Introduction

The colorless, odorless, and transparent chemical 2,3-butanediol (2,3-BD), synthesized as one of the fermentation products, is a promising bulk chemical owing to its wide industrial application. It can be directly dehydrated into 1,3-butadiene, which is extensively used in the manufacture of many synthetic rubbers [6, 13]. Dehydration of 2,3-BD imparts methyl ethyl ketone (MEK), which is an effective fuel additive, and is commonly used as an industrial solvent for resins and lacquers [6, 13, 45, 48]. Acetoin and diacetyl, which serve as flavoring agents for numerous food products, giving a buttery taste, also can be easily manufactured by the dehydrogenation of 2,3-BD. In addition, 2,3-BD has potential applicability in a broad range of industries for producing moistening and softening agents, printing inks, cosmetics, plasticizers and explosives, fumigants, spandex, and pharmaceuticals [6, 13, 48].

In the face of rising petroleum prices and increasing environmental concerns, production methods of bio-based fuels and chemicals are getting more public attention [15, 42, 46]. 2,3-BD can be produced by microbial fermentation, and several bacteria are reported to synthesize 2,3-BD as the major fermentation product under microaerobic conditions. Among them, Klebsiella pneumoniae, K. oxytoca, Enterobacter aerogenes, Paenibacillus polymyxa, Serratia marcescens, and Pseudomonas chlororaphis produce large amounts of 2,3-BD [6, 48]. Klebsiella spp., especially K. oxytoca, show the best 2,3-BD production performance in terms of yield, productivity, and final titer [6, 9, 13, 17, 19–22, 39, 45, 48] as well as carbon-source utilization (glucose, galactose, lactose, xylose, and glycerol) [4, 7, 16–18, 21, 47]. Furthermore, K. oxytoca belongs to Risk Group 1 (RG 1), being categorized as a generally regarded as safe (GRAS) organism (National Institute of Health, Guidelines for Research Involving Recombinant DNA Molecules, 2002), which indicates that K. oxytoca is safe for industrial use.

With an aim to enhance 2,3-BD production by reducing the formation of byproducts, including formic acid, acetic acid, ethanol, lactic acid, succinic acid, and acetoin, various genetic modifications on K. oxytoca have been conducted [21, 24, 47]. In one example, Ji et al. [21] reported that the insertional inactivation of acetaldehyde dehydrogenase gene (aldA) in K. oxytoca led to a significantly improved 2,3-BD production by reducing the formation of ethanol. In other studies, Yang et al. [47] and Kim et al. [24] constructed a lactate-deficient mutant of K. oxytoca, in which the ldhA gene was disrupted, for preventing the formation of lactic acid. However, traditional metabolic engineering approaches have many limitations in developing ideal strains because they focus on local metabolism and do not consider the overall but complex metabolic network of cells. Accordingly, the approaches through systems metabolic engineering, which studies an organism at the systems level, became important [2, 3, 12, 25, 26, 34]. In particular, in silico genome-scale metabolic models and their applications to simulating metabolic behavior have been widely employed as a useful and promising tool to provide a system-level understanding of a cell, as well as systematic strategies for strain improvement [1, 10, 23, 27, 29, 33, 37].

We recently constructed a genome-scale metabolic model of K. oxytoca, KoxGSC1457, containing 1,457 reactions and 1,099 metabolites [38]. In that study, knockout candidates that can overproduce 2,3-BD and reduce byproduct formation at the same time were identified using the KoxGSC1457 model, and K. oxytoca mutant strains were developed based on simulation results. In particular, the effect of agitation speed during 2,3-BD fermentation was carefully studied in order to improve cell growth and 2,3-BD productivity. Then, optimization of agitation speed during fed-batch fermentation was conducted to enhance 2,3-BD production.

