preloader
mascotte1

Modeling




Our modeling strategy aimed to complement our efforts of engineering Pseudomonas fluorescens. By evaluating modifications in silico, we characterized the impact of each modification on native metabolism and increased our understanding of the engineered strains. By testing pathways and comparing alternatives, we supported and improved design. Model predictions contributed to experimental design, and laboratory measurements were integrated in silico to further validate and improve modeling outcomes.


Goals

The goal of BioMoon is to produce an engineered strain of Pseudomonas fluorescens that, through the integration of different modifications, will be able to adequately foster plant growth on lunar regolith.

    To support this goal, the modeling strategy aims at achieving three main objectives:

  • Inform design by implementing the modifications in silico, simulate pathways, and comparing alternative options to refine decisions.

    e.g. testing different oxidative stress detoxification pathways to identify the most efficient pathway to target.

  • Contribute to system understanding by providing insights into the impact of each modification on metabolism and characterizing both quantitatively and qualitatively these impacts.

    e.g. assessing the impact on growth rate and ammonia production of the creatinine metabolization pathway.

  • Generate global insights of system function and predictions of global system behavior by creating a complete in silico representation of our modified strain from the combination of all the designed modifications.
  • e.g. by assessing the impact of oxidative stress resistance on the production of nitrates during growth on creatinine.

These efforts have contributed to driving experiments, and have been further reinforced by the integration of laboratory results into the model. By doing so, our modeling strategy has been efficiently integrated into the engineering cycle, allowing us to obtain insights and predictions of our system beyond what can be solely achieved in the lab.



Strategy



The essence of our project lies in the engineering of different pathways in P. fluorescens to improve its metabolism to support plant needs in the adverse environment of lunar regolith. To effectively complement the project, the modeling strategy reflects this modular structure by testing each modification individually and then integrating them together alternatively to obtain meaningful insights (Figure 1).

Figure 1: overview of the modeling strategy. Each engineered pathway is first assessed by itself to answer key questions which will enable us to obtain meaningful insights and predictions. Then, modules are integrated together alternatively to characterize their combined behavior.


Tools



To simulate each pathway and their impact on metabolism, our strategy centers around metabolic modeling. We have adapted a Genomic Scale Metabolic Model (GSMM) of P. fluorescens and performed simulations using Flux Balance Analysis (FBA) and Flux Variability Analysis (FVA) with the Python package cobrapy1.





A GSMM is an in silico reconstruction of the complete metabolic network of a given organism. It is a comprehensive collection of all the reactions occurring within a cell where each reaction links different metabolites and is governed by specific genes (Figure 2). This type of reconstruction has become fundamental to many applications in life sciences and as such, GSMMs have become a pervasive tool in many fields of biology including metabolic engineering2.


WHY?

GSMM has been chosen as the base for our modeling strategy, because it enables us to obtain a comprehensive overview of how each modification fits into and influences native metabolism of P. fluorescens.


Figure 2: GSMM structure and in silico implementation example. Each metabolite is involved in different reactions regulated by a gene. In silico, each reaction is reported with the relevant information to record the network structure of the GSMM in a single file.

The use of GSMMs allows us to simulate metabolism and gain significant insights into system function and structure, as well as predict behavior. This is possible through FBA, which uses mathematical principles to calculate the fluxes going through each of the reactions present in the GSMM. Simulations can be performed to respect different constraints, metabolic conditions and modifications (Figure 3).


WHY?

FBA is an important tool for our strategy as it enables us to simulate metabolism, obtaining a comprehensive overview of which reactions are active under different conditions and effectively determining metabolic behavior by predicting the flux distribution.


Figure 3: Flux Balance Analysis mathematical background. First, the stoichiometric matrix is constructed from the GSMM. Then, the model assumes metabolism operates at steady state, implying that fluxes are constant over time and metabolites are not depleting or accumulating. Mathematically, this translates into the equation dM/dt = S.v = 0, with M the given metabolite, S the stoichiometric matrix, and v⃗ the (unknown) flux vector. By assuming steady state, this equation can be solved for the unknown flux vector ​​v⃗ , containing the solution fluxes of all reactions in the model, and the solution space can be determined. Conditions are applied to constrain the solution space. Lastly, fundamental to FBA is the objective function, which can be the maximization or minimisation of the flux through a specific reaction. Through the definition of this function, the solution vector can be calculated and the list of fluxes through the system obtained.

Throughout the project, a variation of FBA was also used: Flux Variability Analysis (FVA). This method allows us to calculate the flux through different reactions in the metabolism for a range of maximum and minimal fluxes, therefore allowing us to better evaluate different ways in which metabolism can behave.


WHY?

FVA allows us to strengthen the conclusions made with FBA by extending our predictions for reaction behavior over a range of possible fluxes.


In FBA we solve for the steady state reaction, Sv⃗ =0, subject to maxv (cT⃗ v⃗ ) , to obtain the optimal solution v⃗0, with c⃗ indicating the objective function. In FVA, after this, for each of the fluxes of interest (vi) a range of is calculated as the min and max values of vi in a range γ of the optimal solution γv⃗0 where 0 < γ < 1. Note that for γ ≤ 1 the maxvi= minvi . In this way FVA solves two optimisation problems for each of the vi under the steady state assumption Sv⃗ =0 of FBA for a range, dictated by γ, of the optimal solution v⃗03.

In order to perform FBA and FVA on the GSMM, we will use the python package cobrapy1. This package allows us to work with the metabolic model and perform flux analysis. All the results are provided in the standard units of the model, which are mmol/(gDW h).


