Spirulina-in Silico-Mutations and Their Comparative Analyses in the Metabolomics Scale by Using Proteome-Based Flux Balance Analysis

This study used an in silico metabolic engineering strategy for modifying the metabolic capabilities of Spirulina under specific conditions as an approach to modifying culture conditions in order to generate the intended outputs. In metabolic models, the basic metabolic fluxes in steady-state metabolic networks have generally been controlled by stoichiometric reactions; however, this approach does not consider the regulatory mechanism of the proteins responsible for the metabolic reactions. The protein regulatory network plays a critical role in the response to stresses, including environmental stress, encountered by an organism. Thus, the integration of the response mechanism of Spirulina to growth temperature stresses was investigated via simulation of a proteome-based GSMM, in which the boundaries were established by using protein expression levels obtained from quantitative proteomic analysis. The proteome-based flux balance analysis (FBA) under an optimal growth temperature (35 °C), a low growth temperature (22 °C) and a high growth temperature (40 °C) showed biomass yields that closely fit the experimental data obtained in previous research. Moreover, the response mechanism was analyzed by the integration of the proteome and protein–protein interaction (PPI) network, and those data were used to support in silico knockout/overexpression of selected proteins involved in the PPI network. The Spirulina, wild-type, proteome fluxes under different growth temperatures and those of mutants were compared, and the proteins/enzymes catalyzing the different flux levels were mapped onto their designated pathways for biological interpretation.


Introduction
Growth temperature stress causes biochemical changes in cells and reductions in biomass yield. Proteome analysis in Spirulina (Arthrospira platensis), a cyanobacterium, was performed to explore the changes at the protein level and the protein interaction cascade when the cells undergo many cellular modifications under thermal stress conditions. In previous proteome-wide studies, the identification of proteins regulated at both the translational and post-translational levels, that is, bi-level-regulated, could ideally represent the key signaling proteins that played a significant role in low-and high-temperature responses. The key two-component system (TCS) proteins SPLC1_S041070 (histidine kinase Hik28), SPLC1_S082010 and SPLC1_S230960 were identified as bi-level-regulated proteins and were linked to under the stress conditions reported by our group [17][18][19], to understand the metabolic capabilities of the cells, through computational simulations and prediction of the response of the cyanobacteria to particular growth conditions. The proteome-based constraint method is applied to the newly expanded GSMM of A. platensis C1 to predict the flux distribution under three different temperature conditions to simulate the biological processes that occur under the temperature change. The model was constructed to simulate the genome-scale metabolic fluxes at the optimal temperature (35 • C) and under low-and high-temperature stress conditions (22 • C and 40 • C, respectively). The comparison of simulated metabolic flux under the three growth temperatures was performed to explore the reactions in metabolic pathways affected by the temperature changes. Moreover, the proof of concept was carried out by in silico knockout of the proteins involved in the temperature response mechanism regulated by the key signaling protein reported earlier, histidine kinase Hik28. The simulated fluxes of the knockout mutants were analyzed. The cascade of temperature stress response mechanisms at the metabolome-wide level was revealed. In the near future, the metabolic fluxes simulated from the proteome-based GSMM may lead to flux-oriented manipulation for maximum biomass production under environmental stress.

Proteome-Based Constraint Method
The protein expression level of A. platensis C1 was obtained from a proteomics analysis conducted with the three growth temperatures: an optimal temperature of 35 • C (T35), a low temperature of 22 • C (T22) and a high temperature of 40 • C (T40). After normalization of the quantitative proteome, the log2-fold changes in protein between the experimental temperature condition and control groups were computed. Then, a two-sample t-test was performed for statistical analysis. Proteins with a log-2-fold change greater than 1.2 and p < 0.05 were considered differentially expressed (DE) proteins [17][18][19]. Three DE protein lists were prepared from the comparisons between the stress temperatures, 22 • C and 40 • C, and the control temperature, 35 • C: the T22, T40 and T35 DE lists. The normalized expression level of the identified significantly DE proteins (according to DE protein list) was then used to determine the reaction flux constraints following the E-flux method. To determine the optimal flux distribution, linear programming was used to maximize the biomass, which was the objective function. A schematic of the proteome-based metabolic model construction is shown in Figure 1. cultures grown under the stress conditions reported by our group [17][18][19], to understand the metabolic capabilities of the cells, through computational simulations and prediction of the response of the cyanobacteria to particular growth conditions. The proteome-based constraint method is applied to the newly expanded GSMM of A. platensis C1 to predict the flux distribution under three different temperature conditions to simulate the biological processes that occur under the temperature change. The model was constructed to simulate the genome-scale metabolic fluxes at the optimal temperature (35 °C) and under low-and high-temperature stress conditions (22 °C and 40 °C, respectively). The comparison of simulated metabolic flux under the three growth temperatures was performed to explore the reactions in metabolic pathways affected by the temperature changes. Moreover, the proof of concept was carried out by in silico knockout of the proteins involved in the temperature response mechanism regulated by the key signaling protein reported earlier, histidine kinase Hik28. The simulated fluxes of the knockout mutants were analyzed. The cascade of temperature stress response mechanisms at the metabolome-wide level was revealed. In the near future, the metabolic fluxes simulated from the proteome-based GSMM may lead to flux-oriented manipulation for maximum biomass production under environmental stress.

