Team:Wageningen UR/Software/Testcase validation
The iGEM PIPE
Test case and validation
For Cattlelyst, we are re-engineering metabolic pathways of microbes. To aid our research, we developed a computational pipeline: the iGEM Pipeline for Improved Pathways Engineering (PIPE). The PIPE was developed while working toward a test case that was relevant for the project of WUR iGEM team 2021. Additionally, the tool was validated using three studies and comparing the results given by the PIPE with either computational or experimental results on pathway engineering.
Test Case
For the development of the iGEM PIPE design strategies on the Escherichia coli model iML1515 were investigated for allowing the chassis to grow on methane and produce L-lactate. A bacteria with such a phenotype would have two advantages. First, the uptake of methane and ammonia is the goal of our project and while ammonia can be taken up by E. coli as the preferred source of nitrogen, methane uptake has to be engineered. Second, L-lactate, the precursor of the biodegradable poly-L-lactic acid (PLA) polymer, is a commercially relevant compound [1]. L-lactate can be endogenously produced by the chassis, therefore the iGEM PIPE was used to look for alternative production strategies. We decided against producing L-lactate despite its potential feasibility, due to limiting methane concentrations. We found out that the methane concentration was the bottleneck of our biofilter design when we modelled it as we describe in our Modelling page. Nonetheless this test case shows how useful the PIPE could be when the aim is to synthetically combine the consumption of a substrate with the production of another compound. The results of this optimization are presented in the following subsections, which are organised per module of the PIPE.
The full description of the modules of the iGEM PIPE is can be read at this page and a summary is provided in Figure 1
-
Knock-in identification and evaluation
To identify which reactions have to be added to the model of E. coli for making it able to take up methane some constraints need to be used. The biomass reaction which corresponds to the growth of the microorganism is limited to the wild type growth rate and all the possible carbon sources are made unavailable except for methane. All this was done automatically by reading in a user-created input file. Flux balance analysis (FBA) was then used to simulate the growth of the microbe and as expected the result was infeasible because E. coli cannot naturally uptake methane as the carbon source. Thereby, the PIPE uses another function for searching reactions with the Gapfilling algorithm. This finds combinations of reactions that, if added to the model, enable the simulation of growth.Continues...
arrow_downward
Four solutions were found: all had one different methane monooxygenase each and reaction(s) needed for turning methanol into formaldehyde.Figure 2: Reaction knock-in strategies suggested by the iGEM PIPE for methane uptake in E. coli. The boxes represent the models. Models1-4 are obtained from adding the reactions necessary for methane oxidation to the model of E. coli. The IDs of the reactions that were added are the ones in the middle of the boxes. The numbers above the arrows indicate the fluxes through the reactions are resulted from simulating growth with FBA. When the unit is not given it corresponds to mmol/gDW/h. In this case ammonia (NH3) should be considered in its soluble form.
Figure 2 shows the four knock-in strategies that were identified with this first optimization step. For the number of iteration that was set, one solution (i.e. Model2) was found twice, the PIPE however presents a non-redundant list of results. The results were obtained using four iterations of Gapfilling in about four hours of run time on a single core Linux server. Model1 and Model2 were generated by adding a methane monooxygenase (i.e. R01143 or MMO2 respectively) and the methanol dehydrogenase (ALCD1) to iML1515 model of E. coli. The biomass reactions carried the maximum allowed flux. In correspondence to CH4 the uptake rates of methane are indicated, which are negative fluxes through the exchange reaction of methane. The more negative the value, the better the efficiency of the methane oxidation by the enzymes. In this regard, the methane consumption rates of Model1 and Model2 suggested that reaction MMO2 oxidises methane with a more efficient pathway, since a lower methane uptake led to the same growth rate. Model3 was generated adding three reactions: a methane monooxygenase together with POX2 and PRDX. These reactions were present in the reaction database and led to an inefficient methane consumption: high methane take up but low growth rate. This would be a positive scenario for our biofilter, since our goal is to get rid of the gas rather than produce biomass in a large amount. However, it is good to notice that higher growth rates are associated to more negative fluxes through ammonia’s exchange reaction, indicating higher uptake. Model4 had reaction R01142 as the methane monooxygenase and again reaction ALCD1 for methanol oxidation. L-lactate was not the objective of this optimization therefore it was not produced during this simulation of growth by FBA in either of these models.Let’s consider for a moment the reasonability of the methane uptake rates that were found. Uptake rates for E. coli growing on glucose are between 5 and 10 mmol/gDW/h. A glucose consumption rate of 10 mmol/gDW/h leads to a biomass generation rate of 0.87 h-1. Considering the difference in carbon atoms between glucose and methane [5], uptake rates of 30-100 mmol/gDW/h are reasonable for a model of E. coli growing at the same rate but on methane.