WHY?

Python and cobrapy1 were chosen due the free availability, intuitive application for FBA and FVA on GSMMs, seamless integration into the shareable format of jupyter notebooks, and extensive documentation easily available online for the package.


To achieve our modeling goals we reconstructed a P. fluorescens GSMM from the GSMM of P. putida (iJN1463). This adapted model was modified to reflect species-specific gene differences and successfully provided accurate predictions confirmed by laboratory results. All simulations using this model were carried out with standard exchange reaction values.


Go see our GitHub here


Our modeling strategy requires a GSMM of P. fluorescens. The only available GSMM of this bacterium, named iCW1057, has been published by X. Huang and Y. Lin 5. Once obtained, we first verified the model by reproducing the results reported in the original publication. Unfortunately, in performing these simulations, we could not reproduce any of these results. As another concern, this model did not comply with the expectations of a high quality model.


    Specifically we have identified:
  • Lack of properly defined cellular boundaries,
  • Lack of annotations for metabolites and genes,
  • Formalisms that do not comply with regular GSMM standards (regarding e.g., the exchange reactions),
  • Strong discrepancies between the predictions obtained from the model and the simulation results reported in the paper itself.

The ability to perform reliable simulations on iCW1057, the only available GSMM of P. fluorescens, revealed itself to be impossible.


Given that this model was of low quality and could not reproduce the published results, the decision was made to adapt and utilize the GSMM of Pseudomonas putida instead. Both P. putida and P. fluorescens in fact, share a similar core metabolism as they both belong to the same genus and are closely related.

A high quality GSMM model of P. putida KT2440, named iJN14634, is available in json format compatible with cobrapy1. This model complies with the requirements for successful FBA and has therefore been the base of our project’s modeling strategy.

In light of this choice, an additional step was carried out at the beginning of each modeling phase, where key genes that are relevant to the modifications of the project are checked from the P. putida model against the annotated genome of the P. fluorescens strain specific to our project SBW256. Table 1 summarizes the genes that have been identified throughout the project and the resulting changes in the model.

Table 1. Gene comparison between P. putida and P. fluorescens. Reported are the names of the relevant genes and their corresponding enzymes that were found to differ between the two species, and accordingly what was done in the model to ensure it would reflect the metabolism of P. fluorescens.

Gene

Enzyme

In P. putida

In P. fluorescens

Model

codA Creatinine iminohydrolase Present Absent Removed
soxA Sarcosine oxidase tetrahydrofolate dependent Present Absent Removed
norB Nitric oxide reductase (putative) Absent Present Added
nirS Nitrate reductase Present /td> Absent Removed


Starting from a GSMM of P. putida, we built a GSMM of P. fluorescens. As both species share a similar metabolism, this model is an effective substitute that proved relevant to our project, since we obtained meaningful and accurate predictions that were also corroborated by laboratory results.


Conditions for simulation

All of the conducted simulations use the same basic exchange reactions for the cell (Table 2). If applicable, any changes to these exchange reactions are properly indicated throughout the page. Carbon sources and objective functions are specific to each simulation and are indicated for every result. Following the conventional notations, negative (positive) values for the fluxes of an exchange reaction indicate the specific substance is being uptaken (produced) by the cell.

Table 2. Base exchange reactions used for every simulation. The lower and upper boundaries of the exchanges are listed as well as the reaction names.

Lower bound(mmol/(gDW h)

Upper bound (mmol/(gDW h)

Reaction name

Reaction id

-100 1000 Calcium exchange EX_ca2_e
-100 1000 Chloride exchange EX_cl_e
-100 1000 Co2+ exchange EX_cobalt2_e
-100 1000 Cu2+ exchange EX_cu2_e
-100 1000 Fe2+ exchange EX_fe2_e
-100 1000 H2O exchange EX_h2o_e
-100 1000 H+ exchange EX_h_e
-100 1000 K+ exchange EX_k_e
-100 1000 Mg exchange EX_mg2_e
-100 1000 Mn2+ exchange EX_mn2_e
-100 1000 Molybdate exchange EX_mobd_e
-100 1000 Sodium exchange EX_na1_e
-100 1000 Ammonia exchange EX_nh4_e
-100 1000 Ni2+ exchange EX_ni2_e
-100 1000 O2 exchange EX_o2_e
-100 1000 Phosphate exchange EX_pi_e
-100 1000 Selenate exchange EX_sel_e
-100 1000 Sulfate exchange EX_so4_e
-100 1000 Tungstate exchange EX_tungs_e
-100 1000 Zinc exchange EX_zn2_e


Modules


The detailed steps of the modeling strategy for the single modifications (also called modules) are presented below. For each result, we have described the guiding question (which is the goal or hypothesis that motivates the simulation), the experimental setup (i.e., how we carried out the simulation), the results, and lastly how our findings contribute to the overall project.


To enable creatinine metabolization, initial design aimed at integrating three enzymes into the bacterium to catalyze its conversion to glycine, which can be naturally metabolized by the bacterium (Figure 4).

Figure 4: Creatinine uptake pathway. Creatinine is hydrolyzed to creatine by cyclic amidohydrolase (CrnA), which is then converted to sarcosine by creatine amidinohydrolase (CreA) with the production of urea. Finally sarcosine is metabolized into glycine by sarcosine oxidase (SoxA) with the production of two byproducts: H2O2 and formaldehyde.

