In Silico Analysis

Introduction

Creation of the model

Verifying stoichiometry

Validation of the model

Optimization

Supplementary Material

References

Introduction

In the synthetic biology field, metabolic modulation is a powerful tool for the study of biological networks, providing a non-intuitive insight into the biological system. In particular, genome-scale metabolic models are used to computationally describe stoichiometry and mass-balanced metabolic reactions in an organism using gene-protein-reaction associations¹. Further simulation and optimization of these models are crucial to the development of more efficient microbial cell factories with applications in the pharmaceutic, biofuel and food industries^2,3.

In this project, we aim at developing an efficient spectinabilin producing Pseudomonas putida strain. To do so, the spectinabilin production pathway was added to a P. putida genome-scale metabolic model iJN1462⁴. The model’s stoichiometry was analysed before any further work, to guarantee its stoichiometric balance. Optflux⁵, a metabolic engineering software, was then used to perform optimizations using Evolutionary Algorithm (EA), a widely used optimization tool, to find genotype solutions that maximize spectinabilin production.
Therefore, the in silico analysis team had four main goals:

creation of a Pseudomonas putida model with the required reactions for the production of spectinabilin;
check the stoichiometry balance of the newly developed model (since some original reactions were unbalanced);
validation of the model in different environmental conditions;
optimization of the production of spectinabilin, while maintaining cellular viability.

Creation of the Pseudomonas putida model to produce spectinabilin

The first step was the development of a Pseudomonas putida model able to simulate the heterologous production of spectinabilin. There are already several P. putida genome-scale models so there was no need to start from scratch. The model iJN1462⁴ was selected as the foundation for our own model since it was the most recent, complete and manually curated model for P. putida KT2440⁴, the strain used in the wet-lab experiments. Since (S)-methylmalonyl-CoA is a precursor of spectinabilin, its production pathway was also added to the iJN1462 model, connecting with the P. putida metabolism via the Krebs cycle^6,7.
The reactions added to the iJN1462 model are presented in Tables 1 and 2, with the full model available here for download.

Table 1 - Spectinabilin production reactions added to the *Pseudomonas putida* model.
Reaction Name	Substrates	Products
4-aminobenzoate N-oxygenase	4-aminobenzoate Reduced ferredoxin [iron-sulfur] cluster 2 O₂	4-nitrobenzoate Oxidized ferredoxin [iron-sulfur] cluster 2 H₂O
4-nitrobenzoate-CoA ligase	4-nitrobenzoate ATP Coenzyme A	4-nitrobenzoyl-CoA AMP Diphosphate
Spectinabilin polyketide synthase	4-nitrobenzoyl-CoA Malonyl-CoA 6 (S)-methylmalonyl-CoA 6 NADPH 4 H⁺	Demethyldeoxyspectinabilin 7 CO₂ 8 CoA 6 NADP⁺ 5 H₂O
Demethyldeoxyspectinabilin O-methyltransferase	S-adenosyl-L-methionine Demethyldeoxyspectinabilin	S-adenosyl-L-homocysteine Deoxyspectinabilin H⁺
Deoxyspectinabilin monooxygenase	Deoxyspectinabilin 4 Reduced ferredoxin [iron-sulfur] cluster 2 O₂ 4 H⁺	Spectinabilin 4 Oxidized ferredoxin [iron-sulfur] cluster 3 H₂O
Spectinabilin export	Spectinabilin_cytosol	Spectinabilin_{extracellular}

Table 2 - (S)-methylmalonyl-CoA production reactions added to the *Pseudomonas putida* model.
Reaction Name	Substrates	Products
(R)-methylmalonyl-CoA CoAcarbonylmutase	Succinyl-CoA	(R)-methylmalonyl-CoA
Methylmalonyl-CoA epimerase	(R)-methylmalonyl-CoA	(S)-methylmalonyl-CoA

Verify the stoichiometric balance of the model

After building the spectinabilin-producing model, we verified the stoichiometry of its reactions with Optflux⁵. Optflux is an open-access metabolic engineering software developed in Portugal by our PI when she was at the Centre of Biological Engineering (Universidade do Minho) and supported by SilicoLife Lda⁵. This software’s main features include simulation and optimization of stoichiometric models, but it also possesses many plug-ins with interesting capabilities, such as Balance Checker, which automatically calculates disparities in stoichiometric balance. Using this plug-in, we fixed the stoichiometry of several reactions.