Once it was found how to make E. coli’s model to take up methane, the pipeline looks at how to make it produce the target compound, L-lactate in this case. Since it is produced endogenously, the objective function of FBA optimization is set to its exchange reaction. A positive flux was found after running FBA, that implies that the compound can be produced without the need of adding new reactions to the model. However we wanted to check that the iGEM PIPE was able to find reactions filling in a gap in production. Therefore, as a test, we deleted LCADi, the reaction responsible for L-lactate production, from the model. In this way the PIPE was basically asked to look for alternative production pathways by searching for reaction knock-ins with Gapfilling. For this analysis a maximum methane uptake needed to be fixed to get realistic results. Additionally, a minimal growth rate also had to be indicated to simulate growth of the microorganism while producing the compound of interest, as it would occur in practice. The growth rate had then to be set to a lower value than the maximal, otherwise the microorganism would not have the resources to produce the compound. Therefore, the lower bound of the biomass reactions was fixed to 5% of the growth rate on glucose (i.e. 5% of 0.87 h-1 which equals 0.044 h-1). Three reactions (i.e. LacR, LDH_L and LCADi) were alternatively found to produce lactate in the E. coli models growing on methane see Figure 3.
Figure 3: Reaction knock-in strategies suggested by the iGEM PIPE for the alternative production of lactate in models of E. coli uptaking methane. The boxes represent the models. Models1.1, 2.2 and 3.3 are three examples of the strategies found by the PIPE obtained from adding the altrnative reactions for L-lactate production in models together with the reactions that make the model capable of methane oxydation. The IDs of the reactions that were added are the ones in the middle of the boxes. The numbers above the arrows indicate the fluxes through the reactions are resulted from simulating growth with FBA. When the unit is not given it corresponds to mmol/gDW/h. In this case ammonia (NH3) should be considered in its soluble form. The maximal production rate of 0.826 mmol/gDW/h was obtained with the addition of LacR or LDH_L to this knock-in variant. In Model3 (see Figure 2), one of the transporter reactions among ASPt2r, SUCCt2b, SUCCt2r and ALCD1 were suggested in addition to the L-lactate-producing reaction (an example is provided in Figure 3, Model3.3). This result suggests that in the Model3 other products had to be co-produced with lactate. Those three cases allowed a production rate of 0.200 mmol/gDW/h.
-
Thermodynamic analysis
We wanted to use the thermodynamic analysis of the suggested knock-in strategies as a criterium to evaluate their feasibility. This criterium is commonly used in metabolic engineering to evaluate pathways and flux distributions, and existing tools also rely on it. In our pipeline the thermodynamic feasibility is assessed with the max-min driving force (MDF) algorithm [6]. For the algorithm to work, it needs to know which reactions among the whole pool of reactions included in the model of E. coli are specifically needed for the conversion of carbons from methane to L-lactate. For determining this, parsimonious flux balance analysis (pFBA) calculations were used. For an explanation of how these two algorithm are combined, visit the page on the iGEM PIPE setup.