Materials and methods

Genome-scale metabolic model and constraints-based flux analysis

In order to predict the metabolic characteristics in response to genetic perturbations and to identify knockout candidates for 2,3-BD overproduction, we used the genome-scale metabolic model of K. oxytoca, KoxGSC1457 for this study. The KoxGSC1457 model contains 1,457 reactions (229 transport reactions and 1,228 biochemical reactions) and 1,099 metabolites [38]. The model was thoroughly validated by comparing its output with various experimental data in the previous study. Constraints-based flux analysis, including flux balance analysis (FBA), was carried out under the assumption of a pseudo-steady state [11, 14, 35, 43, 44]. Mass balances in the stoichiometric model can be set up as S ij · v j  = 0, in which S ij is a stoichiometric coefficient of a metabolite i in the jth reaction and v j is the metabolic flux [mmol/g dry cell weight (DCW/h] of the jth reaction. The in silico model, which is an underdetermined system due to insufficient constraints, can be simulated by optimization techniques, including linear programming (LP), subject to the constraints of mass balances, experimental measurements, and thermodynamics, as follows:

$${\text{Maximize}}/{\text{minimize}}\;Z = \sum\limits_{j \in J} {c_{j} v_{j} }$$
$$\begin{gathered} {\text{Subject to}}\;\sum\limits_{j \in J} {S_{ij} v_{j} = b_{i} } ,\forall i \in I \hfill \\ \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\alpha_{j} \le v_{j} \le \beta_{j} ,\;\forall j \in J\; \hfill ,\\ \end{gathered}$$

where Z indicates objective function, which is usually the maximization of cell growth rate, and c j is the weight of the reaction j. b i is the net transport flux of metabolite i, which is zero for intermediate metabolites as in S ij · v j  = 0. α j and ß j are the lower and upper bounds of the metabolic flux of the jth reaction. For in silico simulations, glucose consumption rate and oxygen uptake rate were constrained to 16 and 5 mmol/g DCW/h, respectively. In order to simulate the in silico model more accurately, the limits of uptake and secretion rates for some metabolites, including amino acids, organic acids (acetic acid, formic acid, lactic acid, pyruvic acid, and succinic acid), and alcohols (ethanol and acetoin), were constrained by experimentally measured flux values. Likewise, the limits associated with secretion for some metabolites, which were not produced during fermentation as with amino acids, were constrained to zero [38].

To make a 3D mesh plot graph as a continuous surface for in silico flux solution spaces, the cell growth rate was maximized while gradually increasing the 2,3-BD production rate and byproducts formation rate, respectively, from their minimal flux values to maximal flux values. In order to compare between experiments and in silico predictions, flux variability analysis was carried out. Flux variability analysis computes the ranges of in silico metabolic fluxes by maximizing or minimizing the fluxes of intracellular reactions with additional constraints of 95 % optimal cell growth rate, which is a value to relax the mathematically precise solution down to almost infinite places of decimals [28, 37].

Bacterial strains and mutant construction

K. oxytoca GSC 12206 strain (KCTC12133BP, Korean Collection for Type Cultures, Daejeon, Korea) was used in this study. In-frame deletions based on homologous recombination were carried out using overlap polymerase chain reaction (PCR) products to disrupt the target genes in chromosome. The overlap product of the two fragments, in which each ~500-bp upstream and downstream regions of the target gene was amplified and overlapped by PCR, was transformed in the competent cells of K. oxytoca strain by electroporation. The colonies were then selected in Luria–Bertani (LB) chloramphenicol (25 mg/L) plates at 42 °C, and the integrated cassette was cured by sacB expression under sucrose pressure. Integration and excision were confirmed in all mutants by PCR screening using genome-specific primers.