    As detailed in this section, for this modification, modeling has provided:
  • Support for the initial design by revealing the presence of soxA into the WT bacterium. The prediction was tested in the lab and confirmed in silico results.
  • Predictions of ammonia production and growth rates which were tested in the lab and compared to characterize system optimality.
  • Predictions which increased system understanding on the impact of byproducts on metabolism and on the quantitative relationship between growth and the uptake of each substrate of the pathway.
  • Further confirmation of design through the comparisons of alternatives to creatinine.

Modeling drives design of the creatinine metabolization pathway in conjunction with laboratory experiments

The presence of SoxA is found in the model and verified in the lab

In collaboration with design, the modeling has tested the initial formulation of the creatinine metabolization pathway to verify if upon integration of the given genes, growth was possible.


To address this question, growth simulations were conducted in silico on the substrates of the pathway: glycine, sarcosine, creatine, and creatinine, before and after introducing the specific genetic modifications (addition of soxA, creA, and crnA). Each modification was added sequentially on top of the previous one (i.e., creA was added on top of soxA, and crnA was added on top of both soxA and creA), growth rate was used as the objective function and ammonia uptake was set to 0.


Our results (Table 3) show that growth on sarcosine is possible even without the addition of soxA. In the model, we identified the presence of this gene already in the bacterial genome. We also saw that growth on glycine was possible, whereas on creatinine and creatine the relevant genes needed to be added in order to obtain growth.


In this way, we both confirmed that creatinine can support growth as the sole carbon and nitrogen source with the introduction of creA and crnA, thus confirming this pathway works. Additionally, we identified a way in which design could be simplified by revealing that growth on sarcosine is already possible in the WT strain. This last prediction was also tested in the lab, where growth of WT P. fluorescens on sarcosine, but not on creatine or creatinine, was confirmed. See lab here. In literature, no evidence is recorded about growth of P. fluorescens on sarcosine, so our results represent an important first step in the characterisation of this metabolic pathway.


Table 3. Summary of qualitative assessment of growth on the substrates of the creatinine metabolization pathway. The results of growth before and after adding the relevant genetic modification are reported. Each modification was introduced on top of the previous one. Growth rate was used as the objective function and ammonia exchange was set to 0 such that we could test if each substrate could be the sole carbon and nitrogen source.

Substrate

Condition

Result

Glycine No modification Growth
Sarcosine Before adding soxA Growth
Sarcosine After adding soxA Growth
Creatine Before adding creA No growth
Creatine After adding creA Growth
Creatinine Before adding crnA No growth
Creatinine After adding crnA Growth


Design is confirmed by the modeling

The designed pathway is able to support growth

Once the new design was refined thanks to in silico results, we further exploited the model to generate quantitative predictions of growth on each substrate as the sole nitrogen and carbon source.


Using growth rate as the objective function and setting ammonia uptake to 0, we recorded the maximal growth rate possible for each substrate of the creatinine metabolization pathway for a range of uptakes.


Simulation results (Figure 5) showed that growth rate on creatinine and creatine was the same; on sarcosine, growth was slightly lower and on glycine it was much lower. These predictions further support our design as they prove that this pathway provides a carbon and nitrogen source able to support growth. The findings also suggest an effect on growth caused by the byproducts of the creatinine catabolic pathway (urea and formaldehyde), which could explain the difference in growth rates observed among the different substrates.


These results provide quantitative information on the optimal growth rate of P. fluorescens on each substrate of the creatinine metabolization pathway. These results help confirm design because they show that a significant growth rate can be maintained on creatinine, which also contributes to system understanding.

Figure 5. Growth rate for a range of uptakes of each substrate with the modifications introduced from the design (crnA and creA). The growth simulations were performed in the absence of NH 4 (uptake rate constrained to 0) to further validate creatinine as both a carbon and nitrogen source for P. fluorescens. Growth rate was used as the objective function.

Creatinine can be co-consumed with the two other most abundant metabolites in urine

To further support our design choices, we wanted to verify if creatinine was a better substrate for growth compared to the two other most abundant compounds present in urine: urea and citric acid (citrate in the pH of regolith) and if it could be co-consumed with them.


We simulated growth alternatively on citrate, urea, and creatinine using growth rate as the objective function and without any ammonia uptake to verify the advantage of creatinine as a nitrogen source as well as a carbon source (Figure 6). Then, we repeated the simulation to verify if co-consumption was possible by using growth rate as an objective function with no ammonia uptake. We simulated growth for a range of creatinine uptake and different values of either urea or citrate (Figure 7).


As expected, citrate could not support growth in the absence of NH4 since this molecule can only act as a carbon source. Furthermore, we found that urea could not support growth on its own. Co-consumption of creatinine with both urea and citrate was confirmed, which means that under our engineering design the bacterium could efficiently use all three of the most abundant molecules present in urine to drive growth.


These findings provide additional support to our design choices. Firstly, we confirmed that creatinine is the best molecule to target in urine to act as a substrate. Secondly, we showed that growth on creatinine could be coupled with other very abundant substrates present in urine, thus increasing growth rate further.

Figure 6. Growth rate on three most abundant substrates in urine. All the growth conditions for the simulations were the same, using growth rate as objective function and no NH4 uptake to verify that creatinine was the best choice as both a nitrogen and a carbon source for our bacterium.

Figure 7: Growth on creatinine along with either citrate or urea. Using growth rate as an objective function and setting NH4 uptake to 0, we simulated growth for a range of creatinine uptake values and different values of urea exchange and citrate exchange. We can see from the plots that both substrates (though urea contributes less to growth) can be consumed with creatinine to support growth on human urine.


