In silico profiling of cell growth and succinate production in Escherichia coli NZN111

The constraint-based models

Genome-scale metabolic core models of E. coli downloaded from BiGG database were employed to perform all simulations in silico (Schellenberger et al. 2010). Here, the relation of reactant and product in some reactions was exchanged to simulate positive flux values without affecting the correlation between reactions. The output was termed Model for the wild-type E. coli in this study (Additional file 1). E. coli NZN111 was created by insertional inactivation of LDH and PFL encoded by ldhA and pflB, respectively, to improve succinate production and reduce by-product formation. Strain NZN111 produced very low levels of lactate and formate in LB medium (Lucy Stols MID 1997) or M9 medium (Wang et al. 2009) under anaerobic conditions. Additionally, in LB medium ldhA mutant produced little lactate, while pflB mutant also produced little formate (Lucy Stols MID 1997; Zhu and Shimizu 2004). Accordingly, we obtained the modified model for E. coli NZN111 by directly deleting reactions LDH-D and PFL regulated by ldhA and pflB, respectively, named Model_nzn. Similarly, deleting reactions LDH-D or PFL from Model were constructed for ldhA mutant or pflB mutant, respectively, called Model_del_ldh and Model_del_pfl.

Artificial centering hit-and-run algorithm

Artificial centering hit-and-run (Kaufman and Smith 1998), an unbiased random walk algorithm, was employed to randomly sample the steady-state flux space and shape the solution space, which illustrated the global network properties of constraint-based models or reflected the physiological metabolism of microorganisms under specific conditions (Kaufman and Smith 1998; Occhipinti et al. 2007; Schellenberger and Palsson 2009; Thiele et al. 2005). This algorithm involves three steps: The first step requires the identification of an initial point that lies within the solution space but not at the extremity. The second step of ACHR calculates “warm-up” points from the initial point using several iterations of a basic hit-and-run algorithm. The warm-up points are stored as columns of a matrix W, and an approximate center, s, is calculated. In the final step, the sample points are calculated. The direction for the next iteration from a sample point x _m is chosen by randomly taking one point y out of the matrix W and applying the direction vector of y and s (\({\vec{\text{y}}}\)−\({\vec{\text{s}}}\) ) to x _m . At each iteration, the newly calculated point, x _{m + 1} , was substituted randomly into W in the place of a previously calculated point. The approximate center was also recalculated following iteration. This last step is repeated until a desired number of sample points are reached.

In the present study, we sampled the solution space of the constraint-based models using default settings by loading 2000 points per reaction, i.e. 2000 flux values per reaction were provided for subsequent analysis. Therefore, an unbiased average flux value could be obtained by calculating the average value and standard deviation of each reaction (Almaas et al. 2004), and a histogram for each reaction could be plotted to illustrate the probability distributions of flux values. Moreover, Pearson’s correlations between reactions can be calculated as well.

Metabolite flux-sum analysis

A descriptive variable of flux-sum was defined to represent the turnover rate of a metabolite by summing up all the incoming or outgoing fluxes around the metabolite at quasi-steady state (Chung and Lee 2009). This definition clearly indicates that the unit of flux-sum is equivalent to that of the reaction flux (i.e., mmol/g DW/h). Here, the concept of flux-sum was applied to the average flux distributions that resulted from the samplings of quasi-steady-state flux space using unbiased ACHR algorithm. Still, we let Φ _i  denote the flux-sum of metabolite i, and its mathematical form is given by Φ _i  = 0.5 \(\mathop \sum \nolimits_{j} \left| {S_{ij} v_{j} } \right|\). Metabolite flux-sum analysis elucidates the roles of metabolites in the network. In addition, a change of overall flux-sum of metabolites reflects an alteration of the metabolic network with perturbation. Analysis of the overall turnovers of cofactors (e.g. NAD(P)H) can aid in understanding the roles of the identified genes (or enzymes) (Lakshmanan et al. 2015a, b).

