Applied Microbiology and Biotechnology

, Volume 95, Issue 4, pp 1083–1094

Elucidating and reprogramming Escherichia coli metabolisms for obligate anaerobic n-butanol and isobutanol production

Authors

    • Department of Chemical and Biomolecular EngineeringUniversity of Tennessee
Bioenergy and biofuels

DOI: 10.1007/s00253-012-4197-7

Cite this article as:
Trinh, C.T. Appl Microbiol Biotechnol (2012) 95: 1083. doi:10.1007/s00253-012-4197-7

Abstract

Elementary mode (EM) analysis based on the constraint-based metabolic network modeling was applied to elucidate and compare complex fermentative metabolisms of Escherichia coli for obligate anaerobic production of n-butanol and isobutanol. The result shows that the n-butanol fermentative metabolism was NADH-deficient, while the isobutanol fermentative metabolism was NADH redundant. E. coli could grow and produce n-butanol anaerobically as the sole fermentative product but not achieve the maximum theoretical n-butanol yield. In contrast, for the isobutanol fermentative metabolism, E. coli was required to couple with either ethanol- or succinate-producing pathway to recycle NADH. To overcome these “defective” metabolisms, EM analysis was implemented to reprogram the native fermentative metabolism of E. coli for optimized anaerobic production of n-butanol and isobutanol through multiple gene deletion (∼8–9 genes), addition (∼6–7 genes), up- and downexpression (∼6–7 genes), and cofactor engineering (e.g., NADH, NADPH). The designed strains were forced to couple both growth and anaerobic production of n-butanol and isobutanol, which is a useful characteristic to enhance biofuel production and tolerance through metabolic pathway evolution. Even though the n-butanol and isobutanol fermentative metabolisms were quite different, the designed strains could be engineered to have identical metabolic flux distribution in “core” metabolic pathways mainly supporting cell growth and maintenance. Finally, the model prediction in elucidating and reprogramming the native fermentative metabolism of E. coli for obligate anaerobic production of n-butanol and isobutanol was validated with published experimental data.

Keywords

Elementary mode analysisMetabolic pathway analysisMetabolic pathway designMetabolic pathway alignmentFermentationAdvanced biofuelsIsobutanoln-ButanolEthanolRational strain designCofactor engineering

Introduction

Microbial conversion of lignocellulosic biomass into transportation liquid biofuels can provide one of the promising solutions to address the energy problem by reducing society’s reliance on nonrenewable and unsustainable fossil fuels and achieving energy and environmental sustainability (Alper and Stephanopoulos 2009; Blanch et al. 2008; Herrera 2006; Ragauskas et al. 2006; Schubert 2006). The C4 alcohols, such as n-butanol (linear carbon chain) and isobutanol (branched carbon chain), are valuable advanced biofuels that can be produced from renewable and sustainable lignocellulosic biomass (Atsumi et al. 2008b; Atsumi and Liao 2008; Bastian et al. 2011; Dellomonaco et al. 2011; Lee et al. 2008). However, it is a challenge to engineer optimal strains that can produce C4-alcohols efficiently by fermentation.

Fermentation is the most efficient and economical route to produce biofuels under anaerobic conditions because (1) the reducing equivalent NADH generated from sugar degradation can be recycled by using biofuel-producing pathways to maximize product yields and (2) the supply and precise control of air are not required, which makes scale-up fermentation processes simpler and less expensive (Harvey 2012; Trinh et al. 2011). Native butanogens such as Clostridium acetobutylicum and Clostridium beijerinckii could produce n-butanol anaerobically but not efficiently due to complex growth and regulation and formation of inevitable byproducts (Andersch et al. 1983; Bahl et al. 1986; Durre et al. 2002; Dürre and Hollergschwandner 2004; Hartmanis and Gatenbeck 1984; Jones and Woods 1986; Long et al. 1983; Mitchell 1997; Ravagnani et al. 2000; Sauer et al. 1994). Different metabolic engineering strategies have been implemented to improve n-butanol production in these Clostridia (Formanek et al. 1997; Green et al. 1996; Harris et al. 2000; Jang et al. 2012; Lee et al. 2012; Lehmann et al. 2012; Sillers et al. 2008).

Recent research has also been focused on engineering various recombinant hosts [e.g., Escherichia coli (Atsumi et al. 2008a; Inui 2008; Nielsen et al. 2009), Saccharomyces cerevisiae (Steen et al. 2008), Pseudomonas putida (Nielsen et al. 2009), and Lactobacillus brevis (Berezina et al. 2010)] to optimize n-butanol production by avoiding the complex growth and regulation typically inherent to native butanogens. However, the amount of n-butanol produced (<2 g/L) was low presumably due to the cofactor imbalance and inefficient redirection of carbon flux to the n-butanol biosynthesis pathway. The cofactor imbalance in the heterologous n-butanol-producing pathway of E. coli has recently been addressed, which significantly improved n-butanol production under fermentation (Bond-Watts et al. 2011; Shen et al. 2011). An elegant approach to engineer the reverse β-oxidation cycle in E. coli without using the heterologous enzymes has also been demonstrated to improve n-butanol production under completely aerobic condition with dissolved oxygen controlled at ∼5 % (Dellomonaco et al. 2011).

Like n-butanol, a recombinant E. coli JCL260 was successfully engineered to produce isobutanol from glucose using the valine biosynthesis pathway (Atsumi et al. 2008b). However, this engineered E. coli could not grow anaerobically on glucose because of the redox imbalance resulting from the multiple gene disruption to eliminate competing pathways (Trinh et al. 2011). By addressing the cofactor imbalance in the isobutanol-producing pathway, this strain could produce isobutanol anaerobically at a high yield during the no-growth phase (Bastian et al. 2011).