Byproducts of the creatinine metabolization pathway promote growth and NH4 production while introducing oxidative stress

To address the observed differences in growth observed among the substrates of the creatinine metabolization pathway (Figure 5), we investigated in silico the effect that each byproduct (urea, formaldehyde, and H2O2) had on metabolism.


To test these effects, we forced the excretion of each of the byproducts under study at the rate of which they were formed in the pathway of creatinine metabolization. Doing this, we were able to compare metabolic behavior when the byproducts were metabolized and when not.


From these simulations we showed that formaldehyde (Figure 8) plays a key role in explaining the growth rate differences between glycine and the other substrates that are observed in Figure 6. At the same time, the breakdown of urea was seen to contribute marginally to growth, but significantly contribute to the formation of NH4 and CO2 (Figure 9). On the other hand, the production of H2O2 is expected to induce oxidative stress on the bacterium (Figure 18).


These results provide invaluable insights into our system by revealing the ways that the byproducts of the engineered pathway affect metabolism and contribute to growth and NH4 production.
Furthermore, predictions of the role of formaldehyde in growth served as the basis for in vivo experiments. However, due to practical difficulties, the laboratory results were inconclusive. See lab here. Since also no evidence is reported in literature concerning formaldehyde impact on P. fluorescens growth, the model is an important source of this insight, with the perspective of future experiments corroborating the provided predictions.

Figure 8. Growth rate on substrates of creatinine metabolization pathway without utilization of formaldehyde formed by the pathway.Objective function was growth rate and uptake of NH4 was 0. Formaldehyde was constrained to be excreted at the rate with which it was formed (i.e. flux of sarcosine oxidation to glycine). From the plot we can observe that, when formaldehyde is not metabolized, growth rate is significantly more similar across the four substrates compared to when it is (Figure 5).

Figure 9: Growth rate (A) and NH4 production (B) when urea is metabolized and not (i.e., when urea is excreted upon formation during creatinine conversion to sarcosine). Growth rate was used as an objective function and no NH4 was uptaken. Urea excretion (labeled ‘No Urea’) was constrained to a value equal to the flux of creatine conversion into sarcosine. From the performed simulations, we can see that the metabolization of urea contributes slightly to growth (A) and significantly to the production of NH4 (B).


Prediction of growth optimality and NH4 production is verified in the lab

To quantitatively confirm our predictions of growth on the substrates of the creatinine metabolization pathway, we integrated results from the lab into the model. We used in vivo uptake rates to predict the maximal theoretical growth rate in silico. By comparing the predicted and experimental growth rates, we can make meaningful conclusions concerning the optimality of metabolism, which will allow us to inform design for the next engineering cycle.


To do this, we used laboratory NMR results of growth rates, uptake rates, and ammonia production rates of the WT P. fluorescens and P. fluorescens transformed with creA-crnA using creatinine, creatine, or sarcosine as substrate. See lab here. We simulated growth by constraining the model to the uptake rates measured in the lab, using growth rate as objective function and no ammonia uptake, and predicted growth rate and ammonia production in silico.


Results (Table 4) show that, for a given uptake rate of substrate, growth rate is always higher in silico. This is expected since the model runs with growth rate as the objective function, meaning that it calculates the ideal fluxes to maximize growth rate and this is an idealized abstraction of what realistically happens in vivo. On sarcosine, we see that growth rates are similar in silico and in vivo both for the WT and for the creA-crnA-transformed strain. This indicates that the in vivo conditions are close to the ideal ones assumed by the model and that growth on sarcosine is therefore optimal. We deduce that the transformation does not impact optimality. Similarly, Table 4 reports that growth on creatine is also close to optimality (almost 70% optimal growth rate). On the other hand, on creatinine, growth rate in vivo is 4 times lower than predicted in silico. However, the uptake rate is high, suggesting that the CrnA enzyme works efficiently. We can then hypothesize that the reasons behind the observation of this sub-optimal growth rate might be linked to the presence of alternative pathways, which would convert creatinine into compounds that do not feed growth.


These results contribute to a more precise quantitative understanding of the metabolism of P. fluorescens with the given modifications. By integrating laboratory measurements of growth rates and uptake rates into the model, we were able to assess that the metabolism is optimal for sarcosine and creatine. As no results were previously documented in literature about P. fluorescens WT growth on sarcosine, this insight is vital and further corroborates our modeling choices. We can now refine the design of the next engineering cycle and inform future experiments knowing that we would have to investigate the dynamics of creatinine inside the cell to bring growth closer to optimal for this substrate too.


Table 4. Results of growth rate, uptake rate, and ammonia production of WT and modified (creA-crnA) bacteria. Indicated are the conditions of the experiments: the modification of P. fluorescens, and the substrate of growth. Then, the results obtained in the lab: the measured growth rate, the uptake rate of the substrate of growth, and the ammonia production rate. Then, the modeling results which report the growth rate and the ammonia production rate predicted when constraining the model with the substrate uptake rate measured in the lab. Lastly, we reported what percentage of optimality the growth rates measured in the lab are (using in silico growth rate predictions as 100% optimal). NA: not applicable.

Another key modification introduced in the bacterium aims at increasing its resistance to stress through the integration of the genes hfq, rpoS and katB. Both hfq and rpoS genes play a regulatory function, meaning they do not have an associated reaction and therefore there is no associated flux either. Thus, they fall outside the scope of FBA and will not be considered in the calculations. In contrast, the katB gene has a reaction associated with it. Its enzyme, the catalase KatB catalyzes the conversion of H2O2 to H2O and O2, making its flux measurable and calculations possible with FBA.

The katB gene is already present in P. fluorescens. Our engineering design aims to introduce an additional copy to enhance the bacterium’s oxidative stress defense.