Continues...
arrow_downwardThe ability of the strategy of combining pFBA and MDF calculation was first tested on two models for which the pathway was shorter and better defined. Formate was selected as the target compound for E. coli growing on methane, since it is a mandatory intermediate for the assimilation of methane [1]. Methane is first oxidised to methanol which in turn is oxidised to formaldehyde. The latter undergoes another oxidation step leading to formate [7]. See Figure 4 on the representation of the pathway.
Figure 4: Methane oxidation pathway. Carbon atoms are in green, hydrogen atoms in blue and oxygen atoms in orange. Its production was tested on models of E. coli with the additional reactions for methane oxidation: R01143 and MMO2 respectively. Methane and formate have a 1:1 ratio in the balanced equation:
methane + molecular oxygen + NAD+ = formate + NADH.Therefore this was reflected in the bounds of the exchange reactions of the substrate and the target. Formate can be produced endogenously thus no reactions search was carried on the two models for allowing flux through its exchange reaction. The pruned pathway of both models consisted of five reactions. The MDF value of the pathway was obtained for both models and was a positive value of 14.12 kJ/mole.
Once tested that the designed approach worked, we used it to evaluate the thermodynamic feasibility of L-lactate production pathway in the model of E. coli assimilating methane and producing lactate via the native LCADi reaction (see Figure 3, Model1.1). The raw reaction list consisted 39 active reactions in the bioconversion of methane to L-lactate. The analysis resulted in a MDF value of 3.82 kJ/mol, that is a positive number, hence the conversion can be interpreted as thermodynamically feasible.
In some cases, the calculation of MDF was possible only when the correct KEGG identifier of certain compounds was manually added to the model. This happened for instance in one of the two tests with formate as the target, since ubiquinol-8 metabolite (q8h2_c) part of the reaction MMO2 missed that identifier. The KEGG Compound identifiers are used by the equilibrator-package to interact with other packages in order to retrieve the dissociation constants of the reactants. This error could be avoided by manually adding the information to the model. However, issues were encountered in the automation of the process. One weakness of the thermodynamic evaluation approach is that pFBA is not the optimal pathway identification method. The procedure is based on several assumptions. For the identification of the active reaction in a pathway, pFBA must be performed on a carefully simplified model in order to give valuable results. Notably, we assume that cofactors are either present or absent such that these variables can be removed from the model, and the fluxes can be approximated based on core metabolic components. This model simplification can be time consuming and not always accurate. Depending on the pathway, different fine-tuning methods may be required, therefore it is hard to automate the preparation of the model for pFBA. Additionally, pFBA does not necessarily identify a unique flux distribution [8], which affects the isolation of the active reaction involved in the substrate-to-target conversion. The identification and evaluation of pathways could be facilitated by an algorithm able to enumerate all the possible pathways for a desired phenotype (e.g production of L-lactate from E. coli strain assimilating methane) and calculate their MDF values. This is exactly what OptMDFpathway does [9]. This program has been implemented as function in CellNetAnalyzer toolbox in Matlab [10]. CellNetAnalyser toolbox is being rewritten in Python, but the OptMDFpathway function still has to be implemented in this work.
-
Scoring function
The knock-in strategies allowing the simulation of growth of E. coli model grown on methane and produce L-lactate are evaluated based on five default criteria (see our page on the setup of the PIPE) as listed in Table 1. The results shown in this section compare models of more variants than the ones presented in Figure 3, which was only a selection of the results.
Continues...
arrow_downwardThe normalised weights used for the calculation of the final scores are listed (Table 1). The normalised weight sum up to 1 and are automatically calculated from the user-selected weights by the template input file.
Table 1: Criteria and weights. Two columns are used for weight exemplifying the preference for higher uptake or production capability. Criteria Weights for higher importance to consumption Weights for higher importance to production Uptake rate methane 0.48 0.16 Uptake rate ammonium 0.03 0.03 Production rate 0.16 0.48 Number of reaction knock-ins 0.16 0.16 MDF value 0.08 0.08 Pathway length 0.09 0.09
When higher importance was given to the uptake rate of methane, the weight for that criterion was set to 0.48, while production rate was weighted 0.16. The opposite were applied when we wanted to base the comparison among the strategies on production rates.
