You are on page 1of 11


published online: 2 May 2016 | doi: 10.1038/nchembio.2077

Metabolite concentrations, fluxes and free energies
imply efficient enzyme usage
Junyoung O Park1,2, Sara A Rubin3, Yi-Fan Xu1,3,5, Daniel Amador-Noguez1,5, Jing Fan1,3, Tomer Shlomi4
& Joshua D Rabinowitz1,3*

In metabolism, available free energy is limited and must be divided across pathway steps to maintain a negative DG throughout.
For each reaction, DG is log proportional both to a concentration ratio (reaction quotient to equilibrium constant) and to a flux
ratio (backward to forward flux). Here we use isotope labeling to measure absolute metabolite concentrations and fluxes in
Escherichia coli, yeast and a mammalian cell line. We then integrate this information to obtain a unified set of concentrations
and DG for each organism. In glycolysis, we find that free energy is partitioned so as to mitigate unproductive backward fluxes
associated with DG near zero. Across metabolism, we observe that absolute metabolite concentrations and DG are substantially
conserved and that most substrate (but not inhibitor) concentrations exceed the associated enzyme binding site dissociation
constant (Km or Ki). The observed conservation of metabolite concentrations is consistent with an evolutionary drive to utilize
enzymes efficiently given thermodynamic and osmotic constraints.
© 2016 Nature America, Inc. All rights reserved.

bsolute metabolite concentrations determine enzyme binding For simplicity, Q is typically calculated on the basis of concentra-
site occupancies1 and the thermodynamics of metabolic tions, not activities, using ΔG°′ determined at cellular ionic strength
reactions2. Accordingly, knowledge of absolute metabolite and divalent cation concentrations.
concentrations is valuable for dynamic modeling of metabolism3,4. The ΔG has been applied in metabolic analysis to eliminate flux
Such knowledge is also useful for conducting in vitro enzyme assays loops that violate the second law of thermodynamics13,14 and to
under physiologically relevant conditions5,6, predicting whether constrain the directions of cellular metabolic reactions15. In addi-
antimetabolites will outcompete with endogenous metabolites for tion, it determines the fraction of flux through metabolic reactions
relevant enzyme binding sites7 and engineering enzymes that selec- that is in the forward direction and thus productive and thereby
tively bind desired substrates in the complex cellular environment8. relates to the efficiency of enzyme utilization16. Thermodynamic
A variety of methods exist for measuring cellular metabolite con- analysis also can provide regulatory insights, with enzymes operating
centrations, with LC/MS enabling measurements of many metabo- far from equilibrium being more effective targets for regulating
lites with sensitivity and specificity9,10. A deficiency of LC/MS is that pathway flux17.
absolute signal intensity depends on ionization efficiency and there- Here we explore the potential to infer cellular reaction free ener-
fore does not reliably reflect absolute concentrations. This limitation gies on the basis of relative forward-to-backward flux. Specifically,
can be overcome by comparing the signal of endogenous metabolites where J+ and J− are forward and backward fluxes18–20:
to isotopically labeled standards, whose addition before extraction
also accounts for losses in sample handling11. Because isotopically (2)

ΔG = –RT ln (J +/J –)
labeled standards of many metabolites are not commercially avail-
able, it is often more convenient to feed isotopically labeled sub- We validated that the relative forward and backward flux ratios
strates to cells and compare labeled intracellular metabolites to through reversible reactions like triose phosphate isomerase can be
unlabeled standards12. Through this approach, we have previously measured using isotope tracers. Moreover, we recognized that the
measured the absolute concentrations of 103 metabolites in E. coli1. resulting ΔG measurements can be used to evaluate the metabolite
The concentrations of these metabolites are typically higher than concentrations. For example, given direct measurements of the revers-
the associated Michaelis constants (Km) of the enzymes consuming ibility and all but one of the substrates or products of a reaction, the
them, suggesting that most active sites in E. coli are occupied. missing concentration can be determined by combining equations (1)
Using these absolute concentrations, the thermodynamics (Gibbs and (2). More generally, concentrations, fluxes and thermodynamics
free energy of reaction, ΔG) of metabolic reactions have also been are all interrelated, and we aimed to obtain coherent values of each.
assessed on the basis of the fundamental equation, where ΔG°′ is ΔG This integrative analysis was conducted in E. coli, yeast, and a
at standard biochemical conditions, Q is the reaction quotient (the cultured mammalian cell line. The mammalian cell line (iBMK)
ratio of the product to reactant activities), R is the gas constant, T is was derived by immortalization of baby mouse kidney cells21.
the temperature in Kelvin and Keq is the equilibrium constant (the Each cell type was grown in nutrient-rich conditions including
ratio of product to reactant activities at equilibrium) at standard high glucose. Under such media conditions, we observed that
biochemical conditions: absolute metabolite concentrations are substantially conserved
across these highly divergent cell types, arguing for a common set
ΔG = ΔG°′ + RT ln Q = RT ln (Q/Keq) (1) of constraints on their metabolomes. We propose that the three

Lewis-Sigler Institute for Integrative Genomics, Princeton University, Princeton, New Jersey, USA. 2Department of Chemical and Biological Engineering,

Princeton University, Princeton, New Jersey, USA. 3Department of Chemistry, Princeton University, Princeton, New Jersey, USA. 4Department of Computer
Science, Technion–Israel Institute of Technology, Haifa, Israel. 5Present addresses: DuPont Industrial Biosciences, Experimental Station, Wilmington,
Delaware, USA (Y.-F.X.),Department of Bacteriology and Great Lake Bioenergy Research Center, University of Wisconsin–Madison, Madison, Wisconsin,
USA (D.A.-N.). *e-mail:

482 nature chemical biology | vol 12 | JUly 2016 |

normalized to cell vol- We selected [1. expanded to cover a wide spectrum of central metabolic reactions. we can determine the to obtain informative flux ratios depends on the backwards reac- absolute concentration of GAP (which is less abundant and can be tion producing a distinctive labeling pattern. The integration of experimental measurements of forward-to-backward reaction fluxes and of metabolite concentrations results in more coherent and precise determination of both concentrations and ΔG.04 ±0. which had minimal pletely labeled. Together with To lay the groundwork for integrative analysis of metabolite con. 1b). the parent molecule of DHAP and GAP. All rights reserved. ume. Thus. The absolute glucose uptake rate. we needed a tracer that differentially labels DHAP and GAP.5 3 10 200 O O IPTG (µM) O TPI OH Increasing TPI expression Lower glycolysis Figure 1 | Tracing forward-to-backward flux through TPI. tracer analysis was also conducted with [U-13C5]glutamine. The flux ratio then yields ΔG as in a. the labeling data were centrations. α-ketoglutarate dehydrogenase flux. coli. coli the chromosomal TPI and introduced a plasmid containing TPI under the control of the inducer IPTG. 2a and Supplementary Table 1).nature. Inc. The tracers employed obscured by analytical interferences22) using equation (1). respectively.5 ±0.9 –0. (c) To verify that we can measure different extents of reversibility. Labeling data are shown in Isotope tracers reveal reaction reversibility and thus DG Supplementary Results. nature CHEMICAL BIOLOGY | vol 12 | JUly 2016 | www. Flux of glucose-derived two-carbon units into the we knocked out the chromosomal TPI in E. DHAP is E.03 100 O FBP – O – O O– Percentage DHAP labeling P O P 80 Gibbs energy –O O O O HO 60 M+2 HO OH ∆G = ∆G°’ + RT lnQ M+0 40 + – = –RT ln(J /J ) FBA 20 DHAP GAP –O Reaction flux Concentration O – O O– P P 0 – O OH O O 0. they differed in their utilization of the pentose (Fig. enabled such direct measurement of forward-to-backward flux ratios for selected reactions in the pentose phosphate pathway. At low inducer levels. uptake in E.2-13C2]glucose. Using equation (2).1038/nchembio.3 –0. as expected for complete revers. (b) [1.2-13C2]fructose-1. Malic enzyme flux was substantial in yeast and © 2016 Nature America. we knocked out in E. all DHAP molecules would be labeled (M+2). osmotics and patterns by LC/MS were conducted in E. organisms. fluxes and free energies (Fig. The fraction of unlabeled DHAP progressively increased with IPTG addition.2-13C2]glucose (red atom represents 13C) yields [1. malian cell line iBMK. coli but only 3% and 2% in yeast and mammalian iBMK To verify that we can measure different extents of reversibility.2-13C2]fructose-1. Nature chemical biology doi: 10.2077 article most important constraints are thermodynamics. developed to quantify forward-to-backward flux ratios and their con- the forward-to-backward flux ratios yield ΔGTPI. another RESULTS important carbon source for these cells.d. coli or yeast. nutrient uptake and waste excretion 483 . dynamics of central carbon metabolism. in the absence of backwards flux through TPI. Because DHAP abso. Results of all lytic reaction catalyzed by triose phosphate isomerase (TPI) using of the tracer experiments were integrated into a single unified flux isotope tracers. the parent molecule of DHAP and GAP. The extent of DHAP labeling at pseudo-steady state was used to determine the ratio of forward-to-backward TPI flux by isotopomer balancing. The extent of DHAP labeling at pseudo-steady state the two eukaryotic cell types than across the two microbes (Fig. FBP carbons 1–3 form DHAP. Net flux all of the DHAP molecules would be labeled (M+2). yeast and the mam- efficient enzyme utilization. Overall. Thus. mammalian cells but not E. was about 30-fold lower in the mammalian iBMK cells than in 6-bisphosphate. Reverse flux through TPI results in the appearance of unlabeled DHAP. TPI converts dihydroxyacetone phosphate (DHAP) set for each organism (Fig. Analyzing the broader network advantageously accounts for all Integrating metabolite concentrations and free energies major sources and sinks of each metabolite but requires a diversity With an eye toward a better understanding of the overall thermo- of tracers to quantify fluxes throughout central metabolism. for the mammalian cells. In addition. with reverse through the oxidative pentose phosphate pathway was 20% of glucose flux through TPI resulting in the appearance of unlabeled DHAP. the TCA cycle and folate-mediated serine-glycine The same conceptual approach that was applied to TPI can be interconversion (Fig. (a) The free energy of cellular reactions is independently determined by (i) the reaction standard free energy adjusted for substrate and product concentrations and (ii) the ratio of forward-to-backward fluxes. The ability lute concentration can be directly measured. we aimed to combine the Independent experiments feeding multiple different [13C]glucose flux-based free energy measurements from Figure 3 with direct mea- tracers and measuring pseudo-steady-state metabolite labeling surements of metabolite concentrations by LC/MS into a coherent b [1. with malate dehydrogenase oper- Increasing the inducer level resulted in unlabeled DHAP (up to a ating net in reverse. in the mammalian iBMK cells and 2% in yeast. into glyceraldehyde-3-phosphate (GAP). coli. in the absence of backwards flux through TPI. central metabolism by isotopomer and flux balancing. coli and introduced an TCA cycle consumed 24% of pyruvate made in E. DHAP was almost com.1 ±0. Large-scale determination of net and exchange fluxes glycolysis. an in-house algorithm was basis of isotopomer balancing (Online Methods). phosphate pathway and the tricarboxylic acid (TCA) cycle. Although the largest intracellular flux in each organ- formed from carbons 1–3.6-bisphosphate (FBP).2-13C2]Glucose npg a c Calculated ∆GTPI (kJ/mol) –5. 2b). and calculated ΔG errors represent 95% confidence intervals. which yields [1. 3 and Supplementary Table 2). coli but only 13% inducible TPI plasmid.0 –2. can be converted into a ratio of forward-to-backward TPI flux on the In addition to determining net fluxes. 1c). cells. Substantial differences in net fluxes were observed across the three ibility. 1a). (n = 3). fluxes were more similar in ibility (Fig. we initially set out sufficient to determine pseudo-steady-state net fluxes throughout to validate the ability to measure the reversibility of a single glyco. 1c). fidence intervals and associated ΔG (Online Methods). To determine TPI revers. Supplementary Data Set 1.2 ±0. and GAP is formed from carbons 4–6 ism was glycolysis. Labeling fraction error bars represent s. maximum value of 50% unlabeled). indicating nearly unidirectional forward flux (Fig.