Fermentation

Culture medium used in batch and fed-batch fermentations contained (per liter): yeast extract (Becton Dickinson, Le Pont de Claix, France), 5 g; MgSO4·7H2O, 0.25 g; (NH4)2SO4, 6.6 g; K2HPO4, 8.7 g; KH2PO4, 6.8 g; trace metal solution, 10 ml. The trace metal solution contained (per liter): FeSO4·7H2O, 5 g; ZnSO4·7H2O, 0.1 g; MnSO4·H2O, 0.1 g; CaCl2·2H2O, 0.1 g; HCl 10 ml. For inoculum preparation, the suspended cells from single colonies on LB agar (Difco Laboratories, Detroit, MI, USA) plates were precultured in 20 ml test tubes containing 5 ml culture medium (10 g/L d-glucose) at 37 °C for 5 h. Then, 1 ml of the preculture was transferred to a 500-ml Erlenmeyer flask containing 300 ml culture medium (10 g/L d-glucose) and cultivated to an optical density of 1.5–2.0 at 600 nm (OD600). The tube and flask cultivations were conducted in a rotary shaker at 150 rpm and 37 °C (JEIO Tech. Co. SI-900R). Next, 300 ml of the seed culture was transferred to a 5-L bioreactor. Batch fermentations were performed in a 5-L BIOFLO®& CELLIGEN® 310 bioreactor (New Brunswick. Scientific Co., Edison, NJ, USA) containing 3 L culture medium (90 g/L d-glucose). The bioreactor was continuously aerated using a 0.2-μm membrane filter at a flow rate of 1 vvm (air volume/working volume/min). The temperature was maintained at 37 °C, and pH was controlled at 6.5 ± 0.1 by the automatic feeding of NH4OH. Foaming was controlled by the addition of Antifoam 289 (Sigma, St. Louis, MO, USA). All bioreactor experiments were performed at least three times independently, and the representative results are shown in the figures. Fed-batch fermentations were performed under the same conditions as batch fermentations, except for the feeding of glucose into the fermentor to maintain the concentration at 10–60 g/L. For the fed-batch fermentation, the agitation speed was maintained at 450 rpm until the concentration of acetoin reached about 10 g/L and was then switched to 350 rpm. Samples were periodically taken to measure OD600 to determine metabolite concentrations. After centrifugation at 13,200×g for 5 min, the resulting supernatant was used to determine the concentrations of glucose and metabolites.

Analytical procedures

Concentrations of d-glucose and metabolites, including 2,3-BD, formic acid, ethanol, acetic acid, lactic acid, succinic acid, and acetoin, were determined by a high-performance liquid chromatography (HPLC) equipped with ultraviolet-visible (UV-VIS) and reactive-index (RI) detectors (Agilent 1260 series, Agilent Technologies, Waldbronn, Germany). An Aminex HPX-87H column (300 mm × 7.8 mm, Bio-Rad, Hercules, CA, USA) was isocratically eluted with 0.01 N H2SO4 at 80 °C, and a flow rate of 0.6 ml/min was maintained. The OD600 was measured using a UV-VIS spectrophotometry (DR5000, Hach Company, CO, USA) to monitor cell growth. Cell concentration, as DCW per liter of culture broth, was calculated from the predetermined standard curve relating OD600 to DCW (1 OD600 = 0.3877 ± 0136 g DCW/L) [24].

Results and discussion

In silico prediction of gene knockout candidates for enhanced 2,3-BD production

K. oxytoca wild-type strain produced 2,3-BD along with several byproducts, including formic acid, acetic acid, ethanol, lactic acid, succinic acid, and acetoin during batch fermentation under microaerobic conditions [17, 21, 30] (Figs. 1, 2). For enhanced 2,3-BD production, the formation of byproducts should be reduced, after which the remaining metabolic fluxes from carbon sources should be redirected toward the synthesis of 2,3-BD. In our previous study [38], knockout of the ldhA gene, encoding lactate dehydrogenase, was targeted by in silico simulation, with top priority to eliminate the formation of lactic acid because K. oxytoca wild-type strain formed lactic acid as a main byproduct (32.2 g/L, 71 % of total byproducts) (Figs. 1, 2). The best candidate for gene knockout was found based on criteria of the maximization of 2,3-BD production rate and minimization of byproduct formation rate with viable-cell growth rate (>50 % cell growth rate of base strain). Knockout of the ldhA gene did not disrupt the redox balance of nicotinamide adenine dinucleotide/nicotinamide adenine dinucleotide, reduced (NAD+/NADH). The inactivation of lactate-dehydrogenase-oxidizing NADH seemed to be compensated with 2,3-BD-dehydrogenase-producing 2,3-BD with the oxidation of NADH as well. As a result, the constructed K. oxytoca ΔldhA strain successfully decreased the formation of lactic acid to 1.9 g/L (6 % of lactic acid produced by the wild-type strain) and significantly increased 2,3-BD. However, the K. oxytoca ΔldhA strain still produced large amounts of byproducts, especially formic acid and ethanol. The Pareto diagram in Fig. 2c visually identifies the main byproducts, formic acid and ethanol, in K. oxytoca ΔldhA strain. In silico analysis based on the KoxGSC1457 model also suggests that it left room for further strain improvement (Fig. 1). Although the optimal status of in silico flux solution space of the ldhA knockout strain was shifted to the increase of 2,3-BD production compared with that of the wild-type strain, byproduct formation rates of the ldhA knockout strain were still >5 mmol/g DCW/h (Fig. 1c).