Validation of the Pseudomonas putida model in different media

The second step was the validation of the spectinabilin-producing model by performing wild-type simulations with pFBA⁹, a simulation method that allows maximizing the objective function while minimizing the total sum of the remaining fluxes. Two separate objective functions were maximized: the biomass growth rate and the production of spectinabilin. The simulations were performed with three different medium formulations, all presented in the supplementary material:

Minimal medium predefined in the model;
M9 minimal medium, previously used for Pseudomonas putida;
Luria Bertani (LB) growth medium, also previously used for P. putida.

The results of these simulations are presented below in Table 3.

Table 3 - Values of the production of biomass and spectinabilin in different media.
		Objective Function
		Maximization of the production of biomass (h^-1)	Maximization of the production of spectinabilin mmol.(gDW.h)^-1
Media	Minimal medium	0.737	0.995
	M9 medium	1.430	1.579
	LB medium	1.710	3.449

Since the simulations using LB medium allowed to achieve the highest biomass and spectinabilin fluxes (1.710 h^-1 and 3.449 mmol.(gDW.h)^-1), respectively, all the optimization process was proceeded using this condition.

Optimization of spectinabilin production

Lastly, we decided to perform an under/overexpression (UO) and knockout (KO) optimization task to find an optimum genotype that provides the highest production flux of spectinabilin while maintaining a reasonable value of biomass growth rate. To do so, we used Evolutionary Algorithms (EA)⁸ with the simulation method LMOMA¹⁰ (Figure 1). EA searches for the optimal set of genomic conditions that maximizes fitness, which in our case is the flux for the production of the target compound (spectinabilin)⁷. This method is inspired by biological evolution, and uses candidate solutions to produce “offsprings” which will inherit some of the best characteristics of their “parents'', thus creating a new generation of possible solutions whose fitness will be evaluated. This continues for a set number of iterations/generations. LMOMA is a simulation method that minimizes the Manhattan distance between the flux distribution of the mutant strain and the wild-type strain¹⁰. This simulation method was chosen over the pFBA, since the latter method couldn’t find results.

Figure 1 - Optimization process flow.

The number of genomic alterations or KOs allowed in the optimization was six, to avoid solutions with a relatively higher number of modifications, which would difficult the application of those results in an experimental setting. The objective functions used were BPCY (Biomass-product coupled yield) or Yield (Product Yield with minimum biomass). The first one balances both maximization of the cellular growth and the production of spectinabilin⁸, while Yield targets the maximization of the desired compound but maintains minimum biomass levels⁸. In our optimizations that minimum was 10% of biomass.
The results obtained were filtered based on the following criteria:

spectinabilin production above 0;
biomass production above 0;
The results that have a minimal KO or UO of genes.

The best results obtained are presented in Table 4.

Table 4 - Best genomic conditions obtained by the optimization by Optflux software for the desired objective function.
Solution	Genetic technique	Gene ID	UO Factor	Protein/Enzyme	Biomass (h^-1)	Spectinabilin (mmol.(gDW.h)^-1)
1	KO	G_PP_3124	-	Short chain fatty acid transporter	0.178	1.560
		G_PP_0813	-	Cytochrome bo terminal oxidase subunit I
		G_PP_1444	-	Quinoprotein glucose dehydrogenase
		G_PP_0612	-	FAD-dependent glycine/D-amino acid oxidase
		G_PP_4191	-	Succinate dehydrogenase / fumarate reductase, flavoprotein subunit
2	UO	G_PP_3570	0.03125x	Carbohydrate-selective porin	0.339	1.294
		G_PP_4678	0.03125x	Ketol-acid reductoisomerase
		G_PP_5419	0.5x	F-type H+-transporting ATPase subunit a
		G_PP_4192	0.03125x	Succinate dehydrogenase / fumarate reductase, membrane anchor subunit
3	UO	G_PP_3570	0.03125x	Carbohydrate-selective porin	0.401	1.186
		G_PP_1620	0.03125x	5'-nucleotidase SurE
		G_PP_1665	0.5x	Phosphoribosylformylglycinamidine cyclo-ligase
		G_PP_5419	0.5x	F-type H+-transporting ATPase subunit a
		G_PP_4192	0.03125x	Succinate dehydrogenase / fumarate reductase, membrane anchor subunit
4	UO	G_PP_3570	0.03125x	Carbohydrate-selective porin	0.384	1.136
		G_PP_5419	0.5x	F-type H+-transporting ATPase subunit a
		G_PP_4192	0.125x	Succinate dehydrogenase / fumarate reductase, membrane anchor subunit