Arrow widths indicate absolute magnitudes of fluxes. This approach identified previously unmeasured abso. pentose phosphate pathway. npg unified picture of central carbon metabolite concentrations and free energy change from phosphoglucose isomerase (PGI) to pyru- free energies. phosphofructokinase was the most strongly forward driven central carbon metabolite concentrations that best matched both glycolytic step. To this end. To evaluate further the free energy of lose-5-phosphate. Along with phosphofructokinase. We subsequently obtained upper and lower boundaries for of enzyme utilization16. not repeatedly for each pathway enzyme). 4b). including ribu. cells and monitored the formation of lower glycolytic intermediates. coli iBMK cells Figure 2 | Metabolic flux distributions in mammalian iBMK cells. TCA cycle. Fluxes were determined by integrating direct nutrient uptake and waste secretion rate measurements and data from multiple isotope tracers by metabolic flux analysis. we obtained ΔG°′ by a group contribu. Absolute magnitude of glucose uptake is shown for each organism. 1.3). −46 kJ mol−1 (Fig. DHAP and pentose phosphates) Accordingly. normalized to glucose uptake (GLC UPT).1038/nchembio. For reactions in which free energy was not measured. lower glycolysis is shown once per graph. ME. yeast and E. glycolysis. Such even flux partitioning. implying a forward-to- sure owing to analytical interferences or instability. backward flux ratio of 4. Although the concentration of phosphoenolpyruvate was too low we determined ΔG for most central carbon metabolic reactions to measure its labeling directly. Notably.5 ED G6P 6PGL 6PGC G6P G6PDH 6PGL PGL 6PGC G6P 6PGL 6PGC G6PDH PGL G6PDH PGL pathway 5 PGI PGI GND PGI GND GND 10 E4P E4P E4P TKL F6P TKL Ru5P TKL 20 F6P RPE Ru5P RPE F6P RPE Ru5P 40 PFK Xu5P RPI PFK Xu5P RPI PFK Xu5P RPI 80 FBP TAL TKL R5P FBP TAL TKL R5P FBP TAL TKL R5P 160 S7P S7P S7P FBA FBA FBA DHAP G3P DHAP G3P DHAP G3P GLN Gln Gln TPI TPI TPI GAPD GAPD GAPD ACONT ACONT ACONT AC 13BPG 13BPG 13BPG CIT ICIT GLU CIT ICIT Glu CIT ICIT Glu PGK CS PGK CS PGK CS AcCoA ICDH AcCoA AcCoA ICDH THF Ser THF Ser 3PG ICDH THF Ser 3PG 3PG AST ACITL AST AST OAA ASP AKG OAA ASP AKG OAA ASP AKG SHMT PGM SHMT PGM SHMT PGM GLU AKG Glu AKG GLU AKG MLTHF Gly 2PG Fatty acid MLTHF Gly 2PG ICL MLTHF Gly 2PG ICL AKGDH AKGDH AKGDH MDH synthesis MDH MDH THRA ENO THRA ENO THRA ENO PPCK PPC MAL SUCCoA MAL GLX SUCCoA MAL GLX SUCCoA Thr PEP Thr PEP MALS Thr PEP MALS PDH PDH SUCOAS PDH PYK FUM SUCOAS PYK FUM PYK FUM SUCOAS ME FUM SUCC ME FUM SUCC ME FUM SUCC LAC PYR EtOH PYR PYR SUCD SUCD SUCD PC PC b 101 101 101 r = 0. Plotted data are restricted to linearly independent fluxes (for example. black. has been predicted to maximize the efficiency ments. involving triose phosphate isomerase.51 100 100 100 © 2016 Nature America. pyruvate kinase is gener- gramming. other. we were surprised when our global analysis resulted (Fig. on average.70 r = 0.2077 a Mammalian iBMK cells Yeast E.3-bisphosphoglycerate. confirming some back flux through the phosphoglycerate 484 nature chemical biology | vol 12 | JUly 2016 | www. 4a and Supplementary Table 3). Grey. malic enzyme. (i) the directly observed concentrations for measured metabolites Downstream of phosphofructokinase. Combining these concentrations together with ΔG°′ values. In each tion method (component contribution)23 and searched for sets of case. coli Percent GLC UPT GLC GLC GLC ≤1 2. we were able to in only a modest driving force for pyruvate kinase in the mam- obtain reliable measurements for species that were difficult to mea. with a bias toward Metabolite sets were penalized when either concentrations or free fast reactions (such as the triose phosphate isomerase reaction) being energies fell outside of the 95% confidence intervals of the measure. (b) Comparison of normalized net fluxes across organisms. which was always the reaction ΔG was constrained to be negative in the direction of net flux. we added labeled pyruvate to growing 2-phosphoglycerate and . Current textbooks lute metabolite concentrations and more tightly constrained those of indicate that almost the entire energy drop in glycolysis occurs metabolites that were involved in multiple reactions with measured at the phosphofructokinase and pyruvate kinase steps (Fig. Fluxes were normalized to the organism’s glucose uptake rate. coli. the cumulative glycerate. orange. Normalized fluxes in mammalian AKGDH ME Normalized fluxes in E. malian cell line iBMK (ΔG = −3. (a) Net fluxes. article Nature chemical biology doi: 10. coli Normalized fluxes in yeast 10–1 10–1 10–1 iBMK cells 10–2 10–2 10–2 10–3 10–3 10–3 10–4 10–4 10–4 10–5 10–5 10–5 AKGDH ME 10–6 –6 –5 –4 –3 –2 –1 1 10–6 –6 –5 –4 –3 –2 –1 0 1 10–6 –6 –5 –4 –3 –2 –1 1 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 Normalized fluxes in mammalian Normalized fluxes in yeast Normalized fluxes in E. erythrose-4-phosphate. cellular ΔG values (for example. vate kinase (PYK) was. blue.nature. closest to equilibrium. close to equilibrium. 4b). the available free energy was and (ii) the observed cellular ΔG for measured reactions (Online relatively evenly distributed across pathway steps except the step Methods). α-ketoglutarate dehydrogenase. we observed labeling of 3-phospho- (Supplementary Table 4). the pyruvate kinase reaction. Focusing on glycolysis. ally considered a committed step in glycolysis.39 r = 0. consuming nearly half of the available free energy. metabolite concentrations and reaction free energies by linear pro. Inc. AKGDH. All rights reserved. as shown in the legend.8 kJ/mol.3 mM/min 69 mM/min 76 mM/min 2.