Table 2 presents the final ranking of the ten engineering strategies (i.e. Models) when the weights used are the ones favouring high consumption over high production. The scores ranged between 0 and 1 as expected. The best model had score 0.70. One can notice that some models had equal scores. This has a two-fold explanation: the uptake rate values were shared by the engineering strategies with the same reactions knock-ins for growth on methane and ammonia. Additionally, two L-lactate producing pathways (i.e. the ones involving LacR and LDH_L) resulted in the same flux value through the target’s exchange reaction. Model3.4 had the best score since it has the optimal combination of a high uptake rate (i.e. to -500 mmol/gDW/h) and a production rate that is almost the average of the values among the engineering strategies (0.397 mmol/gDW/h).
Table 2: Criteria and weights. Two columns are used for weight exemplifying the preference for higher uptake or production capability. Position Model representing design strategy Final score 1 Model3.4 0.70 2 Model3.1 0.50 3 Model3.3 0.50 4 Model3.2 0.50 5 Model2.2 0.35 6 Model2.3 0.35 7 Model2.1 0.32 8 Model1.3 0.25 9 Model1.2 0.25 10 Model1.1 0.24
However, it has to be noted that the methane oxidating pathway identified in Model3 (see Figure 2) is not the same finally used in Model3.4, since the reaction knock-ins found with the second step of Gapfilling rerouted the pathway. In Model3, R01142 was combined with PRDX for methane oxidation to formaldehyde, while Model3.4 used ALCD1 reaction for the step of methanol oxidation.
-
Restriction sites-free codon harmonization
An important component of the PIPE is to facilitate the transition from in silico design of pathways to building them in vivo. For this aim, the PIPE first looks for the EC number of the added reactions to each design.
Continues...
arrow_downwardFor instance, in Model2.2 it retrieved data on the EC numbers of reactions ALCD1, MMO2 and LacR as shown in Figure 3 and in Table 3 below.
Table 3: Matching reaction ID with EC number. Example on knock-in reactions of Model2.2. Reactions added to E. coli forming Model2.2 EC number ALCD1 1.1.1.244 MMO2 1.14.18.3 LacR 5.1.2.1 To provide a simple representation of the outcome we applied this functionality of the PIPE to a single case: a gene that did present a restriction site. In particular, Figure 5 shows the result of RS-free codon harmonisation on the sequence of a gene from Lactobacillus casei. The gene encoded for an enzyme performing the reaction with identifier LDH_L and EC number 1.1.1.27. This sequence contained RSs, while the other manually selected coding sequences find in association with reactions’ EC numbers did not contain RSs. Hence, only the CDS from L. casei is shown to exemplify the output of the modified codon harmonisation algorithm. The coding sequence obtained after the codon harmonisation from L. casei to E. coli (see Figure 5A) contained two matches with RSs used in Biobrick assembly: CTGCAG recognised by PstI restriction enzyme and GAATTC recognised by EcoRI. The 6-nucleotides sequences have been changed by substituting the first codon with a synonymous ones with the least effect on the CHI. CTG was substituted with TTA at position 27 of the harmonised sequence while GAG replaced GAA at position 552. The change in codons affected the CHI of the harmonised sequence resulting in a slightly higher value (i.e. 0.0779 instead of 0.0740).
Figure 5: Results of the RSs-free codon harmonisation on a CDS matching the function of reaction LDH_L. This figure represents the output of RSs-free codon harmonisation onL. casei’s gene sequence for LDH_L. Figure 5A indicates the codon harmonised sequence of L. casei gene corresponding to LDH_L reaction. Figure 5B is the codon harmonised RSs-free version. The green nucleotides in Figure 5A correspond to PstI RS, while the yellow sequence is the RS for EcoRI. CHI_0 is the CHI of the native sequence. In Figure 5A CHI is the CHI of the sequence after the harmonisation. While in Figure 5B CHI indicates the CHI of the harmonised sequence in which the RSs have been eliminated by codon substitutions. Abbreviations: Codon Harmonization Index (CHI), Restriction site (RS).