Reaction expression profile

Gene expression data of wild-type E. coli BW25113 and NZN111 strains were downloaded from the GEO database (http://ftp.ncbi.nlm.nih.gov/pub/geo/DATA/supplementary/samples/GSM546nnn/GSM546949/GSM546949.CEL.gz). Cells of the wild-type E. coli BW25113 and NZN111 strains growing microaerobically in M9 medium containing 10 g/L glucose were harvested in the mid-exponential phase. Total RNA was isolated using Qiagen’s RNeasy kit following the manufacturer’s protocol. Expression data of 136 genes existed in Model except for gene s0001 were selected for preprocessing using log2 (Additional file 2). Subsequently, a two-sample Kolmogorov–Smirnov test was performed to compare the two groups of values, viz. the gene expression data for wild-type E. coli and NZN111 strains. Results of the two-sample test showed statistically significant differences (p value = 1.6348 × 10⁻¹³; at 5% significance level) between the two E. coli strains.

Furthermore, reaction expression profile can be mathematically characterized by gene expression data based on logical relations of gene reaction described in the model as follows:

Thus, the ratios of logarithmic reaction expression data for strain NZN111 to that of wild-type E. coli were calculated to better indicate the differential expression reactions.

In silico simulation and analysis of the effects of ldhA and pflB knockout

Escherichia coli NZN111 was constructed to anaerobically produce succinate by the disruption of ldhA and pflB genes. Accordingly, based on the metabolic core model of E. coli, we constructed four constraint-based models: Model, Model_del_ldh, Model_del_pfl, and Model_nzn for wild-type E. coli, ldhA mutant, pflB mutant, and NZN111, respectively (see “Methods” section). These models were employed to randomly sample the solution space using the ACHR method. The sampling errors encountered during the procedure were minimal (<1 × 10⁻⁶), indicating the correctness of these sampling results. On the basis of these sampling results, we calculated the average flux and standard deviation of each reaction and plotted a histogram for each reaction (Additional files 3, 4). Therefore, the average flux distributions consisted of average fluxes for wild-type E. coli, ldhA mutant, pflB mutant, and NZN111 were achieved to shape their global metabolisms. And each histogram illustrated the probability distribution of the values of each reaction. Furthermore, for the purpose of convenient comparison, the four average flux distributions were normalized to the specific glucose uptake rate of 10 mmol/g DW/h (Additional file 3) (Fig. 1). Therefore, the metabolite flux-sum values of cofactors on the basis of the four normalized flux distributions can be evaluated as shown in Table 1.

	Model	Model_del_ldh	Model_del_pfl	Model_nzn
Biomass	0.0098	0.0120	0.0092	0.0105
EX_succ (e)	1.4167	1.4389	4.6584	4.8538
EX_lac_D (e)	1.8804	0	2.9909	0
EX_for (e)	11.7796	13.1273	0	0
EX_pyr (e)	0.9957	1.0634	0.5216	0.5439
EX_acald (e)	1.3206	1.3593	0.5099	0.5239
EX_ac (e)	2.3847	2.8250	0.4452	0.4506
EX_etoh (e)	10.0674	11.2243	9.9215	12.6531
Atp flux-sum	29.4088	29.7098	24.7564	24.9150
Ctp flux-sum	0	0	0	0
Fadh2 flux-sum	0	0	0	0
Nadh flux-sum	28.7541	29.3178	35.1824	38.1361
Nadph flux-sum	5.5522	5.7709	3.8235	3.9129
Q8h2 flux-sum	521.7517	523.4008	520.8582	525.2377
Accoa flux-sum	15.0885	16.8031	11.5046	14.2915