This could To infer enzyme binding site occupancies. coli had focused exclusively on substrate binding and found a propensity npg Conservation of absolute metabolite concentrations for most active sites to be occupied1. of organism on absolute metabolite concentrations is stronger Overall. The most presumably useful only when substrate concentrations rise. differed. the impact the pyruvate kinase reaction can be partially reversible. whereas the concen- malian but not the microbial culture medium. The differences. This larger free energy drop likely facilitates rapid fermentation. the ubiquitous metabolite concentrations. with inhibitor from 40% to 70%. metabolite concentrations seem to be tuned to with the fraction of variance in absolute metabolite concentra. Other abundant intracellular metabolites included reduced inhibitory sites. Central carbon trations of the substrates of degradative enzymes (for example. Substrate and inhibitor enzyme binding site occupancy driven in yeast. © 2016 Nature America. sis. ism (Fig. physiology and culture conditions across organisms and the however. 5b and Supplementary Fig. the details to which they bind. coli. Thus. Thus. whereas only M+3 labeling was observed. nature CHEMICAL BIOLOGY | vol 12 | JUly 2016 | www. Globally. despite high concentration may be required to drive transamination for. with metab- concentrations. Similar prior analysis in E. aldolase was closer to equilibrium. lome across three different carbon sources1. coli metabo- observed labeling was not caused by canonical gluconeogenic flux. absolute metabolite concentrations are net effect being a near doubling of the free energy drop in glycoly.nature. a propensity for ATP. 6b). although to a lesser degree than at substrate ones glutathione and pyruvate (Fig. determined by a combination of direct cofactors ATP. we found that ~75% of as this would have resulted in 3-phosphoglycerate being both M+2 variance was explained by correlation across conditions. 4c). 4d). 6a). Such pathways are (including nucleotide-derived cofactors such as NAD+). In the previous work comparing the E. NAD+ and NADPH were almost always saturating. with the effects on cell physiology25. yeast had potential for metabolite concentrations to change without overt higher glucose-6-phosphate. Binding affinities (the inverse of the dissocia- of free energy partitioning may be optimized to organism-specific tion constants Km and Ki) were taken from the BRENDA database26 metabolic objectives. Nature chemical biology doi: 10. suggest- and M+3 labeled24. perhaps to facilitate efficient lular abundances despite the presence of amino acids in the mam. 485 . not measured based solely on isotope tracer data (reversibility of glycolytic reactions inferred from combined flux and concentration data is shown in Fig. we compared the absolute potentially further favor yeast fermentation by enhancing the free concentrations of metabolites to the affinities of the enzyme sites energy available for downstream glycolytic steps. Compared to mammalian iBMK cells. The greater propensity for substrates than centrations of E. 2). yeast and mammalian iBMK cells were all inhibitors to be saturating was statistically significant in each organ- correlated with each other (Fig. substantially forward driven (ΔG ≤ −5 kJ mol−1). of metabolites were the next most abundant. NAD+ and NADPH to be saturating also at ward. in spite of different life- three organisms (Fig. amino acids comprised the largest site occupancy (Supplementary Fig. trast to the general propensity for substrates to be saturating. All rights reserved. 6c). Nevertheless. mutase–enolase-pyruvate kinase reaction sequence (Fig. the similarity of reaction free energies across spe. followed by nucleotides nucleotide catabolism) were typically below Km. (Supplementary Fig. despite most glycolytic reactions being more forward. coli GLC GLC GLC kJ/mol 0 G6P 6PGL 6PGC G6P 6PGL 6PGC G6P 6PGL 6PGC ED G6PDH PGL G6PDH PGL G6PDH PGL pathway –1 PGI GND PGI GND PGI GND –2 F6P TKL E4P Ru5P F6P TKL E4P Ru5P F6P TKL E4P Ru5P RPE RPE RPE –3 PFK RPI PFK RPI PFK RPI Xu5P Xu5P Xu5P –4 FBP R5P FBP TAL R5P FBP R5P TAL TKL TKL TAL TKL ≤–5 FBA S7P FBA S7P FBA S7P DHAP G3P DHAP G3P DHAP G3P GLN TPI GLN Gln TPI TPI GAPD GAPD GAPD ACONT ACONT ACONT AC 13BPG 13BPG 13BPG CIT ICIT GLU CIT ICIT GLU CIT ICIT Glu PGK CS PGK CS PGK CS AcCoA ICDH AcCoA ICDH AcCoA ICDH THF Ser 3PG THF Ser 3PG THF Ser 3PG AST ACITL AST AST OAA Asp AKG OAA Asp AKG OAA Asp AKG SHMT PGM SHMT PGM SHMT PGM Glu AKG Glu AKG Glu AKG MLTHF Gly 2PG Fatty acid MLTHF Gly 2PG ICL MLTHF Gly 2PG ICL AKGDH AKGDH AKGDH MDH synthesis MDH MDH THRA ENO THRA ENO THRA ENO PPCK PPC MAL SUCCoA MAL GLX SUCCoA MAL GLX SUCCoA Thr PEP Thr PEP MALS Thr PEP MALS PDH SUCOAS PDH PDH SUCOAS PYK FUM PYK FUM SUCOAS PYK FUM PYR FUM SUCC EtOH ME FUM SUCC ME FUM SUCC LAC ME SUCD PYR SUCD PYR SUCD PC PC Figure 3 | Reaction free energy determined with isotope tracers in mammalian iBMK cells. coli. substantially conserved across kingdoms of life. styles.1038/nchembio. reaction free energies were highly conserved across the than that of carbon source. contrast. In con- abundant individual metabolite in each organism was glutamate. olite concentration exceeding Km for roughly two-thirds of sites in cies suggested similarity in the metabolomes. Flux reversibility and ΔG were determined from forward and backward fluxes. yeast and E. This favors efficient enzyme utilization. Inc. Thus. whose tor concentrations tended to be near or below Ki (Fig. NADP+ was not reliably saturating. perhaps reflecting pri- grating free energy measurements. regulation or switching to gluconeogenesis1. gray. concentrations were often near Km. In concentration measurements and the above approach for inte. ADP/ATP and NAD+/NADH. the metabolite con. The details of free energy utilization. The remaining 30% to 60% of variance reflects concentrations near Ki positioning the system for regulation to kick some combination of measurement error and true interorganism in if significant deviations from homeostasis occur. saturate enzyme active sites but not effector binding sites in freely tions explained by the interorganism correlation (r2). Notably. Central carbon metabolite fraction of the metabolome (Fig. A compilation of each case (Fig. is provided in Supplementary oritization of a high NADPH/NADP+ ratio over complete enzyme Table 5. As described previously for E. 5a). inhibi- the amino group donor in many transamination reactions. 5a) and were similar in intracel. 1). which ranged growing cells. equilibrium (ΔG ≈0 kJ mol−1). red. Thus. (Supplementary Data Sets 2 and 3). Here we validated this basic Given the fundamental connection between ΔG and metabolite observation also for yeast and mammalian iBMK cells. ing a nutrient contribution of no more than 25%. Across the organisms. 2).2077 article Mammalian iBMK cells Yeast E. Blue. 4b).