Since our objective is to increase the production of spectinabilin, we decided to focus on solution 1, the solution with the KOs, which presents the highest spectinabilin flow. The results are present in Table 5.

Table 5 - Genes KOs obtained from the optimization with the higher production of spectinabilin and respectively protein and reactions.
Gene ID	Gene KO	Protein	Reaction inactivated
G_PP_0612	thiO	FAD-dependent glycine/D-amino acid oxidase	Glycine + H₂O + O₂ ⇌ Glyoxylate + H₂O₂ + NH₄⁺
G_PP_0813	cyoB	Cytochrome bo terminal oxidase subunit I	4 H⁺_citoplasma + 0.5 O₂ + Ubiquinol-8 ⇌ H₂O + Ubiquinone-8 + 4 H⁺_periplasm
G_PP_3124	PP RS16290	Short chain fatty acid transporter	Acetoacetate_periplasm + H⁺_periplasm ⇌ Acetoacetate_citoplasma + H⁺_citoplasma
G_PP_4191	sdhA	Succinate dehydrogenase flavoprotein subunit	Ubiquinone-8 + Succinate ⇌ Fumarate + Ubiquinol-8
G_PP_1444	gcd	Quinoprotein glucose dehydrogenase	Ubiquinone-8 + D-glucose + H₂O ⇌ Ubiquinol-8 + D-gluconate + H⁺

To better understand the variation of the metabolism between the wild-type and the best mutant obtained in the optimizations (Solution 1), we decided to perform a flux analysis, presented in Table 6.

Table 6 - Reactions with higher flux values in the mutant strain versus the wild type strain.
Reaction	Variation on the flux between the mutant and wild-type strain	Reaction	Variation on the flux between the mutant and wild-type strain
D-lactate transport	39.05	Adenylate kinase	1.60
Glucose-6-phosphate dehydrogenase	15.55	4-aminobenzoate N-oxygenase	1.16
6-phosphogluconolactonase	15.55	Demethyldeoxyspectinabilin O-methyltransferase	1.16
D-glucose transport via ABC	15.41	Deoxyspectinabilin monooxygenase	1.16
Hexokinase D-glucose ATP	15.41	4-Nitrobenzoate-CoA ligase	1.16
H2O transport	15.31	3-5-cyclic nucleotide phosphodiesterase	1.16
Succinyl-CoA synthetase	12.39	R spectinabilin polyketide synthase	1.16
R cytochrome oxidase bd ubiquinol 8	11.08	4-aminobenzoate synthase	1.15
Methylmalonyl-CoA-carbonylmutase	8.09	4-amino-4-deoxychorismatesynthase	1.15
D-lactate dehydrogenase	7.17	Adenosine kinase	1.15
D-lactate transport via proton symport	7.17	Adenosylhomocysteinase	1.15
L-aspartase	3.86	Methionine synthase	1.15
Phosphate transport via ABC system	3.76	5-10-methylenetetrahydrofolate reductase NADH	1.15
L-aspartate transport in via proton symport	3.49	R methionine adenosyltransferase	1.15
L-aspartate transport in via periplasm	3.49	R Glycine Cleavage System	1.14
R NADH dehydrogenase ubiquinone 8	1.20

Through the analysis of these results, we can observe that several reactions of the spectinabilin⁶ and methylmalonyl-CoA⁷ pathway (responsible for the production of the precursor of our desired compound) have a higher metabolic flux in the mutant than in the wild-type.
Then, in order to understand which gene KO has more influence in the production of spectinabilin observed in the best mutant, we performed 5 different LMOMA simulations, each with four of the five KOs from the optimum solution. The results are present in Table 7.