Conclusions
The WUR iGEM team 2021 created a pipeline for strain design that aims a facilitating the experimental application of the in silico suggested strategies. We built the iGEM PIPE following an original test case, that is linked to the objective of our project: reducing emissions of methane and ammonia. This test case more specifically is connected to the “mono-culture” approach, that wants to achieve the reduction of emissions by using a unique strain. The iGEM PIPE is divided into seven modules which have been tested singularly and/or in combination with each other. The PIPE is still missing a graphic user interface, however tutorials and a manual are meant to make up for it.
If you are interested in collaborating for the improvement of the iGEM PIPE feel free to access it at our GitHub repository and get in touch with our team!
Validation
So far we described how we developed our tool following the original engineering objective of an E. coli strain able to grow on methane and produce L-lactate. We had to test whether the tool gave valid results by validating its performance using research questions with known answers. Below we present the three validation cases we used.
-
Growth coupled production of itaconic acid in Escherichia coli
arrow_downwardThis test case replicated the approach used to validate OptCouple tool [11]. The aim was to find design strategies for making E. coli able to produce itaconic acid in a growth-coupled manner while glucose was used as the carbon source. In this case therefore the pipeline had to be used only to find strategies that allow the model of E. coli to produce itaconic acid. This engineering objective did not require reaction knock-ins for the uptake of the substrate since glucose was chosen. A strain with this ability was obtained experimentally by knocking-out reactions corresponding to isocitrate lyase (ICL), succinyl-CoA synthase (SUCOAS), pyruvate kinase (PYK) and phosphotransacetylase (PTAr) in a strain of E. coli expressing aconitate decarboxylase enzyme [12]. For reproducing this output with the PIPE, the reaction aconitate decarboxylase (ACDCX) had been included in the input file for the expansion of the universal model (see the main page of the iGEM PIPE to refresh this concept). This reaction corresponds to the decarboxylation of aconitic acid into itaconic acid. The search for reactions on the universal model suggested the addition of ACDCX for the production of the target (i.e. itaconic acid) in E. coli, in line with research by Harder et al. [12]. This test case resulted in only one knock-in engineering strategy, therefore no ranking was needed.
Successively, strategies for coupling itaconic acid flux to the production of biomass were explored with the Optknock algorithm [4]. As noticed in the validation of OptCouple [11], the complete model of E. coli did not allow sensible results since only transporters were suggested as the knock-outs. Therefore the identification of strain designs was conducted on E. coli’s core model [13]. For this analysis a maximum of twelve reaction knockouts was set for the Optknock algorithm implemented in the PIPE. The found strategies consist in a minimum of 3 and a maximum of 9 reaction knock-outs. The strain design approach used in vivo involved four gene knock-outs corresponding to reactions SUCOAS, PTAr, ICL and PYK. The strain design strategies for production of itaconic acid in E. coli identified by the pipeline (see Table 6) lead to similar observations to the one made by Jensen et al. [11] when evaluating the results of OptCouple. Both tools identified the knock-outs used in the strategy tested in vivo. Overall, the candidate knock-outs identified by Optknock rerouted the Tricarboxylic Acid cycle toward aconitate formation, interfered with gluconate formation, which depletes intermediates of the TCA cycle, and blocked the glyoxylate shunt. Additionally, the elimination of reactions consuming acetyl-CoA, such as ACALD2x, were identified. Such modification is coherent with the aim of blocking the formation of secondary metabolite in order to focus the cellular resource toward the target. These results matched the ones found with OptCouple and the strategy used for the engineering of E. coli [11], [12].