3b) were refined by combining confidence intervals from direct LC/MS measurements of their concentrations (blue) with thermodynamic constraints to more precise obtained values (orange). coli ribose-5P sedoheptulose-7P xylulose-5P DHAP fumarate Directly 0. (n = 3). Pyr.1 0. free tion fluxes and thus ΔG directly.06 measured ΔG 0. Here. Blue and white bars depict negative and positive ΔG.6 0 b Mammalian iBMK cells Yeast E.m. a key con- thermodynamics by determining relative forward-to-backward reac. Error bars represent s.03 0. This was exemplified and 5). (a) The absolute concentrations of metabolites involved in reactions with ΔG determined from reaction reversibility (Fig. similar results were obtained from long relied on the ability to infer the free energies of cellular reactions E. coli and the eukaryotic cells. Prior work has integrated metabolite concentrations and by the formation of unlabeled DHAP via backwards triose phosphate fluxes in dynamic labeling experiments31 on the basis of the rela- isomerase flux after feeding [1.32.nature. Mal. 3PG. respectively.80 r = 0.14 0. pyruvate. tribution of this work is a coherent systems-level set of fluxes. using and free energies is an important future objective. which is not 486 nature chemical biology | vol 12 | JUly 2016 | www.e. malian cell line iBMK (Figs. Measurement of energies and metabolite concentrations for E. coli Textbook d d d d m m m m o o o o p p p p k k k k k k k k a a a a k k k k i i i i i i i i pg pg pg pg pg pg pg pg pg pg pg pg en en en py en py py py ga ga ga ga tp tp tp tp pf pf fb pf pf fb fb fb 0 –10 Cumulative ∆G (kJ/mol) –20 –30 –40 –50 –60 c M+2 d 10 2 10 2 10 2 |∆G| in mammalian iBMK cells (kJ/mol) r = 0. More precise and comprehensive determination of flux revers- Metabolism is the process of converting nutrient inputs into usable ibility should be possible in the future through dynamic 13C labeling energy and biomass building blocks. (c) An unexpected finding from the thermodynamic analysis in b is partial reversibility of pyruvate kinase.4 Concentration (mM) Concentration (mM) Concentration (mM) Concentration (mM) concentration LC/MS with 0.1038/nchembio. the concentration of fumarate is informed also by that of malate. 3-phosphoglycerate.10 0.1 kJ mol−1.020 0.2077 a Mammalian Mammalian Mammalian Yeast E. and upstream and downstream metabolites were analyzed for labeling. Too small a drop results in the associated enzyme single enzyme in one compartment from a pair of enzymes (such as catalyzing unproductive backward flux.02 0. |∆G| in E. Like all chemical networks.45 mM) was added to the medium of growing iBMK cells for 20 min. metabolites turn over (and thus tend to label) rapidly when flux we integrated data from multiple different 13C tracers (Supplementary is large or pool size is small31. into these flux and thermodynamic determinations. despite much more extensive com- (ΔG) on the basis of metabolite concentrations combined with standard partmentation in the eukaryotes. 2). Recently.8 interval After imposing 0. This relationship.02 0. however.010 0. npg DISCUSSION Table 2). To directly demonstrate this reversibility.2-13C2]glucose (Fig. metabolism must obey the second law of thermodynamics: each Compartmentation introduces an unaccounted-for complexity pathway step must cost free energy. To obtain forward tionship between metabolite pool sizes. experiments31 and/or the application of 2H and 15N tracers as well.28.015 0.2 0.0 0.82 r = 0. we augmented the tool set for probing cellular In addition to the methodology for measuring ΔG. [U-13C]pyruvate (0. A fundamental challenge is par. isotope tracers and MS.83 M+3 © 2016 Nature America. Plotted data are for all measured reactions with ΔG < −0. All rights reserved. (b) ΔG for glycolysis based on integration of metabolite concentrations and reaction reversibilities. (d) Comparison of ΔG across organisms. 2–5 and Supplementary Tables 1.3 95% confidence 0. coli. yeast and the mam- reaction reversibility is based on the backward reaction producing a dis. biochemists have mentation concerns. such analyses have been enable compartmental analysis of metabolite concentrations. Mitigating compart- To understand how cells manage this challenge. malate. 4 tinctive isotope labeling pattern in the substrate. For example.30. article Nature chemical biology doi: . as per equation (2). Development of methods that free energies27.5 Metabolite measured with Concentration (mM) 0. Inc. coli (kJ/mol) |∆G| in yeast (kJ/mol) Pyr 1 1 1 10 10 10 3PG 3PG M+3 Mal Mal FBA Asp 10 0 10 0 10 0 Asp 0 2 4 6 0 20 40 60 –1 –1 –1 10 –1 0 1 2 10 –1 0 1 2 10 –1 0 1 2 Percentage labeling 10 10 10 10 10 10 10 10 10 10 10 10 |∆G| in mammalian iBMK cells (kJ/mol) |∆G| in yeast (kJ/mol) |∆G| in E. in combination with the reversibility of fumarase. turnover times and fluxes: and backward fluxes for a substantial fraction of central metabolism. distinct isozymes) in different compartments. Too large reported here is for metabolites collected from whole cells and does a drop in free energy at any one step wastes the energy available in not distinguish forward-and-backward flux occurring through a scarce nutrients.04 1. fluxes conducted for increasingly large metabolic networks29. The analysis titioning the available free energy across pathway steps. as per equation (1). Whiskers show 95% confidence limits (Online Methods). coli (kJ/mol) Figure 4 | Integration of flux and concentration measurements via DG.

there should be mentation concerns. several reversible gly- −25 kJ mol−1. phosphoglycerate kinase and enolase34–37.7 kJ mol−1 but instead at least sixfold neogenesis. (a) Pie chart showing fractional contribution of each measured metabolite in each organism.2. this menting such thermodynamic analysis with the direct flux-based would result in a drop per reaction of only 0.64 –2 –2 10 10 10–2 Concentrations in E. for these enzymes to-forward flux. The free energy drop over this bypassed. The textbook view involves three strongly strong evolutionary pressure to use these enzymes at greater than thermodynamically forward-driven reactions.7% of forward colytic reactions.nature.6-bisphosphate to phosphoenolpyru- free energies to achieve systems-level estimates of both1. associated with a forward-to-backward flux ratio of roughly 10:9. we found that the former is substantially more forward reactions very close to equilibrium. yeast and in mammals and phosphotransferase system in E. providing useful one unit of productive flux for 18 units of exchange 4 87 . Spread evenly across six reactions. in at least some organisms. which results in minimal back flux (<0.17. resulting quantitative values of cellular ΔG for many reactions. Nature chemical biology doi: 10. The lower free energy drop at pyruvate kinase results in a greater could be used to further constrain concentration and flux estimates amount of free energy available to drive the reversible reactions of in future work. 4b). require high enzyme levels are sufficiently forward driven to avoid lation was omitted from our analysis because of our inability to excessive wasteful backwards flux. glyceraldehyde- flux). The more even free energy distribution in lower glyco- dance of media glucose). Consider the high cellular levels of glyceraldehyde-3-phosphate dehydro- glycolysis. All rights reserved. including those involving aldolase. Names are shown for metabolites whose fractional concentration exceeds 1%.3 kJ mol−1. ciently forward driven. phosphofructokinase functions as a classic stress. Indeed. coli (M) Figure 5 | Conservation of absolute metabolite concentrations. and this has been shown to inhibit glycolysis and thereby pro- irreversible step. Prior studies that are more directly related to our lower glycolysis.7 kJ mol−1. is tolerated to achieve energy efficiency.84 r = 0. six-reaction sequence is not 1. 4b). the ‘difficult’ steps of lower glycolysis that phosphofructokinase and pyruvate kinase. vate is only 1. linked by near-equilibrium ones that. Indeed. quantify the intracellular glucose concentration (given the abun. obviating compart. Thus. the actual situation in cells is different from reverse flux direction during gluconeogenesis and thus must be the current textbook description.4–24. Plotted data are for all measured metabolites. More generally. in only ~5% of enzyme contributing productive net flux. protein synthesis38 and cellular space limitations39. the present work mark. in glucose-fed cells. For phosphofructokinase and pyruvate lysis may affect both pathway regulation and dynamics. (b) Comparison of absolute metabolite concentrations across organisms. –6 –6 10 10 10–6 10–7 10–7 10–7 –8 –8 10 10 10–8 –8 10–8 10–7 10–6 10–5 10–4 10–3 10–2 10–1 10–8 10–7 10–6 10–5 10–4 10–3 10–2 10–1 10 10–7 10–6 10–5 10–4 10–3 10–2 10–1 Concentrations in mammalian iBMK cells (M) Concentrations in yeast (M) Concentrations in E.1038/nchembio. with only the triose phosphate committed reactions are glucose phosphorylation (hexokinase in isomerase step having a drop less than 2 kJ mol−1 in all organisms. Thus. where changes in enzyme driven: ΔG for phosphofructokinase ranges from −13 kJ mol−1 to activities do not impact overall pathway flux. coli (M) Concentrations in yeast (M) 10–3 10–3 10–3 cells (M) –4 –4 10 10 10–4 10–5 10–5 10–5 © 2016 Nature America. applicable to the steady-state labeling experiments reported here. coli). whereas some backward flux at pyruvate kinase mote pentose phosphate pathway flux and NADPH production40. glyceraldehyde- from equilibrium for changes in the associated enzyme activities 3-phosphate dehydrogenase is inactivated in response to oxidative to exert flux control.2 kJ mol−1) (Fig.6-bisphosphate Glutamate Glucose-6-phosphate ATP NAD+ Pyruvate CTP dTTP Glutamate Valine Glutamate GTP Pyruvate Glutamine Serine UTP Ornithine Lysine ATP Citrulline Asparagine Glycine Aspartate Fructose-6-phosphate Glutamine Dihydroxyacetonephosphate Serine Arginine Threonine Threonine Alanine Aspartate Pyruvate Valine 6-phospho-D-gluconate Alanine Glutamine Glucose-6-phosphate Aspartate Alanine Fructose-1. Given npg These data shed light on metabolic design principles. The three higher (11. it ranges from approximately 3-phosphate dehydrogenase and phosphoglycerate mutase. Inc. or edly refines the overall thermodynamic estimates. which cannot 5% efficiency. are suffi- −4 kJ mol−1 to −10 kJ mol−1.65 r = 0. Glucose phosphory. which is ΔG measurements for a subset of reactions. during gluco. coli Total concentration of Total concentration of Total concentration Amino acids measured metabolites: measured metabolites: of measured metabolites: Central carbon 176 mM 238 mM 243 mM Degradation Glutathione Phenylpyruvate N-Acetyl-L-aspartic acid Coenzyme-A Nucleotides Glutathione UDP-glucose Coenzyme-A Trehalose UDP-N-acetyl-glucosamine Other NAD+ UDP-N-acetyl-glucosamine Glutathione Fructose-1. resulting in 2% to 20% backward. the expense of studied organisms and localizes to the cytosol. whereas for pyruvate kinase. these observations are consistent with findings from nature CHEMICAL BIOLOGY | vol 12 | JUly 2016 | www.2077 article a Mammalian iBMK cells Yeast E. According to a standard textbook33. flip their direction of net flux (Fig.6-bisphosphate b 10–1 10–1 10–1 Concentrations in mammalian iBMK r = 0. Unlike kinase. which is the highest-flux pathway in each of the genase. although both reactions are far enough to exert physiologically relevant flux control. the net free current approach have integrated metabolite concentration and energy drop from fructose-1. By aug. Concentrations were obtained by the integrative analysis as in Figure 4a.

