Systems-Level Metabolic Flux Profiling Elucidates a Complete, Bifurcated Tricarboxylic Acid Cycle in Clostridium acetobutylicum ^{▿ †}

Media, culture conditions, and metabolite extraction.

C. acetobutylicum ATCC 824 was grown anaerobically at 37°C inside an environmental chamber (Bactron IV Shel Lab anaerobic chamber) with an atmosphere of 90% nitrogen, 5% hydrogen, and 5% carbon dioxide. The minimal medium formulation used in both liquid and filter cultures was 2 g/liter KH₂PO₄, 2 g/liter K₂HPO₄, 0.2 g/liter MgSO₄·7H₂O, 1.5 g/liter NH₄Cl, 0.13 mg/liter biotin, 32 mg/liter FeSO₄·7H₂O, 0.16 mg/liter 4-aminobenzoic acid, and 10 g/liter glucose (20). In the pertinent experiments, acetate, glutamate, aspartate, or ornithine was added at a concentration of 2 g/liter. In addition to the appropriate minimal medium, the plates used in filter cultures contained 1.5% ultrapure agarose.

Detailed protocols for preparing filter cultures and extracting metabolites in Escherichia coli have been published (2, 31), and these methods were adapted for use in C. acetobutylicum. Briefly, for the preparation of filter cultures, single colonies were picked from agar-solidified reinforced clostridial medium (RCM; Difco), resuspended in liquid RCM, heat treated at 80°C for 20 min, and grown to saturation overnight. This overnight culture was then used to inoculate a liquid minimal medium culture to an initial optical density at 600 nm (OD₆₀₀) of 0.03. When this liquid culture reached an OD₆₀₀ of ∼0.1, 1.6-ml aliquots were taken and passed through 47-mm-diameter round hydrophilic nylon filters (HNWP04700; Millipore), which were then placed on top of agarose plates with the appropriate minimal medium. Cellular metabolism was quenched, and metabolites were extracted by submerging the filters into 0.8 ml of acetonitrile-methanol-water (40:40:20) at −20°C (25). The filters were then washed with the extraction solvent, the cellular extractions were transferred and centrifuged in Eppendorf tubes, and the supernatant was collected and stored at −20°C until analysis. To measure growth, filters from parallel cultures were washed thoroughly with 1.6 ml of fresh medium and absorbance at 600 nm was determined.

Metabolite and flux measurement.

Cell extracts were analyzed by reversed-phase ion-pairing liquid chromatography (LC) coupled with electrospray ionization (ESI) (negative mode) to a high-resolution, high-accuracy mass spectrometer (Exactive; Thermo Fisher) operated in full scan mode for the detection of targeted compounds based on their accurate masses. This analysis was complemented with liquid chromatography coupled with ESI (positive and negative modes) to Thermo TSQ Quantum triple quadrupole mass spectrometers operating in selected reaction monitoring mode (1, 19). Hydrophilic interaction chromatography was used for positive-mode ESI, and ion-pairing reversed-phase chromatography was used for negative-mode ESI. Amino acids were derivatized with benzyl chloroformate before their quantitation by negative-mode LC-ESI-tandem mass spectrometry (MS/MS) (13). Absolute intracellular metabolite concentrations were determined using an isotope ratio-based approach previously described (2). Briefly, C. acetobutylicum was grown in [U-¹³C]glucose medium to near-complete isotopic enrichment and then extracted with quenching solvent containing known concentrations of unlabeled internal standards. The concentrations of metabolites in the cells can then be calculated using the ratio of labeled endogenous metabolite to nonlabeled internal standard.

We used kinetic flux profiling (KFP) for measuring metabolite fluxes and elucidating the metabolic network structure (31). Filter cultures were grown on minimal medium plates to an OD₆₀₀ of 0.35 and then transferred to minimal medium plates containing uniformly ¹³C-labeled glucose as the sole carbon source. At defined time points after the transfer (e.g., 1, 2, 4, 7, 10, 15, 30, and 60 min), metabolism was quenched and cell extracts were prepared and analyzed. The multiple isotopomers produced by the ¹³C labeling were monitored simultaneously using LC-MS. Metabolic fluxes were calculated based on the kinetics of the replacement of the unlabeled metabolites with the labeled ones. Similarly, the KFP experiments with uniformly ¹³C-labeled acetate were performed by transferring the cells to glucose minimal medium plates with added [U-¹³C]acetate.