Proteome-Based Constraint Method
The protein expression level of A. platensis C1 was obtained from a proteomics analysis conducted with the three growth temperatures: an optimal temperature of 35 °C (T35), a low temperature of 22 °C (T22) and a high temperature of 40 °C (T40). After normalization of the quantitative proteome, the log2-fold changes in protein between the experimental temperature condition and control groups were computed. Then, a two-sample t-test was performed for statistical analysis. Proteins with a log-2-fold change greater than 1.2 and p < 0.05 were considered differentially expressed (DE) proteins [17][18][19]. Three DE protein lists were prepared from the comparisons between the stress temperatures, 22 °C and 40 °C, and the control temperature, 35 °C: the T22, T40 and T35 DE lists. The normalized expression level of the identified significantly DE proteins (according to DE protein list) was then used to determine the reaction flux constraints following the E-flux method. To determine the optimal flux distribution, linear programming was used to maximize the biomass, which was the objective function. A schematic of the proteome-based metabolic model construction is shown in Figure 1.   To incorporate the proteomic data into the metabolic flux analysis, we followed the process of the E-flux method, as described in Colijn et al. and Brandes et al. [20,21]. For each reaction, the reaction bounds were approximated proportionally to the proteomic level of the genes associated with each reaction. Proteomic data were mapped to corresponding reactions based on the gene-protein-reaction (GPR) association. When a protein has a measured expression level of value p, the allowable flux value through the respective reaction is set to [-p, +p] and [0, +p] for the reversible and irreversible reaction, respectively. When either the proteome or GPR for a certain reaction was unavailable, then the upper bounds were not constrained (a defined value was used in the model). If the reaction was catalyzed by multiple genes in an enzyme complex, the Boolean expression formula corresponding to enzyme complexes was utilized for that reaction. The mathematical operation for the Boolean operators "AND" and "OR" relationships were calculated as follows: X "AND" Y: the minimum value of the two values; Min (X, Y); X "OR" Y: the maximum value of the two values; Max (X, Y).
After constraining the metabolic model using the experimental proteomic data, the flux distribution of the model was predicted by solving the following linear optimization problem: where Z, the objective function, is the biomass production; C' is a row vector of the coefficient that defines the weight of each flux in the objective function; S is the stoichiometric matrix; v is a flux vector of n reactions, i.e., v = [v i ; I = 1, 2, 3 . . . , n]; and L i and U i are the lower and upper flux bounds, respectively, for each reaction i, which is an approximation based on the normalized expression of the gene catalyzed in that reaction.
Since there were multiple feasible solutions within the constraints, in this case, the method found a unique metabolic flux distribution among the alternative optima by minimizing the total sum of absolute flux. Based on the assumption that the cell attempts to achieve its objective function while allocating the minimum amount of resources; where v(θ) i is a vector of reaction flux prediction given the proteome set (θ).

Metabolic Flux Simulation of the in Silico Knockout and Overexpression of Genes
According to the proteome-based genome-scale metabolic model (GSMM), the metabolic flux of A. platensis C1 under various in silico gene knockouts and overexpression was simulated. The gene knockout condition was simulated in the model by restricting the flux through the designated reactions by setting the upper and lower bounds of the corresponding reaction to zero. in silico overexpression was simulated by increasing or doubling the level of the upper and lower bounds of the corresponding reaction.

