Running Head: 2 Steady-state metabolic flux analysis in ... · 3 feature of plant cell metabolism. Ideally, steady-state metabolic flux analysis (MFA), an 4 emerging tool for creating

1

Running Head: 1

Steady-state metabolic flux analysis in Arabidopsis 2

3

4

Corresponding author: 5

Professor R.G. Ratcliffe 6

University of Oxford, 7

Department of Plant Sciences, 8

South Parks Road, 9

Oxford OX1 3RB 10

United Kingdom 11

12

Tel: +44 (0)1865 275000 13

Fax: +44 (0)1865 275074 14

E-mail: [email protected] 15

16

17

Journal research area: 18

Biochemical Processes and Macromolecular Structures19

Plant Physiology Preview. Published on November 25, 2009, as DOI:10.1104/pp.109.151316

Copyright 2009 by the American Society of Plant Biologists

www.plantphysiol.orgon May 11, 2020 - Published by Downloaded from Copyright © 2009 American Society of Plant Biologists. All rights reserved.

http://www.plantphysiol.org

2

Subcellular flux analysis of central metabolism in a heterotrophic Arabidopsis thaliana 1

cell suspension using steady-state stable isotope labeling1 2

3

Shyam K. Masakapalli, Pascaline Le Lay, Joanna E. Huddleston, Naomi L. Pollock, Nicholas 4

J. Kruger2 and R.George Ratcliffe2* 5

6

Department of Plant Sciences, University of Oxford, South Parks Road, Oxford OX1 3RB, 7

United Kingdom 8

9



3

1 This work was supported by the Biotechnology and Biological Sciences Research Council 1

[43/B17210], United Kingdom. 2

3 2 These authors contributed equally to the article. 4

5

* Corresponding author; e-mail [email protected] 6

7

8

9



4

ABSTRACT 1

The presence of cytosolic and plastidic pathways of carbohydrate oxidation is a characteristic 2

feature of plant cell metabolism. Ideally, steady-state metabolic flux analysis (MFA), an 3

emerging tool for creating flux maps of heterotrophic plant metabolism, would capture this 4

feature of the metabolic phenotype, but the extent to which this can be achieved is uncertain. 5

To address this question, fluxes through the pathways of central metabolism in a 6

heterotrophic Arabidopsis thaliana cell suspension culture were deduced from the 7

redistribution of label in steady-state 13C-labeling experiments using [1-13C]-, [2-13C]- and 8

[U-13C6]glucose. Focusing on the pentose phosphate pathway (PPP), multiple datasets were 9

fitted simultaneously to models in which the subcellular compartmentation of the PPP was 10

altered. The observed redistribution of the label could be explained by any one of three 11

models of the subcellular compartmentation of the oxidative PPP, but other biochemical 12

evidence favored the model in which the oxidative steps of the PPP were duplicated in the 13

cytosol and plastids, with flux through these reactions occurring largely in the cytosol. The 14

analysis emphasizes the inherent difficulty of analyzing the PPP without predefining the 15

extent of its compartmentation, and the importance of obtaining high quality data that report 16

directly on specific subcellular processes. The Arabidopsis flux map also shows that the 17

potential ATP yield of respiration in heterotrophic plant cells can greatly exceed the direct 18

metabolic requirements for biosynthesis, highlighting the need for caution when predicting 19

flux through metabolic networks using assumptions based on the energetics of resource 20

utilization. 21

22



5

INTRODUCTION 1

Extensive subcellular compartmentation, with unique locations for many steps and 2

pathways, as well as the duplication of other steps and pathways in different compartments, 3

adds greatly to the structural complexity of the plant metabolic network (Lunn, 2007; Kruger 4

and Ratcliffe, 2008). The need to consider discrete pools of metabolites in specific 5

compartments, and the transporters that link them, complicates the quest for a detailed, 6

predictive understanding of the regulation of plant metabolism; and as a result it remains 7

difficult to manipulate flows of material through the central metabolic network in a 8

predictable way (Carrari et al., 2003; Kruger and Ratcliffe, 2008; Sweetlove et al., 2008). 9

The emergence of steady-state metabolic flux analysis (MFA) as a practicable systems 10

biology tool for generating flux maps of the central metabolic pathways in plants offers new 11

opportunities for analyzing plant metabolic phenotypes (Ratcliffe and Shachar-Hill, 2006; 12

Schwender, 2008; Libourel and Shachar-Hill, 2008; Kruger and Ratcliffe, 2009). In this 13

approach, substrates labeled with stable isotopes are introduced into the network, and fluxes 14

are determined by measuring the labeling of the system after it has reached an isotopic and 15

metabolic steady state. Subcellular compartmentation of steps and pathways can be 16

incorporated into the model that describes the redistribution of the label, and flux maps of 17

central carbon metabolism in plant cells typically aim to distinguish between fluxes in the 18

cytosol, mitochondria and plastids. In principle compartmented fluxes can be deduced from 19

the labeling patterns of metabolites that are synthesized in a single compartment (Ratcliffe 20

and Shachar-Hill, 2005, 2006; Allen et al., 2007), and this property has been exploited with 21

varying degrees of success in different studies. 22

Most recent flux maps, such as those for developing embryos of oilseed rape (Brassica 23

napus) (Schwender et al., 2006; Junker et al., 2007) and sunflower (Helianthus annuus) 24

(Alonso et al., 2007a), or the one for a heterotrophic cell suspension of Arabidopsis thaliana 25

(Williams et al., 2008), have well-defined mitochondrial fluxes. This reflects the high 26

sensitivity of the available label measurements to these fluxes, but it also reflects the arbitrary 27

way in which the models assume that some fluxes, for example from citrate to 2-oxoglutarate, 28

are exclusively mitochondrial, and others, for example between malate and oxaloacetate, 29

occur in an unspecified compartment. There is less consensus in the description of the 30

pathways of carbohydrate oxidation, with the subcellular compartmentation of glycolysis and 31

the oxidative pentose phosphate pathway (PPP) being described at various levels of detail. 32

For example, a recent model of soybean (Glycine max) cotyledons included complete 33

pathways for glycolysis and the PPP in both cytosol and plastids (Iyer et al., 2008); whereas 34



6

models for developing sunflower embryos (Alonso et al., 2007a), maize root tips (Alonso et 1

al., 2007b) and an Arabidopsis cell culture (Williams et al., 2008) confined the oxidative and 2

non-oxidative steps of the PPP to the plastid. The limited compartmentation of the pathways 3

of carbohydrate oxidation in many flux maps can be traced to an analysis of developing 4

oilseed rape embryos (Schwender et al., 2003) which concluded that the labeling data could 5

be satisfactorily explained without duplication of glycolysis and the PPP. Since some of the 6

enzymes of these pathways are known to be present in both compartments, the model implied 7

fast exchange of intermediates across the plastid envelope, giving rise to a situation in which 8

the physically compartmented pathways are functionally uncompartmented (Schwender et 9

al., 2003; Ratcliffe and Shachar-Hill, 2006). 10

Although the extent to which the steps of the PPP are duplicated in the cytosol and 11

plastids has not been fully established, most of the biochemical evidence suggests that the 12

enzymes for the oxidative steps at least are present in both compartments (Kruger and von 13

Schaewen 2003). So modeling the PPP as fully duplicated in both cytosol and plastids, or 14

alternatively as exclusively plastidic, ignores the most likely structure of the network. Models 15

based on all three network structures have now been tested in an Arabidopsis cell culture. 16

Building on a steady-state MFA analysis of the same culture (Williams et al., 2008), similar 17

experiments were conducted with multiple substrates and the labeling datasets were 18

simultaneously fitted to models in which the subcellular compartmentation of the PPP was 19

varied. Acceptable fits to the data were found for all three models, emphasizing the 20

importance of building the most appropriate level of subcellular compartmentation into the 21

model at the outset. Moreover the flux maps showed that the data were consistent with a 22

major contribution from the cytosol to the oxidative steps of the PPP. 23

24

RESULTS 25

Biosynthetic outputs 26

The Arabidopsis cell culture was maintained in the dark with glucose as the sole 27

respiratory substrate, and the principal activities of the metabolic network, together with the 28

associated biosynthetic outputs, were assessed by determining the redistribution of radiolabel 29

following metabolism of [U-14C]glucose (Table I). Overall the pattern of labeling was 30

comparable to that observed previously for heterotrophic cell cultures of Arabidopsis and is 31

similar to that reported for a wide range of non-photosynthetic plant material (Kruger et al., 32