For the long-term labeling experiments using [3-¹³C]glucose, [4-¹³C]glucose, [1,2-¹³C]glucose, [¹³C]glutamate, [¹³C]ornithine, or [¹³C]aspartate, the cells were extracted 2 h after they were transferred to the plates containing each labeled substrate. In all the experiments with ¹³C-labeled amino acids, the usual concentration of nonlabeled glucose was maintained. The ¹³CO₂ labeling experiments were performed by adding increasing concentrations of NaH¹³CO₃ into exponentially growing liquid cultures (OD₆₀₀ = 0.35). After 1 h, the cells were quickly filtered and extracted using acetonitrile-methanol-water (40:40:20) at −20°C.

The labeling of the C₁ unit pool was determined from the labeling patterns of various intermediates in nucleotide biosynthetic pathways that incorporate C₁ units. We used 5′-phosphoribosyl-N-formylglycinamide and IMP, which incorporate C₁ units from 10-formyl-tetrahydrofolate as well as dTMP, which incorporates a C₁ unit from 5,10-methylene-trahydrofolate. In addition, methionine, which incorporates a C₁ unit from 5-methyl-tetrahydrofolate, was used for corroborating data in some instances.

We used metabolic flux profiling to complement the information obtained from KFP and to determine flux ratios in various pathways. We followed the general approach described in reference 9, with the difference that instead of inferring labeling patterns from proteinogenic amino acids, we quantified them directly for most metabolites. In all experiments, the labeling data were corrected for natural abundance of ¹³C in nonlabeled substrates and for the ¹²C impurity present in ¹³C-labeled substrates in a fashion similar to that reported previously (2, 31).

ODE modeling and parameter identification.

We constructed an ODE model for the metabolic network shown in Fig. 4 as well as Fig. S7 in the supplemental material and then identified model parameters (fluxes and unmeasured pool sizes) that reproduce the laboratory data. The procedure was based on methods previously developed (8, 22). The ODEs describe the rates of loss of unlabeled forms of metabolites (and the creation of particular labeled forms) after feeding of [U-¹³C]glucose. The equations are based on flux balance of metabolites and take the form

where metabolite B, which can be in labeled or unlabeled form, is downstream of another metabolite, A_i. The outflux, F_tot, balances the sum of N influxes, F_i, with A_i. A_tot and B_tot are the total pool sizes of the corresponding metabolites (sum of labeled and unlabeled forms).

The unknown model parameters were identified by a genetic algorithm that minimizes a cost function (7, 8). The cost function quantifies the difference between the computational results and the laboratory measurements for the labeling dynamics, together with the additional constraints indicated in Table S1 in the supplemental material. One thousand sets of model parameters that can reproduce the laboratory data were identified, forming a distribution for each parameter (see Fig. S9 in the supplemental material). The median value and the breadth of the distribution then provide a representation of the fluxes consistent with the laboratory data. The C/C++ programs used for modeling and parameter identification are available upon request.

Glucose catabolism to pyruvate.

To probe metabolic flux in growing C. acetobutylicum, we monitored the dynamic (time-dependent) or long-term (steady-state) incorporation of ¹³C-labeled nutrients (glucose, acetate, CO₂, and selected amino acids) into downstream metabolites in glycolysis, the pentose phosphate pathway, the TCA cycle, and amino acid and nucleotide biosynthetic pathways.