It has been reported that an engineered E. coli could grow and produce ethanol as the sole fermentative product (Trinh et al. 2008; Yomano et al. 2009). However, it remains to be elucidated why and whether the recombinant E. coli strains can grow anaerobically and produce either n-butanol or isobutanol as the sole fermentative product and what the underlying differences between n-butanol and isobutanol fermentative metabolisms are. In this study, elementary mode (EM) analysis was applied to answer these fundamental questions. EM analysis is one of the most useful metabolic pathway analysis tools that can decompose a complex metabolic network into all feasible unique and elementary pathways called elementary (flux) modes (EMs) that support cellular functions under steady state conditions (Schuster et al. 2000; Schuster et al. 1994). From the complete set of EMs of a given metabolic network, optimal strains can be rationally designed with desirable phenotypes (Boghigian et al. 2010; Hädicke and Klamt 2011; Kromer et al. 2006; Melzer et al. 2009; Stelling et al. 2002; Trinh et al. 2006; Trinh et al. 2011; Trinh and Srienc 2009; Trinh et al. 2008; Trinh et al. 2009; Unrean et al. 2010).

The goals of this study are to use EM analysis (1) to elucidate the native fermentative metabolism of E. coli optimized for obligate anaerobic production of n-butanol and isobutanol and compare the unique n-butanol and isobutanol metabolisms, (2) to redesign the native metabolism to improve obligate anaerobic production of n-butanol and isobutanol based on gene deletion, addition, over- and downexpression, and cofactor engineering, and (3) to validate the model prediction with experimental data from published literatures.

Materials and methods

E. coli metabolic network model

The E. coli central metabolism was constructed from literatures and public databases as previously reported (Trinh et al. 2008) (Supplementary Table 1). In this study, both n-butanol- and isobutanol-producing pathways were also integrated into the metabolic network. For isobutanol production, five reaction steps were added into the metabolic network, including the following: IBUT1 (acetolactate synthase, AlsSBS): 2 pyruvate = acetolactate + CO2; IBUT2 (2,3-dihydroxy isovalerate oxidoreductase, IlvCEC): acetolactate + NADPH + H+ = 2,3-dihydroxyvalerate + NADP+; IBUT3 (2,3-dihydroxy isovalerate dehydratase, IlvDEC): 2,3-dihydroxy isovalerate = isovalerate + H2O; IBUT4 (ketoacid decarboxylase, KivdLL): isovalerate = isobutanal + CO2; and IBUT5 (isobutanal dehydrogenase, AdhE2CA): isobutanal + NADH + H+ = isobutanol + NAD+. For n-butanol production, six reactions steps were added into the metabolic network, containing the following BUT1 (acetoacetyl CoA thiolase, PhaARE/AtoBEC): 2 acetyl CoA = acetoacetyl CoA + CoASH; BUT2 (3-hydroxy butyryl CoA dehydrogenase, PhaBRE/HbdCA): acetoacetyl CoA + NADH + H+ = 3-hydroxy butyryl CoA + NAD+; BUT3 (crotonase, PhaJRE/CrtRE): 3-hydroxy butyryl CoA = crotonyl CoA; BUT4 (crotonyl CoA reductase, TerTD): crotonyl CoA + NADH + H+ = butyryl CoA + NAD+; BUT5 (butyraldehyde dehydrogenase, AdhE2CA): butyryl CoA + NADH + H+ = butyraldehyde + NAD+ + CoASH; and BUT6 (butanol dehydrogenase, AdhE2CA): butyraldehyde + NADH + H+ = butanol + NAD+. In the metabolic network model, the heterologous NAD+-dependent formate dehydrogenase from Candida boidinii that could generate NADH by converting formate to CO2 was also investigated (Berrios-Rivera et al. 2002), and its use was specified where applicable.

Computation of elementary modes

All EMs were calculated by using the Metatool 5.1 software package (von Kamp and Schuster 2006). The theory of EM analysis is well established and can be found elsewhere (Schuster et al. 2000; Schuster et al. 1994).

Metabolic pathway alignment of designed strains

Since computation of EMs does not require the input of experimental fluxes such as substrate uptake rates or product secretion rates, each EM represents a metabolic flux vector that is unique up to a scalar (Jevremovic et al. 2010). To compare EMs, metabolic flux vector can be commonly normalized by the substrate uptake rate, i.e., glucose flux.

In this study, each designed strain that could produce n-butanol, isobutanol, or ethanol had a reduced desirable subset of EMs (≤6) (Supplementary Table 2) and hence could be grouped into families of EMs. EMs belonging to the same family had the same overall stoichiometry reactions for target product synthesis and hence the same weighting factors (Wlaschin et al. 2006). One family produced the maximum product yield during no-growth phase, while the other coupled cell growth and high product synthesis during the growth phase. To perform the metabolic pathway alignment for the designed strains, first the relative flux vectors were determined by: \( {{\bf P}} = \sum\nolimits_{{i = 1}}^l {{\alpha_i}\frac{{E{M_{\text{i}}}}}{{{r_{{glucose,i}}}}}} \), where P is the relative average flux vector normalized with respect to the glucose flux rglucose; EMi is the EM i; rglucose,i is the glucose flux in EM i; αi is the weighting factor for EM i; and l is the number of EM families (e.g., l = 2 for this study). Depending on the growth condition, a relative average metabolic flux distribution can be expressed as a non-negative linear combination of EMs distributed between the two families with different weighting factors (e.g., αi ∈ [0,1]). Next each reaction in the relative flux vectors was presented to be either “on” (1, if the flux was non-zero) or “off” (0, if the flux was zero). For the metabolic pathway alignment among different designed strains producing different target products, the weighting factors do not affect the flux patterns (e.g., “on” of “off”) of the designed strains.