    As detailed in this section, for this modification, modeling has provided:
  • Support for the initial design through the simulation of growth in the presence of oxidative stress with and without KatB.
  • Quantitative results to improve system understanding on the relation between flux through the catalase enzyme and oxidative stress detoxification.
  • Further assessment of design through the comparison to alternative oxidative stress pathways.

Modeling confirms the role of the catalase in oxidative stress response and supports the design decision of increasing copies of katB gene to improve stress resistance.


In the presence of H2O2 the catalase is essential to maintain optimal growth

The first simulations performed aimed to qualitatively identify the effect on growth of H2O2 and the role of the catalase in oxidative stress detoxification.

To determine this, we compared growth in the absence and presence of the catalase for increasing needs of detoxifying H2O2.

Simulations show that, in the absence of the catalase, the growth rate decreases proportionally with the increase in H2O2. In the presence of the catalase, growth rate is maintained high regardless of H2O2 that is fully detoxified in all cases (Figure 10).


These observations confirmed our design, demonstrating that the catalase plays a crucial role in supporting optimal growth rate in the presence of H2O2.


Figure 10. Growth rate for a growing range of H2O2 uptake rates with and without the catalase (CAT). All growth simulations were performed in the same conditions using glucose as sole carbon source (with an uptake flux set at -5 mmol/(gDW h)) and growth rate as the objective function. The knock_out() function of cobrapy1 was used to ko the CAT when necessary. As the rate of uptake of H2O2 increases, in the absence of CAT, growth rate decreases linearly with H2O2 uptake. On the other hand, in the presence of CAT (dark blue), the growth rate is maintained stable across all the different uptake values.

Increasing the flux through CAT boosts growth in the presence of H2O2

On the other hand, we wanted to obtain a quantitative verification of design that would allow us to confirm that increasing the copies of CAT can indeed improve the response to oxidative stress of the bacterium increasing system understanding.

To calculate this, we simulated growth in the presence of a fixed amount of H2O2 for an increasing flux through CAT. This allows us to effectively determine the effect that an increasing number of copies of CAT would have on recovering growth in the presence of oxidative stress.

From the results (Figure 11) we see that as the flux increases through CAT, more H2O2 is detoxified, which enables an increase of growth rate linearly proportional to the increase of flux through CAT.


These results support the design idea of introducing multiple copies of CAT because this would increase the flux through the reaction catalyzed by the enzyme and improve detoxification, hence growth as shown by the simulation.


Figure 11. Growth rate with a fixed H2O2 uptake and increasing flux through CAT. Using growth rate as objective function, and setting both H2O2 and glucose uptake to a fixed uptake value of -5 mmol/(gDW h), we constrained the flux of CAT for each simulation to see the impact on growth rate. As the flux through CAT increases, the growth rate increases. The flux was capped at 2.5 because beyond that value, all of the uptaken H2O2 has been detoxified.


Alternatives to oxidative detoxification are not as effective as CAT, confirming our design decision


To further support design, we tested alternative detoxification pathways to confirm that the one we engineered was indeed the most efficient for improving resistance to oxidative stress. The main alternative for H2O2 detoxification reaction in the metabolism of P. fluorescens is the glutathione peridoxase (GTHPi). This enzyme catalyzes the conversion of H2O2 to H2O by coupling it to glutathione oxidation.

In order to compare GTHPi efficiency for stress resistance, we compared growth in the presence and absence of CAT for an increasing flux through GTHPi.

From the test performed (Figure 12), we have seen that, for a fixed uptake of H2O2, growth is optimal only in the presence of CAT and that GTHPi is not sufficient to support high growth in the presence of H2O2. These simulations confirm that CAT is the best candidate to target for improving oxidative stress resistance.
The reason behind these results is that glutathione oxidation through GTHPi is energetically expensive for the cell. Indeed, it requires reducing power of NADPH to re-reduce the glutathione oxidized by H2O2. This suggests that the beneficial effect on growth given by H2O2 detoxification is compensated by the negative effect on growth resulting from NADPH regeneration.


These simulations support our design decisions by confirming that CAT is the best target for improvement of oxidative stress response.


Figure 12. Growth rate with a fixed amount of H2O2 uptake and increasing flux through GTHPi in the presence and absence of CAT. In the presence of CAT (light pink), the growth rate is significantly higher than when it is absent (purple). As the flux through GTHPi increases, we observe a linear decrease in the growth rate when CAT is present. When CAT is absent, as the flux through GTHPi increases, the growth rate maintains a steady value equal to when there is no detoxification for H2O2.


The third engineered pathway implemented in the bacterium aims to introduce the ability to produce nitrates (NO3 ) from ammonia (NH3). Nitrates are the preferred form of nitrogen for plants, which means that this implementation will grant P. fluorescens even higher plant growth-promoting abilities.


    For this modification, modeling has provided:
  • Improvement of initial design through the identification of WT enzymes involved in nitrate production.
  • Greater system understanding through the quantification of NO3- impact on growth rate and NH4 uptake.

The finding of nitric oxide dioxygenase contributes to simplification of design


The initial design of this module aimed at integrating three different enzymes able to catalyze in succession the conversion of NH4 to NO3- through the following pathway (Figure 13).

Figure 13: Initial design of nitrate production pathway.

With the model, our goal was to verify the efficiency of the nitrate production pathway.

To do so, modeling results identified which enzymes were able to use as substrate the various molecules involved in the pathway to anticipate any potential difficulties which might be encountered during the engineering.