Comparative Analysis of the Metabolic Fluxes and Pathway Mapping
The present study is the first to present a comparative analysis of the metabolic fluxes obtained from proteome-based GSMM. Thus, the data normalization and the cut-off criteria for the significant change of the flux level have not been determined to date. In the case of flux values equaling zero, the original simulated values were used in the comparative analysis. Otherwise, the simulated fluxes obtained from the model were subjected to a log 2 function. The log 2 -flux values were normalized with the standard deviation (SD) of each dataset. Second, the difference in the log 2 of the simulated flux obtained from Conditions 1 and 2 was calculated as the fold change value. Third, the cut-off values for the significant change of the simulated metabolic fluxes compared between the two conditions were determined based on the proportion of the significant change reactions less than 30% of the total reaction number. Accordingly, in the case of temperature stresses and in silico knockout or overexpression, when the log 2 values of simulated flux obtained from T35, T22 or T40 were used as the basal level, the cut-off fold change for the comparative analysis was 1.206.
Based on the KEGG global metabolic pathway as a template, the simulated metabolic fluxes of each condition were mapped on the designated reactions by using the KEGG Pathway Mapper tool. In the result images, other reactions that were not involved with the fluxes were manually eliminated.

Comparative Analysis of the Simulated Proteome-based Fluxes and Their Correlation with Temperature Stress
The proteome-based constraint method is applied to the newly expanded GSMM as shown in Figure 1 and the metabolic fluxes simulated from proteome-based GSMM are shown in Table S1. The data clearly showed a decrease in the simulated biomass at 22 • C and 40 • C compared to that at the optimal temperature, 35 • C, which was well correlated with the experimental data obtained previously (Table S2). Regarding the overview of metabolic pathways, the simulated fluxes were mapped with the corresponding KEGG reactions, and the changes in global metabolism under the two stress temperatures are illustrated in Figure 2 and Figure S1.
To investigate the correlation of the temperature change and the simulated metabolic fluxes, scatter plots of the flux distribution were plotted between flux distributions at various temperature conditions. All the flux distribution profiles show a positive trend with a high correlation ( Figure S2).

Simulated Flux Distribution Represented by Pathway Mapping
The simulated flux distribution was analyzed by pathway mapping using KEGG pathways as the templates ( Figure S3). The affected flux levels that significantly changed after the temperature shift from 35 to 22 • C or from 35 to 40 • C were mapped onto the designated pathway. Then, the comparative results are summarized in Figure 3.
More than 90% of the flux level in some central metabolic pathways, e.g., glycolysis, the Calvin cycle and the pentose phosphate pathway, was decreased or unchanged when comparing the optimal-and high-temperature conditions. In contrast, an increasing flux level was observed in the central metabolic pathways after the low-temperature shift. However, the comparative analysis showed a decreasing level of fluxes in the TCA cycle, and as a result, 2-OG biosynthesis was possibly reduced under low-temperature stress. Moreover, the flux levels of the two glutamate (Glu)-synthesizing enzymes glutamate synthase and aspartate transaminase, one using 2-OG and the other using oxaloacetate as substrates, were upregulated at 22 • C. Cells 2020, 9, x FOR PEER REVIEW 6 of 12 OG was also significantly enhanced as a result of the in silico overexpression of ICDHy. However, the biomass of these two mutant models was not changed. https://www.kegg.jp/kegg/pathway.html#global, by using the KEGG Mapper tool. The key enzymes (star-shaped nodes) are annotated with numbers, which are in the list of Figure 3. Major metabolites (black-bordered white circles) are labeled in gray boxes, while the pathway names are labeled in violet. The blue, red and light green lines represent upregulated, downregulated and unchanged or insignificantly changed fluxes. The gray line represents the reaction with a simulated flux level of ≤0. In the case of overexpression of the reaction catalyzed by GLUS3 and GLNS, the simulated biomass was increased only in the GLUS3-overexpressing model. Similarly, the simulated fluxes involved in chlorophyll biosynthesis, photosystem II and fatty acid metabolism were induced after in silico knockout and overexpression (Table S3). It should be noted that the doubled level of GLUS3 flux has a positive effect on the majority of metabolic pathways, including cell biomass.

