Metabolic flux profiling of Escherichia coli mutants in central carbon metabolism using GC-MS


U. Sauer, Institute of Biotechnology, ETH Zürich, CH-8093 Zürich, Switzerland. Fax: + 41 1 633 10 51, Tel.: + 41 1 633 36 72, E-mail:


We describe here a novel methodology for rapid diagnosis of metabolic changes, which is based on probabilistic equations that relate GC-MS-derived mass distributions in proteinogenic amino acids to in vivo enzyme activities. This metabolic flux ratio analysis by GC-MS provides a comprehensive perspective on central metabolism by quantifying 14 ratios of fluxes through converging pathways and reactions from [1-13C] and [U-13C]glucose experiments. Reliability and accuracy of this method were experimentally verified by successfully capturing expected flux responses of Escherichia coli to environmental modifications and seven knockout mutations in all major pathways of central metabolism. Furthermore, several mutants exhibited additional, unexpected flux responses that provide new insights into the behavior of the metabolic network in its entirety. Most prominently, the low in vivo activity of the Entner– Doudoroff pathway in wild-type E. coli increased up to a contribution of 30% to glucose catabolism in mutants of glycolysis and TCA cycle. Moreover, glucose 6-phosphate dehydrogenase mutants catabolized glucose not exclusively via glycolysis, suggesting a yet unidentified bypass of this reaction. Although strongly affected by environmental conditions, a stable balance between anaplerotic and TCA cycle flux was maintained by all mutants in the upper part of metabolism. Overall, our results provide quantitative insight into flux changes that bring about the resilience of metabolic networks to disruption.


mass distribution vector


glucose- 6-phosphate




pentose phosphates










phosphoenol pyruvate:glucose phosphotransferase system

PP pathway

pentose phosphate pathway

ED pathway

Entner–Doudoroff pathway

TCA cycle

tricarboxylic acid cycle


cellular dry weight

Comprehensive and quantitative understanding of biochemical reaction networks requires not only knowledge about their constituting components, but also information about the behavior of the network in its entirety. Toward this end, systems-oriented methodologies were developed that simultaneously access the level of reaction intermediates [1] or rates of reactions [2–5], also referred to as the metabolome [6] and the fluxome [7], respectively. The most important property of biochemical networks are the per se nonmeasurable in vivo reaction rates, which may be estimated by so-called metabolic flux analysis that provides a holistic perspective on metabolism.

In its simplest form, metabolic flux analysis relies on flux balancing of extracellular consumption and secretion rates within a stoichiometric reaction model [5]. To increase reliability and resolution of such flux balancing analyses, additional information may be derived from 13C-labeling experiments. In this approach, 13C-labeled substrates are administered and 13C-labeled products of metabolism are analyzed by methods that distinguish between different isotope labeling patterns, in particular NMR and MS [2,3,8]. In the most advanced methodology, a comprehensive isotope isomer (isotopomer) model of metabolism is used to map metabolic fluxes in an iterative fitting procedure on the isotopomer pattern of network metabolites that are deduced from NMR or MS data [2]. This global data interpretation enables integrated and quantitative consideration of all physiological and 13C-labeling data. Typically, protein hydrolysates are subjected to NMR or GC-MS analysis, which provides not only isotopomer pattern of the amino acids but also of their related precursor molecules that are key components of central metabolism. With the presently available models and software, these isotopomer balancing methods have attained a high level of precision and applicability [2,9,10].

In contrast to isotopomer balancing, direct analytical interpretation of 13C-labeling patterns has long been used not only to identify biochemical pathways and reactions but also to quantify individual flux partitioning ratios [3,11,12]. Such analytically deduced flux ratios were also used successfully as constraints for metabolic flux analysis [13–15]. Based on probabilistic equations, a more general methodology was developed to simultaneously identify network topology and multiple flux partitioning ratios [16,17]. This metabolic flux ratio analysis was based on the detection of 13C-labeling patterns in proteinogenic amino acids by NMR analysis, and provides direct evidence for a particular flux. Global isotopic data interpretation by isotopomer balancing and strictly local metabolic flux ratio analysis are largely independent. Hence, the favorable agreement of results obtained by both approaches for the same experimental data provides strong evidence for their reliability [18,19].

Here we develop a novel methodology for metabolic flux ratio analysis based on GC-MS data from [1-13C] and [U-13C]glucose experiments. This methodology is used for metabolic network analysis in Escherichia coli strains with knockout mutations in all major pathways of central carbon metabolism. The analyses presented here provide not only novel insights into central metabolism but represent also experimental verification of the reliability of metabolic flux ratio analysis by GC-MS.