(c) Enzyme active sites are.nature.. olome. The fraction of concentrations exceeding Km or Ki (i. other driving force for all but the most rapid reactions. Because ΔG depends on the ratio of substrate and greater efficiency of enzyme utilization. A cap on metabolite concentrations is most likely enforced Maintaining a sufficient driving force is expected to favor by osmotics.3 0.4 –7 –5 –5 P < 10 P < 10 P < 10 [Metabolite] relative to 0. Within this upper osmotic pathways in which adequate total free energy is available. dances is saturation of enzyme binding sites. which requires a substantial product concentrations. at least for above considerations globally constrain the concentration ratios of cells fed copious glucose.2 Inhibitors 0. –6 –6 –6 10 10 10 –7 –7 –7 10 10 10 –8 –8 –8 10 10 10 –8 –7 –6 –5 –4 –3 –2 –1 0 –8 –7 –6 –5 –4 –3 –2 –1 0 –8 –7 –6 –5 –4 –3 –2 –1 0 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 Ki (M) Ki (M) Ki (M) c 0. Selective pressure is presumably greatest for high-flux salt concentration in human plasma. in general. data points above the line of unity) is shown in the top left of each graph. similar to the metabolism. enzyme level is substantial. (a. substrate-to-product ratios are tied slower return of the pathway intermediates to steady-state levels fol. Enzyme evolution is 488 nature chemical biology | vol 12 | JUly 2016 | www.e. which are themselves to maintain adequate ΔG to be greatest for steps in which the total important precursors for biosynthetic pathways such as serine syn. metabolic control analysis that flux control. thesis.3 0.b) Comparison of absolute metabolite concentrations (y axis) to enzyme binding site affinities (x axis) for binding of substrate-enzyme (a) and regulator-enzyme (b). Indeed. together by thermodynamics. Indeed. Because each metabolite is simul- lowing perturbations and a decreased ability to control pathway flux taneously the substrate and product of one or more reactions. we have previously reported that lower glycolytic reactions uted across glycolytic enzymes.1 0 0 0 –10 –5 0 5 10 –10 –5 0 5 10 –10 –5 0 5 10 ln (metabolite concentration/Km or Ki) npg Figure 6 | Comparison of absolute concentrations to enzyme binding site affinities for substrates and for .2077 a Mammalian iBMK cells Yeast E. but not their absolute magnitude. coli 0 0 0 10 10 10 410/611 = 67% 228/324 = 70% 513/768 = 67% Substrate–Enzyme –1 –1 –1 10 10 10 interactions: [met] versus Km –2 –2 –2 10 10 10 Concentration (M) AT P –3 –3 –3 10 10 10 NAD+ –4 –4 –4 NADH 10 10 10 NADP+ 10 –5 10 –5 10 –5 NADPH –6 –6 –6 10 10 10 Central-carbon –7 –7 –7 Degradation 10 10 10 Other –8 –8 –8 10 10 10 –8 –7 –6 –5 –4 –3 –2 –1 0 –8 –7 –6 –5 –4 –3 –2 –1 0 –8 –7 –6 –5 –4 –3 –2 –1 0 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 Km (M) Km (M) Km (M) b 10 0 10 0 10 0 –1 76/158 = 48% –1 33/78 = 42% –1 67/141 = 48% Regulator–Enzyme 10 10 10 interactions: –2 –2 –2 10 10 10 [met] versus Ki Concentration (M) –3 –3 –3 10 10 10 –4 –4 –4 10 10 10 –5 –5 –5 10 10 10 © 2016 Nature America.1038/nchembio.4 0. A disadvantage of this more even free energy distribution is a Throughout metabolism. Apparently. any reaction of lower glycolysis that is substantially the glycolytic direction) or acetate (where net flow is in the opposite forward driven could be used to control pathway flux or the levels direction). either due to high flux or low kcat.2 0. these disadvantages are outweighed by a all metabolites. an important determinant of absolute metabolite abun- respect. All rights reserved. glycolysis in glucose-fed cells may reflect an extreme case.1 0. is not concentrated in one or two key are closer to equilibrium in cells fed glycerol (where flux remains in steps41–43.1 0. one would also expect evolutionary pressure of upstream and downstream intermediates. In general. article Nature chemical biology doi: 10. the via any single effector-enzyme interaction44.2 0. Inc. In each of the studied organisms. more saturated than inhibitor sites (P values are from the Kolmogorov-Smirnov test). thereby mitigating factors determine the absolute concentration of the collective metab- backward flux and thus the required amount of enzyme.3 binding site affinity (Kmor Ki) Probability Substrates 0.4 0. the sum of all mea- efficient enzyme use not only in glycolysis but also throughout sured metabolite concentrations is around 200 mM. In this bound. though unevenly distrib.