Fig. 1
figure 1

In silico investigation for flux solution spaces of Klebsiella oxytoca wild-type and mutant strains. a Metabolic pathway related with 2,3-butanediol (2,3-BD) production. Three-dimensional mesh plot graph as a continuous surface for in silico flux solution spaces of K. oxytoca, b wild-type, c ΔldhA, and d ΔldhA ΔpflB strains. The maximal point of in silico cell growth rate is generally acceptable to predict the actual state of a cell, which is denoted by a small circle, because the first priority of an actual cell is survival [32, 35, 41]. The dotted arrow indicates the transition of maximal cell growth rate by gene manipulation. Colors in the contour graph indicate the values of cell growth rate. Byproduct contains ethanol and several organic acids: lactic, acetic, succinic, and formic

Full size image
Fig. 2
figure 2

Batch fermentation results. a Klebsiella oxytoca wild-type strain. b K. oxytoca ΔldhA strain. c Byproducts produced by K. oxytoca ΔldhA strain. d K. oxytoca ΔldhA ΔpflB strain. Batch fermentations were performed with a working volume of 3 L at 37 °C. Aeration rate and agitation speed were maintained at 1 vvm and 150 rpm, respectively; pH was controlled at 6.5 by the automatic feeding of NH4OH. Symbols in fermentation profiles indicates the concentration of glucose (filled diamonds), 2,3-butanediol (2,3-BD) (open circles), optical density at 600 nm (OD600 (filled squares), acetoin (open squares), ethanol (filled triangles), succinic acid (asterisks), lactic acid (filled circles), formic acid (open diamonds), and acetic acid (open triangles)

Full size image

In order to further improve the K. oxytoca ΔldhA strain, an in silico single-gene knockout simulation was performed on the criteria of the maximization of 2,3-BD production but minimization of byproduct formation rates. As a result, the pflB gene encoding pyruvate formate lyase was given as a knockout candidate of top priority for 2,3-BD overproduction. The metabolic characteristics of K. oxytoca ΔldhA ΔpflB strain were investigated by in silico flux solution space (Fig. 1d). The additional knockout of the pflB gene dramatically decreased the formation rate of byproducts, particularly those of formic acid and ethanol, <2 mmol/g DCW/h, and redirected the remaining metabolic fluxes into 2,3-BD synthesis. Inactivation of pyruvate formate lyase—which converts pyruvic acid into acetyl coenzyme A (acetyl-CoA) and formic acid—by pflB gene knockout, reduced effectively the formation of both formic acid and ethanol. The reduction of ethanol can be explained by the fact that the inactivation of pyruvate formate lyase decrease the formation of acetyl-CoA, which is the precursor for the formation of ethanol. In addition, the knockouts of ldhA and pflB genes significantly increased the pyruvic acid (an important precursor for 2,3-BD synthesis) pool, which was redirected into 2,3-BD synthesis without the disruption of redox balance. These predictions and observations were validated by the construction of K. oxytoca ΔldhA ΔpflB strain and its application to fermentations.