Materials and methods

Strains, media, and growth conditions

The nomenclature of the employed E. coli knockout mutants indicates the affected genes (Table 1). Unless indicated otherwise, aerobic batch cultures were grown at 37 °C in 500 mL baffled shake flasks with 50 mL of M9 minimal medium on a gyratory shaker at 200 r.p.m. Anaerobic cultures were grown in 100 mL sealed glass flasks containing 50 mL medium that was gassed with N2 prior to sterilization for 10 min. The M9 medium contained per litre of deionized water: 0.8 g NH4Cl, 0.5 g NaCl, 7.52 g Na2HPO4, and 3.0 g KH2PO4. The following components were sterilized separately and then added (per litre of final medium): 2 mL of 1 m MgSO4, 1 mL of 0.1 m CaCl2, 1 mL of 1 mg·L−1 thiamine HCl (filter sterilized), and 10 mL of a trace element solution containing (per litre) 16.67 g FeCl3·6H2O, 0.18 g ZnSO4·7H2O, 0.12 g CuCl2·2H2O, 0.12 g MnSO4·H2O, 0.18 g CoCl2·6H2O, and 22.25 g Na2EDTA·2H2O. Filter-sterilized glucose was added to a final concentration of 3 g per litre. For 13C-labeling experiments, glucose was added either entirely as the [1-13C] labeled isotope isomer (> 99%; Euriso-top, GIF-sur-Yvette, France) or as a mixture of 20% (w/w) [U-13C] (13C, > 98%; Isotech, Miamisburg, OH) and 80% (w/w) natural glucose. The 13C-enrichment of [U-13C]glucose was independently determined to be 98.7% from cells grown exclusively on [U-13C]glucose.

Table 1. E. coli strains used in this study. The original strain designation is given in parentheses.
StrainsRelevant characteristicsReference
MG1655Wild-type K12 strain (λ F rph-1)[44]
W3110Wild-type K12 strain (λ F  IN(rrnD-rrnE)1 rph-1)[44]
JM101[F traD36 lacIqΔ(lacZ)M15 proA+B+supE thiΔ(lac-proAB)][45]
ZwfG6P dehydrogenase-deficient K10 (DF2001)[46]
PgiPhosphoglucose isomerase-deficient W3110 (LJ110)[47]
PfkAPhosphofructokinase-deficient K10 (AM1)[48]
PykAFPyruvate kinase-deficient JM101 (PB25)[49]
Mae/PckMalic enzymes (ScfA and Mae)- and PEP carboxykinase-deficient K12 (EJ1321)[50]
SdhA/MdhSuccinate dehydrogenase- and malate dehydrogenase-deficient MG1655 (DL323)[29]
FumAFumarase A-deficient K12 (EJ1535)[30]

Analytical procedures and physiological parameters

Cell growth was monitored by measuring the optical density at 600 nm (D600). Glucose concentrations were determined enzymatically using a commercial kit (Beckman, Palo Alto, CA, USA). The following physiological parameters were determined during the exponential growth phase in batch cultures as described previously [7]: Maximum growth rate, biomass yield on glucose, and specific glucose consumption rate, using a predetermined correlation factor of 0.44 g cellular dry weight (CDW) per D600 unit.

Sample preparation and GC-MS measurements

Aliquots of batch cultures were harvested during the mid-exponential growth-phase, defined as D600 of 0.8–1.5, and centrifuged at 14 000 g at room temp. for 5 min. Pellets were washed once in 1 mL 0.9% (w/v) NaCl and hydrolyzed in 1.5 mL 6 m HCl at 105 °C for 24 h in sealed glass tubes. The hydrolysate was dried in a vacuum centrifuge at room temperature and derivatized at 85 °C in 50 µL tetrahydrofurane (Fluka, Switzerland) and 50 µL of N-(tert-butyldimethylsilyl)-N-methyl-trifluoroacetamide (Fluka, Switzerland) for 60 min [20]. 1 µL of derivatized sample was injected into a series 8000 GC, combined with an MD 800 mass spectrometer (Fisons Instruments, Beverly, MA, USA), on a SPB-1 column (SUPELCO, 30 m × 0.32 mm × 0.25 µm fused silica) with a split injection of 1 : 20. GC conditions were: carrier gas (helium) flow rate at 2 mL per min, oven temperature programmed from 150 °C (2 min) to 280 °C at 3 °C per min, source temperature at 200 °C and interface temperature at 250 °C. Electron impact (EI) spectra were obtained at −70 eV. GC-MS raw data were analyzed using the software package MassLab (Fisons), avoiding detector overload and isotope fractionation as described [20].

The amino acids analyzed by GC-MS were aspartate, glutamate, glycine, histidine, isoleucine, leucine, phenylalanine, proline, serine, threonine, tyrosine, and valine for [U-13C]glucose experiments and aspartate, isoleucine, leucine, phenylalanine, serine, threonine, tyrosine, and valine for [1-13C] experiments.

Bioreaction network

The considered E. coli bioreaction network was described previously [18] but included additionally the ED pathway [21] and threonine aldolase [22](Fig. 1). The amino-acid-carbon skeletons were derived from the metabolic intermediates PEP, Pyruvate, P5P, E4P, OAA, and OGA as described [16].