Joshi. 1453–1461 (2006). E. Biol. K. Biotechnol. V.. Correspondence and Biophys.. Henry. 5. Minakami.. 3212–3221 (2004).D. pMg. 42. & von Stockar. 7. Chem.A. 30. 93.P. 12. Amador-Noguez.S.R. and other factors on the reactivity and thermodynamic constraints on the metabolome. 90. L. These competing pressures are captured by the the molecular mechanisms regulating tumorigenesis. Nielsen. Beg. Chem. Fleming for their help with 14.. Biotechnol. Biotechnol. Understanding the Control of Metabolism (Portland Press.. Microbiol.E. and US National Institutes of Health 15.. Absolute resulting from optimality principles. Noisommit-Rizzi. 2255–2264 (2007).M.S. J. Schellenberger. developed the computational tools. 40. Berg.-J. Biophys. Noor and R. PLoS One 2. Curtis. & Palsson. Nature 436. H.C. M. Reactions may Krebs cycle with 13C-labeled lactate..M.A. metabolic flux analysis. fragmentation products as cellular metabolites in liquid chromatography- tating higher kcat and enabling more net flux per Xu. A. Acad.-F. 18.L. Lu. The thermodynamic meaning of metabolic exchange fluxes. 41. J. 79. & Lee. concentrations reflects the balance of this tailoring against universal 28. Metabolomic analysis via reversed-phase ion-pairing liquid substrate uptake by Escherichia coli and constrains its metabolic activity. 2012).L. Biol.. M. R. V. Wehner.D. P. conservation of absolute metabolite erythrocyte glycolysis. 121 (2007). J. Proteomics 11. Institute international student research fellowship. et al. active site occupancy in Escherichia coli. 544–553 (2011). B. Biol.. Thermodynamics-based 33. et al.-F. A genome-scale metabolic reconstruction for Escherichia coli R01grant 1R01CA163591 and DRC grant 2P30DK019525-37. D.M. Rev. e1003098 (2013)... 10039–10044 (2013). A new method for accurate quantification of changes in concentrations 43. 9.G.U. Liebermeister. K. S. & Zamboni. Nature chemical biology doi: 10. M. Beard. E. B. 3. Acad. Nat.. level of a protein. C. Schomburg.P. E. Nat. & Mann. of intracellular metabolites..A.F. Biophys. A. Author contributions 16. Measuring enzyme activities under standardized 36. Geiger. 2. A. & Hatzimanikatis. Kinetic flux 1. Xu. available online at http://www. 142. C. & Qian. W. 17. carried out the experiments.D. et al.. 014050 (2012). C. 309–315 (2002). et al. Biol. van Eunen. Fundamentals of Enzyme Kinetics (Portland Press. Protoc. Noor. USA 110. designed the study. Lu. D. Cox.D. 10 (2007). Tran. Appl. Ensemble modeling of metabolic networks. Biol. Dynamic rerouting of the carbohydrate flux is key to 10. R. J.S. J.D. A.. Fan was supported by a Howard Hughes Medical thermodynamically infeasible loops in steady-state metabolic models. J. Relationship between thermodynamic driving force Any supplementary information. 12663–12668 (2007)..O. E. A. The observed 27. R. & Alon. Competing financial interests Mol. 780–795 (2009). 53–73 (2002). 23 (2007). et al. Biol. Funding was provided by Biophys.A. Yuan.M. Biotechnol. Schuster. Y. Escherichia coli achieves Dynamic modeling of the central carbon metabolism of Escherichia coli. 31. Bar-Even.. Silhavy. Hoppe. Time hierarchy in enzymatic reaction chains 12. J. J. there is selective pressure to maintain metabolite pools as large as 23. Liebermeister. Chem. Katz.. quantification of haploid versus diploid yeast. Chem. Thermodynamic considerations on impressive. 529–530 (2006). and J. Proc. Cornish-Bowden. analysis of the thermodynamic feasibility of metabolic pathways. USA 104. 45–50 (2001). N. 41–68 (1990). Bioeng. Bioeng.html. X. Visual account of protein investment in cellular Part III—metabolic reaction rates. et al. Sci. USA 111. functions.P. nature CHEMICAL BIOLOGY | vol 12 | JUly 2016 | www. BRENDA in 2013: integrated reactions. Rizk. 5298–5304 (2006). Comparative Ultrasensitive regulation of anapleurosis via allosteric activation of PEP proteomic analysis of eleven common cell lines reveals ubiquitous but varying carboxylase. 1996).-F.O. Panke..Sauer. K.M. J. 95. L. Nöh. estimation of Gibbs energy using component contributions. K. Mauch. Absolute metabolite concentrations and implied enzyme profiling of nitrogen assimilation in Escherichia coli. & Palsson. M. MIRACLE: mass isotopomer ratio analysis of U-13C-labeled Hall. Lewis. taining substantial thermodynamic driving force. 588–592 (2005). activity coefficients. 2273–2281 (2015). N. J. Anal.. D. R. A functional genomics strategy that uses metabolome data straints or specific environmental challenges. Lu. Avoiding misannotation of in-source concentrations render low Km values unnecessary. R. J. M. L. 37. Raamsdonk. Syst. Henry.nature. Biol. D.L. & Stryer.. A. J. Kimball. Flamholz. & Rabinowitz. & Rabinowitz. Broadbelt. Chassagnole. & Heinemann. Mol. 87. 3. mass spectrometry–based metabolomics. L.C. All rights reserved. Syst. et al. Metabolic flux analysis at ultra short time scale: isotopically non-stationary 13C labeling experiments.O. M. Bioeng. Acknowledgments 13. & White. J.D. Adamberg. Biotechnol..R.N. comparison of methods for quantitative metabolomics of primary metabolism. J. Influence of uncertainties in pH.O. S. 263–283 (2001). Seiman. 249–267 (2007). et al. & Hatzimanikatis. New York. 41. 1997). 25509–25521 (1993). 25. 35. 6. Theor. Nucleic Acids Res..P. and J. Biol. Biochem. Nat. © 2016 Nature America. Bennett. As for main.. V. 4. S. classical metric of enzyme performance. J. 0034 (2006).. H. wrote the paper.J. Theor. Broadbelt. Biol. Including metabolite concentrations Methods into flux balance analysis: thermodynamic realizability as a constraint on flux Methods and any associated references are available in the online distributions in metabolic networks. J. Fowler. Haraldsdóttir.X. Heinrich. such as nutrient scar.R. Clin. New York. U.D. Glycolytic J. available in the online version of the paper. U. Freeman. L.M. J. Biol. Biol.. & Schuster. quantitation of intracellular metabolite concentrations by an isotope ratio-based approach. Commun. K. Ralser. Consistent feasible given thermodynamic and osmotic bounds. de Godoy.A. Kimball. & Rabinowitz. faster growth by increasing catalytic and translation rates of proteins. 189–209 (1987). & Reuss. 29. et al. Brynildsen and White labs for for biochemical networks. 277. Fell. W.2077 article subject to dual pressures to enhance both substrate binding affinity 21.H. Reprints and permissions information is 20. Drug Discov.. Milo.D. improved disease classification: new options and nutrient availability should therefore result in organism-specific contents in BRENDA. Sci. 327–333 (2004). A.J. 489 .F.. 103. Biochemistry (W. & Qian. & Milo. Natl. Anal. Mashego.. 44. E. J. strategy as a tradeoff between energy yield and protein cost. Noor. J.. Wiechert. I. et al. 129. The Regulation of Cellular Systems (Chapman & 11.. J. & Holzhütter.. M. 1995)... published online 2 May 2016 Genome-scale thermodynamic analysis of Escherichia coli metabolism.D. PLoS Comput.M. Biotechnol. W. Biol. Mol. Kümmel. Y. H. E. this pressure is 24. Haraldsdóttir. Wals. Res. 85.. N. Cancer Res. 2344–2358 (2013). 1299–1311 (2008). J.Ø. W. & Fleming. Dekel. R.. Theor.X. L. Vojinović. S. H. T. 8. Chem. J. chromatography coupled to a stand alone orbitrap mass spectrometer. 2. accepted 10 March 2016.nature.Ewald. B. J. A mouse model system to genetically dissect and turnover rate. Hoffmann. J. Cross-platform counteracting oxidative stress. to reveal the phenotype of silent mutations. chemical compound information and source data are and one-way fluxes in reversible processes. US Department of Energy grant DE-SC0012461. J. Reaves.R. Tymoczko. The authors declare no competing financial interests. Nat. Isotopomer studies of gluconeogenesis and the greatest for ‘hard’ reactions in high-flux pathways.A. D. be considered hard because of universal chemical reactivity con. 26. Natl. & Rabinowitz. W. Proc. K-12 MG1655 that accounts for 1260 ORFs and thermodynamic information. 620–628 (2004). J. J.. but incomplete. Feng. strains and cell lines and H.. 6. C. 3. 9. P. and S. W.-N. Feist. U. Cell. 100.. E. Degenhardt. A. T. Metabolic engineering. 8. B. Comprehensive mass-spectrometry–based proteome in vivo-like conditions for systems biology. & Yoshikawa. Higher metabolite 22. Received 30 July 2015. 19. expression of most proteins. S. Protein kinases—the major drug targets of the twenty-first century? 38. 593–599 (2009). K. thereby facili. version of the paper. Thus. 34. Chem. 82. J. 92. 562–568 (2012). 55. Y. D. Q.A.Czernik. & Liao. 2. Y. A. et al. Additional information 19.R. Proc. Optimality and evolutionary tuning of the expression Nat. Natl. FEBS J. S. 81. E. Schaab. 268. Schmid. 5606–5617 (2008).T.. 129. Sci. 1. city. 228.R.. Valgepea. 1792–1805 (2007). Nat.1038/nchembio. e144 (2007). & Heinrich.S. Acad. Biosyst. References 32. & Vilu. Metabolic dynamics in the human red cell. Biophys. W.K. J.R.W. Büscher. kinetic data. Elimination of Component Contribution method. Inc. BMC Syst. 18. extracts. Intracellular crowding defines the mode and sequence of 9.. 1251–1254 (2008). E. H. tailoring of absolute metabolite concentrations. 8488–8493 (2014). Yuan. D764–D772 (2013). Jankowski.D. Thermodynamic constraints The authors would like to thank the Gitai. S. Nature 455. 3212–3221 (2010). M. Cohen. and J. 5. J. W. S. 39. 1. J. 345–349 (1965). metabolite concentrations.R. npg Anal.O. et al. kcat/Km. Beard. Natural variation across organisms in pathway utilization and enzyme function data..H. Mol. Bennett.. 749–760 (2010). Babson. Chem. requests for materials should be addressed to J. Putative regulatory sites unraveled by network-embedded thermodynamic analysis of metabolome data.