2007a). However, extending the analysis to include components that are frequently ignored 33

during similar fractionation procedures, such as the organic phase obtained during 34



7

chloroform/methanol fractionation and the medium following removal of cells, allowed a 1

more comprehensive analysis of the fate of metabolized glucose and measurement of the 2

labeling of lipids and ethanol. The uptake of label was linear over 24 h, and the proportion of 3

radioactivity recovered in each of the major chemical fractions was the same after 12 and 4

24 h incubation (Supplemental Fig. S1). This implies that the pools of metabolic 5

intermediates had achieved isotopic steady-state within this time and that accumulation of 6

radiolabel in the products reported in Table I reflects the rate of synthesis of these compounds 7

rather than just isotopic equilibration within existing pools. The data show that almost 40% of 8

the metabolized glucose was released as CO2, while less than 25% of the carbon was 9

converted into biopolymers (comprising starch, cell wall, protein and lipid), with the 10

remainder accumulating mainly as soluble sugars, organic acids and amino acids. While this 11

analysis defines the major output fluxes associated with growth, further information is 12

required to determine the fluxes that generate the precursors needed for biomass production, 13

and this can be provided by steady-state MFA. 14

15

Metabolic network 16

In MFA, the metabolic network is reduced to a model that can account for the observed 17

redistribution of the label. Typically this leads to a model in which: (i) multiple steps between 18

branch points in the network are represented as a single reversible or irreversible step; and (ii) 19

subcellular pools of metabolites are treated as single pools in the absence of compartment-20

specific information about their labeling. The scope of the model is strongly influenced by 21

the choice of label precursor, and by the extent and quality of the measurements that define 22

the redistribution of the label after the system has reached an isotopic and metabolic steady 23

state. It has already been shown that statistically robust flux maps of central carbon 24

metabolism can be obtained for heterotrophic Arabidopsis cell suspensions using [1-25 13C]glucose as the label source (Williams et al., 2008), and here the aim was to establish 26

whether improved definition of the subcellular compartmention of carbohydrate oxidation 27

could be obtained by combining the results of experiments with [1-13C]glucose, [2-28 13C]glucose and [U-13C6]glucose as label precursors. 29

The existing model, in which the oxidative and non-oxidative reactions of the PPP were 30

confined to the plastids (Fig. 1A), was compared with two alternatives: (i) a model in which 31

the oxidative steps of the PPP were allowed to occur in both the cytosol and plastids, with the 32

non-oxidative steps confined to the plastids (Fig. 1B); and (ii) a model in which the complete 33

PPP was present in both compartments (Fig. 1C). Models were constructed in the format used 34



8

by the steady-state MFA software 13C-FLUX and full details of the model in which the 1

oxidative steps of the PPP were duplicated in the cytosol and plastids (Fig. 1B, 2) are given in 2

Supplemental File S1. In essence the model defines the relationship between the carbon 3

atoms in substrates and products for each step in the network so that the underlying fluxes 4

can be deduced from the redistribution of label in the steady-state labeling experiment. The 5

models used here are directly related in most respects to the one used earlier (Williams et al., 6

2008) and they were constructed in the same way. Several features of the models, and the 7

biochemical networks they represent, should be noted. 8

First, analysis of the subcellular distribution of flux depends on obtaining information 9

about the steady-state labeling of metabolic intermediates in specific compartments. Here the 10

labeling of the plastidic pools was deduced from the labeling of starch, and from amino acids 11

that are known to be synthesized exclusively in the plastid (Cys, Gly, His, Ile, Leu, Lys, Phe, 12

Ser, Trp, Tyr, and Val). Note that although serine, following synthesis in the plastid (Ho and 13

Saito, 2001) can be reversibly converted to glycine via serine hydroxymethyltransferase, the 14

fact that this can potentially occur in several compartments has no effect on the interpretation 15

of the serine labeling pattern. Similarly the labeling of cytosolic intermediates was deduced 16

from the labeling of sucrose and alanine (Miyashita et al., 2007); and arginine, γ-17

aminobutyrate (GABA), glutamate, glutamine and proline were assumed to be derived from 18

mitochondrial 2-oxoglutarate. In contrast it was not possible to deduce information about the 19

subcellular labeling of oxaloacetate and malate, and the synthesis of these metabolites was 20

not assigned to a specific compartment. 21

Secondly, because the model describes the redistribution of label through the network, 22

expected biochemical features of the network may be concealed or omitted. For example, it is 23

possible to synthesise glycine from serine using threonine aldolase (Joshi et al. 2006b), but 24

using this as the sole route to glycine leads to a substantially worse fit to the data, and adding 25

the threonine aldolase reaction as an extra component of the model in Fig. 2 leads to no 26

change in the flux estimates (data not shown). Thus there is no evidence from the labeling 27

data to justify the inclusion of the threonine aldolase reaction in the model. Theoretical 28

analysis also has a bearing on the construction of the model. For example flux through the 29

GABA shunt is combined with the flux through the TCA cycle between 2-oxoglutarate and 30

fumarate because these pathways have identical labeling outcomes in the model. Similarly 31

sucrose cycling is omitted because this can only be correctly analyzed in steady-state MFA 32

with information on the labeling of the cytosolic and vacuolar glucose pools (Kruger et al., 33

2007b). The flux supported by the putative glucose-6-phosphatase (Alonso et al., 2005) is 34



9

omitted for the same reason, because there is no easy way to measure the labeling of the 1

cytosolic and vacuolar glucose pools, and ignoring the compartmentation of glucose is likely 2

to produce very misleading flux values (Kruger et al., 2007b). In contrast, discrete cytosolic 3

and plastidic pools of glucose 6-phosphate were retained in the model because the glucosyl 4

units of sucrose and starch differed appreciably in their patterns of labeling, establishing that 5

they were synthesized from isotopically distinct precursors (Supplemental Fig. S2, A-C). 6

Subsequent in silico analysis established that the calculated flux distribution would produce 7

differential labeling in products derived directly from cytosolic and plastidic hexose 8

phosphate pools (Supplemental Fig S2, D-F). 9

Thirdly, the steps catalyzed by transketolase (TK) and transaldolase (TA) in the non-10

oxidative steps of the PPP are represented in terms of half-reactions to allow for the 11

underlying mechanism of the enzymes and their lack of specificity (Kruger and von 12

Schaewen, 2003; Kleijn et al., 2005, Selivanov et al., 2005). Conventional stoichiometric 13

models of the non-oxidative reactions (Fig. 3A) assume that the label is directed through a 14

subset of the possible reactions and thus misrepresent the underlying fluxes. Adding the extra 15

reactions for an equivalent formulation of the pathway (Fig. 3B) leads to ambiguity in the 16

flux solution (Fig. 3D), but this can be avoided by using the half-reactions based on the ping-17

pong mechanisms of TK and TA (Fig. 3C, E). Theoretical analysis has shown that this 18

formulation is essential for a correct description of the redistribution of label and that it 19

allows the investigation of parallel pathways supported by different TK isozymes (Kleijn et 20

al., 2005). Accordingly the half-reaction scheme is used here to maximize the possibility of 21

resolving cytosolic and plastidic contributions to the PPP fluxes. 22

23

Data handling and mathematical modeling 24

Heterotrophic Arabidopsis cell cultures were incubated with [1-13C]glucose, [2-25 13C]glucose or 10% [U-13C6]glucose, the label redistribution at isotopic and metabolic steady 26

state was quantified by 13C NMR, and flux solutions, constrained by measurements of 27

glucose consumption and biosynthetic outputs (Supplemental Tables S1-4), were calculated 28

in 13C-FLUX using the protocol summarized in Figures 4 and 5. Label was routinely 29

measured in soluble metabolites, including amino acids, citrate, fumarate, GABA, malate and 30

sucrose, in the amino acids derived from protein hydrolysate, and in the glucose derived from 31

starch. The best fit flux solutions presented here were based largely on sucrose, starch and 32

protein hydrolysate labeling data, and full details of the measurements used in the fitting 33

process are given in Supplemental File S2. Measurement errors were generally set at 5% for 34