Yield and experimental flux calculation

Based on the growth kinetics of the engineered strains reported from published literatures, experimental fluxes for substrate uptake and product formation were calculated by the formula: ri = dCi/dt (e.g., i = glucose, ethanol, n-butanol, isobutanol, biomass, etc.) for both growth and nongrowth phases. Yields on glucose were calculated by the formula: Yi = ri/rglucose. For anaerobic isobutanol production, data were collected from the reports by Bastian et al. (2011) and Trinh et al. (2011). For anaerobic n-butanol production, data were collected from the report by Shen et al. (2011). Since the OD600nm instead of dry cell weight was reported by Shen et al., 1 OD600nm = 0.4 g DCW/L (±50 % error) was estimated from the carbon balance of the E. coli strain JCL299 pEL11/pIM8/pCS138 presented in the report of Shen et al.

Results

Can E. coli produce either n-butanol or isobutanol as the sole fermentative product?

Recombinant E. coli can produce n-butanol and isobutanol using distinct and heterologous metabolic pathways (Fig. 1). Unlike ethanol, neither n-butanol- nor isobutanol-producing pathway is homo-fermentative. Therefore, it is important to investigate: Can E. coli metabolism be redesigned to enable the engineered strain to grow and produce either n-butanol or isobutanol anaerobically as the sole fermentative product? This question cannot be answered by intuitive reasoning. To answer this question, EM analysis was applied to calculate all EMs inherent to the metabolic network of E. coli and examine the effect of deleting all competing fermentative pathways on the obligate anaerobic production of n-butanol and isobutanol.
https://static-content.springer.com/image/art%3A10.1007%2Fs00253-012-4197-7/MediaObjects/253_2012_4197_Fig1_HTML.gif
Fig. 1

A simplified metabolic network for converting sugars to n-butanol, isobutanol, and ethanol in E. coli. Stoichiometry reactions, genes, and enzymes involved in these pathways are presented in Supplementary Table 1

n-Butanol fermentative metabolism is NADH-deficient

For the n-butanol metabolic network, a total of 41,618 EMs were identified and 7,347 of which were anaerobic EMs corresponding to the optimal conditions for anaerobic n-butanol production. Table 1 shows the effect of deleting competing fermentative pathways that produced succinate, lactate, acetate, and ethanol on n-butanol fermentative metabolism. In addition, the effect of deleting the key pyruvate dehydrogenase enzyme complex (PDHC) that converted pyruvate to acetyl CoA and generate NADH was investigated. This reaction was known to be inactivated or highly repressed under anaerobic conditions and replaced by the pyruvate formate lyase (PFL) because high accumulation of NADH inhibits PDHC (de Graef et al. 1999). Therefore, with inactivation of PDHC by native regulation, the maximum n-butanol yield was reduced from 0.41 to 0.35 g n-butanol/g glucose during the no-growth phase (Table 1). This result strongly suggests that n-butanol fermentative metabolism was NADH-deficient for optimal anaerobic n-butanol production.
Table 1

Effect of deleting competing fermentative pathways on the feasibility of producing n-butanol as the sole fermentative product

Deleted genesa

Total EMs

Biomass EMs

BuOH EMs

Biomass & BuOH EMs

Range of BuOH yield on glucose (g /g)

min

max

None

7,347

6,007

3,625

2,965

0.0003

0.41

ΔaceF

5,954

4,792

2,934

2,363

0.0003

0.35

Δfrd ΔadhE

1,627

1,361

1,613

1,361

0.0004

0.41

Δfrd ΔldhA ΔadhE

1,318

1,066

1,318

1,066

0.13

0.41

Δfrd ΔldhA Δpta ΔpoxB

1,628

1,236

814

618

0.22

0.41

Δfrd Δpta ΔpoxB ΔadhE

1,090

880

1,076

880

0.0004

0.41

ΔldhA Δpta ΔpoxB ΔadhE

1,946

1,477

1,946

1,477

0.03

0.41

Δfrd ΔldhA Δpta ΔpoxB ΔadhE

814

618

814

618

0.22

0.41

BuOH n-butanol

aΔaceF: removal of pyruvate dehydrogenase enzyme complex (PDHC); Δfrd: removal of succinate-producing pathway; ΔadhE: removal of ethanol-producing pathway; ΔldhA: removal of lactate producing pathway; ΔptapoxB: removal of acetate-producing pathway

Deleting all competing fermentative pathways did not inhibit cell growth (Table 1). Many combinations of deleted fermentative pathways such as succinate-, lactate-, acetate-, and ethanol-producing pathways (Δfrd ΔldhA Δpta ΔpoxB ΔadhE), succinate-, lactate-, and ethanol-producing pathways (Δfrd ΔldhA ΔadhE), or lactate-, acetate-, and ethanol-producing pathways (ΔldhA Δpta ΔpoxB ΔadhE) exhibited unique properties. In particular, the total number of EMs was equal to that of the n-butanol-producing EMs, suggesting that cells were forced to produce n-butanol anaerobically. In addition, the number of biomass EMs was equal to that of biomass- and n-butanol-producing EMs, indicating that cells could couple anaerobic growth and n-butanol production during the growth phase. Even though deleting competing fermentative pathways could force cells to couple anaerobic growth and n-butanol production during the growth phase, the range of n-butanol yield, 0.03–0.41 g n-butanol/g glucose, was still large such as for the mutant strain with ΔldhA Δpta ΔpoxB ΔadhE. The large range of n-butanol yield implies that the engineered strain could obtain low n-butanol yields under a specific condition. Therefore, redesigning the native fermentative metabolism of E. coli is important to optimize n-butanol yield.