C. acetobutylicum can potentially metabolize glucose to trioses via three different pathways: glycolysis (the Embden-Meyerhof pathway), the Entner-Doudoroff pathway, and the pentose phosphate pathway. Homologues of enzymes of each of the above-mentioned pathways, with the exception of the oxidative pentose phosphate pathway, are present in the C. acetobutylicum genome (23). The contribution of glycolysis to pyruvate synthesis relative to that of the Entner-Doudoroff pathway can be determined from cells grown in [1,2-¹³C]glucose (carbons 1 and 2 are ¹³C labeled) or [3-¹³C]glucose (carbon 3 is labeled) since each pathway yields distinct positional labeled forms of pyruvate, which can be distinguished by tandem mass spectrometry (MS/MS). For C. acetobutylicum growing exponentially on glucose as the carbon source, all pyruvate appeared to be derived from glycolysis, with no detectable Entner-Doudoroff pathway flux (see Fig. S1 in the supplemental material).

The pentose phosphate pathway provides essential precursors (ribose-5P and erythrose-4P) for nucleotide and amino acid biosyntheses. Ribose-5P molecules can be produced by the oxidative pentose pathway from glucose-6P, by the nonoxidative pentose phosphate pathway via the transketolase reaction, or by the combined activity of transaldolase and transketolase. Consistent with the lack of oxidative pentose phosphate pathway enzyme homologues in the C. acetobutylicum genome, feeding of [1,2-¹³C]glucose resulted in no detectable production of ribose-5P containing a single ¹³C atom, the hallmark of oxidative pentose phosphate production. Pentoses were instead produced via transketolase (∼80%) and via transaldolase-transketolase (∼20%) (see Fig. S2 in the supplemental material).

The above-described experiments suggest normal catabolism of glucose into pyruvate via glycolysis, and consistent with this, feeding of [U-¹³C]glucose as the sole carbon source resulted in rapid and complete labeling of glycolysis intermediates through phosphoenolpyruvate. Pyruvate, however, appeared in roughly equimolar amounts in its fully labeled form and in an unexpected form with two ¹³C carbons (Fig. 1 B). Glycolysis splits glucose down the middle, converting carbon positions 1, 2, and 3 (and 6, 5, and 4) into the methyl, carbonyl, and carboxyl carbons of pyruvate, respectively. As shown in Fig. 1C, growth of C. acetobutylicum in [1,2-¹³C]glucose (100%) resulted, as expected, in ∼50% of phosphoenolpyruvate and pyruvate each containing two ¹³C carbons. In contrast, feeding of [3-¹³C]glucose (100%) or [4-¹³C]glucose (100%) resulted in 50% labeling of phosphoenolpyruvate (and upstream trioses) but only ∼25% labeling of pyruvate. This suggested that the ¹³C label was being lost specifically from the carboxyl carbon of pyruvate, presumably in an exchange reaction with environmental carbon dioxide (CO₂), which comprises 5% of the anaerobic gaseous environment and is ∼99% nonlabeled. Consistent with exchange of the carboxyl carbon of pyruvate with carbon dioxide, growth of cells in the presence of NaH¹³CO₂ resulted in the formation of [1-¹³C]pyruvate, with the fraction of labeling increasing with increasing concentrations of NaH¹³CO₂ (Fig. 1D). There was minimal or no labeling of upstream metabolites. This confirms the rapid interchange between the carboxylic acid group in pyruvate and environmental CO₂.

In anaerobic organisms, the oxidative decarboxylation of pyruvate to produce acetyl-CoA and CO₂ is catalyzed by pyruvate-ferredoxin oxidoreductase (PFOR, also known as pyruvate synthase). The use of ferredoxin, whose redox potential is close to that of pyruvate, as the oxidant has the potential to make the overall reaction reversible (10, 26). When ¹³C-labeled acetate was added to the medium, however, there was no detectable labeling of pyruvate, even when a significant fraction of the acetyl-CoA pool was labeled (Fig. 1E). In the proposed mechanism of acetyl-CoA synthesis by PFOR, pyruvate is first decarboxylated to form the intermediate hydroxyethyl-thiamine pyrophosphate (TPP, the prosthetic group in PFOR). This intermediate then reacts with CoA (coenzyme A) to produce acetyl-CoA (26). Our data indicate that the carboxylic group in pyruvate interchanges rapidly with CO₂ but that the overall PFOR reaction is essentially irreversible. We accordingly propose that the interchange results from the reversibility of the decarboxylation step of the PFOR reaction.