In silico Knockout and Overexpression of Hik28 Client Proteins and the Proteins at the Interconnection of C-and N-Metabolism
The fluxes of the reactions catalyzed by three enzymes, isocitrate dehydrogenase (ICDHy), ferredoxin-dependent glutamate synthase (GLUS3) and glutamine synthetase (GLNS), which were found in the group of client proteins of Hik28 in the PPI network ( Figure S4), were subjected to in silico knockout and overexpression. These three enzymes are directly involved with the 2-OG level. The changes in the overall simulated flux levels are presented in Table S3.
For example, an overview of metabolic flux analysis after the in silico knockout and overexpression of the reaction catalyzed by ICDHy, which presumably caused the reduction and enhancement of 2-OG, respectively, showed a decreasing flux of approximately 5% of the total metabolic reactions for both ICDHy mutants. However, approximately 4% of the flux level in the ICDHy-knockout and 7% of the ICDHy-overexpressing models was increased, and among these reactions, the flux catalyzed by GLUS3 was increased in the ICDHy-knockout model and vice versa in the ICDHy-overexpressing model. It is worth noting that the fluxes involved in CO 2 uptake and oxygen evolution in photosystem II were induced in the ICDHy-overexpressing model, whereas an increase in citrate exchange flux was observed in both mutant models. Moreover, the exchange of 2-OG was also significantly enhanced as a result of the in silico overexpression of ICDHy. However, the biomass of these two mutant models was not changed.
In the case of overexpression of the reaction catalyzed by GLUS3 and GLNS, the simulated biomass was increased only in the GLUS3-overexpressing model. Similarly, the simulated fluxes involved in chlorophyll biosynthesis, photosystem II and fatty acid metabolism were induced after in silico knockout and overexpression (Table S3). It should be noted that the doubled level of GLUS3 flux has a positive effect on the majority of metabolic pathways, including cell biomass.

Proteome-Based GSMM
To the best of our knowledge, this study is the first to employ experimental fold change proteomic datasets to represent relative change in Spirulina metabolism under an optimal growth temperature and suboptimal stress temperatures. The expression levels of significantly DE proteins were used as constraints for the new Spirulina metabolic flux model to provide an in silico-simulated snapshot of the metabolic changes that occur under different temperature conditions. Whether proteomics data are hard to obtain, and quantitative proteomic data is not widely available, the proteomics experiments are more precise and more suitable for metabolic integration than transcriptomics approaches. As reaction activities will correlate more strongly with protein abundance than with mRNA abundance, flux bound should be more correlated with protein expression than with mRNA expression. The estimated results from the integration of the proteomic dataset with the metabolic model enable more reliable metabolic flux predictions. Moreover, the estimated flux from the integration of a statistically significant differential expressed proteomic dataset into the metabolic model could refine our understanding of biological changes in Spirulina under temperature stress.
Accordingly, by using our proteome selection strategy, the flux predictions were in good agreement with experimental data obtained by our group and others, as described below. The simulated metabolic fluxes under the optimal temperature and the suboptimal stress temperatures obtained from the proteome-based GSMM correspond well to the experimental data obtained earlier. For example, the highest biomass of A. platensis C1 was observed under the optimal temperature compared to that under the stress conditions, and the biomass of the culture grown at 22 • C was higher than that at 40 • C (Table S2). Moreover, the fluxes in fatty acid desaturation under the stress temperatures confirmed the fatty acid composition obtained from the cultures under the designated growth temperatures [22,23].