Anaerobic isobutanol metabolism is NADH redundant

The isobutanol fermentative metabolism was completely different from the n-butanol fermentative metabolism. It was found that either PDHC deletion by disrupting aceF or PDHC inhibition under anaerobic conditions did not affect E. coli to grow and produce isobutanol anaerobically. Many pathways still existed and enabled anaerobic isobutanol production at the maximum theoretical yield, 0.41 g isobutanol/g glucose (Table 2).
Table 2

Effect of deleting competing fermentative pathways on the feasibility of producing isobutanol as the sole fermentative product

Deleted genes

Total EMs

Biomass EMs

iBuOH EMs

Biomass & iBuOH EMs

Range of iBuOH yield on glucose (g /g)

min

max

None

5,621

4,500

1,899

1,458

0.0004

0.41

ΔaceF

4,551

3,624

1,531

1,195

0.0004

0.41

Δfrd ΔadhE

28

0

14

0

0.41

0.41

Δfrd ΔldhA ΔadhE

14

0

14

0

0.41

0.41

Δfrd ΔldhA Δpta ΔpoxB

1,150

900

814

282

0.04

0.41

Δfrd Δpta ΔpoxB ΔadhE

28

0

14

0

0.41

0.41

ΔldhA Δpta ΔpoxB ΔadhE

427

251

427

251

0.06

0.41

Δfrd ΔldhA Δpta ΔpoxB ΔadhE

14

0

14

0

0.41

0.41

iBuOH isobutanol

The key result was that deletion of all competing fermentative pathways that made succinate, lactate, ethanol, and acetate by disrupting genes frd, ldhA, pta, poxB, and adhE completely inhibited cell growth because no existing EMs could produce biomass (Table 2). E. coli could only grow and produce isobutanol anaerobically if and only if it coupled with reduced-metabolite producing pathways that made either succinate or ethanol. This unique characteristic together with the capability of E. coli to maximize anaerobic isobutanol production without using PDHC strongly suggests that the isobutanol fermentative metabolism was NADH redundant. In addition, deleting the competing fermentative pathway by disrupting frd, ldhA, pta, and poxB was not sufficient to force cells to couple anaerobic growth and isobutanol production, with the range of isobutanol yield (0.04–0.41 g isobutanol/g glucose) still being large (Table 2). This wide range of isobutanol yield requires rational strain design to enhance anaerobic isobutanol production.

Rational strain design

Strain design strategy

Based on the complete set of computed EMs, the strain design strategy was to identify the minimum set of reactions that should be deleted from the metabolic network in order to (1) remove as many undesirable pathways as possible that have low product yields, (2) retain a small subset of optimal pathways that can achieve maximum product yield, and (3) tightly couple growth and metabolite production during cell growth (Trinh et al. 2008).

Design of the most efficient n-butanol-producing strain

The n-butanol fermentative metabolism of E. coli needs to be redirected to maximize anaerobic n-butanol production. By deleting six genes zwf, mdh, frd, ndh, adhE, and ldhA, the total numbers of EMs, biomass-producing EMs, n-butanol-producing EMs, and biomass- and n-butanol-producing EMs were reduced from 7,347, 6,007, 3,625, and 2,965 to 17, 5, 17, and 5, respectively (Supplementary Figure 1). These gene disruptions would enable the designed strain to produce only n-butanol and couple both anaerobic growth and n-butanol production during the growth phase. Even though the range of n-butanol yield was constrained from 0.0003–0.41 to 0.13–0.41 g n-butanol/g glucose, it was still wide. If the acetate-producing pathway were deleted by disrupting genes pta and poxB, the range of n-butanol yield would be constrained to 0.29–0.41 g n-butanol/g glucose and the total numbers of EMs, biomass-producing EMs, n-butanol-producing EMs, and biomass- and n-butanol-producing EMs would be reduced to 6, 2, 6, and 2, respectively. However, these high n-butanol-yielding EMs used PDHC to generate acetyl CoA required for anaerobic growth and n-butanol production, which was known to be inhibited under anaerobic conditions due to enzyme inhibition. In this case, additional deletion of acetate-producing pathway would inhibit cell growth. Therefore, it is important to identify an alternative strategy to overcome this problem by cofactor engineering.

Cofactor engineering to enhance obligate anaerobic n-butanol production

Two strategies could be applied to address the NADH-deficient problem inherent to native E. coli fermentative metabolism for anaerobic n-butanol production. The first approach was to use a PDHC* mutant that bypassed the NADH inhibition (Kim et al. 2008) and to avoid the native transcriptional regulation of PDHC under anaerobic conditions by expressing it with an orthogonal expression system such as a T7 promoter. With this approach, the additional deletion of the acetate-producing pathway resulting in a total of eight knockout genes zwf, mdh, frd, ndh, adhE, ldhA, pta, and poxB could overproduce n-butanol and tightly couple anaerobic growth and n-butanol production. The range of n-butanol yield could be constrained to the optimal range as previously stated to be 0.29–0.41 (g n-butanol/g glucose).

