Background

Bacterial glucose metabolism is particularly diverse and involves the orchestrated actions of multiple pathways, including the Embden–Meyerhof–Parnas pathway (EMPP), Entner–Doudoroff pathway (EDP), pentose phosphate pathway (PPP), and oxidative pathway (via gluconic acid) [1, 2]. These pathways produce pyruvate, reducing power (NADH and/or NADPH), and energy (ATP), and play a central role in growth and bio-production [3, 4]. In contrast to Escherichia coli, which primarily uses the EMPP for glucose metabolism, Pseudomonas, and Zymomonas spp. employ the EDP, as they lack phosphofructokinase (Pfk), a pivotal EMPP enzyme. The PPP is a metabolic pathway that functions in parallel with either the EMPP or EDP and supplies the energy requirements for cell growth, NADPH regeneration, and replenishment of essential precursors for nucleotide and amino acid synthesis.

The EMPP comprises 10 enzymatic steps, yielding two pyruvate, two net ATP, and two NADH molecules per one glucose molecule, whereas the EDP utilizes only five enzymes to have two pyruvates (one of which is produced from glyceraldehyde-3-phosphate (G3P) via the lower glycolysis), one net ATP, one NADH, and one NADPH molecule per one glucose molecule. The EDP has been reported to have certain advantages compared with the EMPP [5]. For example, it is the thermodynamically more favorable pathway, owing to lower ATP production, in turn, boosting glycolysis [6]. Considering the number of enzymatic steps involved, the costs of protein synthesis in EDP is lower than that of EMPP [7]. It is also characterized by less allosteric regulation due to the lack of Pfk catalysis, one of the committed steps in the EMPP, which is found to be subjected to intensive regulation at the transcription, translation, and even post-translation and allosteric levels via distal metabolites, as well as energetic precursors [8]. The EDP is also associated with higher NADPH production to provide high reducing power for biosynthesis, resistance to oxidative stress, and the optimal synthesis of both the carbon and redox precursors for terpenoid biosynthesis. For example, Z. mobilis that relies extensively on the EDP shows notably rapid glucose uptake and ethanol production rates [9], and Pseudomonas species that utilizes an incomplete EMPP shows high production of highly toxic chemicals associated with their innate high solvent tolerance. Besides, the absence of a Pfk step facilitates the cyclization of PPP, which regenerates excess NADPH [10, 11].

Nevertheless, although there are several distinct benefits of employing EDP, the industrial strains of E. coli do not use their inherent EDP for glucose metabolism. Several studies have been conducted to engineer E. coli in an effort to activate the intrinsic EDP and to block the EMPP to promote more efficient biosynthesis. Previously, blockage of the EMPP via deletion of the phosphoglucose isomerase gene (pgi) resulted in re-routing the glycolytic flux through the PPP and EDP. The Δpgi recombinant E. coli strains were characterized with enhanced production of polyhydroxybutyrate [12], isoprenoid [13], and hydrogen [14]. However, although the Δpgi mutants remain viable, given that the glycolytic flux is re-routed through the PPP and EDP, excess NADPH production was found to perturb a significant portion of the metabolic network, resulting in an approximately 80% reduction in cell growth in glucose minimal media when compared with the wild-type levels [15]. Disrupting the EMPP by knocking out the phosphofructokinase I (PfkA) gene also resulted in re-routing the glycolytic flux through the PPP (~ 60% of the glycolytic flux) and the native EDP (~ 14% of glycolytic flux) [16]. The ΔpfkA recombinant E. coli strains showed enhanced production of 1,3-diaminopropane [17], hydrogen and ethanol [18], lycopene [19], and methyl 3-hydroxybutyrate [20]. However, these strains showed a more significant decrease in growth rate compared with the Δpgi mutants owing to the partial cyclization of PPP, which may produce excess NADPH [21]. E. coli in which the EMPP was completely blocked by deleting pfkAB showed retarded growth on glucose minimal medium due to redox imbalance and reduced glucose uptake, suggesting the PPP and EDP were not sufficiently activated to alleviate the metabolic defect.

In this study, we isolated and characterized a pfkAB-deleted E. coli mutant strains that can thrive on glucose minimal medium with adaptive laboratory evolution (ALE). Using these strains, we investigated the genotype–phenotype relationship to determine and classify the requirements for the full activation of the EDP in E. coli. Finally, as a proof of concept and based on the preliminary information obtained, we further engineered the evolved mutant to confer it with the ability to produce a higher level of 3-hydroxypropionic acid (3-HP) that requires more NADPH regeneration.

Results and discussion

Adaptive evolution of the ∆pfkAB strain and analysis of mutations identified in the evolved strains

Given that E. coli has two Pfk isozymes, Pfk-I and Pfk-II encoded by pfkA and pfkB, respectively [22], a recombinant E. coli strain with complete inactivation of the EMPP was constructed by deleting both of pfkA and pfkB genes. The ∆pfkAB strain showed slower specific growth rate on minimal medium containing glucose as a sole carbon source (Fig. 1a). ALE experiments were performed to recover the reduced cell growth by activating the EDP. After 50 subcultures of the ∆pfkAB strain in M9 minimal medium supplemented with 5 g/L glucose, the strain thereafter grew more rapidly on glucose. The five evolved ∆pfkAB strains showing recovered cell growth were isolated and denoted as pfk_ALE-1, -2, -3, -4, and -5 strains (Additional file 1: Figure S1a). The ALE-1 exhibited 92% of cell mass accumulation of the wild-type MG1655 strain after culturing for 20 h. This is the first E. coli strain with a completely inactivated EMPP that can grow more rapidly than mutants with partial blockage of the EMPP, such as the ∆pfkA and ∆pgi mutants (Fig. 1a). Besides, the ALE-1 strain showed improved terpenoid biosynthesis via the methylerythritol phosphate pathway, which has been observed in EDP-activated E. coli strains, because the synthetic pathway starts with 1-deoxy-d-xylulose 5-phosphate (DXP) synthase reaction with the two substrates G3P and pyruvate that are the final products of EDP. In addition, the synthetic pathway uses NADPH as sources of reducing regent produced by EDP. Interestingly, the production culture conditions used in the present study were able to rescue cell growth of the ALE-1 strain, with a 115% higher cell mass accumulation than the wild type at 36 h of culturing (Additional file 1: Figure S2).

