Single-level FBA and OptKnock designs for the improvement of 2,3-BD production from glycerol in B. subtilis
We used single-level FBA and OptKnock to redesign metabolic pathways in B. subtilis to enhance 2,3-BD production from glycerol. We created a metabolic pathway of glycerol dissimilation following flux distribution based on the results of single-level FBA and OptKnock in iYO844 GEM with glycerol as a sole substrate (Fig. 1). Single-level FBA design was based on eliminating the fermentative products competing with 2,3-BD production, as described in our previous study [7]. The only objective function was biomass production. According to single-level FBA, the inhibition of four reactions, namely, ACKr (ack), PTAr (pta), LDH_L (lctE), and ACACT1r (mmgA), would enhance 2,3-BD production (Table 1). The removal of ACKr (ack) and PTAr (pta) reaction was predicted to reduce acetate production. The removal of LDH_L (lctE) reaction was predicted to inhibit lactate production. Finally, the removal of ACACT1r (mmgA) reaction was predicted to block the production of acetoacetate. According to single-level FBA, inhibition of these four reactions would improve the 2,3-BD production from 0 to 2.54 mmol/gCDW/h (CDW, cell dry weight), compared with the flux in the wild-type strain.
OptKnock suggested a different strategy, with the improvement of growth (biomass production) coupling with the improvement of 2,3-BD production set as the objective functions. The modelling suggested the removal of the ACKr (ack), PTAr (pta), ACACT1r (mmgA), and G6PDH2r (zwf) reactions. Considering the flux distribution, the biomass production would increase to 0.35 mmol/gCDW/h, with 2,3-BD production at a level of 2.10 mmol/gCDW/h, with the acetate production decreased to 0.15 mmol/gCDW/h (Table 1). Of note, the flux to lactate production became zero in the model, even though the LDH_L (lctE) reaction was not removed. To the best of our knowledge, the knockout of the zwf gene to enhance 2,3-BD production has not been reported to date. Hence, the analysis suggested that deletion of these genes would impede lactate production and improve 2,3-BD production.
We subsequently constructed mutant strains of B. subtilis based on the gene deletions suggested by single-level FBA (strain LM01) and OptKnock (strain MZ02) predictions, and investigated 2,3-BD production during bio-fermentation, with the wild-type strain (W168) as a control. The results of these analyses are described below.
Effect of gene deletions on the bio-fermentation profile of B. subtilis
2,3-BD production from glycerol in B. subtilis strains was evaluated using batch fermentation. The deletion of multiple genes in strains LM01 (ackA, pta, lctE, and mmgA) and MZ02 (ackA, pta, mmgA, and zwf) did not notably negatively affect bacterial growth (Fig. 2). Biomass production during logarithmic (24 h), early stationary (36 h), and late stationary phases (48 h) was similar in all strains. The major fermentation products were acetate, lactate, acetoin, and 2,3-BD, with only a minor production of ethanol, succinate, pyruvate, and diacetyl during the experiment.
We compared the yields of major fermentation products and glycerol utilization by all strains during each growth phase (Table 2). Glycerol utilization in all strains was not affected by the growth phase, and the multiple-gene knockouts did not strongly affect glycerol utilization of the strains. Total biomass production by the W168, LM01, and MZ02 strains was 4.35±0.05, 4.44±0.10, and 4.03±0.19 mmol/L, respectively. The wild-type strain (W168) produced acetate and lactate as the major fermentation products during the logarithmic phase, with these VFAs converted together with glycerol to acetoin and 2,3-BD during the stationary phase.
Considering the deletion of ackA and pta genes in the LM01 and MZ02 strains, acetate production was inhibited during the logarithmic phase; however, acetate was detected when the growth reached the early stationary phase. This was in agreement with our previous observations [7]. Similar, Fu et al. [20] suggested that the knockout of the pta gene does not inhibit acetate production in B. subtilis, which might be generated via pathways for acetate or acetyl-phosphate production from other precursors in this bacterium. Indeed, based on metabolic pathways that operate in B. subtilis 168 (Kyoto Encyclopedia of Genes and Genomes) [21], acetate can be produced via many pathways not included in the iYO844 model, e.g., one involving the ydap gene (BSU04340, EC:1.2.3.3) and yflL gene (BSU07640, EC:3.6.1.7), which convert pyruvate to acetyl phosphate, and then acetate. These missing reactions should be analyzed and the information gap filled to improve the iYO844 GEM.
In addition to the above, the LM01 strain lacks the lctE and mmgA genes. Knocking out lctE suppressed lactate production during fermentation. The product of this gene is the main competitor of the 2,3-BD production reaction, competing for NADH consumption [7]. Accordingly, deletion of the lctE or ldh gene improves 2,3-BD production in different bacteria [20, 22, 23]. Further, improvement of 2,3-BD production by mmgA knockout was shown in a previous study [7]. In the current study, the maximum yield of 2,3-BD was 0.28 mol/mol‧glycerol, which was the highest productivity noted among the compared strains. In addition, the LM01 strain produced only a small amount of acetoin, as NADH availability directly drove the metabolic pathway involved towards 2,3-BD production.
