, another target was to design a kill-switch mechanism that would eliminate our device from bioreactor systems. This design prevents the Cysteinator from overstepping the threshold and from outcompeting the dark fermentative bacteria themselves. Thus, a dual-plasmid system was proposed, with each device-bearing plasmid tested individually before co-transformation into our final chassis, BW25113, which is known to be optimal for L-cysteine overproduction using plasmids similar to our own [7]. In the next section, we explain why the two plasmid system is superior to the current state-of-the-art.

L-cysteine Overproduction Plasmid

The L-cysteine overproduction plasmid is designed to accomplish the overproduction and export of L-cysteine into the medium. The native mode of expression and export of L-cysteine in E. coli functions through a series of complex but well-studied metabolic pathways. There is a significant body of previous work to manipulate and optimize the metabolic production of L-cysteine, which our team drew on to design a plasmid with high L-cysteine yields and low complexity [8]. This combination is important, as it has been found that high L-cysteine yields often come at the expense of cell fitness [7]. Our final plasmid design manipulates cysE, serA, and eamB—three key genes in the native enzymatic pathways of E. coli cysteine production and export [8]. While to our knowledge these genes and their variants have never been used in the same construct, they have all been well studied and used individually in various designs for producing L-cysteine. Our team placed mutant versions of these genes under the control of strong constitutive promoters to build the cysteine overexpression plasmid.

The cysE gene encodes serine O-acetyltransferase, which catalyzes the rate-limiting step of the synthesis of L-cysteine from L-serine in E. coli [8]. This enzyme in its native form is strongly feedback-inhibited by L-cysteine, which posed an issue for our design. Luckily, this problem has been tackled by metabolic engineers. Scientists researching the industrial overproduction of L-cysteine have identified several cysE mutants that are insensitive to feedback inhibition, one of which was even used by a past iGEM team [9, 10]. We decided to use the recently identified A237V mutant, which was used in a chassis optimized for L-cysteine overproduction to produce an L-cysteine concentration of 234.9 mg/L, which was 2.67 times higher than the basal L-cysteine concentration achieved by the optimized strain (88.1 mg/L) [11].

The serA gene encodes phosphoglycerate dehydrogenase, which catalyzes the rate- limiting step of the synthesis of L-serine from 3-phosphoglycerate [8]. This gene is critical to achieve L-cysteine overproduction because it ensures that adequate levels of the precursor, L-serine, exist in the cell. Similar to cysE, native serA is feedback-inhibited by L-serine. Again, mutants have been identified which greatly reduce this feedback-inhibition. Upon reviewing literature detailing these mutants, our team encountered some confusion: a commonly cited T410Stop mutation does not align with the E. coli serA gene, which contains a tyrosine residue at position 410 and only contains 411 residues [7]. We believe this is instead the serA gene native to C. glutamicum, but could not confirm this. We instead decided to use a triplet of mutations, H344A/N346A/N364A, which were shown to entirely remove the ability of phosphoglycerate dehydrogenase to bind serine in E. coli [12].

The EamB gene encodes the EamB inner-membrane transporter formerly known as YfiK, which exports L-cysteine and O-acetylserine in E. coli [13]. This prevents the toxic accumulation of L-cysteine within the cell and will help to avoid the induction of L-cysteine degradation pathways. Our team considered several options for transporters to overexpress, namely a similar inner-membrane exporter EamA (formerly YdeD) and an outer-membrane exporter TolC [8]. Ultimately, we chose to overexpress EamB due to the fact that a mutant version of the gene was shown to increase L-cysteine yields of an overexpressing strain by up to 70% [14]. We implemented this mutation in our design: N157S/G156S.

We considered size constraints while designing this plasmid, as past Virginia iGEM teams have had severe issues with assembly of large parts. We considered entirely removing genes like serA or EamB from the circuit, which would decrease L-cysteine output but still likely result in a functional device while also reducing the metabolic strain. Ultimately, the three-gene system of serA, cysE, and eamB was small enough that all three genes could be retained in a 3428 base pair insert.