Figure 1.

Bioreaction network of E. coli central carbon metabolism. Arrows indicate the assumed reaction reversibility. Solid arrows indicate precursor withdrawal for the amino acid analyzed by GC-MS. Inactivated proteins in the investigated knockout mutants are highlighted in boxes. Abbreviations: 6PG, 6-phosphogluconate; S7P, seduheptulose-7-phosphate; T3P, triose-3-phosphate; PGA 3-phosphoglycerate.

Correction for naturally occurring isotopes

The obtained EI spectral data are sets of ion clusters, each representing the distribution of mass isotopomers of a given amino-acid fragment. For each fragment α, a mass isotopomer distribution vector (MDV):


was assigned, where m0 is the fractional abundance of fragments with the lowest mass and mi>0 the abundances of molecules with higher masses. These higher masses result from isotope signals that originate from (a) natural abundance in non-C-atoms, (b) natural abundance of 13C in the derivatization reagent, and (c) 13C in the carbon skeleton of the amino-acid fragment that were incorporated from naturally or artificially 13C-labeled substrates. To obtain the exclusive mass isotope distribution of the carbon skeleton, MDVα were corrected for the natural isotope abundance of O, N, H, Si, S, and C atoms in the derivatizing agent by using correction matrices as described elsewhere [23], yielding MDV*α. Prior to analysis, the contribution of 13C from unlabeled biomass in culture inocula was subtracted from MDV*α yielding MDVAA according to


where funlabeled is the fraction of unlabeled biomass and MDVunlabeled,n is the mass distribution of an unlabeled fragment of length n. Its elements i can be calculated from the natural abundances of 12C and 13C according to Eqn (3).


c0 and c1 represent the natural abundance of 12C and 13C, respectively, and inline image is a binomial coefficient. The corrected MDVAA now represent the mass distributions of the carbon skeletons due to substrate incorporation (Fig. 2A).

Figure 2.

Example of the information flow from experimentally determined mass distributions in amino acids to metabolites(A) and the calculation of flux ratios(B). Bars illustrating the mass distribution (m0, m1,…,mn) are drawn to scale for the example of an E. coli batch culture grown on a mixture of 20%[U-13C] and 80% unlabeled glucose. Mass distributions of amino-acid fragments (MDVAA) are obtained from the experimentally determined MDVα by correction for natural isotope abundance and unlabeled biomass fraction. Mass distributions of metabolite fragments (MDVM) are calculated from MDVAA by using Eqn (4). (B) MDVM of different metabolites are used to calculate split ratios of diverging pathways and the MDV of CO2 according to Eqn (9).

MDV of metabolites

Amino acids are derived from one or more metabolic intermediates and MDVM of these metabolites (or their fragments) can easily be derived from the MDVAA, as illustrated schematically in Fig. 2A. If we assume that the carbon skeleton of an amino acid originates from the metabolites M1 and M2, the mass distribution vector MDVAA is a combination of the mass distributions MDVM1 and MDVM2 and can be derived by element-wise multiplication according to:


MDVM were obtained from a least squares fit to all MDVAA using the MATLAB function lsqnonlin with the additional constraint that the sum of their element equals 1.

MDV of substrate fragments

A fragment with n carbon atoms of a mixture of uniformly and naturally labeled substrate has the following mass distribution


where l is the labeled fraction and p is the purity of the labeled substrate. A fragment of a substrate that is 13C-labeled at a specific position can either be unlabeled, thus having the mass distribution MDVunlabeled,n (Eqn 3) or it may contain the 13C-labeled position leading to


A summary of all obtained MDV is given in Table 2.

Table 2. Mass distribution vectors used for flux ratio analysis. The carbon atoms included in each considered fragment are specified for each MDVM and MDVAA. MDVS are described by the length n of the fragment and its 13C-content. U, 20%[U-13C] and 80% unlabeled glucose experiment; 1, 100%[1-13C]glucose experiment.
Amino acid

Calculation of metabolic flux ratios

The intracellular pool of a given metabolite can be derived from other metabolite pools through biochemical pathways (Fig. 2B). The fractional contribution f of a pathway to a target metabolite pool with MDV1 was determined as:


where MDV2 and MDV3 are the mass distributions of the source metabolites degraded through the examined and the alternative pathway, respectively. As MDV are vectors and not single data points, f represents the least-squares solution to Eqn (7). Accordingly, using MDV with n elements, up to n alternative pathways can be distinguished. For example, the individual contributions of three converging pathways is determined as:


with f3 = 1 − f1 − f2.

The origin of several intracellular metabolite pools can be determined with Eqns (7) and (8). Specifically, MDVM of six metabolites and MDVAA of two amino acids were used for metabolic flux ratio analysis (Table 2) together with MDVS of substrate fragments. In some cases, however, the metabolic precursors MDV2 or MDV3 were combinations of two MDVM. Eqn (4) was applied to calculate the mass distribution of these combinations.