10

amino acid data, and 15% for other metabolites, but a few measurements were assigned larger 1

deviations as specified in Supplemental File S2. 2

Label measurements for a particular metabolite were initially grouped together for the 3

fitting process, but subsequently it was established that better fits (lower residua) could be 4

obtained by treating 13C peaks with multiplet structure as separate subgroups. Simulations in 5

13C-FLUX showed that the reduction in the residuum could be attributed to the inclusion of 6

additional scaling factors, compensating for discrepancies in the relative intensities of the 7

signals from different carbon atoms within a metabolite, and that the subgrouping procedure 8

did not compromise the accuracy of flux estimation. Subgrouping multiplet signals avoids the 9

need to establish the relative intensities of multiplets from the same metabolite, and it is used 10

routinely when two dimensional NMR methods, which often mask the true relative 11

intensities, are used to analyze [U-13C6]glucose labeling experiments (Sriram et al., 2004). 12

Each of the three labeling experiments was performed twice, and various combinations of 13

experiments were analyzed simultaneously in 13C-FLUX by setting up models in which the 14

reaction network was replicated two, three or six times. Software limitations, arising from the 15

large size of the network and the quantity of data, placed some restriction on this approach, 16

but a three-substrate model, comprising a sub-network for each labeled precursor, i.e. [1-17 13C]glucose, [2-13C]glucose or 10% [U-13C6]glucose, and two datasets for each sub-network, 18

worked well, and a single flux solution corresponding to the entire dataset could be generated 19

by constraining equivalent fluxes in the three sub-networks to be equal in the model 20

definition file. Ultimately this procedure allowed a more reliable definition of the fluxes in 21

the network than could be obtained by analyzing different labeling experiments separately 22

(see below). 23

Preliminary fits of the data to the model were used to refine the model and to identify 24

errors in the labeling data. In this iterative process, 13C-FLUX was used to generate 50-100 25

solutions for a given model and dataset. Changes were then made to either the model or the 26

dataset, for example the addition of a pathway or the correction of an incorrectly annotated 27

data point, and the fitting process was repeated. Comparing the residua for the five best fits 28

provided the basis for accepting or rejecting a change to the model or the dataset. For 29

example the reaction catalyzed by isocitrate dehydrogenase, and the mitochondrial uptake of 30

pyruvate, were both defined as unidirectional because making them reversible had no effect 31

on the fit. Similarly CO2 uptake at natural abundance was removed from the model because it 32

had no discernable effect on the fit. In contrast adding unidirectional plastidic pyruvate 33

uptake improved the fit, as did the inclusion of separate cytosolic and plastidic pools for PEP 34



11

and triose phosphate. However, the latter metabolites were combined to create a single pool 1

of 3-carbon phosphate esters in each compartment since this removed indeterminable fluxes 2

from the network and had no effect on the fit. The iteration process lead to the model in 3

Figure 2, and the validated dataset in Supplemental File S2, and thus provided a secure 4

foundation for testing the extent to which the method could reveal the subcellular 5

compartmentation of carbohydrate oxidation (Fig. 1). 6

7

Best fit fluxes and statistical analysis 8

The six datasets, comprising over 700 positional isotopomer measurements and 29 9

biomass fluxes, were initially analyzed in a three-substrate model based on the flux network 10

in Figure 2. Estimates for the free fluxes were obtained by running the optimization routine 11

1000 times, generating a correspondingly large set of flux solutions, each with a set of 12

predicted labeling measurements. Some of these solutions failed to satisfy constraints in the 13

model, and were identified as infeasible solutions by the software; but the majority were 14

feasible solutions, and these were extracted from the 13C-FLUX output file with a 15

customized software tool (available from the authors on request). The residua for the feasible 16

solutions varied over a wide range (Fig. 6A), and a subset of the solutions corresponding to 17

residua below 3000 was selected for further analysis. This cut-off was justified both by visual 18

inspection of the distribution of residua (Fig. 6A) and by principal components analysis of the 19

mean-centered, unit variance scaled flux solutions (Fig. 6B). The one dimensional scores plot 20

corresponding to the first principal component confirmed the similarity between the selected 21

feasible solutions and provided further justification for rejecting the solutions with residua 22

above the cut-off. 23

Monte Carlo simulations with bootstrap sampling of isotopomer measurements provides 24

an effective method for characterizing the size and shape of the flux space (Wiback et al., 25

2004; Joshi et al., 2006a) and the selected solutions showed considerable variability in the 26

definition of the free fluxes (Supplemental Fig. S3). Some fluxes were tightly defined, for 27

example many of those in the TCA cycle, while others were normally distributed over a 28

wider range of values. To reduce the hundreds of selected feasible solutions to a global best 29

fit, the mean values from the Monte Carlo flux space were used as starting free flux values 30

for a final run of the optimizer with the validated dataset. This run was performed without 31

bootstrap Monte Carlo sampling, and it resulted in a global best fit solution with the lowest 32

residuum (Table II). These procedures were repeated for the models in which the full PPP 33



12

was either confined to the plastids (Fig. 1A) or present in both compartments (Fig. 1C), and 1

the final best fit fluxes for the three reaction networks are given in Table II. 2

Statistically reliable flux values were generated for all three models of the PPP (Table II) 3

with only minor differences in residuum. The flux distribution for the steps involving sugar 4

phosphates differed substantially for the three models (Fig. 7), and it was noticeable that the 5

bulk of the flux through the oxidative steps of the PPP switched to the cytosolic pathway 6

when this option was included in the network structure. Thus while the flux through the 7

oxidative steps was necessarily confined to the plastids in the absence of a cytosolic pathway 8

(Fig. 1A, 7A), the plastidic pathway made only a minor contribution when the cytosolic 9

pathway was included in the model (Figs 1B,C, 7B,C). The degrees of freedom in the models 10

increased with the number of fluxes, and as expected this led to a small improvement in the 11

residuum achieved in the fitting procedure and an increase in the standard deviations for the 12

flux estimates. For example including the full PPP in both cytosol and plastids (Fig. 1C) 13

produced a marginal improvement in residuum, but it produced a flux map in which 28 of the 14

33 fluxes shared by all three models had the largest standard deviations. 15

The validity of the flux maps was tested by comparing the predicted contribution of the 16

carbon atoms of metabolized glucose with the relative specific activity of 14CO2 released 17

from specifically labeled glucose (Fig. 8). Ratios of 14CO2 labeling from [1-14C]-, [2-14C]-, 18

[3,4-14C]- and [6-14C]glucose are typically used to characterize carbohydrate oxidation 19

(Harrison and Kruger, 2008), and the pattern of 14CO2 release predicted by the global best-fit 20

flux estimates for each model was not significantly different from the experimentally 21

determined pattern (Fig. 8), implying that each of the flux maps provides an adequate 22

explanation of the observed data for the release of 14CO2 from positionally labeled glucose. 23

Since the models are indistinguishable, it can be concluded that the subcellular 24

compartmentation of the PPP cannot be determined unambiguously from the available 25

labeling data alone. However on current biochemical evidence the most likely model is the 26

one in which the oxidative steps of the PPP are duplicated in the cytosol and the plastids (Fig. 27

1B, 7B; Kruger and von Schaewen, 2003). The global best-fit solution for this network is 28

presented in detail in Supplemental Table S5, and it provides good agreement between the 29

observed labeling data and the isotopomer abundances predicted from the global best fit flux 30

map (Fig. 9A). There is also excellent correspondence between the measured, scaled and 31

predicted relative fractional abundance of positional isotopomers for individual metabolites – 32

as exemplified by the data on protein-derived aspartate/asparagine (Fig. 9B). More than 80% 33

of the total residuum was confined to 176 measurements (25% of total), and the flux map 34