We also considered alternative methods for overexpressing L-cysteine. Many strains engineered to overexpress L-cysteine include knockouts of key L-cysteine degradation pathway genes such as TnaA and YhaM to achieve much higher rates of L-cysteine overexpression [7, 8, 11]. For instance, one construct using gene knockouts as well as serine acetyltransferase mutants was able to achieve an L-cysteine concentration of 621 mg/L in the medium [15]. Future iterations of our device could include elements such as these knockouts to enhance L-cysteine production.

Cysteine Overexpression Plasmid Timeline:

Build (1)

To save time, we ordered the first build of our cysteine overexpression plasmid ready to transform in the pTwist Amp medium-copy vector from Twist Biosciences. Unfortunately, multiple transformations of this vector in 2 different lab spaces and by multiple people failed, yielding instead a 7736 base pair plasmid with no sequence homology to our insert (except in the antibiotic resistance gene). We believe this plasmid was introduced somewhere in the commercial assembly pipeline.

Build (2)

We then tried ordering the cysteine overexpression genes as a single gBlock from IDT, with the plan to clone the insert into a low-copy pACYC plasmid. However, IDT spent several weeks trying to assemble the insert before canceling the order due to a failed assembly. This was likely due to the length and high complexity of the sequence (although the complexity was within the parameters for assembly).

Build (3)

After speaking with a representative from IDT, we codon-optimized the coding regions to greatly reduce the complexity of the insert and ordered again. Again, the assembly failed and the order was canceled.

Build (4)

Finally, we ordered our insert from Ansa Biotechnologies after being awarded their grant for challenging sequence assembly. They assembled our insert in a large plasmid (pIND) designed for challenging sequences. We amplified this insert and added homologous overhangs via PCR and attempted to clone it into a low-copy pACYC plasmid via In-Fusion assembly. Unfortunately, several ligation attempts yielded high numbers of colonies containing unmodified vector, and a colony PCR was unable to identify any colonies with a successfully assembled construct.

Build (5)

We then attempted to clone our insert via traditional restriction cloning, using cut sites already present in our insert. We performed this assembly with 19, a high-copy plasmid we had on-hand that enabled blue-white colony screening. This, finally, yielded a large number of colonies containing our insert, which we verified by sequencing.

Test

To assay cysteine overexpression, we used a commonly cited protocol known as a ninhydrin assay to screen for amino acid concentrations[16]. Ninhydrin, a yellow crystal, interacts with the free alpha-amino group contained in amino acids through a temperature-induced reaction, and results in the formation of a brilliant deep pink/purple color compound known as Ruhemann's purple [16]. Absorbances are read at 560 nm to quantify the amount of intra/extracellular cysteine present against a blank. Similar protocols have been demonstrated and adapted from prior papers modifying the Gaitonde protocol, and additionally from iGEM teams in past years, such as the iGEM Trento 2012 team [17, 18].

Inoculation and growth conditions were adapted from previous studies that quantified cysteine outputs [14, 15]. SM1 is a minimal media with prior validation as an optimal medium for growth of cysteine-producing chassis, although M9 media was considered as well [13, 7]. The device was cultured in 50 mL of SM-1 in 250 mL Erlenmeyer baffled shake flasks with a thiosulfate additive acting as a sulfur source for L-cysteine synthesis [14].

The figure above demonstrates a ninhydrin assay performed in our lab with varied cysteine concentrations, functioning as a calibration curve for spectrophotometric readings with standard concentrations. This calibration curve demonstrated is qualitatively of weaker coloration than spectrophotometric readings taken, as the pigment is prone to degradation. Optimizing the lack of degradation has been a focus of our lab as well. Future iGEM teams should know that the sensitivity of this assay is greatly enhanced by prepping new Ninhydrin Reagent 2 fresh with every iteration of the assay and minimizing time spent between water bath/ice bucket, along with ethanol addition to just-boiled sample tubes, to reduce degradation (see Experiments page for more details).

Optical densities of our bacteria were measured every 1-2 hours at 600 nm to obtain an approximation of the growth rate and examples of metabolic strain on our chassis, as well as primarily functioning to inform our model. In this way, we can create an understanding of the way in which our device functioned as well as how best to modify our designs for optimal function in future iterations.