Pentose phosphate pathway

E. coli can potentially catabolize glucose to trioses via three different biochemical pathways, i.e. glycolysis, ED pathway, and PP pathway [24] (Fig. 1). Upon growth on a mixture of [U-13C] and unlabeled glucose, introduction and cleavage of bonds between carbon atoms is reflected in the MDVM of PEP, P5P, and E4P. As glucose catabolism through the glycolysis and the ED pathway yields uncleaved trioses, the activity of these two pathways is indistinguishable with [U-13C]glucose. The activity of transketolase and transaldolase in the nonoxidative PP pathway, however, can be accessed.

As exchange fluxes between serine and glycine [16] clearly influence the mass distribution of serine, PEP(1−2) was used to determine the fraction of trioses that were cleaved and rearranged between C1–C2 by the action of transketolase, and compared to the fraction that originates from an unbroken two carbon unit of glucose according to Eqn (7). An upper bound for PEP molecules that were generated from P5P can be calculated assuming that five trioses are produced from three pentoses and that at least two trioses are rearranged by transketolase. It should be noted that the thus calculated fraction of PEP originating from P5P does not necessarily reflect glucose catabolism through the PP pathway, but may likewise result from a reversible exchange flux via transketolase.

Two other metabolites that reflect transketolase and transaldolase activities are P5P and E4P. P5P molecules may be produced either via the oxidative PP pathway from G6P, thus yielding an intact five carbon skeleton from a source molecule of glucose, or via the transketolase reaction, which cleaves between C3–C4. Additionally, P5P may also originate from E4P and a one carbon unit through the combined action of transaldolase and transketolase. The contributions of the three converging pathways are thus calculated using Eqn (8). As transketolase can reversibly cleave P5P and multiple cycling may occur through the PP pathway, P5P from G6P is calculated as a lower bound for the fraction of P5P molecules that were generated via the oxidative PP pathway.

The second PP pathway intermediate, E4P, is either produced from F6P as an uncleaved four carbon unit or via the combined activity of transketolase and transaldolase from P5P. The latter introduces E4P molecules with cleaved C1–C2 bonds originating from the fraction of P5P that was cleaved between C3–C4. The E4P pool was analyzed using Eqn (7).

Anaplerosis and the TCA Cycle

[U-13C]glucose experiments were also used to distinguish OAA produced either from a four carbon unit via the TCA cycle or from PEP and CO2 via the anaplerotic reaction catalyzed by PEP carboxylase (see also Fig. 2). OAA(1−4) can thus be derived from the mass distribution of OGA(2−5) or from a combination of the MDV of PEP with CO2, according to Eqn (4). As the fractional labeling of CO2 (lCO2) is unknown in batch cultures and may be lower than the fractional enrichment of the input substrate, it was treated as an additional unknown using


The fraction of OAA molecules that originate through the TCA cycle is thus determined as 1 − f. The remaining fraction originates from PEP either through PEP carboxylase or through reversible malic enzyme via pyruvate. Additionally, the fraction of OAA(1−4) derived from glyoxylate via the glyoxylate shunt can be detected as a combination of pyruvate(2−3) and OAA(1−2).

Gluconeogenic reactions

Fluxes from the TCA cycle to the lower part of glycolysis via malic enzyme and PEP carboxykinase can be diagnosed as cleaved C2–C3 bonds in pyruvate and PEP, respectively. The interconversion of malate to pyruvate via the malic enzymes (ScfA and Mae) can thus be determined by comparing the pyruvate(2−3) and PEP(2−3) fragments using Eqn (7). As the mass distribution of malate is unknown, a pyruvate(2−3) molecule produced via malic enzyme was assumed to have the mass distribution of two combined one carbon units, each with the fractional 13C-label of the input glucose. This assumption includes (a) that all malate molecules are broken between C2–C3, thus are derived from OGA, and (b) that the fractional enrichment of C2 and C3 in malate does not differ from the fractional enrichment in the input substrate. A dilution of the fractional enrichment might be observed, for example, in positions where CO2 is introduced. This, however, may occur only at C1 or C4 of malate, thus does not affect the present calculation of the lower bound for malic enzyme activity. If the malate pool is in equilibrium with OAA, intact C2–C3 fragments from anaplerosis are present in malate. Thus, an upper bound for pyruvate produced through malic enzyme can be defined for the extreme case of full equilibration of the malate and OAA pools.

Similarly, PEP carboxykinase activity can be detected in the cleaved fraction of PEP(2−3) using Eqn (7). As a cleaved C2–C3 bond in PEP may also result from transaldolase activity, the thus calculated fraction of PEP originating from OAA remains an upper bound on the PEP carboxykinase activity.