13

obtained after removal of these values was essentially identical to that obtained from the 1

complete data set suggesting that the poorly fitting measurements did not have a major 2

influence on the flux solution (Supplemental Fig. S4). Furthermore, there was no consistency 3

in the identities of the isotopomers that contributed most to the residua between duplicate 4

feeding studies involving identically labeled substrates, implying that the discrepancies 5

between the observed and predicted values were due to poor estimates of measurement error, 6

rather than a failure of the model to account adequately for the pattern of label redistribution 7

in a particular section of the network. 8

Finally, in silico analysis of the chosen network revealed that the reliability of the flux 9

estimates improved by combining data from labeling studies using [1-13C]-, [2-13C] and [U-10 13C6]glucose – this was particularly noticeable for the oxidative steps of the PPP in the 11

cytosol and the plastids - and it also emphasized the importance of using well defined 12

labeling measurements (Fig. 10). 13

14

DISCUSSION 15

Choice of experimental system and application of steady-state MFA 16

A plant cell suspension was chosen for the investigation to simplify the task of testing 17

different models of the subcellular compartmentation of the PPP. The Arabidopsis cell 18

suspension is experimentally tractable, it has been used successfully for MFA (Williams et 19

al., 2008), and it avoids the difficulties that might arise from multiple cell types in 20

differentiated tissues. Moreover many of the general metabolic features of the chosen 21

Arabidopsis culture are similar to the heterotrophic metabolism that occurs in roots (Lehmann 22

et al., 2009). 23

The steady-state MFA protocol used here was broadly similar to existing best practice, 24

but the application of this tool in plant biology is still developing and several new features 25

were introduced with the aim of further improving the precision and reliability of the flux 26

estimates. First, half-reactions were used to represent the underlying carbon transitions that 27

occur during the steps catalysed by TA and TK. This approach, which provides a more 28

accurate description of the redistribution of the label by the PPP, has not been used 29

previously in steady-state MFA of plants and has only rarely been used in microbial work. 30

Secondly, Monte Carlo simulations with bootstrap sampling of the isotopomer abundances 31

provided an effective method for exploring the flux space that was consistent with the 32

observed redistribution of label. Stochastic sampling of the starting fluxes is routine in the 33

modeling process (Schwender et al., 2006), and combining this with statistical sampling of 34



14

the data points improves the fitting procedure (Kelly et al., 1990; Namba, 2004). 1

Surprisingly, the availability of a subroutine to exploit this method in the commonly used 2

13C-FLUX software appears to have been overlooked. Thirdly, the output from the Monte 3

Carlo simulations was subjected to principal components analysis to identify the set of best fit 4

feasible solutions. This novel approach provided an objective method for analyzing the 5

outputs of the modeling process and an effective route to the global best fit. Finally, 6

redistribution of radioactivity from [U-14C]substrate was used as a convenient method for 7

ensuring that all major metabolic products had been identified and as a sensitive method for 8

determining the output fluxes. 9

10

Quantifying the compartmented fluxes of central metabolism 11

The subcellular compartmentation of metabolism in eukaryotes complicates steady-state 12

MFA (Roscher et al., 2000), and the problem is acute in plants because of the more extensive 13

duplication of steps and pathways in the metabolic network (Kruger and Ratcliffe, 2008). 14

Capturing the characteristic compartmentation of the pathways of carbohydrate oxidation in a 15

realistic model of heterotrophic plant metabolism exemplifies the problem, and it would be 16

particularly useful if this could be done by comparing models for networks compartmented to 17

different extents, since this would allow an in vivo assessment of the functional 18

compartmentation of the network. Since most steady-state MFA models are based on over-19

determined datasets, reflecting the wealth of stable isotope labeling information that can be 20

obtained from nuclear magnetic resonance (NMR) or mass spectrometry (MS) (Ratcliffe and 21

Shachar-Hill, 2006), it is easy to add extra steps to the model to allow it to match biochemical 22

reality more closely. However the relationship between fluxes and measurements is not 23

always obvious, and even if the extra fluxes are potentially resolvable they may not be 24

determinable with any great confidence. 25

In practice many steady-state MFA investigations of the PPP in plants use models in 26

which the compartmentation of the pathway is pre-defined. For example a single plastidic 27

pathway has been assumed in maize root tips (Dieuaide-Noubhani et al., 1995; Alonso et al., 28

2007b), tomato cell cultures (Rontein et al., 2002) and Arabidopsis cell cultures (Williams et 29

al., 2008); whereas complete duplication of the pathway in the cytosol and plastid has been 30

assumed in the analysis of cultured soybean embryos (Sriram et al., 2004; Iyer et al., 2008). 31

In some tissues, for example oilseed rape embryos (Schwender et al., 2003) and soybean 32

embryos (Allen et al., 2009), the modeling task is made easier because the observed labeling 33

patterns imply that intermediates in glycolysis and the PPP are in fast exchange between the 34



15

cytosol and the plastids, ensuring that duplicate pathways in different physical locations 1

function indistinguishably from a single uncompartmented pathway. However in most of the 2

tissues that have been examined so far, including soybean embryos in some instances (Sriram 3

et al., 2004), the hexose phosphate pools in the cytosol and the plastids are found to be 4

labeled to different extents. Typically this evidence might be based on differences in the 5

labeling patterns of the glucosyl moieties from sucrose and starch (Supplemental Fig. S2), 6

and it is then necessary to consider the subcellular compartmentation of glycolysis and the 7

oxidative PPP explicitly. 8

Inspection of the labeling patterns in products derived specifically from either the 9

cytosolic or plastidic hexose phosphate pools may provide circumstantial evidence for the 10

operation of the full PPP in both compartments (Krook et al., 1999), justifying the 11

construction of a model with duplicate pathways. However, intuitive interpretation of the data 12

is complicated by the multiple metabolic processes that contribute to the labeling of these 13

metabolites (Kruger et al., 2003), and a more robust quantitative approach is needed to 14

compare the results of fitting all the labeling data to models in which the pathways are 15

duplicated to different extents. Here, steady-state MFA was used to compare three different 16

models of the oxidative PPP (Fig. 1) and the analysis indicates that several aspects of the 17

protocol contribute to its effective implementation. First, it is necessary to work with over-18

determined models so that the statistical significance of the flux solutions can be properly 19

evaluated. Secondly, for the PPP, it is important to define the carbon transition network in 20

terms of the half-reactions for TA and TK to maximize the chance of being able to 21

discriminate between parallel events in the cytosol and plastid. Thirdly, it is essential to use 22

high quality data from experiments with several differently labeled substrates in order to 23

deduce fluxes with sufficient confidence to allow a comparison of the models (Fig. 10). 24

Finally, it is necessary to establish a procedure (Figs 4, 5) that leads to an optimal best fit 25

based on clearly defined criteria so that the results for different models can be compared. 26

Analysis of three models for the oxidative PPP (Fig. 1) emphasized the difficulty of 27

determining the flux distribution with sufficient confidence to be able to discriminate 28

between the models. There was little to distinguish a plastid-only model (Fig. 1A) from the 29

model in which the oxidative steps were duplicated in the cytosol (Fig. 1B), even though the 30

latter is more plausible on the basis of molecular evidence. The increase in the degrees of 31

freedom in the second model (Fig. 1B) produced only a marginal improvement in the 32

residuum obtained for the global best fit, and the main observations that support the model 33

are the non-random allocation of flux between the oxidative steps in the two compartments 34



16

(Supplemental Fig. S3), and the reliability of these flux estimates, as reflected in their 1

confidence intervals (Table II). Adding further degrees of freedom by duplicating the 2

complete oxidative PPP in the two compartments (Fig. 1C) led to a further, modest, 3

improvement in the fit, and a radically different flux distribution through the pathways of 4

carbohydrate oxidation (Table II), but it also reduced the confidence in the flux values, with 5

most of the fluxes showing an increase in standard deviation. Thus this model provides a less 6

reliable flux map and can be rejected on the basis that the fit to the available data is less 7

convincing. Such a conclusion is always subject to the proviso that data might become 8

available that would put further constraints on the assessment of the compartmentation of the 9

pathway (Kruger and Ratcliffe, 2009), but currently the available data imply only a plastidic 10

location for the non-oxidative steps of the PPP in Arabidopsis cells. 11

In silico analysis of the sensitivity of cumomer abundance to changes in flux through the 12

irreversible oxidative section of the PPP can identify the isotopomer measurements that are 13

best able to define the fluxes through this step in the cytosol and plastids (Supplementary File 14