Fig. 1
figure 1

Comparison of cell growth on glucose among EMPP-disrupted strains and the effect of mutations on cell growth of the evolved strains and their background strains. a Growth rate profiles (OD600) were compared for three groups of E. coli strains: wild type with MG1655, complete disruption of EMPP with background strain (∆pfkAB), and evolved strain (ALE-1), and partial disruption of EMPP with the phosphofructokinase I step (∆pfkA) and phosphoglucose isomerase step (∆pgi). Error bars indicate standard deviations of measurements for three independent biological replicates. b Genetic characterization of five independent evolved ∆pfkAB isolate strains with genome resequencing analysis. The number of unique mutations per gene is noted within brackets. c The effect of mutations on cell growth were determined by introducing single or multiple mutations in the background strain ∆pfkAB. For the mutation in ptsG, the IS5 insertion disrupted the protein, and consequently, PtsG would lose its functionality. Thus, we knocked out the ptsG gene instead of knocking-in the IS element. Error bars indicate standard deviations of growth measurements for three independent biological replicates

Full size image

We sequenced the whole genome of ALE-1 and compared the sequence with the parental strain, E. coli MG1655. Resequencing identified six mutations in ALE-1 (Table 1). To confirm enriched and major mutations during ALE, the mutations detected in the ALE-1 were analyzed in the rest of four evolved mutants by PCR amplification and DNA sequencing. Mutations were found in the gntR, galR, ptsG, and ugd-gnd-wbbL-2 sequences of all five pfk_ALE strains, whereas mutations in the talB and crp genes were detected only in one or two strains (Fig. 1b). Notably, galR showed three unique mutations during adaptive evolution, but all resulted in truncated and non-functional protein (Additional file 1: Figure S1b, c). This suggests that the galR mutation could be critical for the growth of the evolved strain. In contrast, mutation in the talB gene was only observed in the ALE-1 strain and crp mutation was detected in the ALE-1 and ALE-4 strains, probably indicating that these mutations might not considerably affect the physiological background during ALE [23,24,25].

Table 1 List of mutations identified in the evolved strain pfk_ALE-1
Full size table

To compare the effects of the above mutations on the growth of ALE-1 strain, single mutations or combinations of the multiple mutations were introduced into the background ΔpfkAB strain. The growth rates of the resulting strains were compared with that of the background and evolved strain. The single mutation in the gntR, galR, crp caused an increase in the specific growth rate of 1.2-fold, 1.1-fold, and 1.2-fold, respectively, relative to the background ΔpfkAB strain. The single mutation in talB and the ptsG deletion mutant of the background strain showed a slight decrease in the specific growth rate. The background strain carrying a deletion of the ugd-gnd-wbbL-2 region showed an even slower cell growth. When each of the six mutations was introduced independently, the restored cell growth levels of the resulting mutants were considerably lower than that of the evolved strain. Eventually, when the mutant was harboring all six mutations, the growth rate was comparable to ALE-1 (Fig. 1c). Based on these observations, we deduced that the enhanced growth differentiating ALE-1 from the background mutant was attributed not only to the occurrence of a single mutation in gntR, galR, ugd-gnd-wbbL-2, ptsG, crp, and talB, but also to their synergistic interaction.

GntR is a DNA-binding transcriptional regulator that represses the expression of genes involved in gluconate metabolism in the absence of the effector molecule D-gluconate [26]. Gluconate is phosphorylated by gluconate kinase to produce 6PG, which is also produced from glucose. It is finally metabolized to G3P and pyruvate via the 6-phosphogluconate dehydratase (Edd) and 2-keto-3-deoxy-6-phosphogluconate aldolase (Eda) enzymes of EDP. Knockout of gntR has previously revealed that the expression of edd and eda genes can occur even in the absence of gluconate [27]. Therefore, we hypothesized that the loss-of-function mutation in the gntR gene de-represses the originally latent genes of the EDP, thereby providing a major route for glucose metabolism in ALE-1 with an inactive EMPP (ΔpfkAB) and an inactive oxidative PPP (Δgnd) (Table 1) [28]. In fact, ALE-1 Δedd strain could not grow on glucose, indicating that ALE-1 relies on the EDP on glucose as the sole carbon source (Additional file 1: Figure S3).

GalR is a transcriptional regulator that down-regulates the genes associated with galactose metabolism when this sugar is unavailable [29]. The regulon of GalR harbors the galP gene encoding galactose permease, which also has glucose uptake activity [30]. Wild type E. coli mainly takes up external glucose by a phosphoenolpyruvate-dependent phosphotransferase system (PTS), which consists of enzyme I (PtsI), HPr (PtsH), and glucose-specific PTS EIIA (Crr) and EIICB (PtsG) components [31]. Thus, we anticipated that the evolved mutant that lost the PTS activity due to PtsG inactivation by insertion of the IS5 and SgrS regulation of sugar transporter mRNAs (ptsG and manXYZ) due to the accumulation of glucose-6-phosphate, and consequently, the strain can co-metabolize glucose and xylose, which was the case in the ptsG-deleted E. coli strains (Additional file 1: Figure S7d–f) [32,33,34]. It has also been reported that the loss of PtsG function results in retarded cell growth due to the limited capacity of glucose import, and non-PTS glucose transporters such as GalP can restore the glucose uptake rate in ptsG null mutants [35, 36]. Based on this assumption, we hypothesized that the expressed GalP, which is depressed by the inactivation of galR, rescues the glucose uptake capacity of ALE-1.