The second approach was to identify a new pathway that was equivalent to the function of PDHC, but could function anaerobically. This approach could be achieved by using the heterologous NAD+-dependent formate dehydrogenase (FDH*) from C. boidinii to convert formate into CO2 and yield NADH (Berrios-Rivera et al. 2002). For the native pathway, the reaction GG13 (PDHC) was pyruvate + CoASH + NAD+ = acetyl CoA + CO2 + NADH + H+, which is inhibited under anaerobic conditions. For the modified pathway, the function of PDHC was equivalent to two enzymatic steps that can operate under anaerobic conditions, e.g., the reaction FEM1 (PFL): pyruvate + CoASH = acetyl CoA + formate and the reaction FEM4* (FDH*, NAD+-dependent formate dehydrogenase): formate + NAD+ = CO2 + NADH + H+. With the modified FEM4*, all EMs inherent to the modified metabolic network were recalculated. By deleting genes zwf, mdh, frd, ndh, adhE, ldhA, pta, poxB, and aceF, the total numbers of EMs, biomass-producing EMs, n-butanol-producing EMs, and biomass- and n-butanol-producing EMs were reduced from 5,558, 4,626, 2,772, and 2,313 to 5, 1, 5, and 1, respectively. The range of n-butanol yield was also constrained to 0.29–0.41 (g n-butanol/g glucose) (Supplementary Figure 2). The designed strain with the modified n-butanol fermentative metabolism could achieve the redox balance and the highest n-butanol yield.

Figure 2 shows the metabolic flux ratios between the two designed n-butanol-producing strains that used the original metabolic network and the modified one containing heterologous NAD+-dependent formate dehydrogenase FDH*. The strain designed from the modified metabolic network also had the additional disruption of the acetate-producing pathway. The NADH regeneration by FDH* and additional deletion of the acetate-producing pathway significantly increased the metabolic flux through the n-butanol biosynthesis pathway (∼2.3-fold) at the expense of the decreased metabolic fluxes toward the tricarboxylic acid cycle (∼0.23-fold), nonoxidative pentose phosphate pathway (∼0.23-fold), and biomass synthesis (∼0.23-fold). The increase in NADH supply by forcing the designed strain to use FDH* also resulted in the decrease of the transhydrogenase flux (∼0.23-fold) converting NADH to NADPH. In addition, it was observed that the glycolytic flux remained unchanged for the designed strains that used both the original and modified metabolic networks.
https://static-content.springer.com/image/art%3A10.1007%2Fs00253-012-4197-7/MediaObjects/253_2012_4197_Fig2_HTML.gif
Fig. 2

Metabolic flux ratios between the designed n-butanol-producing strains in the modified and original metabolic networks. The modified metabolic network used FDH* (NAD+-dependent formate dehydrogenase.) For the modified metabolic network, the designed n-butanol-producing strain has the following knockout genes zwf, mdh, frd, ndh, adhE, ldhA, pta, poxB, and aceF (see Suppplementary Figure 2.) For the original metabolic network, the designed n-butanol-producing strain has the following knockout genes zwf, mdh, frd, ndh, adhE, and ldhA (see Supplementary Figure 1.) The asterisk denotes null flux through the acetate-producing pathway in the modified metabolic network, which results in a relative flux ratio of zero. Other reactions whose relative flux ratios are null have null fluxes in both the modified and original metabolic networks. Reaction notation can be found in the Additional file 1

Design of the most efficient isobutanol producing strain through cofactor engineering

EM analysis was applied to design the most efficient isobutanol-producing strain as described in the previous study (Trinh et al. 2011). Briefly, by deleting eight knockout genes zwf, mdh, frd, ndh, pta, poxB, ldhA, and aceF, the designed isobutanol-producing strain could couple anaerobic growth and isobutanol production during the growth phase and maximize anaerobic isobutanol yield (0.41 g isobutanol/g glucose) during the no-growth phase (Supplementary Figure 3).

This study focused on manipulating the appropriate NADH and NADPH availability to enhance obligate anaerobic isobutanol production. EM analysis was used to investigate the effect of changing cofactor preference from NADPH to NADH of 2,3-dihydroxy isovalerate oxidoreductase (IlvC) and of NADPH-dependent isobutanal dehydrogenase (YqhD) on the isobutanol fermentative metabolism and strain design. The enzyme IlvC converts 2-acetolactate to 2,3-dihydroxy isovalerate while YqhD converts isobutanal to isobutanol (Fig. 1). The result shows that changing the cofactor preference slightly affected the metabolic network structure. For the case of using the modified NADH-dependent IlvC*, the total number of EMs decreased from 38,219 to 38,002, and for the case of using the NADPH-dependent isobutanal dehydrogenase, the total number of EMs increased from 38,219 to 39,570. The result also shows that the number of EMs remained unchanged in the designed strain (e.g., six EMs) for both the original and modified networks even though the intracellular fluxes were slightly different between them (Fig. 3). It was found that relative intracellular fluxes were similar for both the referenced and perturbed states under no-growth and growth phases, except for one reaction PntAB converting NADH to NADPH. The result implies that the change in cofactor preference for either 2,3-dihydroxy isovalerate oxidoreductase or isobutanal dehydrogenase mitigated the expression level of PntAB. Therefore, the control of appropriate expression level of PntAB could play a significant role in optimizing the anaerobic isobutanol production.
https://static-content.springer.com/image/art%3A10.1007%2Fs00253-012-4197-7/MediaObjects/253_2012_4197_Fig3_HTML.gif
Fig. 3