4). Although the sensitivities are influenced by the label distribution in the supplied substrate 15

and by the flux distribution through the network, estimates based on the biochemically 16

preferred flux map (Fig. 7B) indicate that isotopomers of cytosolic and plastidic hexose 17

phosphates and compounds derived directly from these pools are amongst the most diagnostic 18

measures of these PPP fluxes. This discrimination arises in part because, in the absence of 19

plastidic fructose-1,6-bisphosphatase, cytosolic glucose 6-P is the only substrate replenishing 20

the plastidic glucose 6-P pool, and any difference in labelling must therefore stem from 21

plastidic PPP activity. However intermediates within the reversible non-oxidative section of 22

the PPP, and their derivatives such as histidine and aromatic amino acids, are also relatively 23

sensitive to perturbations in PPP flux. Thus, measurements of label distribution in these 24

compounds would be expected to further improve the flux estimates and could lead to greater 25

discrimination between the different models. 26

A similar, but less comprehensive, analysis of the compartmentation of metabolism in 27

sunflower embryos found that models of the PPP that included cytosolic steps were no 28

improvement over a model in which the PPP was confined to the plastids (Alonso et al., 29

2007a). However the flux distributions in the alternative models were not discussed, and it is 30

perhaps surprising that the additional degrees of freedom in the models with cytosolic steps 31

led to no improvement in the fit. Irrespective of these points, since all the models gave 32

effectively the same fit to the data then it is unclear on what basis one of them was preferred. 33



17

The difficulties encountered in analyzing the compartmentation of the oxidative PPP by 1

steady-state MFA highlight some of the practical limitations of the approach. The objective is 2

to capture a realistic metabolic phenotype, but in practice the flux map is a representation of 3

the movement of the chosen label through the system. Resolving the contribution of 4

chemically distinct pathways between two metabolites in the same compartment, or of 5

duplicated steps and pathways in distinct compartments, is not always possible or even 6

attempted. For example the steps from citrate to 2-oxoglutarate are routinely assigned to the 7

mitochondrion in steady-state MFA even though there may be parallel fluxes in the cytosol. 8

Thus if a function of MFA is to guide metabolic engineering (Kruger and Ratcliffe, 2008; 9

Libourel and Shachar-Hill, 2008) then current flux maps are unlikely to be useful in 10

predicting the result of manipulating a compartment-specific enzyme activity. Similarly, even 11

if MFA provides evidence that a physically compartmented pathway is functionally 12

uncompartmented as a result of fast exchange of intermediates (Schwender et al., 2003; Allen 13

et al., 2009), the resulting metabolic phenotype still conceals the associated compartmental 14

demands for reducing/oxidizing power and energy. It can be concluded that it is essential to 15

push steady-state MFA to its limits to obtain the most informative model for the flux 16

distribution. 17

18

Metabolism in a heterotrophic Arabidopsis cell suspension 19

The network structure used to determine the biochemically preferred flux map (Fig. 7B) 20

differs from the one used in the previous analysis of Arabidopsis cell suspension cultures 21

(Williams et al., 2008) in two main ways. First, the models differ in their treatment of the 22

lower section of glycolysis. In steady-state MFA of plant metabolism there is usually 23

insufficient labeling information to support a model with separate cytosolic and plastidic 24

pools of both glyceraldehyde 3-phosphate and phosphoenolpyruvate. In the earlier model 25

(Williams et al., 2008) each metabolite was assumed to be in a common uncompartmented 26

pool, implying fast exchange across the plastid envelope; whereas here it was assumed that 27

the two metabolites were at isotopic equilibrium in a three-carbon phosphate ester pool (C3-28

P; Fig. 2) within each compartment. This assumption allowed the degree of 29

compartmentation of the pools to be tested by not arbitrarily assuming fast exchange between 30

the cytosol and plastid and it led to a marked alteration in the flux distribution, since all three 31

flux maps showed a significant net flux from the plastid to the cytosol (Fig. 7) at the level of 32

C3-P and only a limited exchange flux (Table II). Secondly, the biochemically preferred 33

model that emerged from the comparison of three different models for the oxidative PPP 34



18

(Fig. 1) allowed the oxidative steps to operate in both the cytosol and plastids. This 1

compartmentation is supported by estimates of transcript abundance based on publically 2

available microarray data for genes encoding isozymes of the PPP (Supplemental Fig. S5A) 3

and associated plastid envelope translocators (Supplemental Fig. S5B). 4

A striking feature of the flux map generated from the biochemically preferred model is 5

that the oxidative steps of the PPP are almost completely confined to the cytosol. Since most 6

biosynthetic processes requiring NADPH are plastidic, this flux distribution implies extensive 7

exchange of reducing power equivalents across the plastid envelope. However, this is 8

unlikely to be problematic since previous studies on maize root tips suggest that the oxidative 9

PPP in the cytosol and plastids can co-operate in providing NADPH for biosynthesis 10

throughout the cell (Averill et al., 1998). Nevertheless, the dominance of the cytosolic 11

pathway in catalyzing flux through the oxidative section of the PPP is unexpected, and it is 12

unclear whether this is a general characteristic of the organization of carbohydrate 13

metabolism in higher plants since compartmentation of the PPP is not included in most flux 14

maps. The only other plant tissues for which information is available are rosy periwinkle 15

hairy root cultures and developing soybean embryos, and in both of these only 25-50% of 16

flux through the oxidative section of the PPP is cytosolic (Sriram et al., 2004, 2007; Iyer et 17

al., 2008). However, in these systems total flux through the oxidative PPP is considerably 18

higher than that observed in Arabidopsis cells, and when expressed as a proportion of the rate 19

of sugar consumption the cytosolic fluxes in all three systems are similar (rosy periwinkle, 20

42.6%; soybean, 47.4%; Arabidopsis, 37.2%) implying that variation in compartmentation of 21

PPP flux is due to fluctuations in plastidic activities rather than to a decrease in the 22

significance of the cytosolic flux. Moreover, irrespective of the exact proportion of the 23

oxidative PPP flux occurring in the cytosol, two recent studies demonstrate the importance of 24

the cytosolic activity. First, disrupting the expression of the two genes encoding cytosolic 25

isozymes of glucose 6-phosphate dehydrogenase (G6PD5 and G6PD6) in Arabidopsis 26

through insertional inactivation alters seed lipid content (Wakao et al., 2008). Secondly, 27

supplementation or replacement of cytosolic glucose 6-phosphate dehydrogenase with an 28

alternative isoform improves resistance of tobacco leaves to Phytophthora infestans, and this 29

is attributed to changes in the provision of NADPH and stimulation of the hypersentitive 30

defense response (Scharte et al., 2009). Both observations highlight a requirement for the 31

oxidative section of the PPP in the cytosol to support normal metabolic activity. 32

33

Energy and redox balance in plant metabolic networks 34



19

The metabolic activity of the Arabidopsis cell culture appears to be broadly representative 1

of a wide range of heterotrophic plant systems and one measure of this can be found in its 2

efficiency in converting sugar to biomass components. The proportion of the [U-14C]glucose 3

metabolised to macromolecules and soluble compounds (Supplemental Table SI) indicates a 4

carbon use efficiency of 60%, in good agreement with the 65% calculated from a comparison 5

of substrate uptake and estimated CO2 output in the previous study of Arabidopsis cells 6

grown under equivalent conditions (Williams et al., 2008). These estimates of carbon-use 7

efficiency are within the range obtained for other heterotrophic plant systems, including a 8

rosy periwinkle hairy root culture (24%; Sriram et al., 2007), maize root tips (42-47%; 9