In doing so, the enzyme nitric oxide dioxygenase was identified in the model of P. putida and its presence in P. fluorescens was confirmed in the annotated genome (Table 1). This enzyme is able to catalyze the oxidation of NO to NO3, therefore making the addition of nxr for the production of nitrates no longer necessary.


In light of this finding the design of this pathway was simplified to include only amoCAB and hao.



Modeling characterizes the relation between NO3 production and growth, the consumption of NH4 and NO3 production, and determines limiting factors in NO3 production.

Growth is robust with respect to NO3- production up to a threshold value.

After improving the initial design, the next goal of the modeling with respect to this pathway is to >characterize how growth is affected by the production of NO3.

To do so, we simulated growth while maximizing NO3- production and constraint growth for a range of values.

From the results obtained (Figure 14) we see that up to a specific threshold value, high levels of NO3 production are sustained for a growing range of growth rates. This means that growth is robust with respect to NO3 production and the pathway can be integrated without significantly impacting growth.


These results allow us to characterize how the metabolism responds to the engineering of the new pathway, highlighting that within a specific range of growth rate values, NO3 production is efficient with respect to growth.


Figure 14: Optimizing for nitrate production for a range of growth rates for a fixed amount of glucose intake. Simulations were performed for a fixed uptake of glucose of -10 mmol/(gDW h) using NO3 production as objective function and constraint growth rate over a range of different values. Nitrate production remains at maximum levels up to a growth rate threshold of 0.8. Before this value, nitrate production is equivalent to when all energy resources are only dedicated to NO3 production (growth rate = 0). Beyond this value, energy becomes insufficient to support both high growth rate and nitrate production, leading to the observed sharp decline in NO3- production.

FVA reveals that a wide range of NH4 uptakes can support maximal NO3 production.

FVA reveals that a wide range of NH4 uptakes can support maximal NO3 production.

In doing so, we use FVA which allows us to compute the minimum and the maximum of the possible range of NH4 uptake values that allow NO3 production to remain in its maximal range.

Results indicate that a wide range of NH4 uptake rates can support maximal NO3 production.However, this range narrows as growth rate increases due to the rising demand for NH4 necessary to support growth (Figure 15).


This step further describes the system behavior in the context of our metabolic engineering efforts. We deepened our understanding of how the metabolism processes nitrogen to produce nitrates, while supporting growth by quantifying how the range of NH4 uptake changes according to growth rate and NO3 production rate.


Figure 15: FVA results for NH4 and N2O within 10% of maximum range of NO3 production. FVA is carried out for two reactions: NH4 uptake and N2O exchange for a 10% optimal value of NO3 production as the objective function. Simulations are performed with a fixed value of glucose uptake equal to -10 mmol/(gDW h). The range for possible uptakes of NH4 gets smaller as growth rate increases, to demonstrate the increasing energetic constraints which aim to reduce NO3 production and shut down any pathway that might not be strictly necessary for growth. Additionally, as growth rate increases the production of NO3 becomes too energetically expensive and another pathway for the consumption of NO is turned on which converts it into N2O at a lower energetic expense for the cell.

Production of NO3 is limited by O2 availability

In previous experiments we have noticed an upper bound for NO3 production around 30.9 mmol/(gDW h) and aimed at identifying the reason for this to characterize even further the pathway.

We investigated the reasons for this upper limit by alternatively modifying the uptake rates of different key compounds (including ammonia, H+, and O2 ).

We were able to establish, in silico, that the major limiting factor in NO3 production is the availability of O2 (Figure 16).


This result provides further understanding of the pathway by showing that high oxygen supply is needed to maximize NO3 production for the given design.


Figure 16: Relationship between O2 uptake and NO3 formation for a fixed amount of growth rate. We simulated NO3 production by constraining glucose uptake for a range of values and constraining growth rate to 5 1/h. We increased the lower bounds of oxygen exchange to report how much oxygen was needed to maximize NO3- production. Nitrate production is highly linked to O2 uptake, with high levels of O2 necessary to produce NO3.



Combination of the engineered pathways

    After assessing each engineered pathway on its own, and answering relevant key questions, the modeling strategy aimed at combining the engineered pathways to obtain holistic insights into system function by:
  • quantifying the relationships between NO3- production and creatinine uptake,
  • quantifying the effect of oxidative stress resistance during growth on creatinine,
  • characterizing the combined effects of oxidative stress resistance on NO3 production during growth on creatinine.


Production of NO3 is robust with respect to growth on creatinine in the absence of NH4 in the medium.

With our modeling strategy, we wanted to quantify the relationship between growth and NO3- production when modified P. fluorescens was using creatinine as its sole carbon and nitrogen source.


We performed simulations of NO3 production for different values of growth rate with creatinine being used for energy. Furthermore, these tests were performed without uptake of NH4 by the cell so that all the nitrogen necessary for NO3 production had to be produced from the metabolization of creatinine as planned by the design.


The results show the same curve shape as the simulations on glucose and NH4 (Figure 17). Such evidence highlights both the efficiency of creatinine as a source of NH4 and the ability of the bacterium to sustain growth robustly while producing NO3 using creatinine as growth substrate. Additionally, during experiments the specific pathways leading to the production of NH4 are identified to determine exactly how creatinine uptake contributes to ammonia production, thus further increasing our understanding of the proposed system.


These simulation results confirm our design with respect to the beneficial effect of the combined pathways, confirming that they complement each other because creatinine metabolization enables NO3 production. Furthermore, characterisation of the pathways that lead to NH4 production during creatinine metabolization increased our understanding of the engineered strains.