Possible Biological Meaning Add Link Between Stories
Our previous proteome analysis of A. platensis C1 in response to temperature stress led to the hypothesis that the low-temperature stress represented a stress response pattern similar to that of nitrogen starvation due to changes in protein expression levels involved in nitrogen assimilation, e.g., GLUS3 [17][18][19]. According to the present proteome-based GSMM simulated using proteome data obtained from T22, the flux of GLUS3, aspartate transaminase (AspTA) and nitrite exchange involved in N assimilation were increased. Whereas the fluxes of the enzyme were reported to play a role under carbon limitation, ribulose-bisphosphate oxygenase and -carboxylase were reduced [1]. Moreover, under both temperature stresses, the simulated fluxes in chlorophyll biosynthesis and photosynthesis were reduced (Table S1), which corresponded well to the biomass reduction found in silico and in experiments. The data demonstrated that applying the proteome data obtained under stress to GSMM led to the reliable biological meaning of the simulated fluxes.
Proteome-based GSMM was then applied for in silico knockout and overexpression to mimic the biological conditions of interest. Under nitrogen starvation, the 2-OG level synthesized by a reversible enzyme, iCDHy, was reported to play a critical role as the signal molecule, which was reported to control the C/N balance, resulting in the N assimilation of the cells [24]. Thus, the increasing level of 2-OG was mimicked by in silico overexpression of the reaction that synthesizes the 2-OG performed by iCDHy, which presumably caused the accumulation of 2-OG (Table S3). The resulting fluxes of GLUS3 and ribulose-bisphosphate oxygenase and ribulose-bis carboxylase were upregulated under T22; however, the simulated fluxes of chlorophyll biosynthesis and photosynthesis were increased, and the simulated biomass was not affected by the in silico mutation. Moreover, it is well-known that under N depletion, arginine (Arg) biosynthesis is reduced, whereas fatty acid metabolism is induced [1]. The down-and upregulation of simulated fluxes in arginine biosynthesis and fatty acid metabolism, respectively, including desaturases, were detected only in the model of in silico overexpressed GLUS3, similar to that observed under nitrogen stress (Figure 3, Table S3). Li et al. reported that the control of C/N balance under N depletion occurred through the activity of N-acetyl-L-glutamate kinase (NACGK), which was regulated by the nitrogen regulatory protein PII. When the level of 2-OG was low, the PII interacted with NAGK and abolished the inhibitory effect of Arg on NACGK, thereby promoting its activity. In contrast, under N starvation, the increasing level of 2-OG, as a result of inhibited Arg biosynthesis, led to nitrogen assimilation and facilitated lipid biosynthesis [1].
After a comparative analysis of the T22, T40, iCDHy-overexpression and GLUS3-overexpression reactions, the results demonstrated that the reaction catalyzed by ribulose-bisphosphate carboxylase (RBPC), a key enzyme in carbon dioxide fixation, was the only downregulated reaction, whereas pyruvate kinases, PYK1, PYK5 and PYK6, and adenosylcobinamide kinase, ACBIKGTP, were the upregulated reactions found in common among the four models ( Figure S5). Notably, the four upregulated metabolic fluxes, as mentioned above, were reported to be involved in the oxidative stress response [25,26].
Moreover, it was previously reported that the uncoupling of catabolism and anabolism could be an indicator of the delayed adaptation of bacterial cells to sudden unbalanced conditions [27,28]. In our proteome-based model, a reduction in biomass was found after temperature downshift and upshift, whereas a higher percentage of anabolic and catabolic reactions was downregulated than upregulated in the T40 model. This result showed the coupling of anabolism and catabolism and thus indicated the immediate adaptation of the cells to high-temperature stress. In the T22 model, the decrease in biomass was uncoupled, with a higher percentage of catabolic reactions being observed to be upregulated than downregulated; therefore, this finding might indicate the delayed adaptation of the cells to low-temperature stress. In the case of the in silico mutagenic model, biomass was increased in the model of GLUS3 overexpression in combination with anabolic and catabolic reactions. The results suggested immediate adaptation to the changing level of GLUS3 (Table S4).
An analysis of the ratio of the affected reactions in C-and N-metabolism after pathway mapping, as shown in Table S5, clearly shows that both temperature stresses caused greater upregulation of the reactions in nitrogen metabolism than that in carbon metabolism; thus, the C/N ratio was >1. The simulated data were well correlated with the findings of our previous quantitative proteome analysis [19]. It is worth noting that in silico overexpression of GLUS3 showed the highest number of upregulated reactions in both C-and N-metabolism. However, the highest number of downregulated reactions was observed in the in silico GLNS knockout. Moreover, the overexpression of iCDHy in the TCA cycle at the interconnection of C-and N-metabolism caused a more negative effect on the reaction fluxes in C-metabolism than in N-metabolism, resulting in a notably high C/N ratio.