The correlation between the perturbed and referenced relative fluxes in the designed isobutanol-producing strains. a No-growth phase. b Growth phase. The perturbed relative fluxes are referred to the modified metabolic network where the 2,3-dihydroxy isovalerate oxidoreductase IlvC is engineered to be NADH-dependent. The designed isobutanol-producing strains in both the modified and original metabolic networks are identical and have the following knockout genes zwf, mdh, frd, ndh, pta, poxB, ldhA, and aceF

Model validation with experimental data from published literatures

By calculating the complete set of all EMs, the designed strains were constrained to operate according to the most efficient routes in a reduced subset of EMs (<6) to convert glucose into n-butanol and isobutanol anaerobically based on gene deletion, addition, over- and downexpression, and cofactor engineering (Supplementary Table 2). The designed strains could couple anaerobic growth and C4-alcohol production during the growth phases. To compare with experimental data from published literatures, the phenotypic space of the designed strains was mapped on the 2-D space (Fig. 4).
https://static-content.springer.com/image/art%3A10.1007%2Fs00253-012-4197-7/MediaObjects/253_2012_4197_Fig4_HTML.gif
Fig. 4

Metabolic flux alignment of designed strains for producing n-butanol, isobutanol, and ethanol anaerobically. The metabolic flux distribution of the designed n-butanol-producing strain was calculated by using the modified metabolic network that used NAD+-dependent formate dehydrogenase FDH*

Obligate anaerobic n-butanol production

Figure 4a and b shows the phenotypic space of the wild type and designed strain spanned by the n-butanol and biomass yields on glucose for the modified metabolic network replacing the function of PDHC by that of PFL and FDH*. The phenotypes of the wild type were constrained on or within the space connecting groups G1–5 of EMs (Fig. 4a). The phenotypes of the designed strain had the following knockout genes zwf, mdh, frd, ndh, adhE, ldhA, pta, poxB, and aceF and were constrained on the line connecting groups G1–2 of EMs (Fig. 4b). Group G1 was referred to EMs that could produce n-butanol at the theoretical yield but not support anaerobic growth. However, group G2 contained an EM that could support anaerobic growth and n-butanol production.

Shen et al. developed the engineered strain JCL299/pEL11/pIM8/pCS138 that had the knockout genes ldhA, adhE, frd, and pta in the host JCL299, the knockin gene fdhCB in the plasmid pCS138 that used the FDH* to generate NADH, and the n-butanol-producing pathway plac::atoBEC::adhE2CA::crtCA::hbdCA in the plasmid pEL11 and plac::terFS in the plasmid pIM8 (Shen et al. 2011). The phenotypes of the engineered strain were constrained in the space connecting groups G1, G2, G6, andG7. Based on the model prediction, even though the phenotypic space of JCL299/pEL11/pIM8/pCS138 was larger than the designed strain, it was still able to couple anaerobic growth and n-butanol production during the growth phase.

To compare published experimental data with model prediction, anaerobic isobutanol production of JCL299/pEL11/pIM8/pCS138 was classified into both growth (open square) and no-growth (filled square) phases (Fig. 4b). Consistent with the model prediction, the physiological state of the engineered strain lied in the expected phenotypic space. However, the engineered strain could be further optimized to enhance anaerobic n-butanol production toward the phenotypes of the designed strain spanned by groups G1–G2 of EMs.

Different from JCL299, the engineered strain JCL166/pEL11/pEM8/pCS138 still contained the acetate-producing pathway (Shen et al. 2011). Based on the model prediction, this strain would not be able to tightly couple anaerobic growth and n-butanol production. The engineered strain was constrained in the phenotypic space connecting groups G1, G2, G3, and G8 (Fig. 4b). During the growth phase, the engineered strain JCL166/pEL11/pEM8/pCS138 operated in the expected phenotypic space. However, during the no-growth phase, the physiological state of the strain operated outside the expected phenotypic space.

Obligate anaerobic isobutanol production

Figure 4c and d shows the 2-D phenotypic space of the wild type and designed strain spanned by the isobutanol and ethanol yields on glucose. The phenotypes of the wild type were constrained on or within the space connecting groups G9–G11 of EMs (Fig. 4c). The phenotypes of the designed strain had the following knockout genes zwf, mdh, frd, ndh, pta, poxB, ldhA, and aceF, and were constrained on the line connecting groups G9 and G13 (Fig. 4D). These groups did not use PDHC and could overproduce isobutanol. Group G9 did not support anaerobic growth but group G13 did and produced both isobutanol and ethanol.

The engineered strains 1993m IlvC/YqhD + PntAB, 1993m 6E6/AdhA, and 1993m 6E6/RE1 were derived from JCL260 that had the following knockout genes adhE, fnr, pflB, frdBC, pta, and ldhA (Bastian et al. 2011). Under anaerobic conditions, EM analysis shows that these engineered strains could not grow anaerobically on glucose but produce isobutanol at the theoretical yield during the non-growth phase, corresponding to the group G9 (Fig. 4d). The strain 1993m IlvC/YqhD + PntAB overexpressed the transhydrogenase PntAB to increase the availability of NADPH for the NADPH-dependent 2,3-dihydroxy isovalerate oxidoreductase IlvC and NADPH-dependent isobutanal dehydrogenase YqhD reactions in the isobutanol-producing pathway. The strain 1993m 6E6/AdhA had 6E6 derived from IlvC and was engineered to be NADH-dependent, while the strain 1993m 6E6/RE1 had both 6E6 derived from IlvC and RE1 from AdhA, and engineered to be NADH-dependent. Consistent with the model prediction, the engineered strains 1993m IlvC/YqhD + PntAB, 1993m 6E6/AdhA, and 1993m 6E6/RE1 were able to produce isobutanol anaerobically only during the no-growth phase, matching with group G9 of EMs (Fig. 4D).