Figure 17: NO3 produced and contribution of each pathway to the production of NH4 for a range of growth rates. The simulation was performed using NO3 production as the objective function, without exchange of NH4 by the cell, constant uptake of creatinine equal to -15 mmol/(gDW h), and constraining the growth rate at increasing values for each simulation. On creatinine, growth rate is robust with respect to the production of NO4 up to a threshold value, as previously noted during growth on glucose. Additionally, we can identify that the main sources of NH4 from creatine uptake are urea degradation and the glycine cleavage system. Some of the produced NH4 is excreted, this amount increases when the energy is not sufficient to support both growth and nitrate production.

Lab measurements for NH4 production during growth on sarcosine enable predictions of NO3 production

The initial prediction of NH4 production during growth on creatinine was tested in the lab (see lab here), which provided real-world data on NH4 produced when the creA-crnA-transformed bacterium is grown on creatinine (Table 4). To verify the amount of NO3 that could be produced by the bacteria, we used this result from the lab.


Because of the stoichiometry of the process, we know that 1 mol of NO3 is produced per mol of NH4 used.


As a result, we predict a maximal production rate of NO3 equal to the production of NH4 measured in the lab: 1.51 mmol/(gDW h).


This information allows us to expand the evidence obtained in the lab with results that can both serve as predictions for future experiments (once the amoCAB-hao-transformed bacterium is obtained), and provide quantitative information on how many nitrates our modified bacterium can provide for the plants.

CAT helps growth on creatinine by detoxifying H2O2


During creatinine uptake, H2O2 is formed, which we know leads to oxidative stress. For this reason, the presence of CAT should be essential to allow optimal growth on creatinine. Using our model, we aim to characterize this beneficial effect from the combination of these two modifications.

To test our hypothesis, we compare growth on creatinine in the presence or absence of CAT. We repeat the simulation for growth on glycine to verify that the difference in growth between those conditions is due to H2O2 produced during sarcosine metabolization into glycine.

The obtained results confirm our hypothesis in showing that in the presence of CAT, growth on creatinine is higher than in its absence. They also show that there is no difference between growth with or without CAT on glycine, highlighting that the oxidative stress is indeed originating from the pathway artificially engineered by us (Figure 18).


These results allow us to identify ways in which the different designed modifications can fit together and quantify the benefits they provide to each other. They demonstrate that growth on creatinine is optimized by an efficient oxidative stress resistance pathway. The findings support design decisions by highlighting that the improvement of oxidative stress resistance supports growth on creatinine. Moreover, they increase system understanding by allowing us to quantify the benefit provided by one modification to the other.


Figure 18: Growth rate for a range of uptakes of creatinine (A) and glycine (B) in the presence and absence of CAT. Simulations were performed without NH4 uptake, constraining each time the consumption of either creatinine or glycine, and using the knock_out()[] function of cobrapy[1] to ko the CAT when necessary. Growth rate was used as an objective function for all experiments. Growth on creatinine (A) in the absence of CAT is significantly lower than when CAT is present. To confirm that this difference is due to the H2O2 produced during the creatinine uptake pathway that was engineered, we compare the differences in growth to glycine (B). Since no difference is highlighted between the presence and absence of CAT during growth on glycine, we confirm that the oxidative stress comes from the pathway modified into P. fluorescens, confirming that CAT is necessary to optimize the engineering of the creatinine metabolization pathway.

Without CAT the maximum amount of nitrate produced is lower


Finally, the last goal of the modeling strategy was to combine all the modifications to obtain a complete in silico representation of our engineered system. This allows us to quantify the effect of oxidative stress resistance on NO3- production and growth using creatinine as the sole carbon and nitrogen source.

We achieved this and were able to simulate the effect of each modification on the others by simulating how oxidative stress resistance is important to support high levels of NO3- production during growth on creatinine.

The results indeed show the beneficial effect of the activity CAT when the bacterium is grown on creatinine to produce NO3 (Figure 19).


Overall this last step allows us to simultaneously characterize the effect of each modification in combination with the others. The fact that NO3 production is supported on creatinine and optimized in the presence of a good oxidative stress response, is significant for the optimality of our combined design. The quantifications performed and the insights obtained on how these modifications complement each other further increases our understanding of our system and allows us to make predictions of nitrate production for a given amount of creatinine uptake and optimal oxidative stress resistance.


Figure 19: NO3 production for different values of growth rate on creatinine in the presence or absence of CAT. For each simulation, NH4 was not uptaken by the cell and creatinine consumption was held fixed at -15mmol/(gDW h). Growth rate was constrained at a specific value and CAT was knocked out using the knock_out() function of cobrapy1 when necessary. The objective function used was NO3 production. The curves observed have the same shape as the one observed in figure 17; notably, in the absence of CAT, the threshold value for constant NO3 production is much lower and the cell cannot produce as many nitrates as when CAT is active. This shows how each designed modification fits together to contribute to maximize the optimality of each engineered pathway.

Figure 20. Key results of the modeling strategy. For each of the single modifications we addressed the key questions for their characterization. We then combined each modification and obtained relevant insights on their joint behavior.


Conclusion

In conclusion, our modeling strategy supported and complemented the BioMoon project and proved to be an invaluable resource to its unfolding.