The interchange between the carboxyl group in pyruvate and atmospheric CO₂ could also be explained by reverse flux through pyruvate dehydrogenase or pyruvate formate lyase. However, the presence of a pyruvate dehydrogenase complex has never been reported in C. acetobutylicum or in any other Clostridium species (5). Additionally, isotopic tracer experiments with aerobically and anaerobically grown E. coli strains indicate that neither pyruvate dehydrogenase nor pyruvate formate lyase is capable of producing the exchange between the carboxyl group in pyruvate and CO₂ that we observe in C. acetobutylicum (see Fig. S3 in the supplemental material).

Complete bifurcated TCA cycle.

After gaining an understanding of the pathways that catabolize glucose to trioses, we examined the TCA cycle (Fig. 2). Feeding of [U-¹³C]glucose (100%) resulted in labeling patterns of oxaloacetate, malate, and fumarate which closely matched the labeling pattern of pyruvate, with the appearance of close to equimolar amounts of isotopomers with two or three ¹³C carbons (Fig. 2B). This observation is consistent with the synthesis of oxaloacetate from pyruvate and atmospheric CO₂ (which is nonlabeled) and with the production of malate and fumarate from oxaloacetate by running the TCA cycle in the reductive (counterclockwise) direction. Succinate's labeling pattern, however, did not fully agree with that of pyruvate. Although succinate showed the same predominant labeled forms, their ratios were different, with nearly twice as much succinate containing three ¹³C carbons than two ¹³C carbons (Fig. 2B). This suggested that although succinate may be produced from fumarate, there must also be another source to account for the enhanced triple ¹³C labeling.

The labeling pattern for α-ketoglutarate differed from that of pyruvate or succinate. If, as previously hypothesized (14, 23), α-ketoglutarate is produced from succinate and CO₂ by running the TCA cycle reductively, α-ketoglutarate containing two and three ¹³C carbons should have appeared. However, the predominant form of α-ketoglutarate had four ¹³C carbons. The actual route of α-ketoglutarate production was revealed by the labeling patterns in citrate (Fig. 2B). Despite the putative lack of citrate synthase, there was a measurable intracellular pool of citrate that labeled rapidly after feeding of [U-¹³C]glucose. Citrate (a molecule with six carbons) was produced in two major isotopic forms, containing either four or five ¹³C carbons. This labeling pattern is consistent with citrate's production from oxaloacetate (with two or three ¹³C carbons) and acetyl-CoA (where the 2-carbon acetyl moiety is fully ¹³C labeled). The labeling pattern of α-ketoglutarate was then readily explained based on its production via citrate. Citrate containing either four or five ¹³C carbons produces α-ketoglutarate with four ¹³C carbons because the additional ¹³C carbon in citrate corresponds to the carboxyl group that is lost during oxidative decarboxylation of isocitrate to α-ketoglutarate.

The labeling of glutamate matched that of α-ketoglutarate, consistent with its formation by reductive amination of α-ketoglutarate driven by either ammonia or glutamine. To rule out the previous hypothesis that glutamate could be synthesized from ornithine by running the arginine biosynthesis pathway in reverse (28), we added [U-¹³C]ornithine to the medium. While arginine pathway compounds downstream of ornithine became labeled, we observed no production of labeled glutamate (see Fig. S4 in the supplemental material).

The lack of production of succinate containing four ¹³C carbons initially suggested that there was no production of succinate via α-ketoglutarate. However, experiments with additional ¹³C-labeled nutrients proved that this does occur. When cells were grown in unlabeled glucose plus [U-¹³C]acetate, ¹³C was assimilated into acetyl-CoA. Consistent with turning of the TCA cycle in the oxidative direction, citrate, α-ketoglutarate, and glutamate with two ¹³C carbons were produced, but the cycle was incomplete; there was no detectable labeling in oxaloacetate, malate, or fumarate. Interestingly, however, we observed the production of succinate with one ¹³C carbon (Fig. 2C).