The reversible exchange of the serine and glycine pools was quantified by determining the fraction of serine(1−3) originating from glycine(1−2) and a one carbon unit vs. the fraction that is identical with PEP(1−3) (Eqn 7). Additionally, the fraction of glycine(1−2) derived from serine(1−2) was attained assuming that the remaining glycine fraction with two independent C atoms originates from CO2 and a one carbon unit through the reversible glycine cleavage pathway or through threonine cleavage catalyzed by the threonine aldolase. The latter enzyme was reported to be active in E. coli under some conditions, albeit not those used here [22,25].

Calculation of metabolic flux ratios from [1-13C]glucose experiments

To obtain more precise information about the in vivo activities of the PP and ED pathway and the PEP carboxykinase, positional labeling was detected from cells grown exclusively on [1-13C]glucose. As the MDV of PEP could not be obtained in [1-13C]glucose experiments, serine was used instead to quantify the relative contribution of glycolysis to triose-3P synthesis, compared to the PP and ED pathways. The exchange flux with glycine does not change the label content in serine, unless substantial fractions of glycine or the one carbon unit are produced from sources other than serine. The oxidative PP or the ED pathway both yield unlabeled triose-3P, while glycolysis yields 50% unlabeled and 50% triose-3P that is 13C-labeled at C1 (Eqn 7).

If the ED pathway is active, additional label is introduced at the level of pyruvate, resulting in different MDV of serine(1−3) and pyruvate(1−3), which can be used to assess the relative contribution of this pathway to pyruvate synthesis using Eqn (7). Additionally, pyruvate derived through the ED pathway is labeled at C1, while pyruvate originating from glycolysis is labeled at C3. The fraction of pyruvate molecules labeled at C1 can be calculated from the difference between pyruvate(1−3) and pyruvate(2−3). This information is used to verify that the label is indeed introduced through the ED pathway and not through a gluconeogenic reaction.

Finally, PEP(1−2) originating from OAA(1−2) via the PEP carboxykinase was quantified using Eqn (7) assuming that the remaining fraction is identical to serine(1−2).

Error consideration

The experimental measurement error was determined by comparing the MDVα of amino acids with identical carbon skeletons, and the standard deviation of these redundant data was used for calculation of the covariance matrix Cm of the measured individual mass intensities. Standard deviations of the calculated flux ratios were determined applying the law of error propagation Cr = J*Cm*JT where J is the jacobian matrix and Cr the covariance matrix of the output variables. J was obtained numerically for MDVM after the least-squares fitting step and calculated analytically for the final flux ratios.


Sensitivity of metabolic flux ratio analysis using different mixtures of [U-13C] and unlabeled glucose

For economical reasons, low fractions of expensive 13C-labeled substrates are desirable for labeling experiments, provided that analytical resolution and sensitivity are maintained. To identify an optimal compromise, we grew E. coli MG1655 batch cultures in 5 mL M9 medium with different mixtures of [U-13C] and unlabeled glucose. While fully 13C-labeled or unlabeled biomass contained no information on metabolic fluxes, mixtures of 20/80, 40/60, 60/40, and 80/20 of [U-13C] and unlabeled glucose, respectively, allowed to determine flux ratios that were consistent within the experimental error (data not shown). Although the lowest experimental error is achieved at around equimolar fractions of [U-13C] and unlabeled glucose, the 20%[U-13C]glucose mixture enabled very reliable determination of intracellular flux ratios and was thus used in the further experiments.

Metabolic flux ratio analysis of E. coli under different environmental conditions

While exponentially growing cells are initially in a physiological pseudo steady state, metabolic switches occur upon oxygen limitation or accumulation of metabolic byproducts. To identify reproducible conditions that faithfully reflect the physiological state of unlimited, exponentially growing cells, biomass aliquots were harvested at different time points from wild-type batch cultures in shake flasks growing on 100%[1-13C]glucose or on a 20%/80% mixture of [U-13C] and unlabeled glucose. Overall, the determined origin of metabolite pools did not change significantly with the time of harvest (data partly shown in Fig. 3). The sole exceptions were increasing fractions of serine derived through glycolysis and OAA derived through the TCA cycle upon approaching stationary phase (Fig. 3), as was observed earlier [7]. Hence, all further analyses were performed with biomass aliquots harvested at D600 values between 0.8 and 1.5.

Figure 3.

Influence of harvest time on METAFoR analysis of E. coli MG1655 in aerobic shake flask batch cultures. The line indicates the exponential fit with a growth rate of 0.6 h−1 to the D600 readings (closed circles). Fractions of OAA through the TCA cycle (open circles), serine from glycine (open triangles), and pyruvate from malate (ub) (open squares) were obtained from 20%[U-13C] and 80% unlabeled glucose experiments. Serine through glycolysis (open diamonds) was obtained from 100%[1-13C]glucose experiments. Error bars indicate standard deviations of triplicate experiments.