Table 7 - Values of biomass and spectinabilin when we perform simulations with four of the five KOs obtained from the optimizations.
Active gene from the optimum genotype	thiO	cyoB	PP RS16290	sdhA	gcd
Biomass (h^-1)	0.250	1.092	0.194	1.204	0.837
Spectinabilin Fux (mmol.(gDW.h)^-1)	1.016	0.241	1.138	0.00013	0.479

The results show that the spectinabilin production is almost null if the sdhA gene is not knocked out, while in the remaining simulations the continuous activity of the other individual genes still leads to spectinabilin production. From this we can deduce that the KO of the sdhA gene is essential for the production of spectinabilin.

The gene sdhA codifies the succinate dehydrogenase flavoprotein subunit (Table 5) that is involved in the synthesis of fumarate from succinate of the tricarboxylic acid (TCA) cycle in the carbohydrate metabolism. Two of the reactions more active in the mutant strain compared with the wild-type strain present in Table 2 - the succinyl-CoA synthetase and methylmalonyl-CoA-carbonyl mutase - consume succinate to produce, respectively, succinyl-CoA and methylmalonyl-CoA. So, when the sdhA gene is inactivated, fumarate is no longer produced from succinate, creating a pool of succinate that can then be consumed by other pathways including the secondary metabolic pathways previously mentioned leading to a higher flux in the formation of (S)- methylmalonyl-CoA. Since this compound enters in the spectinabilin production pathway, an increase of (S)- methylmalonyl-CoA leads to an increased production of spectinabilin, as can also be observed by the spectinabilin pathway presenting a higher metabolic flux in the mutant strain. A similar result was observed by McCloskey et. al, when genes of the SUDCi reaction were knocked out there was increased flux out of the TCA, in E.coli¹².

A second result to analyse is the KO of the gene cyoB, that codifies the cytochrome bo terminal oxidase subunit I. While its activation still allows the production of spectinabilin, the values of the desired compound are lower than those of the remaining solutions. Fluxomic studies about the KO of this gene were not found, so it is unclear its role in the mutant obtained.
The results obtained from the computational analysis will support the further development of our engineered bacteria by manipulating its metabolism in order to maximize the flux production of spectinabilin. In order to implement this solution we will start by the KO of the gene sdhA, followed up by the KO of both sdhA and cyoB.

Supplementary Material

Minimal Medium

Metabolite ID	Metabolite Name	Lower Bound Value	Upper Bound Value
M_arg__L_e	L_Arginine	-2.68	999999.9
M_glu__L_e	L_Glutamate	-15.79	999999.9
M_glc__D_e	D_Glucose	-16.65	999999.9
M_pi_e	Phosphate	-6.90	999999.9
M_h_e	H	-100.0	999999.9
M_asp__L_e	L_Aspartate	-6.44	999999.9
M_o2_e	O₂	-18.50	999999.9
M_thr__L_e	L_Threonine	-3.45	999999.9
M_his__L_e	L_Histidine	-1.91	999999.9
M_ser__L_e	L_Serine	-6.18	999999.9

M9 medium

Metabolite ID	Metabolite Name	Lower Bound Value	Upper Bound Value
R_EX_glc__D_e	D_Glucose	-10.00	0.0
R_EX_o2_e	O₂	-18.50	0.0
R_EX_h2o_e	H₂O	-999999.9	999999.9
R_EX_na1_e	Na	-999999.9	999999.9
R_EX_co2_e	CO₂	-999999.9	999999.9
R_EX_cobalt2_e	Co	-999999.9	999999.9
R_EX_nh4_e	NH₄	-999999.9	999999.9
R_EX_ni2_e	Ni₂	-999999.9	999999.9
R_EX_so4_e	SO₄	-999999.9	999999.9
M_h_e	H	-999999.9	999999.9
M_fe3_e	Fe₃	-999999.9	999999.9
M_fe2_e	Fe₂	-999999.9	999999.9