0 OD600 = 0. (i) the net and the exchange flux for each ured after derivatization with S-methyl methanethiosulfonate (MMTS) on a reaction were each. cells were washed with warm PBS twice to remove exogenous amino ν is flux. alanine and ammonia reversibility of mammalian pyruvate kinase. Extraction continued for the Kyoto Encyclopedia of Genes and Genomes (KEGG)52. BAX−/−/BAK−/−–immortalized baby mouse kidney package53.22. and we assumed salt) was spiked into culture medium (which had been changed 3 h before.45 μm.3 for E. ONLINE METHODS (for example. the acetyl group labeling in acetyl- (IPTG) was introduced by electroporation. Organism-specific biomass hand) to a final concentration of 0. 15N]amino acids as internal standards. coli and yeast medium samples were analyzed by proton nuclear FBS. (iii) any balanced flux raphy (HILIC) coupled to tandem MS in positive ion mode (Quantum Ultra. E. the chro. yeast in [1. a high-resolution orbitrap mass spectrometer by electrospray ionization operating in negative ion mode (Exactive.6 × 10−14 l aqueous volume per cell49.1 M formic acid46). coli. Metabolism of expo. boring compounds in the networks. coli were cultured in medium containing magnetic resonance (1H-NMR) spectroscopy. in Gutnick minimal medium45 containing 10 mM NH4Cl and 0. the BioCyc database collection (for example. Inc. from prototrophic strains S288C and W303 were grown at 30 °C in 2% (w/v) D-glucose medium containing 6. Using the carbon mapping networks and 13CFLUX2 (http://www. Microbes were labeled starting from over. we obtained cumulated isotopomer models54 for further and later used in calculating cellular metabolite concentrations. and 2–3 h thereaf.6 for yeast) were vacuum-filtered onto nylon HumanCyc.0 OD600 = 0.47 gCDW/l. To determine confidence intervals. night cultures and mammalian cells were labeled for at least 24 h. In this n  siso n case. To obtain 95% confi- rities (and also CO2 incorporation for absolute metabolite quantification). coli filter culture and liquid culture are uptake and excretion rates using the interior-point algorithm: similar (Supplementary Fig. and metabolism was precursor effluxes were calculated using the measured growth rates and biomass quenched after 20 min. To measure cellular concentrations of amino 2 2 acids that are present in DMEM. successively increased or decreased.2-13C2]glucose. coli. and an ion of identical molec- Strains and culture conditions. All rights reserved. and metabolites were extracted in and carbon mappings were manually curated on the basis of genome-scale precooled extraction solvents. from the culture plates using cell lifters (iBMK cells).62 gCDW/l. and 5. the exchange flux was constrained to be media as above.  isoexp − iso(n )   n exp − n  min ∑   + ∑  s  beled DMEM and used [U-13C.9 × 10−11 gCDW per cell isotopic steady states were reached. we cultured iBMK cells in standard unla. sets where the variance-weighted sum of squared residuals did not increase by Thermo)47. calculated from the resulting net and exchange fluxes. Nucleotides. The metabolomes of E. lactate. Under net flux balance constraint. Organism-specific metabolic networks nentially growing cells was quickly quenched. (ii) the LC/MS/MS instrument performing hydrophilic interaction liquid chromatog. and extraction used solvents containing known concentra.3.005% (0. tion rates were determined by the rates of disappearance and appearance of lial cells (iBMK)21 were grown at 37 °C and 5% CO2 in Dulbecco’s modified metabolites in the medium after accounting for the amount of cells and their Eagle medium without pyruvate (DMEM) supplemented with 10% dialyzed growth rate. The supernatants were dried under nitrogen gas and reconsti. For iBMK cells. cells were grown in [U-13C6]glucose balanced flux distributions (a vector with two entries per reaction.2077 . (3 ml at OD600 ≈0. we used the following conversion factors: for E. YeastCyc) and standard textbooks. with the labeled media diffusing through agarose and filters until als between the simulated and experimental (i) isotope distributions and (ii) extraction12. 3 × 10−13 gCDW per cell and 7 × 10−16 l aqueous glucose or [U-13C5]glutamine (99%. forward fluxes. and metabolism was immediately quenched by the networks were mapped to their corresponding identification numbers in adding −80 °C 80% methanol to the culture dishes. mosomal tpiA was knocked out by P1 phage transduction of a deletion allele with To provide additional constraints for flux analysis. EcoCyc. Cysteine was meas. Metabolic flux analysis including determination of flux ratios with confidence tuted in HPLC-grade water for LC/MS analysis. BRaunschweig 20 min at −20 °C. one-by-one. Briefly. Each flux distribution simulated all isotope tracer studies simultaneously in E. and s is the measurement standard devia- acids before adding the extraction solvent. 3). The full set of reac- membrane filters (0. Saccharomyces cerevisiae derived CoA can be computed given the labeling of both acetyl-CoA and CoA).03 to 0. Maximum and Many LC/MS ion-specific chromatograms were complicated by multiple peaks minimum values (among the accepted flux distributions) of each individual nature chemical biology doi:10. coli or yeast culture metabolic reconstructions51. the membranes were immediately submerged in precooled Care was taken to avoid reaction inclusion bias by considering all known −20 °C 40:40:20 acetonitrile/methanol/water in petri dishes. dence intervals. [U-13C6] 1. backward fluxes and their ratios for all reactions were for natural 13C abundance as well as incomplete labeling due to 12C impu. that 70% of packed cell volume was aqueous volume. initial flux sets until similar flux sets that best simulated the experimental data Samples were analyzed by reversed-phase ion-pairing LC coupled to were obtained repeatedly. coli and yeast) and detaching the cells Knowledgebase (BiGG)51 to facilitate integration among these databases. coli and 0. small portions of E. 1.84. The extracts were centri- fuged at 4 °C. non-convex optimization problem was solved starting from over a thousand The extracts were neutralized with 15% NH4HCO3 before centrifugation. To create the inducible triose phosphate isomerase strain. For iBMK cells at routes to and from experimentally measureable metabolites and their neigh- ~80% confluency. and packed cell volume (iBMK cells) at the time of quenching were measured 13cflux.7 g/l Difco Yeast Nitrogen Base (YNB) with. Resulting mass spectra and chromatograms were processed using more than a fixed value (χ2) were accepted55 and (iv) for those accepted flux Metabolomic Analysis and Visualization Engine (MAVEN)48 and corrected sets. Uptake and secretion rate measurements.2-13C2]glucose. coli as well as NADH and NADPH in E. Cambridge Isotope Laboratories) until volume/cell. iso(ν) is simulated labeling.2-13C2] changes to cellular fluxes. yeast and mammalian for individual organisms. other fluxes were again optimized with one flux fixed. Metabolite measurements and absolute quantification. OD600 (E. Once tions included in the flux modeling is shown in Supplementary Table 6. nucleosides and bases tion. coli K-12 strain NCM3722 was grown at 37 °C ular formula to fumarate is produced by in-source degradation of malate). To account for other potential local minima. coli and S. E. [U-13C3]pyruvate (as the sodium excretion and fatty acid biosynthesis were previously reported50. the cells were measured using acidic extraction solvents (0. and metabolite extracts were collected into Eppendorf tubes ENzyme DAtabase (BRENDA)26 and a Biochemical Genetic and Genomic after thoroughly washing the filters (E.1038/nchembio. For steady-state flux analysis. analysis in MATLAB (MathWorks). cerevisiae) intervals. © 2016 Nature America. glucose-6-phosphate has isomers. some metabolite functional a kanamycin resistance cassette from the Keio collection. the rates of glucose. coli was cultured atop filters from OD600 were obtained by minimizing the variance-weighted sum of squared residu- 0. optimal npg For absolute metabolite quantification. Nutrient uptake and product secre- out amino acids. Reactions and metabolites included in ter the medium was aspirated. and a pAC24N::tpiA group labeling fractions were computed via the inverse Cauchy product of ASKA plasmid under the control of isopropyl β-D-thiogalactopyranoside other measured metabolites (for example. Organism-specific metabolic models. Thermo)9.45 mM). cells were loaded. one for the (and also [U-13C5]glutamine for the mammalian cells) but otherwise the same net flux and another for exchange flux. glucose or 50% [U-13C6]glucose. [3-13C1]glucose or 50% [U-13C6]glucose. E. standard injection and not simply the largest peak of the correct exact mass. and iBMK in [1. Millipore) resting on a vacuum filter support. the χ2 cutoff (1 degree of freedom) was 3. To convert medium concentration [1. cells were supplied with fresh medium. 1. negligible for known strongly thermodynamically forward-driven reactions) tions of authenticated standards12. To verify the glutamine and serine uptake as well as pyruvate. E. composition incorporated in respective genome-scale models51. and for yeast.4% (w/v) Care was taken to pick the peak at the correct retention time as validated by D-glucose.