Some of the knock-out reactions suggested by Optknock were unexpected. Transport via diffusion, reactions involved in cofactor balancing and ATP synthase are examples of reactions that should be excluded from Optknock’s results, since it is not reasonable to eliminate them from the microorganism, which would then not survive. The PIPE has then been improved by modifying the function call of Optknock and making it disregard by default the mentioned types of reaction as well as exchange reactions.
-
Pseudomonas putida growing on D-xylose
arrow_downwardThe aim of this validation case was to find engineering strategies allowing P. putida to grow on D-xylose as the sole carbon source. Experimentally, this was done by Elmore et al. [14] knocking-in only three genes corresponding to the following reactions: D-xylose isomerase (XYLI1), xylulokinase (XYLK ), D-xylose:H+ symporter (XYLt2). The analysis did not produce results after more than 64 hours of run, probably because of the many combination of reactions from the universal model able to yield D-xylose production. Therefore, this validation strategy was conducted on a mock universal model. This included:
- the three reactions matching the design used in vivo,
- a second candidate pathway
- and unrelated reactions (methane oxidation to formaldehyde)
The alternative D-xylose degradation pathway consisted of three reactions. These allow the reduction of xylose to xylitol (XYLR) which then reconnects to the pentose phosphate pathway when converted into xyluslose-5-phosphate (XYLTD_D). The unrelated pathway was also included to test the specificity of the iGEM PIPE in distinguishing relevant solutions among the reactions in the universal model. Both D-xylose degrading strategies were found when the analysis was carried out. The search of reactions enabling assimilation of D-xylose was done setting three iterations of Gapfilling. The results are summarised in Figure 6. The two alternative strategies are indicated as P. putida D-xyl1 and P. putida D-xyl2.
Figure 6: Reaction knock-in strategies suggested by the iGEM PIPE for growth of the model iJN1463 of P. putida on D-xylose. The boxes represent the models derived by the two different suggested strategies: P. putida D-xyl1 and P. putida D-xyl2. The IDs of the reactions that were added are the ones in the middle of the boxes. The numbers above the arrows indicate the fluxes through the reactions are resulted from simulating growth with FBA. When the unit is not given it corresponds to mmol/gDW/h. The production rate of 2-ketoglutarate (the number above the blue arrow) is obtained by setting the objective of the model to be the target instead of simulating growth of the model.
The strategy that was proved to work experimentally [14] corresponded to P. putida D-xyl1, whereas the second solution identified the alternative D-xylose degradation pathway. The former had a higher uptake rate indicated as a lower negative flux through the exchange reaction. The transporter reaction XYLt2 was added to the model by default before the optimization analysis, thereby it is not indicated among the reactions knock-ins. As expected, the reactions for methane and methanol oxidation were not included among the candidate solutions for the engineering objective.
In the study where the pathway was engineered in vivo, the metabolite that was considered to be the end product of D-xylose degradation was 2-ketoglutarate (akg) [14]. Therefore, 2-ketoglutarate was indicated as the target in the input file, however, if the assimilation pathway of D-xylose is successful, this compound is produced without the need of further reaction knock-ins. Indeed, once it was set as the objective of the maximisation problem, FBA could find a solution (see column H) without the need of Gapfilling.
-
P. putida growing on L-arabinose
arrow_downwardThe third validation case aimed at reproducing the engineering design for a strain of P. putida growing on L-arabinose as the only carbon-source. This had been experimentally achieved in the same study that engineered the degradation pathway of D-xylose. The modifications were studied on a strain already able to utilise D-xylulose, hence P. putida D-xyl1 from Figure 6 was the reference model for this validation case. In the study of Elmore et al. [14] two arabinose degradation pathways were compared in vivo. They were characterized by:
- Isomerase enzyme and 3 gene knock-ins
- Dehydrogenases in an oxidative pathway requiring 4 gene knock-ins
L-arabinose isomerase (ARAI),
L-ribulokinase (RBK_L1)
L-ribulose-phosphate 4-epimerase (RBP4E)
L-arabinose-1-dehydrogenase-rnx (ARABD),
L-arabinolactonase (LARLC),
L-arabinoate dehydratase (LARNDH),
L2keto3-deoxyarabinoate dehydratase (L2K3dARNDH).