In the MZ02 strain, in addition to ackA and pta, the mmgA gene was also deleted to promote 2,3-BD production. However, together with the deletion of the zwf gene, this did not improve 2,3-BD production (Table 2). The highest yield of 2,3-BD was 0.19 mol/mol‧glycerol, which was equal to the highest yield of 2,3-BD obtained with the wild-type strain. Surprisingly, lactate was produced during the logarithmic phase at a remarkably high level, compared to that in other strains, with the highest yield of 0.22 mol/mol‧glycerol, in the early stationary phase. In a previous study, lactate yield of B. subtilis XZ7 strain was improved by the deletion of the alsS gene (encoding α-acetolactate synthase), which inhibited the production of 2,3-BD [24]. Apart from genetic engineering in B. subtilis, modifications of some genes in E. coli were tested for improved lactate production, e.g., by removing reactions competing for pyruvate, such as the ethanol/acetate metabolic pathway, or precursor reactions converting phosphoenolpyruvate into succinate [25]. To the best of our knowledge, we report here for the first time that the knockout of the zwf gene, which is related to glycerol dissimilation to the pentose phosphate pathway, enhances the production of lactate. Considering the single-level FBA predictions, the increase in lactate production might be driven by a decreased metabolic flux from dihydroxyacetone phosphate (dhap) to d-fructose 1,6-bisphosphate (fdp) as a result of the zwf gene deletion, with lactate produced by the conversion of dihydroxyacetone phosphate (dhap) to glyceraldehyde 3-phosphate (g3p) (Fig. 3). Hence, the improved flux of glyceraldehyde 3-phosphate dehydrogenase reaction (EC:1.2.1.12; BSU33940) increased the amount of NADH in the system. Together with the increase of metabolic flux from glyceraldehyde 3-phosphate to pyruvate, this resulted in increased lactate production.
Finally, in the MZ02 strain, acetate and lactate were converted into acetoin and 2,3-BD during the stationary phase, similar to their conversion in the wild-type strain. Since we observed that the gene deletion suggested by OptKnock did not increase 2,3-BD production, we next proceeded to identify the reason for this inaccuracy.
Comparison of FBA and OptKnock predictions with experimental data for MZ02 strain
OptKnock focuses on two objective functions, i.e., cell growth and 2,3-BD production, suggesting non-essential reactions and those competing with biomass and 2,3-BD production for removal. The modelling suggested that deletion of four specific reactions (genes) would enhance the biomass and 2,3-BD production in B. subtilis, to 0.35 and 2.10 mmol/gCDW/h, respectively (Table 3). However, the experimental data did not confirm these predictions. We hence used a two-step FBA simulation to identify the reason for this discrepancy.
Two-step FBA simulates metabolic distribution before gene deletion. We set objective functions similar to those of OptKnock, by maximizing biomass production as the first target and maximizing 2,3-BD production as the second target (Table 3). Biomass production by the wild type predicted by using two-step FBA was the same as that predicted using single-level FBA (0.3 mmol/gCDW/h, Tables 1 and 3). However, the major produced VFAs predicted by two-step FBA (acetate and 2,3-BD) were different from those predicted by single-level FBA (acetate and lactate). This revealed the underlying reason for the inaccurate prediction of OptKnock. Specifically, since acetate and lactate were detected as the major VFA produced in single-level FBA, the reactions for their production were given priority for removal. On the other hand, two-step FBA considered acetate and 2,3-BD as the major products and, consequently, only suggested the reactions for acetate production for removal. LDH_L (lctE) reaction was not considered for removal because the flux of this reaction was zero. This strategy might have led to a bias when assessing biomass and 2,3-BD production, and contributed to the erroneous identification of reactions for removal.
We analyzed the MZ02 strain by single-level FBA, with the biomass production as the only objective function. The predictions corresponded to the experimental data, especially for the logarithmic phase of growth (Tables 2 and 3). The production of acetate was negligible, whereas the production of lactate was higher than that in the wild-type strain. The predictions of single-level FBA for strains W168 and LM01 were also in line with the experimental results (Tables 1 and 2). Hence, we suggest that single-level FBA be accepted as a reliable approach for the design and evaluation of metabolic pathways, as OptKnock was not be applicable to the GEM in the current study. In the case of the current study, a recombinant B. subtilis strain should be constructed to allow the conversion of the entire intracellular pyruvate pool to 2,3-BD, not lactate.
Many factors might impact the predictions of OptKnock, such as the degree of completion of the entire metabolic pathways in GEM, the use of solvers for problem function, and so on. Nonetheless, the successful use of OptKnock to optimize biochemical production of native products, especially in the E. coli model, was shown in many studies [10, 13, 14]. Overall, we recommend evaluating the OptKnock predictions by using single-level FBA to ensure the precision of the designed metabolic pathway.