Next, we investigated the metabolic impact of different levels of aeration from fully aerobic (500 mL baffled shake flask) to suboptimally aerated (15 mL vials) and anaerobic E. coli batch cultures (Fig. 4). With decreasing oxygen availability, most prominently, the fraction of OAA originating through the TCA cycle decreases from 44% to 5%. This reveals a branched, noncyclic operation of the TCA cycle to fulfill exclusively biosynthetic requirements, as was also shown earlier [7,16,26]. Although the oxidative PP pathway is still active under anaerobic conditions (serine through glycolysis), its relative contribution to glucose catabolism is decreased from 19% to 5% (Fig. 4), which concurs with most [7,16] but not all [26] reports. The frequently reported upper bound on in vivo PP pathway activity obtained from [U-13C]glucose experiments, in contrast (PEP from P5P), is not sensitive to this decrease. Unexpectedly, suboptimally aerated conditions promote relatively high in vivo malic enzyme activity (pyruvate from malate). Likewise, the of CO2 originating from air in the [U-13C]glucose experiments decreased with decreasing oxygen availability from 20% to 0%. Thus, introduction of unlabeled CO2 via carboxylation reactions can be neglected in vials or anaerobic cultures, but is significant in the better aerated shake flask cultures. To ensure fully aerobic conditions, all further experiments were conducted in shake flasks.

Figure 4.

Origin of metabolic intermediates in E. coli wild-type during aerobic (white bars), suboptimally aerated(gray bars), and anaerobic(black bars) growth. The experimental error was estimated from redundant mass distributions. Asterisks indicate results obtained from 100%[1-13C] glucose experiments. All other results were from 20%[U-13C] and 80% unlabeled glucose experiments. The fractions of pyruvate originating from malate and PEP originating from OAA could not be determined under anaerobic conditions because the OAA pool is derived exclusively from PEP.

Metabolic flux ratio analysis of E. coli mutants of central metabolism

The above developed metabolic flux ratio analysis by GC-MS was used for metabolic flux profiling of nonlethal mutations in all major pathways of E. coli central metabolism (Fig. 1). For this purpose, aerobic batch cultures were grown in shake flasks with M9 medium containing either [1-13C]glucose or a 20/80 mixture of [U-13C] and unlabeled glucose, which were identified above as reliable experimental conditions. Based on the physiological data obtained from three different wild-type strains, maximum specific growth rates of 0.5–0.7·h−1, biomass yields of 0.4–0.5 g(CDW)·g(glucose)−1, and specific glucose uptake rates of 6.5–8.5 mmol·g(CDW)−1·h−1 may be considered as normal for E. coli(Table 3). Hence, only the Pgi, PfkA, and Mae/Pck mutants exhibited clear physiological phenotypes with significantly reduced growth and glucose uptake rates.

Table 3. Aerobic growth parameters of exponentially growing E. colistrains in [1-13C] and [U-13C]glucose (in parentheses) experiments.
StrainGrowth rate (h−1)Biomass yield (g·g−1)Glucose uptake rate (mmol·g−1·h−1)
MG16550.61 (0.60)0.39 (0.39)8.5 (8.6)
W31100.55 (0.53)0.41 (0.43)7.3 (6.8)
JM1010.69 (0.68)0.49 (0.49)7.7 (7.7)
Zwf0.68 (0.65)0.53 (0.52)8.8 (8.8)
Pgi0.17 (0.15)0.39 (0.40)2.5 (2.0)
PfkA0.08 (0.08)0.41 (0.41)1.4 (1.5)
PykAF0.60 (0.59)0.41 (n.d)8.1 (n.d)
Mae/Pck0.41 (0.44)0.40 (0.42)5.7 (5.8)
SdhA/Mdh0.50 (0.51)0.43 (0.40)6.5 (7.1)
FumA0.67 (0.65)0.46 (0.45)8.2 (8.3)

While the flux profiles were similar in the three wild-type strains with small differences in the fractions of serine originating from glycine and OAA originating through the TCA cycle (Fig. 5), major changes were seen in the mutants (Fig. 6). Consistent with its severely reduced growth rate, the phosphoglucose isomerase-deficient Pgi mutant exhibited a very different flux profile without any glycolytic flux (serine through glycolysis in Fig. 6). Unexpectedly, the ED pathway was found to contribute about 30% to glucose catabolism in the Pgi mutant (pyruvate through ED pathway), so that the remaining 70% are contributed by the PP pathway, which is consistent with the upper bound of 80% PEP from P5P (Fig. 6).

Figure 5.

Origin of metabolic intermediates in the E. coli wild-type strains MG1655(white), JM101(gray), and W3110(black) during aerobic exponential growth. The experimental error was estimated from redundant mass distributions. Asterisks indicate results obtained from 100%[1-13C]glucose experiments. All other results were from 20%[U-13C] and 80% unlabeled glucose experiments.

Figure 6.