Moreover, another possible reason for the expression of a non-PTS glucose permease GalP in EMPP-disrupted strains could be the low availability of phosphoenolpyruvate (PEP), which is essential for the PTS-dependent glucose uptake. Previous studies showed that the ∆pfkA mutant had reduced PEP concentration [15]. Besides, an increase in flux through the EDP can further lower the intracellular PEP pool [16]. It is supported by the stoichiometry of the EDP, where PEP generation per glucose is half of that of the PPP or EMPP [13]. Limited glucose importing capacity due to low PEP availability can be supplemented by the expression of PEP-independent glucose permease and glucokinase [35].

cAMP receptor protein (CRP) is a global transcription regulator that directly or indirectly regulates the expression of over 100 genes [37,38,39]. The mutation site of CRPA85S was very close to the cAMP binding site, suggesting the possible effect on the binding affinity. However, most of cAMP-independent CRP mutants contained mutations away from the binding site such as I112, T127, and A144 residues [40]. Therefore, it is hard to establish that the CRPA85S is a cAMP-independent mutant. Since no significant difference in the cell growth was observed by this mutation, the activity of this enzyme was not affected (Fig. 1c). However, the exact mechanism underlying the crp mutation-mediated growth rate enhancement in the evolved strain remains unclear. We discovered that crp mutation occurred in the later period of adaptation (during 30–50 repeats of culture) while ptsG mutation was detected in the earlier period of adaptation (during 0–30 repeats of culture). Although PTS inactivation by ptsG mutation is an efficient way to trigger carbon catabolite repression (CCR) release like galP expression, the modification of binding between CRP and cAMP might further contribute the release of CCR and provide a strong selective pressure during adaptation [41].

TalB is an enzyme involved in the non-oxidative PPP and provides a reversible reaction between the PPP and the upper EMPP [42]. The A247 residue is located in an α-helix and away from the active center of E. coli TalB. The B-factor value of the mutated residue was kept considerably lower, indicating that the mutation stably maintains its α-helix structure. In addition, the substrate binding affinity of the TalBA247T in E. coli W3110 strain was not modified in a previous study [43]. Moreover, the expression level of the talB gene in ALE-1 was not significantly changed (1.3-fold) by this mutation compared to that in MG1655 (Additional file 1: Figure S5). Since the ALE-1 strain lacks the gnd gene, a change in the non-oxidative PPP flux might be directed by a genetic mutation in talB during adaptation to supply precursors essential for growth.

Functional analysis of the genetic mutations with transcriptomic and phenotypic data

Functional analysis of the mutations detected in the ALE-1 strain was conducted to elucidate the molecular mechanisms underlying the effects of these mutations. We performed transcriptome analysis with RNA-seq to investigate the transcriptomic effects of mutations involved in the glycolytic pathway and global transcriptional regulation. Transcriptomic analysis revealed the expression levels of 4497 genes on the chromosome of MG1655 and ALE-1 strains. Among those genes, we first focused on the glycolytic pathways including the EMPP, EDP and PPP. Complete deletion of the pfkA gene was confirmed with no pfkA expression. In ALE-1, the expression of the edd and eda genes, which are involved in the EDP, were 11.2- and 5.1-fold higher than in MG1655, respectively. This indicates that the latent EDP was activated during adaptation due to loss of function for GntR. Additionally, the expression level of the GntR regulon in ALE-1 was higher compared to MG1655. There was no expression of gnd, involved in the oxidative PPP, due to its complete deletion in ALE-1 (Fig. 2a). Taking these observations into consideration, the ALE-1 strain utilizes glucose via the EDP rather than the EMPP and oxidative PPP since ALE-1 lacks functional PfkA, PfkB, GntR, and Gnd. The RNA-seq analysis also showed lower expression level of ptsG in ALE-1 compared to MG1655, caused by degradation of ptsG mRNA. Instead, the expression level of the non-PTS glucose transporter, galP, in ALE-1 was 13.7-fold higher than MG1655. This phenomenon was also confirmed by quantitative RT-PCR, demonstrating 7.6-fold increased expression level (Additional file 1: Figure S5). Notably, we found that the expression levels of the galR regulon in ALE-1 was higher than that in MG1655 due to the inactivation of GalR (Fig. 2a). These indicate that the limited glucose uptake capacity was rescued by the GalP via the inactivation of GalR. However, only a small change was observed in the gntR and galR expression level between the MG1655 and ALE-1 strains, which suggests that the transcriptional levels of gntR and galR were not affected by the genetic mutation. Instead, mutations in gntR and galR might result in loss of protein function since the expression levels of their regulon were down-regulated. During evolution, ALE-1 activated the latent pathways, including the glyoxylate shunt and phosphoenolpyruvate carboxykinase (Pck) reaction. Since the glyoxylate shunt does not regenerate NADPH unlike the isocitrate dehydrogenase (Icd) reaction, the evolved strain might enhance the glyoxylate shunt flux rather than the Icd reaction to reduce the NADPH regeneration capacity. Overall, transcriptome analysis suggested that the glycolytic flux was re-routed through the activation of latent pathways and rescued the glucose uptake capacity in ALE-1. On the other hand, even though glucose can be utilized via mannose PTS [41], the expression levels of manX, manY, and manZ in ALE-1 were 2.9-, 2.0-, and 1.5-fold reduced compared to those in MG1655, respectively (Fig. 2a).