L-cysteine regulated kill-switch

Our L-cysteine regulated kill switch is designed to induce cell death once enough L-cysteine has been exported into the medium. This is accomplished with a plasmid consisting of three genes: the CcdB toxin, the CcdA antitoxin, and the CcdR transcription factor.

CcdR is a small ligand-binding transcription factor in the family of feast/famine regulatory proteins [19]. It is natively found in P. ananatis, although it has a homolog, DecR, in E. coli [20]. CcdR exists as a tetramer but will dimerize upon binding cysteine to form an octamer that can act on binding motifs present in the promoter of the (unfortunately named) CcdA gene [11], which must not be confused with the CcdA antitoxin present in our kill-switch device. The CcdA gene in P. ananatis encodes cysteine desulfhydrase—a necessary enzyme for the degradation of L-cysteine [19]. It is referred to as CcdA* in this project to avoid confusion with the antitoxin CcdA. Interestingly, in P. ananatis the CcdR and CcdA* genes are located antisense to each other and are separated by an intergenic region only 117 nucleotides in size [11]. We used a mutant of CcdR, V84I, in our construct, which was identified to have a lower Km than the wild type, and a higher signal-to-noise ratio [11]. It was also shown that a combination of the low-strength promoter and RBS, J23114 and B0034, displayed the highest signal-to-noise ratio among combinations tested, so we used these elements in our construct as well [11].

CcdB is a well-studied toxin in E. coli, and has been used both in kill switches and, more commonly, in gateway cloning [21, 22, 23]. CcdB kills E. coli by inhibiting DNA gyrase, leading to double strand breaks during replication [21]. Natively, this toxin exists in the F’ plasmid, where it works in conjunction with the CcdA antitoxin to select against plasmid loss [21]. In our construct, CcdB is placed under the control of the CcdA* promoter to make it L-cysteine-inducible. Because the regulatory regions of the CcdA* promoter are not perfectly understood, this was accomplished by placing the CcdB coding sequence under the control of the entire CcdR-A* intergenic region. A similar strategy was used to create an L-cysteine mediated biosensor by allowing the intergenic region to control GFP expression, and we implemented this method following discussion with one of the scientists who created this biosensor [11]. We chose this toxin because it is relatively well-studied, and because there are several E. coli strains on the market that are CcdB insensitive and can therefore be used to propagate our kill-switch before it has been tuned [25]. We took this coding sequence from the iGEM registry (BBa_K3512001) and synthesized it with a silent point mutation, V73V, to remove an illegal BsaI cut site.

The CcdA antitoxin negates CcdB action by directly complexing with the toxin protein, preventing it from binding DNA gyrase [21]. In the native system, cell death is prevented only if CcdA on the F’ plasmid is being expressed. If the F’ plasmid is lost, residual CcdB remains active longer than CcdA and leads to cell death [18]. We included this antitoxin in our kill-switch design to ensure that “leaky” expression of CcdB does not lead to cell death in the absence of L-cysteine. Inclusion of this antitoxin also allows us to tune our kill switch to different cysteine concentrations. We included BsaI restriction sites flanking the CcdA promoter, allowing us to quickly swap Anderson promoters of different strengths into our construct. We will use this to perform parallel testing to identify what level of basal CcdA expression will lead to cell death at our desired L-cysteine concentration of 300 mg/L. We took this coding sequence from the iGEM registry (BBa_K4907032).

Cysteine-Induced Kill-Switch Plasmid Timeline:

Build (1)

We intended to order our kill switch plasmid from Twist Biosciences in similar fashion to the cysteine overexpression plasmid, but Twist was unable to accept the order due to the presence of the CcdB toxin in the construct.

Build (2)

We then ordered the kill switch plasmid as a single gBlock from IDT to conserve time. We added homologous overhangs via PCR, cloned the insert into pUC19, and transformed it into competent cells that contained the F’ plasmid to avoid CcdB-associated toxicity. This also required several attempts due to high numbers of colonies containing unmodified vector or vector containing a primer-dimer, but screening via restriction digest allowed us to identify colonies that contained the insert. However, upon sequencing we noticed that all colonies contained 2-3 frameshift mutations in the CcdB or CcdR coding regions.