This labeling of succinate is consistent with its production from α-ketoglutarate, but the stereospecificity of citrate synthase was the opposite of that found in common bacterial model organisms and eukaryotes. This unusual Re-stereospecificity of citrate synthase was confirmed by examining the positions of ¹³C carbons within glutamate and proline by MS/MS analysis (see Fig. S5 in the supplemental material). Production of succinate from α-ketoglutarate explains the succinate labeling patterns in the [U-¹³C]glucose labeling experiments; succinate was synthesized both via fumarate (producing succinate containing two and three ¹³C carbons) and via α-ketoglutarate (producing succinate containing three ¹³C carbons). With glucose as the sole carbon source, the relative contributions from each route to succinate production are ∼60% and 40%, respectively.

Aspartate and glutamate can be deaminated to oxaloacetate and α-ketoglutarate, respectively, to enter the TCA cycle. When [U-¹³C]aspartate is added to the medium (in the presence of unlabeled glucose), a large fraction (>80%) of the malate, fumarate, succinate, and citrate pools becomes quadrupoly ¹³C labeled. α-Ketoglutarate and glutamate become triply ¹³C labeled (Fig. 2D). When cells are grown in the presence of [U-¹³C]glutamate plus unlabeled glucose, both α-ketoglutarate and succinate become fully ¹³C labeled. In this case, oxaloacetate, malate, and fumarate are not ¹³C labeled (Fig. 2E). These observations corroborate the existence of a complete bifurcated TCA cycle in C. acetobutylicum.

Amino acid biosynthetic pathways and C₁ metabolism.

By analyzing the labeling patterns of amino acids and key biosynthetic intermediates, we were able to resolve most of the amino acid biosynthesis pathways in C. acetobutylicum. The observed labeling patterns were consistent with those expected based on canonical amino acid biosynthesis pathways, with the exception of isoleucine and glycine production (see Table S2 in the supplemental material). The labeling patterns in isoleucine indicated that it is not synthesized by the canonical pathway via threonine but are instead consistent with its production from acetyl-CoA and pyruvate via the citramalate pathway (see Table S2).

For glycine, there are two alternative pathways (Fig. 3 A). The more common pathway involves synthesis of glycine from serine by the enzyme serine hydroxymethyltransferase, which transfers the methanol group from serine to tetrahydrofolate (THF). The resulting methyl-folate species provide C₁ units for the biosynthesis of purines, thymidine, and methionine. Alternatively, in Saccharomyces cerevisiae and some bacteria, glycine can be synthesized by degradation of threonine, e.g., into acetaldehyde and glycine (12, 21). When cells were grown on [U-¹³C]glucose, serine (synthesized via 3-phosphoglycerate) became fully labeled, whereas threonine (synthesized via pyruvate) was ∼50% triply ¹³C labeled and ∼50% doubly ¹³C labeled. Consistent with its predominant formation from threonine but not from serine, glycine was ∼50% fully labeled and ∼50% singly ¹³C labeled (Fig. 3B). The synthesis of glycine from threonine was further corroborated by growing cells in [U-¹³C]glucose plus nonlabeled aspartate and observing that both the threonine and glycine pools are mostly nonlabeled while serine was largely fully labeled (Fig. 3C).

Since glycine is synthesized primarily from threonine, C₁ units must be obtained from a precursor other than serine. Glycine is also commonly used as a precursor of C₁ units, but we also found that this route is nearly inactive in C. acetobutylicum. When cells were grown in [1,2-¹³C]glucose, less than 5% of C₁ units were ¹³C labeled, even though the methylene group in glycine (the precursor of C₁ units) was ∼50% labeled (Fig. 3D). Conversely, when cells were grown in either [3-¹³C]glucose or [4-¹³C]glucose, the methylene group in glycine was nonlabeled but ∼25% of the C₁ unit pool was ¹³C labeled (Fig. 3E). The possibility that C₁ units could be synthesized from the carboxyl group of glycine via some noncanonical pathway was ruled out by the observation that the percentage of ¹³C-labeled C₁ units is essentially unchanged between cells grown in [U-¹³C]glucose and cells grown in [U-¹³C]glucose plus nonlabeled aspartate (Fig. 3E), even though the ¹³C label in the carboxyl group of glycine decreases from ∼50% to ∼10% (Fig. 3B and C). These experiments show that there is minimal production of C₁ units via serine or glycine. We found, however, a strong correlation between the labeled fraction of the carboxylic group of pyruvate and the labeled fraction of C₁ units across all these labeling experiments (Fig. 3E). In addition, when NaH¹³CO₂ was added to the medium, ¹³CO₂ was incorporated into C₁ units. The fraction of labeled C₁ units did not correspond directly to the fraction of labeled CO₂ but did correspond to the fraction of labeled CO₂ that was incorporated into pyruvate (Fig. 3F). Our data therefore indicate that in C. acetobutylicum, C₁ units are derived primarily (>90%) from the carboxylic group of pyruvate, likely through the combined action of pyruvate formate lyase and formate-tetrahydrofolate ligase.