Fig. 2
figure 2

Functional analysis of major mutations found in the evolved strains. a RNA-seq transcriptome analysis illustrated the expression level of genes involved in glucose uptake system, glycolytic pathway, TCA cycle and transcriptional regulons including gntR and galR, crp regulons and sugar phosphate stress. Expression change (log2 fold change) of pfkA and gnd genes was unable to be calculated due to loss of expression and it was marked with an asterisk (*). Unexpected expression level of the pfkB gene of ALE-1 was affected by a remained coding sequence of pfkB (Additional file 1: Figure S8). b Functional validation experiments identified effective target genes among mutated genes from ALE-1: lacking the Gnd function, and loss of function for GntR and de-repression of its target genes, edd and eda. Error bars indicate standard deviations of growth measurements for three independent biological replicates. c Combined fluorescence histograms of MG1655, ALE-1, MG1655 Δpgi ΔgntR strains carrying pHexR-GFP. d The schematic diagram illustrates altered glucose metabolism in ALE-1 through the EDP, via enzymes encoded by edd and eda. Transcriptional repression by GntR is indicated with red marks

Full size image

To further confirm our hypotheses, we initially deleted the gntR, galR, and ptsG genes in the ALE-1 strain. We found that the deletion of either the gntR, galR or ptsG gene did not affect the growth of ALE-1 cells, whereas restoration of these mutations resulted in notable adverse effects on the growth of ALE-1 cells (Fig. 2b). These results imply that GntR and GalR mutants may be unable to repress the expression of their target genes, including edd, eda, and galP. As expected, the same phenotype was observed in the ALE-1 strain that had a transposon inserted in ptsG or had a deleted ptsG mutation, confirming that the ptsG gene with the IS insertion lost its activity. To verify the target genes of the two TFs, we expressed the plasmid-encoded endogenous edd-eda or galP gene in the gntR- and galR-restored ALE-1, respectively, resulting in increased growth from 36 and 55% up to 82% and 86% of ALE-1, respectively. These observations thus indicate that the plasmid-based expression of edd-eda and galP genes can functionally recover cell growth defects caused by the restored active GntR and GalR in ALE-1 (Fig. 2b and Additional file 1: Figure S4a). To verify the effect of the large deletion of the ugd-gnd-wbbL-2 region carrying the two functional genes ugd and gnd on ALE-1 cell growth, we re-introduced the deleted region into the chromosome of ALE-1. The restored strain showed reduced cell growth, and then the deletion of the gnd gene, but not the ugd gene, resulted in a substantial recovery of the growth defect caused by the restoration of the ugd-gnd-wbbL-2 region (Fig. 2b and Additional file 1: Figure 4a).

According to the transcriptome analysis, we hypothesized that the ALE-1 uses EDP instead of the EMPP or oxidative PPP as the major glycolytic pathway. Since 2-keto-3-deoxy-6-phosphogluconate (KDPG) is an intermediate of EDP, intracellular KDPG level should be higher in ALE-1 compared to its wild-type strain. It was previously reported that the transcription of hexR, that originated from P. putida KT2440, was repressed in the presence of KDPG produced from glucose or gluconate [44]. To analyze the EDP flux, a KDPG-responsible biosensor using a transcriptional regulator (HexR) and GFP was constructed to sense the intracellular level of KDPG. Using an appropriate gate, M1, 85.8% of the population of ALE-1 and 9.5% of the wild-type MG1655 were detected (Fig. 2c). Previously, the glycolytic pathway of Δpgi ΔgntR double-knockout E. coli strain was redirected from EMPP to EDP [28]. In fact, 64.2% of the population of the Δpgi ΔgntR double-knockout cells were detected with M1 gate, which was higher than wild-type strain. This KDPG-responsible biosensor successfully supported our hypothesis that the EDP of ALE-1 was more active than that of the wild-type.

In conclusion, the growth of the EMPP-deficient mutant was facilitated by a number of factors that includes, 1) the re-direction of the glucose flux from the oxidative PPP to EDP by the deletion of gnd, 2) the gain of the EDP activity by the lack of gntR, and 3) the recovery of glucose uptake by the inactivation of GalR (Fig. 2d and Additional file 1: Figure S4b).

Prioritization between the EDP and PPP for optimal cell growth

Notably, the evolved strain had the gnd mutation resulting in the inactivation of oxidative PPP, an alternative glycolytic pathway. The ALE-1 strain thus employs only the EDP for glycolysis, indicating that an active PPP might harm the growth of bacterial cells using the EDP as the main glycolytic pathway. The availability of the ALE-1 strain with plasmid-based Gnd expression enabled us to assess the effect of glucose catabolism via both oxidative PPP and EDP on its cell growth (Fig. 3a, b). As expected, expression of Gnd conferring oxidative PPP activity together with the EDP had an adverse effect on the growth of ALE-1 cells, leading to a 23% decrease in the growth rate of the evolved strain. We further investigated the effect of the further increased glycolytic flux from the EDP to oxidative PPP on the growth of ALE-1 cells by expressing both gnd and gntR on a plasmid, where Gnd reactivates oxidative PPP and GntR represses the EDP (Fig. 3a, c). As a result, high glycolytic flux to oxidative PPP resulted in more severe growth defects, resulting in a 39% reduction in cell growth, but exhibiting a 120% higher NADPH/NADP+ ratio (Fig. 3d, e). Using HexR-based KDPG-responsible biosensor, we further examined the activity of EDP in ALE-1 with plasmid-based expression of the gnd and/or gntR genes to confirm that expression of those genes led to redirect glycolytic flux. As a result, the concentration of intracellular KDPG was decreased when only gnd was expressed and further decreased when both the gnd and gntR were expressed (Additional file 1: Figure S6). This observation led us to speculate about why the oxidative PPP function was lost instead of being conserved to handle the glycolytic flux together with the EDP. It has been suggested that one of the reasons why the E. coli Δpgi mutant, which is primarily dependent on the PPP for glucose catabolism, shows reduced glucose uptake rate is due to a redox imbalance and/or an excess of NADPH regenerated from the PPP [45]. When glucose is metabolized via the EDP, one less NADPH is regenerated than via the PPP in this mutant [46]. However, the PPP of the pfkAB knockout mutant can regenerate excess NADPH even more intensively due to its cyclized architecture compared with the pgi-null mutant [10]. Taking these observations into consideration, the effect of re-routing the glycolytic flux to the EDP on the reduction of the regeneration of NADPH would be more significant in the latter mutant. Despite partial relief afforded by the EDP, the NADPH formed via the cyclized PPP might exceed the demand for cell growth.