Build (3)

Suspecting selection against toxic genes could be the issue, we performed the assembly again and transformed into Invitrogen’s Oneshot CcdB Survival 2 T1R competent cells, which are designed to propagate plasmids containing CcdB. Sequencing of these colonies yielded mutations in similar locations to previous attempts.

Build (4)

Another IDT representative suggested that these mutations could be caused by the PCR step and offered to resynthesize our sequence for free, with the overhangs required for In-Fusion already added. We accepted this offer and removed an internal cut site that will make cloning easier should it become necessary to explore other cloning strategies.

Test

We cultured chassis containing our kill switch in 15 mL culture tubes, inoculating at a standard optical density and taking a final O.D at 600 nm after roughly 4 hours. To quantify cell death over time, we used a fluorescent bacterial live/dead screening assay [20]. Initial modeling and literature predict that cell death occurs at around 3 hours after CcdB concentration exceeds CcdA.

Our modeling also relies on quantification of the toxin/antitoxin CcdB/CcdA system, and so we are removing small aliquots for Western blotting to quantify the levels of CcdB protein in the cell over time. These readouts give insight into how our toxin/antitoxin system works, and are extremely useful to characterize for future iGEM teams and future applications over a broad range of use, as both a mechanism for controlling concentration, and also for safety and biocontainment purposes.

Learn

The iterative device assembly process has taught us about the complexities of in-fusion assembly and toxin synthesis, insight that is valuable for future teams. While DNA synthesis companies might offer to synthesize entire constructs, the cost of failure can greatly outweigh the benefits of success. Whenever possible, teams should order individual genes and subclone them in-lab. This is relatively simple with assembly methods such as Golden Gate and allows for faster turnaround time of DNA orders. More than that, this lowers the chances of failed syntheses and prevents one failed synthesis from setting a project back by months.

Although our planning was meant to offset any experimental errors that arose, the volume of difficulties with our assemblies has led to delays in our Build phase. The incorrect vectors received and ligation attempts with high mutation rates did slow down the impact and depth of data we were able to collect. As we are still in the process of acquiring full understandings of our functioning device, there is much to be gained from the further study of these parts through future testing schemes.

Currently, our team is continuing to test the now-ligated cysteine overexpression plasmid while simultaneously working on the assembly of the toxin/antitoxin kill switch. Our new method is based on what we have now learned from success with traditional restriction cloning methods, along with cloning into CcdB-resistant cells to minimize any potential selection occurring prior to our tuning of the kill switch. We will perform our ninhydrin assay scheme to determine the L-cysteine concentration our overproduction plasmid can achieve. Once our kill-switch plasmid is successfully ligated, promoters on CcdA can be swapped to tune the kill switch. In this way our project will undergo further iterations of the DBTL cycle leading up to the Jamboree.

The plasmid design process as a whole required significant creativity, and we acknowledge that there is future potential for optimizing design as further variants are discovered with higher rates of efficiency that may be better for certain kinds of dark fermentative application over other bioreactor systems. For example, the team abstained from using gene knockouts throughout this process, which would even further increase the efficiency of L-cysteine production in future iterations of this design [2]. The preliminary data collected will already better inform further design or build schemes.

Once our initial goals have been reached, the Cysteinator’s interactions with other bacteria in true bioreactor systems should be tested. Coculture interactions are imperative to study, either through wet lab or modeling analytics, to further characterize the device’s behavior in real-world environments. Additional future steps will investigate growth-phase and timing optimization in some capacity along with testing the quantity of bacteria needed for a bioreactor of a particular size to obtain a target concentration in a specific timeframe or even testing impact on optimal feedstock or inoculation conditions. The device will also be further optimized to prolong plasmid maintenance and decrease metabolic strain. Future modeling applications can also be implemented to examine these parameters, which will optimize the final product ultimately generated for the bioreactor environment.