Metabolic flux quantitation.

Among the most important characteristics of a biochemical network are the in vivo reaction rates. To achieve a quantitative understanding of the fluxes in the newly proposed metabolic network, we developed an ordinary differential equation (ODE) model that describes the isotope labeling kinetics of metabolites following the addition of universally labeled [¹³C]glucose (see Materials and Methods and Fig. S7 in the supplemental material). Given the model equations, we employed a nonlinear global search algorithm to identify the fluxes that can quantitatively reproduce the laboratory data (8). In addition to the labeling kinetics, inputs to the model included intracellular concentrations of glycolysis and TCA cycle intermediates and amino acids (see Table S3 in the supplemental material), nutrient uptake rates, excretion rates (see Fig. S6 in the supplemental material), and specific flux branch point data obtained from the steady-state labeling experiments discussed previously. The details of the cost function used for model fitting are presented in Table S1 in the supplemental material. To avoid overfitting the data, any simulations which fell within the 95% confidence limits of the laboratory data were considered acceptable; only more severe misfits were penalized during the search. A total of 1,000 well-fitting sets of fluxes were identified and used to estimate flux confidence intervals.

Figure 4 shows representative results for the ODE model fitting and a map of the identified median flux values in central metabolism. The complete results are presented in Fig. S8 and S9 in the supplemental material. The ODE model fits all of the observed data. Most of the identified fluxes, with the exception of several exchange fluxes, were tightly constrained, indicating that they are reliably defined by the available laboratory data. The results show that glycolytic flux predominates and is directed primarily toward acid production. Other significant fluxes include aspartate production via pyruvate/oxaloacetate, fatty acid production from dihydroxyacetone phosphate, and ribose-phosphate production from glycolytic intermediates. Within the TCA cycle, the flux through the oxidative branch is slightly larger than through the reductive branch. The production of succinate from succinyl-CoA can occur via succinyl-CoA synthetase but is also expected to occur via the canonical methionine and lysine pathways. The computational results indicate that the median contribution of the succinyl-CoA synthetase flux to the total succinyl-CoA flux into succinate is about 25%, while the methionine and lysine pathways combined contribute to ∼75% of the total flux (see Fig. S9).

In addition to providing quantitative flux values, the ODE model also helped to resolve an ambiguity in the network structure that was not adequately addressed by qualitative analysis of the isotope labeling patterns alone. Malate and fumarate production can occur directly via the reductive TCA cycle or alternatively from passage of carbon from aspartate to fumarate, which would then be oxidized to malate (see the alternative pathway in Fig. S7 in the supplemental material). Both pathways result in qualitatively indistinguishable labeling patterns. To distinguish them, we constructed ODE models for the two alternative pathways and performed flux identification using the procedure described above. Both models were able to describe the quantitative dynamic data following [U-¹³C]glucose labeling. However, in the second model, because fumarate is partly used for the production of malate, the contribution of fumarate to succinate production is smaller than that in the first model. Quantitatively, the percentage of succinate produced from fumarate is ∼54% for the first model and ∼6% for the second model. Compared with the experimentally measured value of ∼60%, the computational results indicate that the second model is inaccurate and the first one is correct, meaning that malate is produced primarily reductively from oxaloacetate rather than oxidatively from fumarate.