Alonso et al., 2007b), developing sunflower embryos (50%; Alonso et al., 2007a) and a 10

tomato cell culture (52-68%; Rontein et al., 2002). Although considerably higher values have 11

been reported for developing embryos of oilseed rape (85-95%) and soybean (82-83%), in 12

these systems there is appreciable reassimilation of respired CO2 through Rubisco, and light 13

makes significant contributions to the provision of ATP and/or reductant required for 14

biosynthesis (Allen et al., 2009; Goffman et al., 2005). In contrast, the energy and reducing 15

power required for biosynthesis to support growth in heterotrophic Arabidopsis cells must be 16

generated exclusively by oxidation of glucose, the sole respiratory substrate, through the 17

central pathways of carbon metabolism. 18

Comparison of the oxidation state of the substrates in the culture medium and the 19

products generated by the Arabidopsis cells confirms that the metabolic network is able to 20

operate without an input of reducing power. The primary substrates for cell growth were 21

glucose, nitrate plus ammonium, and sulphate in which the average oxidation states of C, N 22

and S are 0, +1 and +6, respectively. The analysis in Supplemental File S5 shows average C 23

oxidation states of 0 for carbohydrates, -1.56 for lipid, +0.92 for organic acids, and -1.02 for 24

protein and soluble amino acids in which N and S have oxidation states of -3 and -2, 25

respectively. The corresponding values for C in CO2 and ethanol, the principal respiratory 26

end-products, are +4 and -2, respectively. It follows from the relative partitioning of substrate 27

during growth (Table 1) that the average change in oxidation state in metabolic end-products 28

is +1.33 per C (Supplemental File S5). The analysis also shows that assimilation of nitrate 29

used in amino acid production represents the greatest demand for reducing power in the 30

Arabidopsis cell culture under the growth conditions used in this study. 31

However this simple redox analysis ignores the coenzyme specificity of different 32

processes, and does not consider the preferential involvement of NADPH in many reductive 33

biosynthetic steps. This limitation can be overcome by using flux maps to evaluate the 34



20

coenzyme balance in the network. In contrast to the situation in developing soybean embryos 1

(Iyer et al., 2008), inspection of the fluxes determined for the Arabidopsis cells establishes 2

that the oxidative PPP is insufficient to meet the NADPH requirements for biosynthesis 3

predicted by the network (Table III). This observation is the result of two assumptions: (i) the 4

assimilation of equal amounts of nitrate and ammonium (Table III); and (ii) the NADPH-5

dependence of glutamate synthesis (Table S6). In this situation reductive biosynthesis will 6

require other NADP+-dependent processes, and in the network considered here the demand 7

could be met by the operation of NADP+-isocitrate dehydrogenase (Table III). 8

In contrast, the production of NADH is more than enough for its utilization in reductive 9

processes, including the demands for ATP synthesis. Indeed, if it is assumed that excess 10

NADH (and FADH2) are re-oxidised through the mitochondrial electron transport chain to 11

produce ATP through oxidative phosphorylation (at stoichiometries of 2.3–2.5 and 1.4–1.5 12

ATP per NADH and FADH2, respectively; Hinkle, 2005), then the maximum yield of ATP 13

generated by the metabolic network is over seven times that required to support the flux to 14

biosynthetic products. The actual yield of ATP will depend on the extent of engagement of 15

the alternative oxidase and the activity of mitochondrial uncoupling proteins, but the 16

comparison suggests the metabolic network has a considerable over-capacity for ATP 17

production. Of course, these calculations ignore the energy requirements of other metabolic 18

processes such as substrate cycling, turnover of proteins and other macromolecules, as well 19

as synthesis and turnover of RNA that are not included in the present network. Moreover, 20

ATP is used in other cellular processes such as substrate uptake, transport of metabolites 21

across intracellular membranes, and maintenance of trans-membrane ion and electrical 22

gradients. Such activities can account for over half of the cellular ATP consumption in 23

microorganisms (Nielsen et al., 2003), and the findings presented here suggest that an even 24

greater proportion of the potential chemical energy released during metabolism in actively 25

growing plant cells could be needed to meet these requirements. Recently, analysis of energy 26

demands in a genome-scale stoichiometric model of Arabidopsis has led to a similar 27

conclusion that only a relatively small proportion of the potential ATP generated by the cell is used 28

for biosynthesis (Poolman et al., 2009). 29

This conclusion has implications for determination of flux through plant metabolic 30

networks by flux balance analysis (FBA). This approach depends on the selection of an 31

appropriate objective function to obtain an optimum solution within the potential flux space 32

defined by the stoichiometry of the biochemical reaction network, plus other constraints, such 33

as the metabolic outputs of the system (Kauffman et al., 2003). Typically, the optimisation 34



21

criterion in microorganisms is maximisation of yield of biomass, ATP or a product of 1

interest, or alternatively maximisation of flux, i.e. maximisation of the rate of production of 2

one or more components required for growth (Schuster et al., 2008). A recent study on 3

developing endosperm of barley used a refinement of this approach in which the objective 4

function was maximisation of growth per unit flux – this is a combination of maximization of 5

biomass yield, and minimisation of overall flux (Grafahrend-Belau et al., 2009). The logic 6

behind this optimisation criterion is that cellular metabolism has evolved to fulfil its functions 7

with the most economic use of available resources (Holzhütter, 2004), but application of this 8

criterion requires that the major energy-consuming processes have been adequately 9

incorporated into the metabolic model. The demonstration that a large proportion of the 10

potential biochemical energy released through respiration in heterotrophic plant cells is 11

dissipated, or used by processes other than those directly associated with the synthesis of 12

macromolecules and soluble components that accumulate during growth, means that ATP 13

expenditure for such purposes must be accurately assessed if FBA is to generate meaningful 14

flux maps of the network of central carbon metabolism in plants. 15

16

CONCLUSION 17

Steady-state MFA of multiple labeling experiments in an Arabidopsis cell suspension 18

shows that the experimental observations can be explained by any one of three different 19

models of the subcellular compartmentation of the oxidative PPP. Other biochemical 20

evidence favors the model in which the oxidative steps are duplicated in the cytosol and 21

plastids, and this leads to the conclusion that the bulk of the flux through the oxidative steps 22

is cytosolic. This illustrates the point that interesting features of the metabolic phenotype can 23

be missed with an inappropriate model and emphasizes the importance of basing the model 24

on all the available biochemical evidence. 25

26

MATERIALS AND METHODS 27

Cell culture 28

Cell suspensions of A. thaliana (ecotype Landsberg erecta) were maintained on an orbital 29

shaker at 150 rpm in 250 ml Erlenmeyer flasks sealed with a double layer of aluminum foil 30

under a 16 h light, 8 h dark cycle at 22°C (Williams et al., 2008). Every 7 d, 15 ml cell 31

suspension was subcultured into 100 ml fresh Murashige and Skoog (MS) medium 32

supplemented with 3% (w/v) glucose, 0.5 mg/L 1-naphthaleneacetic acid, and 0.05 mg/L 33



22

kinetin. Heterotrophic cell suspensions were produced by subculturing 15 ml of a 7 d old 1

light-grown cell suspension into 100 ml fresh medium and incubating in the dark at 22°C. 2

3

Chemicals 4

[1-14C]glucose (specific activity 2.00 GBq mmol-1), [6-14C]glucose (specific activity 2.07 5

GBq mmol-1), and [U-14C]glucose (specific activity 10.4 GBq mmol-1) were purchased from 6

GE Healthcare Life Sciences (http://www1.gelifesciences.com/). [3,4-14C]glucose (specific 7

activity 1.11 GBq mmol-1) was purchased from American Radiolabeled Chemicals Inc. 8

(http://www.arc-inc.com/) and [2-14C]glucose (specific activity 1.67 GBq mmol–1) was 9

obtained from PerkinElmer NEN (http://las.perkinelmer.com/). [2-13C]glucose (99 atom %) 10

was purchased from Omicron Biochemicals Inc. (http://www.omicronbio.com/), and [1-11 13C]glucose (99 atom %) and [U-13C6]glucose (99 atom %) were obtained from Isotec Inc. 12

(http://www.sigmaaldrich.com/chemistry/stable-isotopes-isotec.html). All enzymes were 13

from Roche Diagnostics Ltd. (http://www.roche.com/). General chemicals and 14

chromatography resins were purchased from Sigma-Aldrich (http://www.sigmaaldrich.com/) 15

or Merck (http://www.merckbiosciences.co.uk/). 16

17

Incubation of Arabidopsis cell culture with 13C-labeled glucose 18

Cells were labeled by subculturing light-grown cells into media in which the unlabeled 19

glucose was replaced with either [1-13C]glucose, [2-13C]glucose or a mixture of 10% [U-20 13C6]glucose and 90% unlabeled glucose. Cells were grown in the dark for 7 d at 22°C, then 21

harvested by vacuum filtration, rinsed with 80 ml of MS medium, frozen by immersion in 22

liquid nitrogen and stored at -80°C. This procedure has been shown to have no discernable 23

effect on the flux distribution in the core metabolic network of Arabidopsis (Kruger et al., 24