Application to System Biology
The proteomic data obtained under growth temperature stress conditions were integrated to develop the proteome-based model in the present study. The model was subjected to biological applications in terms of gene manipulation and flux alteration in in silico knockout and overexpression.
In cyanobacteria, there have been intensive studies on 2-OG, which is the intermediate in the TCA cycle at the interconnection of C-and N-metabolism, as the signal molecule for nitrogen stress [24]. It is well-known that under nitrogen limitation, the level of 2-OG is increased and leads to the nitrogen assimilation of the cells via the control of nitrogen regulatory protein PII, and the reverse is observed under nitrogen oversupply. Thus, the C/N balance of the cells is maintained. GlsF and GlnA are enzymes that catalyze the following reactions of 2-OG biosynthesis. Moreover, these enzymes were found to be the client proteins of Hik28 in A. platensis, as mentioned earlier ( Figure S4). Therefore, the proteome-based model was applied to simulate the fluxes in the cells where these two enzymes were missing or overexpressed.
The in silico knockout and overexpression of the reactions of interest can be performed by using proteome-based GSMM to simulate the systematic-scale fluxes that may have occurred after genetic manipulation. According to our previous proteomic analysis, several differentially expressed proteins under low-temperature stress were the same as those under nitrogen depletion, e.g., proteins involved in N assimilation and fatty acid desaturation [24]. Moreover, we found that Hik28, which played a role in the low-temperature stress response, showed a protein-protein association with several proteins involved in N assimilation, as shown in its PPI network ( Figure S4). Protein binding experiments, yeast two-hybrid system, for Hik28 and its client proteins were performed by our group, and positive results were revealed [29]. Therefore, the proteome-based model was applied in this study to mimic the increasing level of the key signal molecule, 2-OG, under N-stress and the reaction fluxes that catalyzed the compounds surrounding the key reaction located at the interconnection of C-and N-metabolism. Metabolic changes associated with the in silico overexpression of iCDHy, GLUS3 and GLNS were analyzed as described earlier. The data obtained from the in silico mutant models was well corresponded to the effect of Hik28-deletion mutant to fatty acid analysis and photosynthetic activity under temperature stress, examined by our group (Kurdrid et al. [29] and Figures S6 and S7).

Conclusions
The GSMM is a computational analysis through FBA, which is used to predict and explain the performance and behavior of microorganisms. The mathematical approach is able to simulate metabolism providing information difficult to obtain in laboratory or field cultivations to maximize the production yield of microalgal biomass. In the present study, proteome analysis data obtained under growth temperature stress conditions of A. platensis C1 were integrated into our existing GSMM by using an optimization method. Taken together with the pathway mapping and analysis, the metabolic fluxes simulated from the proteome-based model led to prospective biological interpretation and enabled the prediction of genome-scale metabolic fluxes under the designated conditions, including those generated from in silico knockout and overexpression. Therefore, by using the model, the cultivated conditions for biomass production and production of the desired biochemical compounds, as well as the strategy for genetic manipulation for biochemical production, can be anticipated. However, the inability to directly integrate the cell regulatory mechanisms into the model remains a limitation regarding whether the proteome data obtained from stress response studies can be integrated into the GSMM. Moreover, the approach presented in this study is applicable to the in silico modification of a metabolic flux catalyzed by an irreversible enzyme.
Supplementary Materials: The following are available online at http://www.mdpi.com/2073-4409/9/9/2097/s1, Figure S1: Global metabolic pathway mapping of the results obtained from the comparison of simulated metabolic flux when the cells were grown at (A) low temperature (22 • C) and (B) high temperature (40 • C), Figure S2: Scattering plot representing the correlation of the growth temperature and the simulated flux level, Figure S3: Bar graph illustrating the relative % of the affected reaction fluxes compared to the total reactions (100%), simulated at 22 • C and 40 • C in the designated pathway map, Figure S4: Protein-protein interaction network of Hik28 constructed by using STRING, Figure S5: Venn diagram showing the number of the affected reactions under/of the simulated conditions/in silico mutants, Figure S6: Graph of the fatty acid composition and GC profiles of total fatty acids, Figure S7: The amounts of chlorophyll a and photosynthetic oxygen-evolving activity in wild type and mutant, Table S1: Metabolic reactions and their simulated flux of the Spirulina-GSMM under the optimal growth temperature of 35 • C (T35), low temperature of 22 • C (T22) and high temperature of 40 • C (T40), Table S2: The effect of temperature on specific growth rate and biomass concentration of Arthrospira platensis strains, Table S3: Metabolic reactions of Spirulina-GSMM and their simulated flux of the in silico knockout/overexpress mutants, Table S4: Classification of reactions into catabolism and anabolism, Table S5: Ratio of the relative % of affected reaction in the carbon and nitrogen metabolism under/of each condition/in silico mutant.

Conflicts of Interest:
The authors declare no competing interests.