The engineered strain BFA7.001(λDE3) pCT013 had seven knockout genes zwf, mdh, frdA, ndh, pta, poxB, and ldhA in the host BFA7.001(λDE3) and the heterologous isobutanol-producing pathway pT7::alsSBS::ilvCEC::ilvDEC pT7::kivdLL::adhEEC in pCT013 overexpressed under the strong T7 promoter (Trinh et al. 2011). According to the model prediction, the designed strain was constrained in the phenotypic space connecting groups G9, G10, G12, G13 (Fig. 4d). Groups G10 and G12 of EMs used PDHC and mainly produced ethanol. It should be noted the designed strain that could grow and produce isobutanol anaerobically were required to couple with the ethanol-producing pathway. It was of great interest to constrain the designed strain to operate on the line connecting groups G9 and G13 to maximize isobutanol production in both growth and no-growth phases. The published experimental data shows that the designed strain BFA7.001(λDE3) pCT013 was able to couple anaerobic growth and isobutanol production, and its physiological state lied on the predicted phenotypic space (Fig. 4d). However, the isobutanol yield was lower than the ethanol yield because the isobutanol flux was limiting, associating with the use of unspecific isobutanal dehydrogenase AdhE (Trinh et al. 2011). Additional deletion of PDHC and using more specific isobutanal dehydrogenase (e.g., AdhE2CA or YqhD/PntAB) could constrain the phenotypic state of the engineered strain BFA7.001 (λDE3) pCT013 toward the region connecting groups G9 and G13.

Metabolic pathway alignment

Figure 5 shows the metabolic flux alignment of the designed strains that produced n-butanol and isobutanol. Metabolic pathways were aligned according to their functions and could be classified into two types—“core” and “auxiliary” metabolic pathways. The “core” metabolic pathways were necessary but not sufficient to support cell growth and maintenance because they were designed to couple with either the n-butanol- or isobutanol-producing pathways to be functional under anaerobic conditions. The “auxiliary” metabolic pathways were programmed to produce target products, n-butanol and isobutanol. It was observed that the metabolic flux distributions of the designed strains had very similar metabolic flux patterns except for reactions that participated in n-butanol- and isobutanol-producing pathways even though they had complete different fermentative metabolisms (Fig. 5). It should be noted that the definition of “core” and “auxiliary” pathways in this study was completely different from the concept of “core” and “side” EMs introduced by Kenenov et al. (2010) where either a “core” or “auxiliary” pathway could not stand as an EM based on the strain design.
https://static-content.springer.com/image/art%3A10.1007%2Fs00253-012-4197-7/MediaObjects/253_2012_4197_Fig5_HTML.gif
Fig. 5

The phenotypic spaces of the wild type and designed strains for anaerobic n-butanol and isobutanol production. a The phenotypic space of the n-butanol-producing wild type spanned by groups G1-5 of EMs. Filled circle anaerobic EMs. b The phenotypic space of the butanol-producing mutants. The designed strain with knockout genes zwf, mdh, frd, ndh, adhE, ldhA, pta, poxB, and aceF is constrained by G1-2, the engineered strain JCL299 (Shen et al. 2011) with knockout genes ldhA, adhE, frdBC, and pta constrained by groups G1, G2, G6, G7, and the engineered strain JCL199 (Shen et al. 2011) with knockout genes ldhA, adhE, and frdBC constrained by G1, G2, G3, and G8. Symbols (open growth phase, filled no-growth phase) are referred to experimental data, e.g., square JCL299/pEL11/pIM8/pCS138, and triangle JCL166/pEL11/pEM8/pCS138. c The phenotypic space of the isobutanol-producing wildtype spanned by groups G9-11 of EMs. d The phenotypic space of the isobutanol-producing mutants. The designed strain with knockout genes zwf, mdh, frd, ndh, pta, poxB, ldhA, and aceF is constrained by G9 and G13, the engineered strain 1993m (Bastian et al. 2011) with knockout genes adhE, fnr, pflB, frdBC, pta, and ldhA constrained by group G9 (no-growth), and the engineered strain BFA7.001(λDE3) with knockout genes zwf, mdh, frd, ndh, pta, poxB, and ldhA constrained by G9,10,12,13. Symbols are referred to experimental data, e.g., square BFA7.001(λDE3) pCT13; triangle 1993m IlvC/YqhD + PntAB, circle 1993m E6E/AdhA, and diamond 1993m 6E6/RE1

In a recent study, Trinh et al. (2008) designed, constructed, and characterized an efficient ethanol-producing strain that could couple anaerobic growth and ethanol production and produce ethanol close to the theoretical yield, 0.51 g ethanol/g glucose. When the metabolic flux distribution of the designed ethanol-producing strain was aligned with those of the designed isobutanol- and n-butanol-producing strains, all the designed strains had the same metabolic flux distribution pattern in the “core” metabolic pathways (Fig. 5). However, the metabolic flux patterns only differed in the “auxiliary” pathways because each designed strain was engineered to produce different types of biofuels (Fig. 5).