High-yield production of 2,3-BD by K. oxytoca ΔldhA ΔpflB strain

According to the in silico predictions, K. oxytoca ΔldhA ΔpflB strain was constructed by a marker-free and homologous recombination knockout method. Batch fermentation of K. oxytoca ΔldhA ΔpflB strain was performed on the culture medium containing 90 g/L of d-glucose at 150 rpm and 37 °C. Fermentation results were compared with those of the wild-type and ldhA knockout strains (Fig. 2). K. oxytoca ΔldhA ΔpflB strain was able to produce about 40.9 g/L of 2,3-BD, which is more than 2.3 times and 1.3 times higher than those produced by the wild-type (17.3 g/L of 2,3-BD) and ldhA knockout (29.9 g/L of 2,3-BD) strains, respectively (Fig. 2; Table 1). As the in silico predictions, the formation of ethanol and formic acid (<1 g/L) was decreased in K. oxytoca ΔldhA ΔpflB strain, whereas the ldhA knockout strain formed >3 g/L of ethanol and 6 g/L of formic acid. In particular, the 2,3-BD production yield from glucose by K. oxytoca ΔldhA ΔpflB strain was much higher (0.44 g/g, about 90 % of theoretical maximum yield) than that of the wild-type (0.18 g/g) and ldhA knockout (0.32 g/g) strains (Table 1; Fig. 2). It can be speculated that the substantially increased pool of pyruvic acid, 2,3-BD precursor, by the inactivation of pyruvate formate lyase, enhanced the synthesis of 2,3-BD. In addition, we compared the experiments with in silico predictions (Fig. 3). Given the fact that multiple equivalent solutions of intracellular metabolic fluxes can be computed for the same state [28, 36, 40], we should investigate the ranges of in silico metabolic fluxes by flux variability analysis. Flux patterns predicted by in silico simulation for the wild-type and mutant strains showed good agreement with experimentally measured metabolic fluxes. As represented in Fig. 3, the knockouts of ldhA and pflB genes dramatically decreased byproduct formation rate and increased 2,3-BD production rate in comparison with wild-type and ldhA knockout strains.

Table 1 Batch fermentation results of Klebsiella oxytoca wild-type and mutant strains
Full size table
Fig. 3
figure 3

Comparison between experiments and in silico predictions. a Klebsiella oxytoca wild-type strain. b K. oxytoca ΔldhA strain. c K. oxytoca ΔldhA ΔpflB strain. Bars indicate ranges of in silico metabolic fluxes computed by flux variability analysis. Squares indicate the experimentally measured metabolic fluxes based on fermentation results

Full size image

However, overall 2,3-BD productivity (0.51 g/L/h) of K. oxytoca ΔldhA ΔpflB strain was lower than that of the wild-type (0.58 g/L/h) and ldhA knockout (1.07 g/L/h) strains (Fig. 2; Table 1). This is mainly due to the severe inhibition of cell growth resulting from the inactivation of pyruvate formate lyase, which considerably decreased overall cell glucose uptake and 2,3-BD production rates (Fig. 2d). As mentioned earlier, the conversion of pyruvic acid into formic acid and acetyl-CoA is catalyzed by pyruvate formate lyase. In bacteria, acetyl-CoA is essential to energy generation, as it is the substrate that forms citric acid to initiate the tricarboxylic acid (TCA) cycle. It also contributes to biomass synthesis by serving as a precursor to fatty and amino acids. Another way to convert pyruvic acid into acetyl-CoA is conduction by the pyruvate dehydrogenase complex (PDC)—a complex of pyruvate dehydrogenase, dihydrolipoyl, and dihydrolipoyl dehydrogenase—in the presence of oxygen with reduction of NAD+. However, PDC activity seemed to be insufficient to replace the reaction of pyruvate formate lyase under the oxygen-limiting condition (150 rpm and 1 vvm) [5, 8]. Therefore, we examined the effect of agitation speed on 2,3-BD fermentation in order to improve overall 2,3-BD productivity of the K. oxytoca ΔldhA ΔpflB strain (Table 2).

Table 2 Summary of batch fermentations at different agitation speeds by Klebsiella oxytoca ΔldhA ΔpflB
Full size table

Strategy to improve 2,3-BD productivity by changing agitation speeds

In a previous study, we examined the response of the 2,3-BD production rate of K. oxytoca ΔldhA strain to variations in agitation speed during fermentation based on in silico genome-scale metabolic model (KoxGSC1457) [38]. The prediction showed that the optimal agitation speed existed on 2,3-BD production. The 2,3-BD productivity was proportional to agitation speed to 450 rpm during fermentation. However, further increase of the agitation speed gradually decreased 2,3-BD productivity [17, 38]. In order to improve overall 2,3-BD productivity of K. oxytoca ΔldhA ΔpflB strain, the effects of agitation speed on 2,3-BD production and cell growth were examined from 150 to 450 rpm, keeping the aeration rate at 1 vvm.