Figure 1 shows that the carbon source for producing succinate under anaerobic conditions is mostly derived from phosphoenolpyruvate (pep) pool through the reductive TCA pathway in all four models. Respective deletion of reaction LDH-D or PFL all increased the specific production rate of succinate and reduced the biosynthesis of by-products (Table 1). In strain NZN111, simultaneous deletion of reactions LDH-D and PFL enforced carbon flow towards the reversed TCA cycle via reaction PPC catalyzed by phosphoenolpyruvate carboxylase encoded by the gene ppc and reduced biosynthesis of by-products such as lactate, acetaldehyde, acetate, and ethanol. Similarly, researchers have validated that strengthening the reductive TCA pathway by overexpressing the ppc gene enabled succinate as a major product in E. coli under anaerobic conditions (Millard et al. 1996). In addition, the results from our simulations were also consistent with the gene expression data of wild-type E. coli BW25113 and NZN111. Compared to the wild-type strain, expression data of reactions PPC, MDH, and FUM in strain NZN111 indicate significant up-regulation (Fig. 2; Additional file 2).

Furthermore, in strain NZN111, a high ratio of NADH/NAD⁺ (imbalanced redox state) resulted in growth defects on glucose under anaerobic conditions. (Singh et al. 2009). In addition, a high flux-sum value of NADH was found in Model_nzn (Table 1), which we considered to be mainly a result of knockout of reaction PFL. The reason for this assumption was that the knockout of reaction PFL caused a much higher flux-sum value of NADH than that caused by knockout of reaction LDH-D, although knockout of reaction LDH-D resulted in a slightly enhanced flux-sum value of NADH as well. The quantities of increased NADH turnover resulting from the inactivation of PFL were about 11-fold higher than that of inactivation of NADH-dependent LDH. That is to say that in strain NZN111 a high ratio of NADH/NAD⁺ may mainly result from the inactivation of PFL rather than the inactivation of LDH, which can probably be attributed to the two NAD⁺ formed from ethanol production versus one NAD⁺ from lactate formation for each molecule of pyruvate (Yang et al. 1999b).

The ethanol and lactate production pathways are two of the most robust anaerobic fermentative pathways in E. coli for the regeneration of NAD⁺ from NADH for cell growth. (Zhu and Shimizu 2004). Experimental results have shown that the anaerobic cell growth is not affected by deficiency of the ldh gene, while the pfl mutant could only grow in glucose medium supplemented with acetate (Graef et al. 1999; Lucy Stols MID 1997). Numerous studies investigated the two fermentative pathways that are utilized to regenerate NAD⁺ from NADH in E. coli. For example, overexpression of the NAD⁺-dependent formate dehydrogenase in E. coli resulted in a significant redistribution of metabolic fluxes, as evidenced by a dramatic increase in the ethanol-to-acetate ratio and a decrease in the flux to lactate (Berrios-Rivera et al. 2002a). In addition, the experimental results have also shown that ackA-pta mutant has a reduced acetate level and a much higher lactate formation rate with simultaneously lower formate and ethanol synthesis rates. The observed diversion of flux to lactate is mainly due to the necessity of NAD⁺ regeneration (Yang et al. 1999a). Moreover, a mutant with severely impaired ethanol production under anaerobic conditions was found to have generally improved the production of lactate and succinate (Yun et al. 2005). Thus, we concluded that lactate and ethanol production pathways can be an alternate way to regenerate NAD⁺ from NADH in order to maintain a proper redox balance of cofactor NADH/NAD⁺. However, in strain NZN111, the inactivation of LDH and PFL due to the disruption of genes ldhA and pflB, respectively, blocked the two alternate pathways and failed to regenerate sufficient NAD⁺ to oxidize glucose under anaerobic conditions.