This validation was also conducted using a mock reaction database composed from the input file. All the mentioned reactions were included in the input file, as well as the reactions for methane oxidation to formaldehyde, to check for the specificity of the reaction search. The starting strain for the optimization was the one growing on D-xylose, therefore the reactions EX_xyl__D_e, XYLI1, XYLK, XYLt2 where directly included in the model iJN1463 of P. putida.
Figure 7: Reaction knock-in strategies suggested by the iGEM PIPE for growth of the model iJN1463 of P. putida on L-arabinose. The boxes represent the models derived by the two different suggested strategies: P. putida L-ara1 and P. putida L-ara2. The IDs of the reactions that were added are the ones in the middle of the boxes. The numbers above the arrows indicate the fluxes through the reactions are resulted from simulating growth with FBA. When the unit is not given it corresponds to mmol/gDW/h. The production rate of 2-ketoglutarate (the number above the blue arrow) is obtained by setting the objective of the model to be the target instead of simulating growth of the model.
The PIPE was used to find the reactions of L-arabinose degradation pathways with Gapfilling. The results with the two knock-in strategy are shown in Figure 7. Two solutions were found, one involving the addition of the isomerase pathways for L-arabinose degradation (P. putida L-ara1), represents the variant with the oxidative pathway (P. putida L-ara2). The latter has a higher uptake rate of D-xylose (i.e. -27 mmol/gDW/h) compared to the flux recorded when the isomerase pathway was used (i.e. -20 mmol/gDW/h). 2-ketoglutarate was the target in this case too. As before, FBA for the maximisation of product formation led to flux values through the exchange reaction without searching for reaction knock-ins.
Conclusions
The first validation case compared the performance of our pipeline with another tool for the identification of reaction deletions for coupling growth with production. This approach help up specifically in improving the implementation of OptKnock algorithm in the PIPE.
The second and third validation test cases confirmed the ability of the iGEM PIPE of finding experimentally proven solutions. The strategy to make the strain of P. putida grow on D-xylose was identified. In the MetaCyc database [15] there are six pathways for D-xylose degradation, among which one is synthetically engineered. An alternative D-xylose degradation pathway was included in the input file to assess the ability of the PIPE of identifying other combinations of reactions conferring the same selected phenotype. The alternative pathway involved the conversion of the substrate to D-xylitol and the successive xylitol degradation steps. This pathway required three reaction knock-ins, instead of the two needed when the isomerase was used. Therefore, the solution was found only when the number of iterations of the algorithm was increased to three. A similar observation can be done on the results of the third validation case for P. putida growing on L-arabinose. The first solution, the isomerase pathway for arabinose degradation, consisted of three reactions; whereas the second strategy, the oxidative pathway, was composed of four reactions and resulted from the third iteration. The specificity of the strategy for the search of reaction was also proved by the last two validation cases. In none of the two the reactions for methane oxidation were identified as candidate solutions.
-
References
arrow_downward- C. A. Henard, H. Smith, N. Dowe, M. G. Kalyuzhnaya, P. T. Pienkos, and M. T. Guarnieri, “Bioconversion of methane to lactate by an obligate methanotrophic bacterium,” Sci. Rep., vol. 6, no. 1, pp. 1–9, Feb. 2016, doi: 10.1038/srep21585.
- V. S. Kumar and C. D. Maranas, “GrowMatch: An Automated Method for Reconciling In Silico/In Vivo Growth Predictions,” PLoS Comput Biol, vol. 5, no. 3, p. 1000308, 2009, doi: 10.1371/journal.pcbi.1000308.