Origin of metabolic intermediates in E. coli mutants during aerobic exponential growth. The experimental error was estimated from redundant mass distributions. Asterisks indicate results obtained from [1-13C]glucose experiments. All other results were from 20%[U-13C] and 80% unlabeled glucose experiments.

The PfkA mutant is deficient in the allosterically regulated, major isoform of phosphofructokinase that constitutes about 90% of the total phosphofructokinase activity [27,28]. As phosphofructokinase is required for glucose catabolism via both glycolysis and PP pathway, the very low specific glucose uptake rate of the PfkA mutant and, as a consequence, the low growth rate on glucose are expected (Table 3). Consistently, the major fraction of serine is still generated through glycolysis (Fig. 6), probably catalyzed by the intact minor isoform phosphofructokinase B. However, the flux partitioning into the PP pathway (PEP from P5P) is significantly increased.

Flux profiles of the Zwf and PykAF mutants defective in G6P dehydrogenase and both pyruvate kinase isoforms, respectively, were somewhat similar to that of the wild-type. Significant flux changes in the Zwf mutant were seen in the reactions related to the PP pathway (data partly shown in Fig. 6). A 93% fraction of serine originating through glycolysis indicates residual PP pathway and/or ED pathway fluxes for glucose catabolism in the range of 7%. Consistent with the previously described metabolic bypass of pyruvate kinase knockout via PEP carboxylase and malic enzyme [7,18], the PykAF mutant exhibited lower fractions of OAA originating through the TCA cycle and higher fractions of pyruvate originating from malate (Fig. 6).

During the growth on glucose investigated here, simultaneous inactivation of the two gluconeogenic reactions catalyzed by malic enzyme and PEP carboxykinase had no significant effect on the flux profile of the Mae/Pck mutant (Fig. 6). This result was expected, as the fractions of pyruvate originating from malate and PEP originating from OAA that are indicative of in vivo malic enzyme and PEP carboxykinase activity, respectively, were already at detection level in the wild-type strains (Fig. 5). Disruption of the TCA cycle in the Sdh/Mdh and FumA mutants [29,30] reduced primarily the fraction of OAA generated through the TCA cycle (Fig. 6). This fraction is zero in the double knockout mutant in malate dehydrogenase and succinate dehydrogenase, which reveals complete inactivation of the TCA cycle and exclusive origin of OAA through the anaplerotic PEP carboxylase. Although knockout of the major fumarase isoform in the FumA mutant strongly reduced TCA cycle fluxes, a residual TCA cycle contribution to OAA synthesis of about 16% remains.


We introduce here metabolic flux ratio analysis by GC-MS as a novel methodology for flux profiling from 13C-labeling experiments. This methodology is based on probabilistic equations that relate mass distributions in amino acids to metabolic activities, and quantifies the relative contribution of converging pathways or reactions to metabolic intermediates. While MS data were used previously to analytically deduce individual flux ratios, for example at the G6P node [13,19,31] and in gluconeogenesis [32], the generalized methodology presented here simultaneously quantifies 14 flux ratios in central metabolism during growth on glucose. The thus obtained metabolic flux profile provides comprehensive information on in vivo activities of all major pathways in central carbon metabolism, hence concomitantly identifies the network topology. Although similar in scope to previously described metabolic flux ratio analysis by NMR [16,17], GC-MS-based analysis provides a significant advance in handling and sensitivity, so that biomass samples as low as 1 mg cellular dry weight may be analyzed. Without the need for time-consuming quantitative physiological analysis, this methodology thus paves the road to rapid diagnosis of metabolic changes in culture volumes below 1 mL.

Using metabolic flux ratio analysis by GC-MS, we dissect here flux responses of E. coli central metabolism to environmental and genetic modifications for two reasons: to (a) experimentally verify the accuracy of the new methodology and to (b) identify novel metabolic response. Estimation of in vivo PP pathway activity has received considerable attention, due to its variability with environmental conditions and relevance for NADPH metabolism. For aerobic batch cultures of E. coli, the relative contribution of the PP pathway to glucose catabolism has long been a matter of debate, yielding values between less than 10% to about 50% of glucose consumption [26,33]. For three different E. coli wild-type strains, we show here that the PP pathway contribution to fully aerobic glucose catabolism varies between 14% and 20%(Figs 5 and 7 A and 7B). This contribution does not change significantly upon mutations downstream of triose 3-phosphate. When forced to serve as the primary route for glucose catabolism in the phosphoglucose isomerase knockout (Fig. 7A), the PP pathway supports only a significantly lower growth rate than that observed for the wild-type. The strong reduction of PP and ED pathway fluxes upon knockout of G6P dehydrogenase (Fig. 7B) reveals the nonessential nature of both pathways for growth on glucose, as the growth physiology of the Zwf mutant was indistinguishable from that of the wild-type. Noticeably, a fraction of about 7% of the serine molecules does not originate from glycolysis in the Zwf mutant. The 13C labeling pattern of serine is instead consistent with a low but significant flux through either the PP or ED pathway. A similar observation was made with other, independently generated G6P dehydrogenase mutants (data not shown). Such a bypass of the inactivated G6P dehydrogenase may be catalyzed for example by the periplasmic glucose dehydrogenase, which produces glucono-δ-lactone that can be further converted to gluconate [24].