Fig. 3
figure 3

The effect of the redistribution of metabolic flux between PPP and EDP on cell growth and the NADPH/NADP ratio. a Schematic diagrams for possible glycolytic flux redistribution in ALE-1 pRFP (active EDP), ALE-1 pGnd (active both PPP and EDP), and ALE-1 pGnd-GntR (more active PPP and weaker EDP). The effect of gradually increasing glycolytic flux redistribution from the EDP to the PPP in the evolved ALE-1 strain on (d) cell growth and e [NADPH]/[NADP+] ratio. Error bars indicate standard deviations of growth measurements for three independant biological replicates. An asterisk (*) indicates a significant difference compared to the ALE-1 pRFP (p < 0.05)

Full size image

Moreover, the EDP has been known to be thermodynamically favorable, even when compared with the native glycolytic route EMPP. This thermodynamically favorable property enables the EDP to sustain a glycolytic flux comparable to that of the EMPP [6, 7]. Indeed, Z. mobilis, which extensively employs the EDP for glycolysis, takes up glucose at a notably rapid rate [9]. This observation indicates that the capacity and turnover rate of the EDP would be substantial [47, 48]. Considering the foregoing observations, establishing the EDP as the sole glycolytic route, rather than sharing the flux with the PPP, would benefit the growth of the Pfk-deficient mutant in terms of both reducing NADPH production and expanding glycolytic capacity during selection. Consistent with this hypothesis, the soil bacterium P. putida KT2440, which lacks a functional EMPP, is known to be almost extensively dependent on the EDP for glycolysis, with less than 10% of the glucose flux entering the PPP [2].

Dynamic regulation of the EDP and PPP for biotechnological applications

To facilitate optimal growth, cells must stringently balance their intracellular redox. However, the microbial production of value-added chemicals often requires an abundant supply of reducing powers in the form of NADPH. Accordingly, there exists a fundamental difference in the optimal architecture of the core metabolic pathways to sustain normal growth and those required for metabolite overproduction. For the successful application of the evolved mutant described herein for bio-production, further manipulation of the core metabolic pathways would be necessary. For proof of concept, we introduced changes into the evolved strain ALE-1 machinery for it to attain an abundant supply of NADPH necessary for the production of 3-HP, an intermediate for many compounds. The bifunctionality of 3-HP enables it to be further transformed into a variety of high-value compounds [49,50,51]. From 1 mol of glucose, it is theoretically possible to generate 2 mol of 3-HP with a consumption of 4 mol of NADPH (Fig. 4a). However, its redox equivalent demand is far beyond the endogenous catabolic supply based on glucose catabolism via the EDP or the linear PPP, which regenerates 1 or 2 mol of NADPH per mol of glucose [13]. According to the stoichiometry of the EDP and PPP, restoration of cyclized PPP activity and the reduction of EDP activity might yield larger amounts of 3-HP [10, 13]. To address this issue, we demonstrated the conditional flux control towards the EDP and PPP at the 6PG node for the enhanced production of 3-HP.

Fig. 4
figure 4

Rebalancing carbon flux through the PPP and EDP for enhanced 3-HP production. a A diagram for genetic strategy for the efficient production of 3-HP. A heterologous 3-HP synthetic pathway was introduced by expressing two dissected subparts, N-terminus and C-terminus ones, from the mutant malonyl-CoA reductase gene (mcr) derived from C. aurantiacus. ACC was overexpressed for malonyl-CoA accumulation while fatty acid synthesis was reduced by expressing a temperature-sensitive FabI at 37 °C. b 3-HP production titer at 48 h post-induction. Error bars indicate standard deviations of 3-HP titer for three independent biological replicates, respectively

Full size image

Before dealing with the redox issue, we attempted to eliminate a potential limiting factor, namely, the low availability of malonyl-CoA for 3-HP production. Efficient microbial production of chemicals such as flavonoids and fatty acids using malonyl-CoA as a precursor are often constrained by the low intracellular concentrations of this compound [52]. In the present study, we overexpressed acetyl-CoA carboxylase (ACC) to enhance the supply of malonyl-CoA, and in conjunction performed site-directed mutagenesis designed to increase the temperature sensitivity of endogenous enoyl-[acyl-carrier-protein] reductase (FabI), thereby reducing the malonyl-CoA consumption via fatty acid synthesis [53, 54] (Fig. 4a). At 48 h post-induction, the evolved strain ALE-1 produced 1.34 g/L of 3-HP, whereas the background strain produced 0.83 g/L. Although the evolved mutant produced a 1.6-fold higher 3-HP titer than the background strain under the same conditions, the fold-increase was subtle, considering there was a substantial difference in biomass accumulation between the two strains (Fig. 1a and Fig. 4b).

To achieve conditional flux control towards the EDP and PPP, the E. coli gnd and gntR genes were expressed under the control of an IPTG-inducible promoter, and the resulting pGnd and pGnd-GntR plasmids were transformed into ALE-1 harboring the 3-HP-producing machinery. Induction of flux redistribution to both the EDP and PPP in the 3-HP-producing ALE-1 mutant harboring pGnd led to a considerable enhancement in the 3-HP titer (1.99 g/L) at 48 h post-induction. Furthermore, when the EDP was repressed by additional expression of its repressor GntR, 3-HP production in the evolved mutant increased further to 2.65 g/L, which is 3.2- and 3.3-fold higher than the titer obtained from the background mutant and wild-type E. coli strains, respectively, harboring the same 3-HP synthesis machinery (Fig. 4b and Additional file 1: Figure S2). These findings thus indicate that facilitating glucose flux via the PPP and reducing the flux through the EDP are critical for generating an abundant NADPH supply, resulting in higher 3-HP production.