As shown in Fig. 4, the specific cell growth rate (µ) and final cell density (OD600) at 450 rpm were increased by more than 10- and 3.5-fold compared with at 150 rpm, respectively. At higher rpm, glucose consumption rate also increased considerably, and the fermentation period was shortened from 90 to 11 h. As a result, the final 2,3-BD productivity drastically increased from 0.43 to 2.7 g/L/h by raising the agitation speed to 450 rpm; no further increase in 2,3-BD production was observed at agitation speeds >450 rpm (data not shown). It seems that PDC recovered its enzyme activity to yield enough acetyl-CoA to support cell growth by energy generation and biomass synthesis under more aerobic conditions of 450 rpm compared with that of 150 rpm. Accordingly, the glucose uptake rate improved with increased aeration by activating the TCA cycle and the metabolism of amino acids and fatty acids, resulting in an increase of overall 2,3-BD productivity. However, activation of the TCA cycle suggests that more carbon dioxide is released as carbon waste. Indeed, the final 2,3-BD titer decreased from 38.6 to 30.8 g/L with increasing agitation speed to 450 rpm; therefore, 2,3-BD yield reduced from 0.44 to 0.35 g/g. Consequently, further optimization of the fed-batch fermentation process should be carried out to maximize 2,3-BD titer, yield, and productivity.

Fig. 4
figure 4

Batch fermentations of Klebsiella oxytoca ΔldhA ΔpflB strain at different agitation speeds. Fermentation results of a 150 rpm, b 250 rpm, c 350 rpm, and d 450 rpm. Batch fermentations were performed with a working volume of 3 L at 37 °C. Aeration rate was maintained at 1 vvm, and pH was controlled at 6.5 by the automatic feeding of NH4OH. Symbols in fermentation profiles indicate glucose concentration (filled diamonds), 2,3-butanediol (2,3-BD) (open circles), optical density at 600 nm (OD600) (filled squares), acetoin (open squares), ethanol (filled triangles), succinic acid (asterisks), lactic acid (filled circles), formic acid (open diamonds), and acetic acid (open triangles)

Full size image

High yield and productivity strategy for 2,3-BD production by fed-batch fermentation

We found that the agitation speed during fermentation was a significant factor for 2,3-BD production in the K. oxytoca ΔldhA ΔpflB strain (Fig. 4). Increasing the agitation speed favored biomass formation, which resulted in the increase of 2,3-BD productivity. Whereas decreasing the agitation speed could increase 2,3-BD yield, it decreased 2,3-BD productivity. Therefore, it was indispensable to establish a strategy of controlling the agitation speed during fermentation for the efficient production of 2,3-BD to high concentration, high yield, and high productivity. Consequently, we adopted the agitation-speed-control strategy of a fed-batch fermentation, consisting of cell growth and 2,3-BD production phases. In the first (cell growth) phase, agitation speed was maintained at 450 rpm for fast cell growth. In the second (2,3-BD production) phase, agitation speed was decreased to 350 rpm, which was beneficial for establishing the 2,3-BD conversion rate and overall yield.

Prior to fed-batch fermentation by controlling the agitation speed, fed-batch fermentation of K. oxytoca ΔldhA ΔpflB strain with constant agitation speed of 450 rpm was performed to determine the optimal point at which to switch the agitation speed. As a result, the maximum concentration of 2,3-BD reached 57.7 g/L, with a yield of 0.29 g/g and productivity of 1.65 g/L/h (Fig. 5). Fed-batch fermentation was finished before the remaining glucose consumption because the capability of 2,3-BD production was severely depressed after 21 h. As described in Fig. 5, the slope of the trend line for 2,3-BD concentration was maintained until 21 h, which indicated that cellular viability was actively preserved until that time, when the concentration of acetoin reached about 10 g/L. Acetoin, which was mostly produced in the presence of excess oxygen, is known as a toxic metabolite for cellular viability [31]. Accordingly, it is assumed that the disruption of capability for 2,3-BD production was caused by acetoin toxicity. The constant agitation speed of 450 rpm might supply excessive oxygen to inactive cells disrupted by acetoin toxicity. As discussed earlier, the excessive oxygen supply could activate respiration, releasing carbon dioxide as a carbon waste, which is not beneficial for efficient 2,3-BD production. Consequently, a fed-batch fermentation comprising a cell growth and a 2,3-BD production phase was carried out by switching the agitation speed to maintain the concentration of acetoin at <10 g/L and to achieve a high yield and productivity of 2,3-BD (Fig. 6).