Our assembly difficulties, particularly with the cysteine-induced kill switch, highlight the complexity associated with cloning toxic genes and transcription factors. It is likely that the high mutation rates in the kill switch were at least in part caused by either the toxicity of CcdB or the off-target effects of CcdR. While we negated the effects of CcdB toxicity by cloning in CcdB resistant E. coli, a previous study that used a CcdR-based L-cysteine sensing circuit described difficulties in achieving a plasmid containing CcdR expressed at high levels, possibly due to off-target effects of the transcription factor [6]. While our construct did use the same low-strength promoter and RBS that were successfully used in this study, future iGEM teams should take note of these difficulties. Transcription factors must be considered in the context of the entire chassis, not just the synthetic gene circuit. Future iterations of this L-cysteine sensing circuit would be greatly improved by enhancing the orthogonality of the cysteine-sensing transcription factor.

Design Modeling

Similar to the circuitry, we decided that our model should also be segmented to model the L-cysteine overproduction and kill-switch insert separately. This was done for two reasons, 1) so that we could build, design, and tune a model using different assumptions for both parts, and 2) we could predict how design decisions reflect themselves using the independent parts of the model. We used constraint-based modeling to model the effects of L-cysteine overproduction on the metabolic network of E.Coli and mechanistic/stochastic modeling to understand the effects of the kill-switch. Afterwards, we integrated the previous models into a dynamic flux balance analysis (dFBA) to understand how the genetic circuits implemented in one plasmid would function together and predict device behavior in a bioreactor.

The constraint-based model with flux balance analysis (FBA) played a large role in validating our device design and the genetic modifications made. We wanted to understand how L-cysteine production pathways would change as a result of the modifications made with the ultimate goal of increased L-cysteine export. The mechanistic model was used to aid in the prediction of testing schemes in wet lab procedure allowing for us to predict with some degree of accuracy the desired promoter on CcdA to reach our target concentration upon kill switch activation as well as how long our device would take to reach this target concentration.

Build + Test

We began with the baseline iML1515 genome scale metabolic model (GEM) for E.Coli K-12 MG1655 and then incorporated enzyme constraints using the ECMpy workflow to increase the accuracy of the predictions made with FBA. The media conditions were set to reflect the conditions used to culture the cells in wet lab with SM1, Luria-Bertani broth (LB), and thiosulfate. We ran FBA, optimizing for L-cysteine export, while requiring a certain level of biomass growth, with the initial conditions to use as a point of comparison. Afterwards, the GEM was altered to reflect the genetic modifications made experimentally. Kcat values and gene abundance values were specifically modified to reflect mutated enzymes and modified promoters. The FBA was run with the corresponding modifications to observe fluxes through relevant pathways involved in L-cysteine production and understand its impact on biomass growth. Separate simulations were run and used to validate L-cysteine overproduction. The exact concentrations of L-cysteine produced cannot be determined from the FBA due to underlying steady state assumptions but were approximated concentrations in later parts of the model.

Learn

From testing the device design using FBA, we learned that our design leads to a 1.93 fold increase in L-cysteine production. We also understood the limitations of the constraint-based modeling, specifically due to steady state assumptions. The model was not able to account for dynamic behavior over time or predict accurate concentrations of relevant metabolites and media components. Thus, we iteratively developed a hybrid multiscale model design to incorporate other models to characterize the various aspects of the device design.

Design

In addition to validating L-cysteine overproduction, we wanted to ensure our kill switch mechanisms and toxin-antitoxin system would work as expected. The kill switch was designed under the assumption that intracellular L-cysteine accumulation would bind to form a transcription factor with CcdR octamers and then activate the CcdB toxin. Mechanistic modeling with stochasticity was implemented to model the processes involved in activating the kill switch over time. The design of the model was developed using Hill-like approximations to reduce the number of kinetic parameters needed. Furthermore, we used molecular docking simulations to approximate the binding affinity values required for the differential equations used.