2007a) and to lead to the establishment of an isotopic and metabolic state in the cell culture 25

(Williams et al., 2008). 26

27

Extraction procedures after stable isotope labeling 28

Soluble metabolites were extracted from frozen cells (approximately 8 g fresh weight) 29

using perchloric acid (Kruger et al., 2007a, 2008). Subsequently NMR spectra were recorded 30

from samples dissolved in 3.2 ml 2H2O containing 10 mM EDTA, 25 mM 1,4-dioxane, and 31

10 mM KH2PO4/K2HPO4 pH 7.5. Starch and protein were extracted from the insoluble 32

residue using procedures that were similar to those described elsewhere (Williams et al., 33

2008). Starch was gelatinized by autoclaving a washed aliquot of the residue in 25 mM 34



23

sodium acetate for 3 h, followed by enzymic digestion with α-amylase and amyloglucosidase. 1

The supernatant, containing glucose, was freeze-dried and redissolved in 3.2 ml 2H2O with 25 2

mM 1,4-dioxane for NMR analysis. Protein in the perchloric acid extraction pellet was 3

hydrolyzed directly by adding 2 ml 6 M HCl and heating under vacuum at 110°C for 24 h in 4

a 5 ml Pierce hydrolysis tube (Pierce, http://www.piercenet.com/). After digestion, the protein 5

hydrolyzate was centrifuged and the supernatant was freeze-dried. The sample was 6

redissolved in water and freeze-dried again after adjusting the pH to 7.5. Finally the sample 7

was redissolved in 3.2 ml 2H2O containing 25 mM 1,4-dioxane for NMR analysis. 8

9

Quantification of biosynthetic outputs 10

Fluxes to biosynthetic products, and the rate of glucose consumption, were obtained from 11

the redistribution of label following metabolism of [U-14C]glucose by cell suspensions. 12

13

Incubation of Arabidopsis cell culture with [U-14C]glucose 14

Cells were grown with unlabeled glucose under the conditions used for the stable isotope 15

labeling experiments. After 6 d, 5 ml aliquots of the cell suspension were transferred to 16

sealed 100 ml conical flasks and supplemented with 185 kBq [U-14C]glucose (specific 17

activity 10.4 TBq mol-1). After 12 or 24 h the cells were harvested by filtration, washed with 18

two volumes of MS medium, and then the cells and medium were frozen in liquid nitrogen. 19

The production of 14CO2 was determined in replicate flasks in which the incubation was 20

terminated at the appropriate time by injection of 2 ml 12 M formic acid to stop metabolism 21

and release dissolved CO2 from the medium (Harrison and Kruger, 2008). Released 14CO2 22

was collected in 1.0 ml 10% (w/v) KOH in a vial suspended within each sealed culture flask. 23

24

Extraction of Arabidopsis cells after 14C-labeling 25

Frozen cells were combined with 1 ml 100% methanol at -20°C and frozen in liquid 26

nitrogen. The sample was thawed on ice for 2 h and then centrifuged at 8,500 g for 10 min at 27

5°C. The supernatant was removed and stored on ice. The pellet was resuspended in 1 ml 28

cold methanol and after storage on ice for 5 min, the suspension was centrifuged and the 29

supernatant combined with the supernatant from the initial extraction. The insoluble cell 30

residue was then extracted twice with 1 ml 100% methanol for 5 min at 70°C. Samples were 31

centrifuged to separate methanol-soluble material from the insoluble residue, and the 32

supernatants were combined with that from the cold extractions. Finally, the insoluble residue 33



24

was washed with two further 1 ml aliquots of 100% methanol and the insoluble material 1

stored at -80°C. 2

3

Ethanol extraction from the incubation medium after 14C-labeling 4

[14C]ethanol was recovered from the cell culture medium by distillation after adding 2 ml 5

100% ethanol to the sample to aid recovery. 6

7

Fractionation of methanolic cell extracts 8

Distilled water and chloroform were added to the methanolic fraction to generate a 9

CHCl3:CH3OH:H2O mixture (4:4:2, by volume). The mixture was centrifuged at 3,500 g for 10

10 min at 20°C to promote phase separation. The aqueous methanol phase was retained and 11

the chloroform phase was rinsed with a further two washes of 2 ml H2O plus 4 ml methanol 12

before storage at -20°C. The aqueous methanol fractions from each wash were pooled with 13

the initial aqueous methanol fraction, and this combined aqueous methanol fraction was then 14

dried by rotary evaporation at 25°C, and resuspended in 5 ml H2O. Total radioactivity in this 15

fraction was measured before further separation into basic, acidic and neutral fractions by 16

solid-phase extraction on Dowex ion-exchange columns (Kruger et al., 2007a). The aqueous 17

basic fraction, consisting of amino acids, was freeze-dried, resuspended in 1 ml H2O and 18

fractionated using an anion exchange column to separate acidic amino acids from 19

neutral/basic amino acids (Cossins and Beevers, 1963). Each neutral fraction was freeze-20

dried, resuspended in 1 ml H2O and separated into sucrose, glucose and fructose using thin 21

layer chromatography (TLC) (Scott and Kruger, 1995). 22

The chloroform (lipid) fraction was separated into its components by reverse phase 23

chromatography on an aminopropyl column (LC-NH2 SPE, Supelco, Poole, Dorset, UK) as 24

described by Kim and Salem (1990). 25

Total radioactivity in the methanol-insoluble material was determined by incubating a 26

100 μl aliquot of insoluble residue overnight with 400 μl Solvabletm tissue solubiliser (Perkin 27

Elmer, http://las.perkinelmer.com) followed by liquid scintillation counting. Insoluble 28

material was further analyzed by autoclaving, digestion with amylase, amyloglucosidase and 29

pronase, and separation into fractions by ion-exchange chromatography (Malone et al., 30

2006). The resulting aqueous, acidic and basic fractions represented starch, cell wall material 31

and protein respectively. Radioactivity in the remaining undigested residue was determined 32

by washing the residue to remove the supernatant including digested material and enzymes, 33

followed by overnight incubation with tissue solubiliser (Kruger et al., 2007a). 34



25

1

Monitoring release of 14CO2 by Arabidopsis cell culture metabolising positionally-2

labeled [14C]glucose 3

A 5-day-old Arabidopsis cell suspension was incubated in the dark at 22°C in a 100 ml 4

conical flask in a total volume of 5 ml culture medium supplemented with 3.7 kBq of either 5

[1-14C]-, [2-14C]-, [3,4-14C]- or [6-14C]-glucose. Each flask was then sealed with a rubber 6

bung and aerated on an orbital shaker at 100 rpm. Released 14CO2 was collected in 0.5 ml of 7

10% (w/v) KOH in a vial suspended in the flask. The KOH solution was replaced at 2-h 8

intervals for 12 h and again 24, 36 and 48 h after the beginning of the incubation. Ratios of 9 14CO2 release from positionally-labeled glucose were determined as described by Harrison 10

and Kruger (2008). 11

12

Determination of radioactivity 13

Radioactivity was determined by liquid scintillation counting in a Beckman LS 6500 14

scintillation counter. An appropriate aliquot of each fraction was mixed with 4 volumes of 15

Optiphase HiSafe scintillation fluid. Counting efficiency was typically >90%. 16

17

NMR spectroscopy 18

One-dimensional 1H-decoupled 13C-NMR spectra were recorded at 20°C and 150.9 MHz 19

on a Varian Unity Inova 600 spectrometer equipped with a 10 mm diameter broadband probe. 20

Waltz 16 1H decoupling was applied during the acquisition time, and spectra were acquired 21

with either a 6 s relaxation delay and low power frequency modulated decoupling to generate 22

the nuclear Overhauser effect (NOE), or a 29 s relaxation delay without the NOE. Large 23

numbers of scans were generally required to generate spectra of sufficient quality for 24

quantitative analysis of the peak intensities and spectra were accumulated in blocks to check 25

for unwanted spectral changes. For example a total acquisition time of 30 h was used for the 26