Conclusions

In this study, we generated and characterized an engineered E. coli using EDP rather than EMPP as the major glycolytic pathway, whose pathway is not functional in the wild-type strain. Based on genetic and biochemical studies, we demonstrated that optimal cell growth of E. coli strain through the EDP requires three main metabolic changes: (1) de-repression of the EDP by inactivation of GntR, (2) inactivation of the oxidative PPP by deletion of Gnd, and (3) an increase in glucose uptake by the inactivation of GalR. Reverse engineering of the evolved strain was applied to enhance the production of 3-HP by increasing glycolytic flux from the EDP to PPP for the supply of excess NADPH during the production stage rather than the growth stage. The engineered strain and its derivatives have certain advantages with respect to specific bio-production purposes, such as 3-HP and lycopene production, which demand NADPH (Fig. 4b and Additional file 1: Figure S7a, b).

Two EMPP-disrupted E. coli mutants, in which the pathway was blocked at the Pfk or Pgi step, resulted in distinct genotypic changes after ALE with glucose as the sole carbon source [18, 23, 55]. The evolved E. coli strain lacking pfkA showed a mutation in the promoter region of pfkB coding for a minor isozyme of Pfk. The expression of Pfk increased with time and this enabled the strain to restore glucose metabolism through the EMPP and could support cell growth. This observation indicates that, among three glycolytic pathways (EMPP, EDP, and PPP), the EMPP may be the most conducive for maximal cell growth in E. coli strains. The loss of Pgi led to a significant increase in relative flux through the PPP. None of the evolved Δpgi mutants could influence the absolute EDP flux. Instead, independent studies have revealed that mutations leading to enhanced activity of transhydrogenase have repeatedly been detected in adapted Δpgi mutants, suggesting that these mutations may reduce NADPH imbalance [23, 55,56,57]. In the present study, we found that the activation of the latent EDP by de-repressing downregulation was one of the key fitness-enhancing factors during the adaptation of the Pfk-null mutant. These findings imply that there are fundamental differences in the metabolic apparatus of the Pfk- and Pgi-null mutants, although both are widely used to redirect the glycolytic flux to the EMPP. One of them is that PPP can be cyclized and produce more NADPH in the Pfk-null mutant.

Although slightly slower growth was found in ALE-1 compared with the wild-type strain that employs the EMPP as the main glycolytic pathway, the evolved strain generated in this study grew well on glucose minimal medium compared with pgi- or pfkA-deleted strains that are characterized by a partially inactive EMPP. An additional factor that may contribute to the suboptimal growth of the evolved strain is that an imbalance or excess of NADPH produced from EDP can reduce cell growth when glucose is used as the sole carbon source. Notably, E. coli expressing G6PDH is strictly NADP+-specific compared with that in other bacteria that primarily employ the EDP as the glycolytic pathway [58,59,60]. Previous studies have reported that the growth of the pgi mutant could not be restored fully to that of the wild type rate by overexpressing UdhA for the re-oxidation of NADPH, indicating that the transhydrogenase system is not sufficient to overcome an imbalance in NADPH [45]. When NADPH-consuming pathways (lycopene and 3-HP production) were engineered in ALE-1, we observed the recovery of cell growth; the lycopene- and 3-HP-producing ALE-1 cell showed higher cell growth than the wild type. These findings suggest that the evolved strain could be a suitable host cell for NADPH-dependent bio-production, as slightly reduced cell growth minimally affects production.

The selection of a suitable host cell for the production of a biotechnology-based product is crucial to ensure economic viability. The bacterium, E. coli, is one of the best-characterized microorganisms from the genetic and metabolic perspectives and has a variety of tractable traits for biotechnological production. However, tolerance to adverse environmental stress, or unfavorable redox potentials may impede the use of E. coli as a platform host organism. The selection of host cells for production should be based on several factors, including the availability of genetic tools for expressing target products, efficient utilization of alternative carbon sources, high redox potential, and a tolerance to the metabolic products. Furthermore, unlike Pseudomonas and Zymomonas, ALE-1 can co-metabolize glucose and xylose, which was the case in the ptsG-deleted E. coli strains (Additional file 1: Figure S7d–f). ALE-1 strain even grows under anoxic conditions using nitrate as a final electron acceptor (Additional file 1: Figure S7c).

In conclusion, we have demonstrated that the normally latent EDP in E. coli can be almost fully activated by enhancing glucose uptake through an inactivation of GalR, inactivating the EDP regulatory system, and blocking the glycolytic flux via the PPP. The resulting strain and its derivatives have advantageous traits that are found in microorganisms using the EDP and PPP as the major glycolytic route, and it is amenable to host strong redox reactions. The engineered strains could expand the utility of E. coli as a platform for biotechnological production.

Materials and methods

Reagents, strains and media

Unless otherwise indicated, the reagents used in this study were purchased from Sigma–Aldrich (St. Louis, MO, USA).

The bacterial strains and plasmids used in this study are listed in Additional file 1: Table S1. The wild-type E. coli MG1655 strain was used as a parent strain for the generation of mutant strains. The E. coli DH10B strain was used for bacterial transformation and plasmid amplification during the construction of the expression plasmids.