In silico simulation and analysis of succinate accumulation in E. coli NZN111

Additionally, we found that the accumulated NADH enforced the conversion of ubiquinone-8 (q8) to ubiquinol-8 (q8h2) by strengthening reaction NADH16, while q8h2 facilitated the conversion of fumarate to succinate (Fig. 3). This observation was consistent with the gene expression data of the wild-type E. coli BW25113 and NZN111. Compared to the wild-type strain, gene expression of reaction NADH16 was significantly up-regulated and that of reaction NADTRHD was significantly down-regulated (Fig. 2). Thus, we speculated that strengthening reaction NADH16 by overexpressing the corresponding genes may be an effective way not only for decreasing the intracellular ratio of NADH/NAD⁺ but also for enhancing succinate production. Reaction NADH16 is catalyzed by enzyme complexes encoded by the nuoA-N gene cluster (14 subunits), which encode a transmembrane NADH:ubiquinone oxidoreductase (NDH-I) responsible for coupling redox chemistry to proton-motive force generation (Yang et al. 1999b). Singh et al. (2009) reported that overexpressing nuoC coding for the connecting fragment of the NDH-I (1 subunit of NDH-I) effectively decreased NADH/NAD⁺ ratio in the cell and improved cell growth and succinate production under anaerobic conditions.

In silico simulating correlation of succinate production in E. coli NZN111

To simulate what correlates to cell growth and succinate production in E. coli NZN111 under anaerobic conditions, Model_nzn stepwise setting distinct upper and lower bounds of succinate exchange reaction [i.e., reaction EX_succ(e)] were performed to sample the solution space using ACHR method. Meanwhile, histograms were plotted and the normalized average flux distributions were evaluated as described above (Additional files 5, 6). As with the elevated specific production rate of succinate [i.e., the fluxes of reaction EX_succ(e) gradually elevated], more carbon source flow towards the TCA cycle via reaction PPC and less acetyl-coA synthesis was apparent (Fig. 4; Table 2). Meanwhile, the higher the accumulation of NADH, the smaller the specific growth rate was, i.e., the accumulated NADH inhibited cell growth. However, the flux-sum value of NADH gradually decreased because accumulated NADH was used to facilitate the regeneration of q8h2, which also improved succinate production. As shown above, the accumulation of NADH facilitated reaction NADH16 and restrained reaction NADTRHD, thus decreasing the turnover of NADH and NADPH gradually (Table 2).

	Model_nzn_A	Model_nzn _B	Model_nzn _C	Model_nzn _D
Bounds	[0, 3.4764]	[3.4764, 6.9529]	[6.9529, 10.4293]	[10.4293, 13.9058]
Biomass	0.0119	0.0085	0.0071	0.0036
EX_succ (e)	2.9783	5.0800	7.6984	10.7240
atp flux-sum	26.4170	24.6409	22.8017	23.5086
ctp flux-sum	0	0	0	0
fadh2 flux-sum	0	0	0	0
nadh flux-sum	40.8157	37.9489	34.3208	30.4042
nadph flux-sum	4.5915	4.0034	3.1865	1.8573
q8h2 flux-sum	515.0937	514.8774	509.5948	511.4309
accoa flux-sum	16.3295	13.9790	11.2263	8.6963

Additionally, based on the 2000 fluxes for each reaction obtained using the ACHR method in Model_nzn [without constraint to reaction EX_succ(e)], a matrix of Pearson’s correlations between reactions was calculated and is shown in Fig. 5. Reaction NADH16 negatively correlated with pyruvate metabolism (ACALD, ALCD2x) because of NADH consumption, i.e., inactivation of pyruvate metabolism (ACALD, ALCD2x) due to the disruption of ldhA and pflB genes suppressing the biosynthesis of acetyl-coA facilitated reaction NADH16 to convert q8 to q8h2. This further confirms that strengthening reaction NADH16 may be an effective strategy for decreasing accumulation of NADH and increasing succinate production in strain NZN111.

Similarly, reaction NADH16 negatively correlated with reaction MDH. In addition, cell growth was found to be positively correlated with reaction MDH due to regeneration of NAD⁺, and with reaction ATPS4r due to the supply of ATP. Moreover, researchers reported that overexpressing malate dehydrogenase (MDH) encoded by the gene mdh improved cell growth and succinate production in E. coli NZN111 (Wang et al. 2009). Furthermore, overexpression of ATP-forming PEPCK from Actinobacillus succinogenes in an ldhA ⁻, pflB ⁻, ptsG ⁻, and ppc ⁻ quadruple mutant strain resulted in a 60% increase in biomass and succinate formation (Singh et al. 2011).