Figure 7.

Ratios of metabolic fluxes(solid arrows) to the synthesis of boxed metabolites in E. coli MG1655(top values), the Pgi mutant(second values), the Zwf mutant(third values), and the Sdh/Mdh double mutant(bottom values). The values are based on the data shown in Fig. 6. (A) Relative contributions of catabolic pathways and PEP carboxykinase to PEP formation from [U-13C]glucose experiments. (B) Relative contribution of the catabolic pathways to the formation of the serine pool from [1-13C]glucose experiments. (C) Relative contribution of the catabolic pathways and malic enzyme to the formation of the pyruvate pool from [1-13C] and [U-13C]glucose experiments. (D) Relative contributions of anaplerosis and the TCA-cycle to the formation of the OAA pool from [U-13C]glucose experiments. Dashed arrows symbolize reactions that are not considered for a given flux ratio.

Consistent with the reported gluconate induction [21], in vivo activity of the ED pathway was low but not completely absent in wild-type E. coli during aerobic growth on glucose (Figs 4,5, and 7C). In knockout mutants of glycolysis and TCA cycle, however, the ED pathway catalyzes up to 30% of glucose catabolism (Figs 6 and 7C). This is surprising because the inducer of this pathway is not present and, at least for the example of the Pgi mutant, in vitro ED pathway enzyme activities are not significantly higher [34]. In the Pgi mutant, this flux rerouting through the ED pathway reduces concomitant excess NADPH formation from exclusive glucose catabolism via the PP pathway, which generates two NADPH compared to one in the ED pathway per catabolized glucose. This overproduction of NADPH is deleterious, as limited capacity for reoxidation of NADPH is one reason for the low growth rate of phosphoglucose isomerase-deficient E. coli[34]. However, exclusive glucose catabolism via the ED pathway does not support growth of E. coli, as double mutants in both isoforms of phosphofructokinase cannot grow on glucose as the sole carbon source [27].

As may be expected from the known genetic regulation, low or absent in vivo activity of the gluconeogenic reactions catalyzed by PEP carboxykinase and malic enzyme was seen in our batch cultures. Consistent with previous flux analyses based on NMR data [7,18], the sole exception was the PykAF mutant, which bypassed the pyruvate kinase reaction by redirecting carbon flow via PEP carboxylase and malic enzyme (Fig. 6).

A very important flux ratio characterizing the metabolic state of a culture is the fraction of OAA originating through the TCA cycle, which quantifies the proportion to which the TCA cycle is used for energy generation vs. biosynthetic precursor supply via the anaplerotic PEP carboxylase (Fig. 7D). Consequently, this ratio is influenced by environmental factors such as growth phase (Fig. 3), aeration (Fig. 4), and overflow metabolism, but to some extent also by the genetic background of the wild-type strains (Fig. 5), as was noted previously for different organisms [7,16,26,35,36]. Generally, anaplerosis is high under conditions that invoke overflow metabolism, as acetate formation reduces the fraction of intact two carbon units entering the TCA cycle. Metabolic flux ratio analysis by GC-MS successfully captures the effective disruption of the TCA cycle in the Sdh/Mdh mutant (Figs 6 and 7D). Although the major fumarase isoform is inactivated in the FumA mutant, its respiratory TCA cycle flux is still at about one third of that in the wild-type (Fig. 6). This reveals that the two remaining fumarase isoforms are also important during growth on glucose.

Despite the different genetic backgrounds of the mutants in the upper part of central metabolism and their variations in growth rate, however, we observed surprisingly small deviations in this fraction of OAA originating through the TCA cycle. Thus, all mutants that were not related to the TCA cycle maintained a similar balance between anaplerosis and energy generation during exponential growth.

Most prominently among the presented data, this last result provides experimental evidence for metabolic network resilience to disruption [37–40]. While this was partly predicted for E. coli from computational network analysis [41] and is obvious from the fact that the investigated mutants grow in minimal medium, the flux results presented here reveal how metabolism manages intracellular flux redistribution upon disruption of all major pathways. These results are particularly valuable for the verification/falsification of hypotheses generated from in silico analyses such as flux balancing [42] or elementary flux mode analyses [43], and will ultimately contribute to a quantitative understanding of metabolic networks.

Supplementary material

The following material is available from EJB3448sm.htm

Table S1. Mass distributions of metabolite fragments in E. coli mutants grown on [1-13C]glucose.

Table S2. Mass distributions of metabolite fragments in E. coli mutants grown on 20%[U-13C] and 80% unlabeled glucose.