Unless otherwise stated, as growth media, we used M9 minimal medium (BioShop Canada Inc., Burlington, Canada) supplemented with 2 mM MgSO4, 0.1 mM CaCl2, and glucose with appropriate antibiotics, if required, and Luria–Bertani (LB) medium. The LB medium was prepared using 5 g/L yeast extract, 10 g/L peptone and 10 g/L NaCl. Antibiotics were used at the following concentrations: ampicillin (100 mg/L), kanamycin (50 mg/L), and chloramphenicol (30 mg/L).

Cell culture

For growth measurements, a single colony from an agar plate was inoculated in 3 mL of LB medium at 37 °C and shaken at 200 rpm. Overnight-grown cells were 1:10 diluted into 5 mL of M9 medium supplemented with 4 g/L glucose. After growing for 12 h, cells were inoculated into 25 mL of M9 medium in 250-mL shaking flask at 37 °C and shaken at 200 rpm to achieve an initial optical density at 600 nm (OD600) of approximately 0.1. For functional analysis, a single colony was cultivated in 3 mL of LB medium with appropriate antibiotics at 37 °C and shaken at 200 rpm overnight. A seeded culture was 1:10 diluted into 5 mL of M9 medium supplemented with 4 g/L glucose and appropriate antibiotics. After 12 h in culture, cells were diluted into 25 mL of M9 medium in a 250 mL shaking flask to achieve an initial OD600 of 0.1. The expression of galP and edd-eda was induced by the addition of 6.25 µM of isopropyl-β-D-thiogalactopyranoside (IPTG) at the beginning of seeding, subculturing, and culturing. The expression of pRFP, pGnd, and pGnd-GntR plasmids was induced by the addition of 50 µM of IPTG after 3 h in culture. OD600 was measured using a Libra S22 spectrophotometer.

For 3-HP production, a single colony from an agar plate was used to inoculate 3 mL of LB medium at 30 °C and shaken at 200 rpm. Overnight-grown cells were 1:20 inoculated into M9 medium supplemented with 20 g/L glucose, 1.0 g/L yeast extract, 2 mM MgSO4, 0.1 mM CaCl2, and trace elements. The composition of the trace element solution used was as follows: 2.4 g of FeCl3·6H2O, 0.3 g of CoCl2·6H2O, 0.15 g of CuCl2·2H2O, 0.3 g of ZnCl2, 0.3 g of Na2MO4·2H2O, 0.075 g of H3BO3, and 0.495 g of MnCl2·4H2O per liter [54]. After growing at 30 °C and shaken at 200 rpm for 9 h, 50 µM of IPTG and 100 nM of tetracycline were added to induce the expression of mcrC*-mcrN, and accDABC, after which the temperature of the culture was increased to 37 °C.

Adaptive laboratory evolution

To isolate rapidly growing mutants, even those lacking a functional EMPP, a mutant in which both the pfkA and pfkB genes had been deleted was adapted by sequential transfer. Initially, an overnight culture of ∆pfkAB mutant cells in LB was transferred to M9 minimal medium supplemented with 2 mM MgSO4, 0.1 mM CaCl2 and 5 g/L glucose and grown at 37 °C and shaken at 200 rpm. At an OD600 of approximately 1.0, the cells were passaged to a 250-mL shaking flask containing 25 mL of the same fresh M9 medium. This procedure was repeated 50 times, after which, the adapted cells were plated onto an LB agar plate. Five evolved mutants showing the most rapid biomass accumulation were selected based on OD600 values monitored using a 96-well microplate reader and named pfk_ALE-1, -2, -3, -4, and -5 (henceforth referred to as ALE-1, -2, -3, -4, and -5).

Genome sequencing

Genomic DNA was isolated from the evolved ALE-1 strain using a DNA isolation kit (GeneAll Biotechnology Co., Korea), and subsequently sequenced via next-generation sequencing using an Illumina HiSeq 2000 sequencer (Macrogen Inc., Korea). The genomic sequence of E. coli MG1655 (NC_000913.3) was used as a reference. Read sequence data of wild-type and ALE-1 strains were deposited in the Sequence Read Archive (SRA) of the NCBI under BioProject accession number PRJNA818725, and BioSample accession number SAMN26878847 for wild-type strain, SAMN26878848 for ALE-1 strain. The occurrence of mutations was confirmed by DNA sequencing of the PCR amplicon of target regions in the background strain and evolved mutant. To analyze the mutation frequency, the presence of mutations that were either the same or equivalent to the mutations discovered in ALE-1 was analyzed in the other four ALE mutants by PCR amplification and DNA sequencing.

Genetic manipulation

Gene knockout was conducted by homologous recombination via a λ RED recombination system, which was mediated by a kanamycin resistance (KmR) cassette flanked by Flp recognition target (FRT) knock-in and flippase-based curing [61, 62]. Knockout constructs were designed with 50-bp overhangs homologous to upstream and downstream sequences of the deletion site. Cells were screened for antibiotic sensitivity, and for those lacking antibiotic resistance, removal of the KmR gene was confirmed by colony PCR and sequencing.

Point mutations were generated via oligonucleotide-mediated genome editing combined with a tetA-based dual selection system. The tetA cassette was inserted into the target site and replaced with oligos containing target mutations and short 50-bp overhangs homologous to sequences upstream and downstream of the insertion site. The recombinants were isolated by negative selection using 50 μM NiCl2. Cells were screened for antibiotic sensitivity, and for those lacking antibiotic resistance, the correct allelic replacement was confirmed by colony PCR and sequencing.

Plasmid construction

All plasmids constructed in this study were derived from BioBrick plasmids [63]. The primers used for plasmid cloning are listed in Additional file 1: Table S2. The primers were synthesized by Macrogen Inc. (Seoul, Korea). Plasmid construction was conducted using Pfu-X DNA Polymerase for the PCR reactions, and all restriction enzymes used were purchased from Fermentas (Vilnius, Lithuania). For plasmid isolation, we used a LaboPass plasmid miniprep kit (Cosmo Genentech, Korea). DNA purification (Cat. no.103-102) and gel extraction (Cat. no.102-102) kits were purchased from GeneAll Biotechnology (Seoul, Korea).