- J. L. Reed et al., “Systems approach to refining genome annotation,” Proc. Natl. Acad. Sci. U. S. A., vol. 103, no. 46, pp. 17480–17484, Nov. 2006, doi: 10.1073/pnas.0603364103.
- A. P. Burgard, P. Pharkya, and C. D. Maranas, “OptKnock: A Bilevel Programming Framework for Identifying Gene Knockout Strategies for Microbial Strain Optimization,” 2003, doi: 10.1002/bit.10803.
- T. J. Lawton and A. C. Rosenzweig, “Methane-Oxidizing Enzymes: An Upstream Problem in Biological Gas-to-Liquids Conversion,” J. Am. Chem. Soc., vol. 138, no. 30, pp. 9327–9340, Aug. 2016, doi: 10.1021/jacs.6b04568.
- E. Noor, A. Bar-Even, A. Flamholz, E. Reznik, W. Liebermeister, and R. Milo, “Pathway Thermodynamics Highlights Kinetic Obstacles in Central Metabolism,” PLoS Comput. Biol., vol. 10, no. 2, p. e1003483, Feb. 2014, doi: 10.1371/journal.pcbi.1003483.
- H. J. M. Op den Camp, S. S. Mohammadi, A. Pol, and P. F. Dunfield, “Verrucomicrobial methanotrophs,” in Methane Biocatalysis: Paving the Way to Sustainability, Springer International Publishing, 2018, pp. 43–55.
- N. E. Lewis et al., “Omic data from evolved E. coli are consistent with computed optimal growth from genome‐scale models,” Mol. Syst. Biol., vol. 6, no. 1, p. 390, Jan. 2010, doi: 10.1038/msb.2010.47.
- O. Hädicke, A. von Kamp, T. Aydogan, and S. Klamt, “OptMDFpathway: Identification of metabolic pathways with maximal thermodynamic driving force and its application for analyzing the endogenous CO2 fixation potential of Escherichia coli,” PLoS Comput. Biol., vol. 14, no. 9, p. e1006492, Sep. 2018, doi: 10.1371/journal.pcbi.1006492.
- A. von Kamp, S. Thiele, O. Hädicke, and S. Klamt, “Use of CellNetAnalyzer in biotechnology and metabolic engineering,” Journal of Biotechnology, vol. 261. Elsevier B.V., pp. 221–228, Nov. 10, 2017, doi: 10.1016/j.jbiotec.2017.05.001.
- K. Jensen, V. Broeken, A. S. L. Hansen, N. Sonnenschein, and M. J. Herrgård, “OptCouple: Joint simulation of gene knockouts, insertions and medium modifications for prediction of growth-coupled strain designs,” Metab. Eng. Commun., vol. 8, p. e00087, Jun. 2019, doi: 10.1016/j.mec.2019.e00087.
- B. J. Harder, K. Bettenbrock, and S. Klamt, “Model-based metabolic engineering enables high yield itaconic acid production by Escherichia coli,” Metab. Eng., vol. 38, pp. 29–37, Nov. 2016, doi: 10.1016/j.ymben.2016.05.008.
- J. D. Orth, B. Ø. Palsson, and R. M. T. Fleming, “Reconstruction and Use of Microbial Metabolic Networks: the Core Escherichia coli Metabolic Model as an Educational Guide,” EcoSal Plus, vol. 4, no. 1, Sep. 2010, doi: 10.1128/ecosalplus.10.2.1.
- J. R. Elmore et al., “Engineered Pseudomonas putida simultaneously catabolizes five major components of corn stover lignocellulose: Glucose, xylose, arabinose, p-coumaric acid, and acetic acid,” Metab. Eng., vol. 62, no. February, pp. 62–71, 2020, doi: 10.1016/j.ymben.2020.08.001.
- R. Caspi et al., “The MetaCyc database of metabolic pathways and enzymes-a 2019 update,” Nucleic Acids Res., vol. 48, pp. 445–453, 2019, doi: 10.1093/nar/gkz862.