LB medium

Metabolite ID	Metabolite Name	Lower Bound Value	Upper Bound Value
R_EX_lys__L_e	L_Lysine	-5.75	999999.9
R_EX_glc__D_e	D_Glucose	-16.65	0.0
R_EX_val__D_e	D_Valine	-5.82	999999.9
R_EX_thr__L_e	L_Threonine	-4.37	999999.9
R_EX_phedca_e	10_Phenyldecanoic_acid	-5.45	999999.9
R_EX_leu__D_e	D_Leucine	-2.92	999999.9
R_EX_gly_e	Glycine	-4.34	999999.9
R_EX_asp__L_e	L_Aspartate	-7.07	999999.9
R_EX_tyr__L_e	L_Tyrosine	-1.63	999999.9
R_EX_ser__L_e	L_Serine	-6.19	999999.9
R_EX_met__L_e	L_Methionine	-1.88	999999.9
R_EX_ile__L_e	L_Isoleucine	-4.48	999999.9
R_EX_glu__L_e	L_Glutamate	-15.79	999999.9
R_EX_arg__L_e	L_Arginine	-2.68	999999.9
R_EX_trp__L_e	L_Tryptophan	-0.71	999999.9
R_EX_pro__L_e	L_Proline	-9.16	999999.9
R_EX_his__L_e	L_Histidine	-1.91	999999.9
R_EX_cys__D_e	D_Cysteine	-0.64	999999.9
R_EX_ala__L_e	L_Alanine	-5.57	999999.9
R_EX_o2_e	O₂	-18.05	0.0

References

Gu, C., Kim, G.B., Kim, W.J. et al. Current status and applications of genome-scale metabolic models. Genome Biol 20, 121 (2019). doi:10.1186/s13059-019-1730-3
Wu G, Yan Q, Jones JA, Tang YJ, Fong SS, Koffas MAG. Metabolic Burden: Cornerstones in Synthetic Biology and Metabolic Engineering Applications. Trends Biotechnol. 2016;34(8):652-664. doi:10.1016/j.tibtech.2016.02.010
He Fei, Murabito Ettore and Westerhoff Hans V. (2016) Synthetic biology and regulatory networks: where metabolic systems biology meets control engineeringJ. R. Soc. Interface.13: 20151046. doi: https://doi.org/10.1098/rsif.2015.1046
Nogales J, Mueller J, Gudmundsson S, et al. High-quality genome-scale metabolic modelling of Pseudomonas putida highlights its broad metabolic capabilities. Environ Microbiol. 2020;22(1):255-269. doi:10.1111/1462-2920.14843
Isabel Rocha , Paulo Maia , Pedro Evangelista , Paulo Vilaça , Simão Soares , José P Pinto , Jens Nielsen , Kiran R Patil , Eugénio C Ferreira and Miguel Rocha. OptFlux: an open-source software platform for in silico metabolic engineering. BMC Systems Biology 2010, 4:45
Gross F, Ring MW, Perlova O, et al. Metabolic engineering of Pseudomonas putida for methylmalonyl-CoA biosynthesis to enable complex heterologous secondary metabolite formation. Chem Biol. 2006;13(12):1253-1264. doi:10.1016/j.chembiol.2006.09.014
Choi YS, Johannes TW, Simurdiak M, Shao Z, Lu H, Zhao H. Cloning and heterologous expression of the spectinabilin biosynthetic gene cluster from Streptomyces spectabilis. Mol Biosyst. doi:10.1039/b923177c
OptFlux: Metabolic Engineering Workbench.URL:http://www.optflux.org
Ponce-de-Leon, Miguel & Cancela, Héctor & Acerenza, Luis. (2008). A Strategy to Calculate the Patterns of Nutrient Consumption by Microorganisms Applying a Two-Level Optimisation Principle to Reconstructed Metabolic Networks. Journal of biological physics. 34. 73-90. 10.1007/s10867-008-9067-2.
Rocha, M., Maia, P., Mendes, R. et al. Natural computation meta-heuristics for the in silico optimization of microbial strains. BMC Bioinformatics 9, 499 (2008). doi: 10.1186/1471-2105-9-499
Schellenberger J, Que R, Fleming RM, et al. Quantitative prediction of cellular metabolism with constraint-based models: the COBRA Toolbox v2.0. Nat Protoc. 2011;6(9):1290-1307. doi:10.1038/nprot.2011.308
McCloskey, D., Xu, S., Sandberg, T.E. et al. Evolution of gene knockout strains of E. coli reveal regulatory architectures governed by metabolism. Nat Commun 9, 3796 (2018). https://doi.org/10.1038/s41467-018-06219-9