For the construction of pEdd-Eda, the gene encoding the red fluorescent protein (RFP) in the pBbA6c-rfp vector was excised using restriction enzymes, EcoRI and BamHI. The edd-eda gene cluster was amplified from E. coli MG1655 genomic DNA using a forward primer containing an EcoRI site and a reverse primer containing a BamHI site. Following excision with EcoRI and BamHI, the purified edd-eda fragment was ligated with the gel-eluted pBbA6c backbone.

For cloning the pGalP, the RFP-encoding gene in the pBbS6k-rfp vector was excised using restriction enzymes, BglII and XhoI. The E. coli galP gene was amplified from E. coli MG1655 genomic DNA using a forward primer containing a BglII site and a reverse primer containing an XhoI site. Following excision with restriction enzymes, BglII and XhoI, the purified galP fragment was ligated with the gel-eluted pBbS6k backbone.

To construct pGnd, the RFP-encoding gene in the pRFP vector was excised using EcoRI and XhoI. The E. coli gnd gene was amplified from the E. coli MG1655 genomic DNA with a forward primer containing an EcoRI site and a reverse primer containing an XhoI site. Following excision with EcoRI and XhoI, the purified gnd fragment was ligated with the gel-eluted pRFP backbone.

For the construction of pGnd-GntR, the E. coli gntR gene was amplified from E. coli MG1655 genomic DNA with a forward primer containing a BglII site and a reverse primer containing the XhoI site. Following excision with restriction enzymes, BglII and XhoI, the purified gntR fragment was ligated with the purified pGnd.

For the construction of the pMCR plasmid, a mcrC*-mcrN fragment including genes encoding the dissected C-terminal domain carrying three mutations (N391V/K557W/S565N) and the N-terminal domain of malonyl-CoA reductase from Chloroflexus aurantiacus was extracted after cutting the pCDF-mcrC*-mcrN plasmid at EcoRI and XhoI sites. The pBbE2k-rfp plasmid was digested with restriction enzymes, EcoRI and XhoI to excise the gene encoding RFP, after which the DNA fragment and the gel-eluted pBbE2k backbone were ligated.

For the construction of a pACC plasmid, a fragment harboring the accDABC gene cluster encoding the ACC complex was obtained by elution after cutting pBbA2k-accDABC at BglII and XhoI sites. The pBbA6c-rfp plasmid was digested with restriction enzymes, BglII and XhoI to excise the gene encoding RFP. After gel-elution, the pBbA6c backbone was ligated with the eluted accDABC fragment.

Transcriptome analysis (RNA-seq)

Cell culture samples were harvested at the mid-log growth phase (OD600 ~ 0.5 for MG1655 and ALE-1 strain). The RNA was stabilized using by RNAprotect® Bacteria Reagent (Qiagen #76506) and total RNA was isolated using RNeasy® Plus Mini Kit (Qiagen #74134). Ribosomal RNA (rRNA) was removed from isolated RNA samples through riboPooL Probes (Pan-prokaryote) siTOOLS BIOTECH and Dynabeads™ MyOne™ Streptavidin C1(Invitrogen #65001). For purification of RNA samples, RNA clean & concentrator (zymoresearch R1013) was used. RNA library was constructed from rRNA-depleted RNA using the KAPA stranded RNA-seq kit (Roche KK8400). The sequencing libraries were sequenced by Nextseq 550, NextSeq 500/550 High Output Kit v2.5 (024906).

Using bowtie software, the sequence reads were mapped onto the wild-type reference genome (NC_000913.2) with the maximum insert size option of 1500 bp (for Additional file 1: Figure S8, 500-bp option was used) [64]. Fragments per kilobase of exon per million fragments (FPKM) and log2 fold change value were calculated by cufflinks [65]. All the datasets were visualized using Metascope and NimbleGen’s Signalmap software. The transcriptomic data of wild-type and ALE-1 strains were deposited in NCBI’s Gene Expression Omnibus (GEO) under GEO Series accession number GSE199153.

On the other hand, the unexpected expression level of the pfkB gene in ALE-1 was affected by a remaining coding sequence of pfkB (Additional file 1: Figure S8). Considering that the active site of PfkB was deleted, we anticipated that the function for PfkB in the ALE-1 strain was lost.

Flow cytometric analysis

For cultivation, a single colony was inoculated in 3 mL of LB medium at 37 °C and shaken at 200 rpm. Overnight-grown cells were 1:50 diluted into 5 mL of M9 medium supplemented with 4 g/L glucose and 5 g/L yeast extract. After growing for 9 h, cells were collected and transferred to PBS. The KDPG-responsible biosensor was induced by the addition of 50 µM of IPTG at the beginning of seeding and culturing. The green fluorescent protein (GFP) fluorescence intensity of individual cells was measured by BD FACSCalibur Flow Cytometer (BD Bioscience, USA) with wavelength of 488 nm excitation. Nonbacterial particles were excluded by applying forward-scattered characteristics and side-scattered characteristics. Gate M1 was set to differentiate the population of cells based on fluorescence intensity. BD CellQuest Pro software was used to analyze the data and to create images.

Analytical methods

Culture media were treated using previously reported methods with minor modifications [66]. After centrifuging for 30 min at 13,200 × g, the supernatant was heat-treated at 80 °C for 1 h. Thereafter, the supernatant obtained by centrifuging again for 30 min at 13,200 × g was used for analysis.

The concentrations of glucose and 3-HP were measured by HPLC using a previously described method with slight modification [67]. After a 20-fold dilution, 20 µL of samples were injected into a 300 mm × 7.8 mm Aminex HPX-87H (Bio-Rad, USA) column at 35 °C using 5.41 mM H2SO4 as the mobile phase.