Discussion

EM analysis was applied to elucidate and reprogram the native fermentative metabolism of E. coli optimized for obligate anaerobic production of n-butanol and isobutanol. The main difference between n-butanol and isobutanol fermentative metabolisms was that the n-butanol fermentative metabolism was NADH-deficient while the isobutanol fermentative metabolism was NADH redundant. As a result, E. coli could grow and produce n-butanol as the sole fermentative product. Since the n-butanol fermentative metabolism was NADH-deficient, E. coli could not achieve the maximum theoretical n-butanol yield anaerobically. On the contrary, E. coli could not produce isobutanol as a sole fermentative product as in the case of ethanol production. In order to grow and produce isobutanol anaerobically, E. coli had to couple with either ethanol- or succinate-producing pathway to alleviate the NADH redundancy. For the purpose of biofuels production, it was of great interest to couple anaerobic growth with ethanol and isobutanol production.

The n-butanol fermentative metabolism was NADH-deficient because it required PDHC to maximize the n-butanol yield. However, PDHC was known to be inhibited anaerobically, which severely affected the optimal anaerobic n-butanol production. Therefore, cofactor balancing that controlled the optimal ratios of NADPH/NADP+ and NADH/NAD+ could play a significant role in regulating n-butanol and isobutanol fermentative metabolisms and hence product yields. EM analysis suggests that the use of the NAD+-dependent formate dehydrogenase could fix the NADH-deficient butanol fermentative metabolism. For instance, the engineered strain JCL199/pEL11/pIM8 that did not use FDH* to replenish cofactor NADH had a lower n-butanol yield than the engineered strain JCL299/pEL11/pIM8/pCS138 containing FDH* (Shen et al. 2011), consistent with the model prediction (Fig. 4B). EM analysis also suggests that additional gene deletion of zwf, mdh, ndh, and aceF in JCL299/pEL11/pIM8/pCS138 could further improve anaerobic n-butanol production by constraining the engineered strain to operate in the designed phenotypic space toward the group G1 in Fig. 4b.

A recent study has elegantly engineered the reverse β-oxidation pathway in E. coli to produce n-butanol and other higher alcohols under completely aerobic conditions (Dellomonaco et al. 2011). Since the reverse β-oxidation pathway used the same precursors (e.g., acetyl CoA and NADH) to elongate the carbon chain like the fermentative n-butanol-producing pathway of C. acetobutylicum, the NADH-deficient problem could still apply and need to be resolved by using either the modified PDHC* or both PFL and FDH* to replace PDHC for enhanced alcohols production under anaerobic conditions.

The isobutanol fermentative metabolism was NADH redundant because the designed strain could achieve high isobutanol yield without using PDHC to generate NADH and was required to couple ethanol- and n-butanol-producing pathways to recycle NADH. Since the designed strain was engineered not to use the oxidative pentose phosphate pathway to minimize the carbon loss by deleting Zwf, EM analysis shows that the pool of NADPH required for anabolism and anaerobic isobutanol production was produced primarily by the transhydrogenase enzyme PntAB converting NADH to NADPH. Interestingly, the published experimental data suggests that NADPH supply by PntAB was the rate-limiting step for the optimal anaerobic isobutanol production (Bastian et al. 2011).

Different strategies can be applied to relieve the metabolic burden of the sufficient supply of NADPH by PntAB for the efficient anaerobic isobutanol production. The first strategy is to engineer both 2,3-dihydroxy isovalerate oxidoreductase IlvC and isobutanal dehydrogenase YqhD to be NADH-dependent in the isobutanol-producing pathway (Bastian et al. 2011). The second strategy is to increase NADPH by recycling the abundant NADH, e.g., overexpression of PntAB (Bastian et al. 2011), reengineering GADPH in the glycolysis to be NADP+ dependent (Martínez et al. 2008), or using NADH kinase to convert NADH to NADPH at the expense of ATP (Mori et al. 2005). It should be emphasized here that according to the model prediction, switching cofactor use from NADPH to NADH by the above strategies would neither affect the strain design nor fix NADH-redundant isobutanol fermentative metabolism.

Among the engineered strains, the host strain BFA7.001(λDE3) shows a great potential to improve anaerobic isobutanol production because it was able to couple anaerobic growth and isobutanol production. Currently, BFA7.001(λDE3) pCT013 did not achieve high isobutanol yield anaerobically during the growth phase because it (1) utilized the unspecific isobutanal dehydrogenase (AdhE), (2) had low NADPH supply likely due to disruption of oxidative pentose phosphate pathway and perhaps undersupply of NADPH from PntAB, and (3) had PDHC not yet been deleted. Integration of the optimal anaerobic isobutanol producing pathway in BFA7.001(λDE3) using specific isobutanal dehydrogenase and supplying sufficient NADPH will improve anaerobic isobutanol production during both growth and no-growth phases.

The metabolic pathway alignment led to a key finding that the metabolisms of the designed strains for overproducing n-butanol, isobutanol, and ethanol were very similar in the “core” metabolic pathways but not the “auxiliary” pathways. This result suggests that it is possible to construct a universal optimal modular cell that contains only “core” metabolic pathways. According to the design, this modular cell must couple with “auxiliary” pathways to perform a novel function such as biofuel production. This modular approach will speed up the strain development process because it is possible to systematically and rapidly design an array of optimal cells all at once.

Acknowledgment

This research was supported in parts by the lab startup fund and the SEERC seed fund for the author at the University of Tennessee, Knoxville, TN, USA.

Supplementary material

253_2012_4197_MOESM1_ESM.pdf (409 kb)
ESM 1(PDF 408 kb)

Copyright information

© Springer-Verlag 2012