Fig. 5
figure 5

Fed-batch fermentation with constant agitation speed of 450 rpm by Klebsiella oxytoca ∆ldhA ∆pflB strain. Fed-batch fermentation was performed with a working volume of 3 L at 37 °C. The aeration rate was maintained at 1 vvm; pH was controlled at 6.5 by the automatic feeding of NH4OH. Dotted line represents the trend of 2,3-butanediol (2,3-BD) production. Symbols in fermentation profiles indicate glucose concentration (filled diamonds), 2,3-BD (open circles), optical density at 600 nm (OD600) (filled squares), acetoin (open squares), ethanol (filled triangles), succinic acid (asterisks), lactic acid (filled circles), formic acid (open diamonds), and acetic acid (open triangles)

Full size image
Fig. 6
figure 6

Fed-batch fermentation by controlling agitation speed (450 → 350 rpm) by Klebsiella oxytoca ΔldhA ΔpflB strain. Fed-batch fermentation was performed with a working volume of 3 L at 37 °C. Aeration rate was maintained at 1 vvm, and pH was controlled at 6.5 by the automatic feeding of NH4OH. Agitation speed was maintained at 450 rpm until acetoin concentration was reached at about 10 g/L and then switched to 350 rpm. Symbols in fermentation profiles indicate glucose concentration (filled diamonds), 2,3-butanediol (2,3-BD) (open circles), optical density at 600 nm (OD600) (filled squares), acetoin (open squares), ethanol (filled triangles), succinic acid (asterisks), lactic acid (filled circles), formic acid (open diamonds), and acetic acid (open triangles)

Full size image

In fed-batch fermentation, agitation speed was maintained at 450 rpm until acetoin concentration reached ~10 g/L, after which the agitation speed was switched to 350 rpm to decrease acetoin formation and to maintain the capability for 2,3-BD production. It was observed that the maximum concentration of 2,3-BD reached 113 g/L with a yield and productivity of 0.45 g/g and 2.1 g/L/h, respectively, in the fed-batch fermentation (Fig. 6). Results were 95.2 % (concentration), 55.2 % (yield), and 27.3 % (productivity) higher than those of the fed-batch fermentation with a constant agitation speed of 450 rpm. In particular, the final 2,3-BD yield from glucose achieved up to 90 % of the theoretical maximum yield (0.5 g/g glucose). Controlling the agitation speed during fed-batch fermentation could not only significantly improve the final concentration and yield of 2,3-BD but also increase the productivity of 2,3-BD efficiently.

Conclusion

In order to improve the efficiency of K. oxytoca for the production of 2,3-BD and to improve its potential for industrial applications, a K. oxytoca ΔldhA ΔpflB strain was developed based on in silico aided metabolic engineering. In silico analyses predicted that inactivation of ldhA encoding lactate dehydrogenase and of pflB encoding pyruvate formate lyase effectively reduced byproduct formation and enhanced 2,3-BD production. The mutant was successfully evaluated by a batch fermentation with a high yield but low 2,3-BD productivity compared with wild-type and ldhA knockout strains. In order to improve 2,3-BD productivity, the effect of agitation speeds from 150 to 450 rpm was investigated. Then, the fed-batch fermentation with constant agitation speed of 450 rpm was performed, first to examine the optimal time for converting the agitation speed. Subsequently, fed-batch fermentation by controlling the agitation speed was performed by switching the agitation speed from 450 to 350 rpm. Finally, the high concentration, yield, and productivity of 2,3-BD were achieved by fed-batch fermentation up to 113 g/L, 0.45 g/g, and 2.1 g/L/h, respectively. This is an example showing the effectiveness of system metabolic engineering for strain improvement and the importance of agitation speed during fermentation for 2,3-BD production. These results demonstrate one of the best concentration, yield, and productivity of 2,3-BD by fermentation of glucose with K. oxytoca, compared with values (e.g., 130 g/L, 0.48 g/g, and 1.64 g/L/h; Ji et al. [21]) previously reported [4, 9, 17, 21, 39]. The strategies of in silico aided metabolic engineering and fermentation process optimization could be applied to overcome the difficulties and hurdles of industrial application of K. oxytoca. Likewise, it is expected that the strategies employed in this study will apply to the development of industrial processes on high concentration, high yield, and high productivity of the other products.