respectively. For the eukaryotic cells.D. ionic strength (M) = [0. Because ΔG is linear with respect to ln[met]. Calvo. The sensitivity of ΔG°′ on pH and min ln[met]∈CCM ln[met] . electric potential relative lower bounds were obtained by solving the following linear programming to cytosol (mV) = [0. flux and of the forward-to-backward flux ratios are reported as the 95% con. problem17: ‘m’]. and for iBMK cells. pH and electric potential. ∆G = (∆ f G and ln[met]) are vectors of 46. The code used for flux and flux ratio calculation. respectively. Acidic acetonitrile for cellular metabolome free energy of reaction. s refers to the standard errors of the measurements or component contribution for met rxn fidence intervals. Supplementary Fig. Km and Ki. 0. experimentally measured metabolite concentrations and PrincetonUniversity/flux-ratio-based-gibbs-energy. Accordingly. with the optimization objective of minimizing linprog function (interior point algorithm). compartment = [ ‘c’. max ln[met]∈CCM ln[met] was solved with the same ated with central carbon metabolism (CCM.15.14. ionic strength and the standard error of ΔG°′ are shown in Supplementary subject to ln[met] ∈[ln[met]exp − 2smet . a quadratic programming problem was formulated with independ. Inorganic phosphate concentrations ological ionic strength serves to incorporate the activity coefficients of reac. respectively. 7. where ∆G = ST (∆ f G + RT ln[met]) . 8.4.0. The lower and upper bounds for individual metabolite concentrations were tric potential were employed. pH and elec. subject to ln[met] ∈[ln[met]exp − 2smet . The uncertainties were generally greater than the perturbations of ΔG°′ resulting from changes up to ± 0.25. the SOAPpy script.1 M ionic strength. and ∆G ∈[∆Gexp − 2S∆G . J. Chem. ln[met]exp + 2Smet ]. ‘e’ for extraorganism. N exp and N exp are the number of input metabolite on Linux clusters and is available as open-source code at https://github.0.25]. ∆Gexp + 2S∆G ] 45. and the values of ΔfG°′ According to extended Debye-Hückel theory. The Python imple- min 2 ′ ′  2 mentation of the Simple Object Access Protocol (SOAPpy) was employed to 1  ln[met]exp − ln[met] 1  ∆ f Gexp − ∆f G In[met].2077 nature CHEMICAL BIOLOGY . For upper bound calculation. not activities. −155]57. The inorganic phosphate and coenzyme A concentrations were into account the effect of [Mg2+] on the ΔG°′ of ATP hydrolysis by using the allowed to vary within 20% of these values. were input as follows: 20 mM in E. The exported data were further processed to remove f 2 1  ∆Gexp − ∆G  entries for mutant enzymes. When multiple entries for the same enzyme-metabolite pair were available.15. musculus whenever possible but otherwise are from Homo sapiens. determination of ΔG°′ at physi. tial relative to cytosol (mV) = [0.. the impact of pH and ionic strength variation within these ranges should minimally affect the subsequent integrative analysis. extraction from Escherichia coli. T. chondria owing to pH difference across compartments. concentrations that also resulted in ΔG falling within its measured 95% con- tial relative to cytosol (mV) = [0. ∆Gexp + 2s∆G ] npg prior estimates of ΔfG°′ on the basis of the component contribution method (which itself incorporates prior literature data on ΔG°′) and our experimental observations of metabolite concentrations and cellular reaction free energies. For upper bound calculation. matched (i) the directly observed concentrations for measured metabolites Similarly. The parameters for mammalian cells are from + rxn N exp ∑  S∆G  enzymes in M. lite concentration and Gibbs energy integration. cerevisiae. where ‘c’ stands for cytosol.N. 128.2].14. In PrincetonUniversity/flux-ratio-based-gibbs-energy. 50 mM in yeast5 and 5 mM in mam- tants and products into ΔG°′. Rabinowitz. Bacteriol 100. max ln[met]∈CCM ∆G was solved with the same set To this end. The set of metabolite concentrations ([met]) and reaction free energies (ΔG) associ. Tables 7 and 8. ‘e’. All rights reserved. metabo- subject to ln[met] ∈[ln[met]exp − 2smet .∆ f G ′ ∈ CCΜ ∑  + for ∑   extract the Km and Ki values in E. −155]5.5]. concentration.2.15]. ‘e’. fidence interval.14].2. calculated for mitochondria were used. concentration pH = [ 7. D. of constraints. ∆ f Gexp + 2S∆ f G ] models are available on the GitHub public repository: https://github.4 pH and ± 0. compartment = [ ‘c’. S. ‘e’.25. we set out to optimize both metabo. coli58. & Ames. MATLAB to calculate standard Gibbs free energy of formation (ΔfG°′) and ΔfG°′ of TCA metabolites depended on whether they are in cytosol or mito- reaction (ΔG°′) at physiological ionic strength. lite concentrations and formation energies so as to maximize consistency with and ∆G ∈[∆Gexp − 2s∆G . the carbon mapping and cumulated isotopomer ′ ′ ∆f G ∈[∆ f Gexp − 2S∆ f G . The linear programming problem was solved using MATLAB ent variables ln[met] and ΔfG°′. 6167–6173 (2007).. therefore. concentration 95% confidence interval for the minimum and maximum ‘p’]. ∆Gexp + 2s∆G ]. these ΔG°′ values are   o′  interrelated. 7. for yeast. Anal. thereby allowing calculation of Q on the basis malian iBMK cells59. 5. For reactions whose ΔG values were not precisely determined. Inc. min ln[met]∈CCM ∆G. E.0. The algorithm was scripted in MATLAB to run in parallel estimates. coli. Component contribution23 was performed in strained to be negative in the direction of net flux.2. electric poten.0]. 90. compartment = [ ‘c’. the departure from the expected ΔfG°′ and measured ln[met] and ΔG: Michaelis and inhibitor dissociation constants. pH = [ 7. Mus musculus and Homo met N exp Smet  N exp  s∆ G  sapiens from BRENDA26.ln[met]exp + 2smet ]. coli15.7. ionic strength (M) = [0. S is the stoichiometric matrix with rows and columns representing individual   o′  J. 90]56. 4) that best set of constraints. 215 (1969). electric poten- © 2016 Nature America. 0. Mitochondrial coenzyme A concentration was input as of concentrations.0. 7. pH = [ 7. and and ∆G ∈[∆Gexp − 2s∆G .M. such that ΔG°′ for any sequence of metabolic reactions must be where ∆G = S T (∆ f G + RT ln[met]) . ΔG was con- Standard Gibbs free energy. ‘m’]. ionic strength (M) = [0. given by the sum of the formation energies of the products minus the forma- tion energies of the reactants. Klopotow. For example. its   o′  Km or Ki was represented by their geometric mean.ln[met]exp + 2smet ]. T  o′  where ∆G = S (∆ f G + RT ln[met]) Integration of metabolite concentrations and Gibbs free energies. 7. data from the NIST Thermodynamics of Enzyme-Catalyzed Reactions data- base regarding the ATP hydrolysis equilibrium constant at typical cellular Mg2+ Confidence intervals of metabolite concentrations and reaction free energies. B. J. component contribution takes 5 mM60. An important consideration in this calculation is that literature val. 79.4. 30. Gutnick. metabolites and reactions. ΔG. Compartment-specific values for ionic strength. reaction free energy lower bounds were obtained by solving: ([met]exp) and (ii) the observed cellular ΔG for measured reactions (ΔGexp) were computed. N exp .com/ formation energies. ues for ΔG°′ may themselves contain error. confidence interval determi- o′ o o nation. standard free energy of formation and log concentrations. Compounds which serve as sole source of carbon or nitrogen for Salmonella typhimurium Lt-2. doi:10. determined by searching within the experimentally measured metabolite ‘p’ for periplasm and ‘m’ for mitochondria: for E. Code availability. & Kimball.1038/nchembio.

S. metabolism in KEGG. Determination of with MAVEN: a metabolomic analysis and visualization engine. R.. Metab. 66. Glutamine-driven oxidative phosphorylation is a major ATP 57.M. Chem.. Leonardi. J.2077 . 8. to Escherichia coli. metabolites by hydrophilic interaction chromatography-tandem mass Bidirectional reaction steps in metabolic networks: III. Clasquin. & Fleming.M. G. H.A. 102.1–14. Jacomb. M. LC/MS data processing 55. N. 48..1.O. 145. et al. & Nasheuer. 50. Prog. measurements. Syst. Biochem. Park. Thiele. & Palsson.M. Inc.. npg nature chemical biology doi:10. relational database to coordinate and facilitate in silico reaction directionality in constraint-based models of metabolism: application modeling of Escherichia coli. Rosenberg. 143–145 (2013). knowledge and principle: back to 59. 56. et al. I. 76–88 (2006). J.. M. Schellenberger.K. 32. L. Eng. BiGG: 58. Separation and quantitation of water soluble cellular 54. Wiechert.M. 1703–1711 (2012). confidence intervals of metabolic fluxes estimated from stable isotope Bioinformatics 37. Melamud. Zhang. 69–85 (1999). 47.A. Nucleic Acids Res. Biol. 42. 49.U. Protoc.1038/nchembio. H. R. BMC Bioinformatics 11. J. S. Biophys.. J. Fan. Isermann. Physiological concentrations of purines and pyrimidines. & Chegwidden. 51. Explicit solution and spectrometry. Kanehisa. All rights reserved. Cell. Conrad. J. S. Wurzel. H. W. J. 123–130 (1982). analysis of isotopomer labeling systems. 149. Bajad.O. Möllney. A. Rock. Sundararaj. Antoniewicz. reconstruction. J. © 2016 Nature America. T.T. 44. 53. Quantitative assignment of source in transformed mammalian cells in both normoxia and hypoxia. & Rabinowitz.. C. Phosphate a biochemical genetic and genomic knowledgebase of large scale metabolic exchange in the pit transport system in Escherichia coli. Haraldsdóttir. I. S. Biotechnol. P. & Jackowski.W. Bacteriol. 213 (2010). Weitzel.D. 52. 1–22 (1994). J.. A 1125. Traut. Russell. 125–153 (2005). 140. Bioinformatics 29.P. information. R. action. Lipid Res. M. et al. D199–D205 (2014).. T. M.Ø. et al. 324–337 (2006). Data. reconstructions. Y.T.. & de Graaf. Bioeng. E. D293–D295 (2004). 712 (2013). The CyberCell Database (CCDB): a comprehensive. M. M.M.F. Fleming. 14. 47–56 (2009). K. Curr. Biophys.11. reaction directionality in a multicompartmental human metabolic Mol. 9.23 (2012). Kelleher. 13CFLUX2--high-performance software suite for 60. Nucleic Acids Res. Quantitative assignment of self-updating. Coenzyme A: back in 13C-metabolic flux analysis. et al. Mol. B...R. & Stephanopoulos. Chromatogr.. Thiele.