Build + Test

The formation of equations used in the mechanistic model were split into sections based on the various components involved in activating the toxin. We iteratively built and rewrote the equations incorporating assumptions from literature and receiving feedback from professionals in the field. The three sections of equations can be split into CcdR oligomerization/transcription factor formation, stochastic equations for the binding of the transcription factor to the CcdA promoter on the CcdB toxin, and CcdA/CcdB pathways for the toxin-antitoxin system. We originally intended to incorporate Michaelis Menten kinetics, however due to the lack of characterization on the CcdR system, we developed modified Hill equations instead. While working through the processes involved, we realized that we cannot assume the CcdB toxin is being expressed at any given time because of low expressions of the transcription factor, requiring us to factor in stochasticity. We tested the model by running simulations and evaluating the number of CcdA, CcdB, TA complexes, TAT complexes over time.

Learn

Through the process of designing the mechanistic and molecular docking models, we were able to understand the activation of the kill switch as well as the specific mechanisms involved in each component of this insert to further characterize the novel composite part. We found that the average passing time for the kill-switch activation was around 30 minutes. Through molecular docking simulations, we were able to explain binding patterns for CcdR, which was previously not well characterized. We found the binding affinity values for each step in the octamerization process and that 8 L-cysteine molecules are required to bind to form the final transcription factor.

Goal 2: Understanding Device Implementation

The objectives of our second goal are to 1) identify potential roadblocks in future implementation and 2) ensure that our device is compatible with the current process flows and bioreactor infrastructure. Since the Cysteinator is meant to be integrated into current DF process flows, it is important that we assess how easily that can be done. To do this we decided to use the FBA from the model in Goal 1 to assess the maximum flux of L-cysteine production in different feedstock environments. Combining this with continued stakeholder engagement and background research we developed a thorough understanding of the challenges that may arise with the Cysteinator’s integration into DF process flows.

Design Model Experiment

The final design of the device incorporates the L-cysteine overproduction insert and kill-switch insert in one plasmid. The previous models model these inserts separately, but it does not reflect the overall device design. Thus, we designed a final dynamic flux balance analysis (dFBA) model as a method to model and predict the behavior of the overall device over time in a bioreactor setting. Furthermore, we used the previous FBA model to test various conditions characteristic of wastewater used in dark fermentation bioreactors to inform how our device is expected to behave in varying environments. The average passing time determined from the mechanistic model was used to understand how long the device would need to produce the required levels of L-cysteine and then lyse before dark fermentative bacteria is added to the bioreactor. We also designed the mechanistic model to be able to reverse engineer based on our desired concentration and determine the most effective promoter strength to influence wet lab design before device implementation.

Build

The dark fermentation bioreactors our device is meant to be implemented in will likely contain wastewater and other components. We altered and tested the media conditions in the FBA to understand these effects. We selected two representative media conditions for real/synthetic wastewater and solid waste. These sources mainly differ from our media conditions used in lab as they contain other carbon sources and could impact growth and L-cysteine export. The real/synthetic wastewater conditions varied in that the main carbon source was sucrose and the main carbon source in solid waste was fructose and xylose. The dFBA was built using time steps based on Euler’s method to resolve the FBA at various time points while updating extracellular metabolite and biomass concentrations. This was done to be able to approximate concentrations of relevant metabolites in the system, which was previously not possible. The dFBA was integrated with the mechanistic model by using the calculated intracellular L-cysteine concentrations to update the intracellular L-cysteine concentrations used to form the transcription factor.

Test

From initial testing of the effects of glucose and thiosulfate uptake on L-cysteine export and biomass growth, we noticed that the uptake values plateaued and no longer contributed to increase flux through the pathway. While testing the various wastewater conditions, we found that E.Coli K-12 derivatives and similar strains are not capable of utilizing sucrose as a source for growth because they lack the mechanisms to metabolite it. However, fructose was shown to lead to increased growth and L-cysteine export. Even in conditions without carbon sources present, the bacteria are able to utilize amino acids and other components for growth. After running the mechanistic model, it was found that the average kill switch activation time is around 30 minutes. The dFBA is currently still in testing, but similar results to the mechanistic model are expected.