    We were able to achieve all the goals set up in the beginning, specifically:
  • Inform design: by testing in silico all the pathways that we implemented in the lab:
    • simplify design of nitrate production pathway to include only amoCAB and hao;
    • simplify design of creatinine metabolization pathway to include only crnA and creA, informing laboratory experiments and including laboratory results in the model;
    • confirm the choice to include multiple copies of katB to improve the oxidative stress resistance.
    We also considered alternative options to validate our decisions by:
    • comparing growth on creatinine to growth on other compounds abundant in urine, verifying that creatinine is the best option;
    • testing different oxidative stress detoxification pathways, which confirmed that CAT is the best target.
  • Contribute to system understanding by providing invaluable insights, namely:,
    • during the creatinine metabolization pathway, the production of formaldehyde contributes to growth and the formation of urea to NH4 production;
    • the quantitative relationship between creatinine and growth was established;
    • we quantified the relation between NH4 and NO3 production, and predicted that NO3 production is limited by O2 availability;
    • predict and verify the optimality of the engineered strains.
  • Generate global insights of system function and predictions of global system behavior, which showed us that, when combined, the engineered pathways positively contribute to each other, specifically:
    • oxidative stress resistance contributes to higher growth rate on creatinine;
    • creatinine metabolization supports production of NO3 and growth on creatinine is robust with respect to NO3 production;
    • oxidative stress resistance is necessary to obtain high growth rate and NO3- production when the bacterium is using creatinine as its sole carbon and nitrogen source.


In conclusion, the proposed modeling strategy was vital to the design stage by testing pathways and comparing choices against alternatives. Experimental growth rates and uptake rates were integrated to validate the predicted growth and ammonia production rates, strengthening the conclusions of the modeling. Furthermore, our strategy significantly contributed to increasing our understanding of the native and engineered strains by generating both qualitative and quantitative predictions that characterized and clarified the metabolic impact of each modification.



Future outlook

In the context of our project, an additional engineering effort has been proposed, one which aims to optimize P. fluorescens biofilm formation abilities to increase water retention of lunar regolith. The modeling strategy could contribute to this effort using FBA and metabolic modeling in a way that fits with the other simulations performed.

Once known the composition of P. fluorescens’ biofilm composition, a function would be implemented in the model to simulate all the molecules required to produce a certain amount of biofilm. Initially, we can simulate how the flux through this reaction affects growth rate to simulate how growth and biofilm formation balance each other


    Later, we can combine it alternatively with the other engineered pathways to:
  • Quantify the relationship between biofilm formation and growth on creatinine
  • Quantify the relationship between NO3 production and biofilm formation on glucose and NH4 and then on creatinine
  • Characterize the effect of oxidative stress on biofilm formation
  • Combine all four modifications to quantify combined effects.

This comprehensive modeling effort will ultimately allow us to efficiently inform how the addition of biofilm overexpression affects metabolism and how this modification complements the others.

    The modeling strategy achieved the initially set goals, which aimed at efficiently complementing the experimental part while remaining within the time constraints of an iGEM project. Following is a list of future experiments that could be performed to deepen and expand our analysis. These could provide a way to both expand in the future the results of BioMoon or serve as a base for future iGEM projects.
  • Verify in vivo the predictions relative to the role of formaldehyde in increasing the growth rate on sarcosine, creatine, and creatinine (Figure 8).
  • Implement future experimental results on nitrate production into the model. These could include the rate of consumption of NH4 and production of NO to confirm optimality of the pathway similarly to what was done for the creatinine metabolization pathway.
  • Perform the in silico experiments relative to biofilm as detailed in the above section.
  • Combine the GSMM of P. fluorescens with GSMM of the target plant species to specifically characterize the beneficial effects to plant growth brought about by our modified bacterial strain.
  • Adapt our simulations to reflect the conditions of lunar regolith. This could allow us to fully establish in silico the metabolic behavior of P. fluorescens in the specific conditions we aim to introduce it in. Additionally, this step could be combined with the one above, obtaining a full simulation of behavior of our bacterium-plant system on lunar regolith. This would provide meaningful insights which can contribute to the improvement of current design and the pinpointing of additional modifications to introduce into P. fluorescens to improve the system even more.
  • Eventually, we can design a fully complete in silico versatile pipeline to characterize behavior of bacteria and plants on a certain substrate. This would allow us to better inform our entrepreneurial efforts of using synthetic biology to develop multiple strains able to complement the specific needs of a wide range of plant species growing on different inhospitable substrates.










  1. Ebrahim, A., Lerman, J. A., Palsson, B. O., & Hyduke, D. R. (2013). COBRApy: COnstraints-Based Reconstruction and Analysis for Python. BMC Systems Biology, 7(1).Click
  2. Passi, A., Tibocha-Bonilla, J. D., Kumar, M., Tec-Campos, D., Zengler, K., & Zuniga, C. (2021). Genome-Scale metabolic modeling enables In-Depth understanding of big data. Metabolites, 12(1), 14. Click
  3. Gudmundsson, S., & Thiele, I. (2010). Computationally efficient flux variability analysis. BMC Bioinformatics, 11(1). Click
  4. Nogales, J., Mueller, J., Gudmundsson, S., Canalejo, F. J., Duque, E., Monk, J., Feist, A. M., Ramos, J. L., Niu, W., & Palsson, B. O. (2019). High‐quality genome‐scale metabolic modelling of Pseudomonas putida highlights its broad metabolic capabilities. Environmental Microbiology, 22(1), 255–269. Click
  5. Huang, X., & Lin, Y. (2020). Reconstruction and analysis of a three‐compartment genome‐scale metabolic model for Pseudomonas fluorescens. Biotechnology and Applied Biochemistry, 67(1), 133–139. Click
  6. Pseudomonas primary sequence Click
  7. Cobrapy Python package manual: Section 6 - Simulating deletions. Retrieved from Click