Learn

Overall, the various types of models informed different aspects of how we expect the device to behave in industrial application. From the FBA, we learned that our model will not continue to increase L-cysteine export and continue growing but continually adding media components like thiosulfate and glucose. After reaching a certain threshold, the media components are no longer the determining factors for the efficacy of the device. Given the cost-effective appeal of the device, we can use this to inform downstream application by optimizing the concentration of carbon sources and thiosulfate added to the media while minimizing costs. Furthermore, the model highlighted the flexibility of our device to continue functioning over a wide range of conditions, even when no carbon sources were present. The time for kill switch activation predicted by the mechanistic model/dFBA informs the period of time necessary for the device to be in the bioreactor in the pretreatment phase before the dark fermentation bacteria can be added. The results from the dFBA integrated with the mechanistic model will be used to predict the time it takes to reach the desired L-cysteine concentration to further refine the predicted time necessary for the device to function as expected.

Goal 3: Evaluating Device Utility

Our final goal aims to ensure that the device’s intended impact is realistic. Specifically, we evaluated whether our device was truly sustainable and can increase the commercial viability of DF and its conjugate systems. To do this, we segmented our analysis into four parts: 1) a comprehensive meta-analysis of the Cysteinator’s alignment with the UN Sustainable Development Goals (SDGs), 2) a life-cycle analysis to analyze the end-to-end emissions of our device, 3) a techno-economic analysis to determine the specific economic impact of the Cysteinator on the commercial market viability and scalability of the Cysteinator integrated DF process flows and, 4) continued literature review and stakeholder engagement to ensure that our assessments were reasonable and consistent with best practices.

Design LCA + TEA

During our meeting with Dr. Lisa Colosi-Peterson, we were introduced to a tool for assessing resource consumption and their resulting emissions, Life Cycle Assessment (LCA). She also specializes in a tool to evaluate the economic feasibility and viability of a technology, a Techno-Economic Assessment (TEA). The combination of these assessments allows us to make estimations on both the environmental and economic impacts of the Cysteinator. We worked with Dr. Colosi-Peterson throughout the process of conducting our LCA and TEA to implement her feedback in our data sourcing and calculations. See section SDG 13.2 on the SDG page for a more detailed overview of her support.

Build

We conducted our LCA and TEA based on existing studies that model similar assumptions as those of the Cysteinator. For the LCA, the base study was constrained by a Dark Fermentation-Microbial Electrolysis Cell (DF-MEC) system that uses wastewater as a feedstock. For the TEA, we considered the application of E. coli in producing high value chemicals. To better model our device, we factored into our calculations the additional predicted H2 production as well as the avoided emissions from bio-based L-cysteine production and the need for wastewater treatment.

Test

From the LCA, we saw significant reductions in CO2 emissions per kg of H2 for a DF-MEC system compared to all current forms of H2 production, including steam methane reforming (SMR) and two forms of electrolysis (PEM and SOEC). It was also clear that wastewater utilization as a substrate makes a major impact in reducing emissions. Meanwhile, while bio-based L-cysteine supplementation makes a positive impact on H2 yields, the way in which it is sourced does not cause a prominent reduction in CO2 emissions. The TEA, furthermore, presented a 17% decrease in cost per kg of hydrogen with the implementation of the Cysteinator compared to a standard DF-MEC system. Hence, the boost in H2 production is significant enough to not only mitigate but reduce the net costs of our optimized system.

Learn

As a result of the similarities in direct emissions associated with industrial production, which relies on the process of keratin hydrolysate, and Cysteinator-derived L-cysteine production, which relies on metabolic engineering strategies, the LCA led us to further consider the ethical and environmental factors that are not accounted for by its scope. For instance, the keratin used in industrial processes is derived from animal sources and regulating their treatment is not well documented nor controlled. Further, it releases a by-product of hydrochloric acid waste, which is energy-intensive to treat and environmentally deleterious when untreated. The LCA ultimately validated the net positive impact of Cysteinator in terms of sustainable H2 production, as well as allowing us to realize factors of sustainability that go beyond the units of CO2 emissions per kg of H2. While it will take time for industrial development to make DF-MEC H2 production cost-effective in comparison to current methods, such as SMR and PEM electrolysis, the TEA demonstrates that the Cysteinator offers a promising tool for both sustainable and economical H2 production.