6 s spectra of the soluble metabolites, and over 40 h was used for the 29 s spectra of protein 27

hydrolyzates. Spectra of starch digests were obtained much more quickly, reflecting the high 28

glucose concentration in the samples. Spectra were processed using NUTS software (Acorn 29

NMR Inc., Livermore, USA) and in some instances relative peak intensities in 6 s spectra 30

were corrected using relative intensities in the fully relaxed spectra. Spectra were processed 31

after a Lorentzian-Gaussian transformation, with a line broadening of -1 Hz and a Gaussian 32

factor of 0.1, and signal integration was performed after base-line correction. Chemical shifts 33

were measured relative to the 1,4-dioxane signal at 67.30 ppm, and peaks were assigned on 34



26

the basis of literature values and authentic standards. Positional isotopomers were specified 1

using 0 to denote 12C, 1 to denote 13C, and X to denote either 12C or 13C. 2

3

Metabolic modeling 4

Metabolic modeling was performed using 13C-FLUX (version 20050329; Wiechert et 5

al., 2001) according to the protocol outlined in Figures 4 and 5. After defining the metabolic 6

network, the free fluxes were fitted to the biomass measurements and labeling data using the 7

Donlp2 (P. Spelluci, Technische Universität Darmstadt, Germany) routine in 13C-FLUX. 8

This sequential quadratic programming algorithm is superior to the CooolEvoAlpha 9

evolutionary algorithm in 13C-FLUX, both in terms of computational efficiency and 10

avoidance of local minima (Wiechert et al., 2001). The optimiser was used in conjunction 11

with Monte Carlo simulations and bootstrap sampling of the isotopomer abundances based on 12

the measured values and their deviations to generate solutions for the flux map. The bootstrap 13

Monte Carlo sampling method was invoked by the command: Donlp2 “model file name”.ftbl 14

-a 0.00001 –m “number of simulations” -bootstrap -logAllF > “output file name”.txt&watch 15

“grep Resi “output file name”.txt|tail -20”. This strategy of combining Donlp2 with bootstrap 16

Monte Carlo sampling was first tested on the network model distributed with 13C-FLUX, and 17

on a published model for developing sunflower embryos (Alonso et al., 2007a), to confirm 18

that it was capable of generating the expected fluxes. Each simulation generated a set of free 19

fluxes after minimizing the sum of squared standardized differences - the residuum - between 20

the experimental measurements and their predicted values. Principal components analysis of 21

the free fluxes obtained from 1000 simulations was performed using SIMCA-P 11.5 22

(Umetrics AB, Umeå, Sweden) to identify the set of best fit feasible solutions, with low 23

values of the residuum. This set of outputs was used to generate a mean flux solution that was 24

then used as a starting point for a further optimization run with Donlp2. EstimateStat, the 25

statistical analysis component of 13C-FLUX, was used to compare the merits of different 26

modeling schemes and to assign errors to the flux values deduced by the numerical analysis. 27

28

Supplemental Data 29

The following material is available in the online version of this article. 30

Supplemental File S1. 13C-FLUX model of the central metabolic network in 31

Arabidopsis thaliana cell suspension culture (MS Excel file). 32



27

Supplemental File S2. Positional isotopomer data sets obtained following metabolism of 1

[1-13C]glucose, [2-13C]glucose and 10% [U-13C6]glucose by cell suspension cultures 2

of Arabidopsis thaliana (MS Excel file). 3

Supplemental File S3. 13C-FLUX metabolic models of carbohydrate oxidation 4

containing different formulations of the oxidative pentose phosphate pathway (MS 5

Excel file). 6

Supplemental File S4. Sensitivity analysis of oxidative pentose phosphate pathway flux 7

in the biochemically preferred flux map of central metabolism in Arabidopsis thaliana 8

cell suspension culture (MS Excel file). 9

Supplemental File S5. Redox analysis of the biosynthetic outputs of heterotrophic 10

Arabidopsis cells (MS Excel file). 11

Supplemental Tables S1-4. Determination of metabolic network output fluxes. 12

Supplemental Table S5. Metabolic fluxes in heterotrophic Arabidopsis cells. 13

Supplemental Table S6. Coenzyme stoichiometries for generation of biosynthetic 14

products from intermediates of the central metabolic network. 15

Supplemental Figure S1. Metabolism of [U-14C]glucose by heterotrophic Arabidopsis 16

cells. 17

Supplemental Figure S2. Comparison of relative fractional abundance of isotopomers in 18

carbohydrates derived from cytosolic and plastidic hexose phosphate pools. 19

Supplemental Figure S3. Distribution of flux estimates for selected steps in the 20

metabolic network from Monte Carlo simulations. 21

Supplemental Figure S4. Assessment of robustness of global best-fit flux solution. 22

Supplemental Figure S5. Relative expression levels of genes encoding isozymes of the 23

pentose phosphate pathway and associated plastid envelope translocators in 24

Arabidopsis cells. 25

26

ACKNOWLEDGEMENTS 27

We thank W. Wiechert (Lehrstuhl für Simulationstechnik, Universität Siegen, Germany) 28

for permission to use 13C-FLUX and T.C.R. Williams (Department of Plant Sciences, 29

University of Oxford) for helpful advice. 30

31

32



28

LITERATURE CITED 1

2

Allen DK, Shachar-Hill Y, Ohlrogge JB (2007) Compartment-specific labeling information 3

in 13C metabolic flux analysis of plants. Phytochemistry 68: 2197-2210 4

5

Allen DK, Ohlrogge JB, Shachar-Hill Y (2009) The role of light in soybean seed filling 6

metabolism. Plant J 58: 220-234 7

8

Alonso AP, Goffman FD, Ohlrogge JB, Shachar-Hill Y (2007a) Carbon conversion 9

efficiency and central metabolic fluxes in developing sunflower (Helianthus annuus L.) 10

embryos. Plant J 52: 296-308 11

12

Alonso AP, Raymond P, Rolin D, Dieuaide-Noubhani M (2007b) Substrate cycles in the 13

central metabolism of maize root tips under hypoxia. Phytochemistry 68: 2222-2231 14

15

Alonso AP, Vigeolas H, Raymond P, Rolin D, Dieuaide-Noubhani M (2005) A new 16

substrate cycle in plants. Evidence for a high glucose-phosphate-to-glucose turnover from 17

in vivo steady-state and pulse-labeling experiments with [13C]glucose and [14C]glucose. 18

Plant Physiol 138: 2220-2232 19

20

Averill RH, Bailey-Serres J, Kruger NJ (1998) Co-operation between cytosolic and 21

plastidic oxidative pentose phosphate pathways revealed by 6-phosphogluconate-deficient 22

genotypes of maize. Plant J 14: 449-457 23

24

Carrari F, Urbanczyk-Wochniak E, Willmitzer L, Fernie AR (2003) Engineering central 25

metabolism in crop species: learning the system. Metab Eng 5: 191-200 26

27

Cossins E, Beevers H (1963) Ethanol metabolism in plant tissues. Plant Physiol 38: 375-380 28

29

Dieuaide-Noubhani M, Raffard G, Canioni P, Pradet A, Raymond P (1995) 30

Quantification of compartmented metabolic fluxes in maize root tips using isotope 31

distribution from 13C- or 14C-labeled glucose. J Biol Chem 270: 13147-13159 32

33

34



29

Goffman FD, Alonso AP, Schwender J, Shachar-Hill Y, Ohlrogge JB (2005) Light 1

enables a very high efficiency of carbon storage in developing embryos of rapeseed. Plant 2

Physiol 138: 2269–2279 3

4

Grafahrend-Belau E, Schreiber F, Koschützki D, Junker BH (2009) Flux balance 5

analysis of barley seeds: a computational approach to study systemic properties of central 6

metabolism. Plant Physiol 149: 585–598 7

8

Harrison PW, Kruger NJ (2008) Validation of the design of feeding experiments involving 9

[14C]substrates used to monitor metabolic flux in higher plants. Phytochemistry 69: 2920-10

2927 11

12

Hinkle PC (2005) P/O ratios of mitochondrial oxidative phosphorylation. Biochim Biophys 13

Acta 1706: 1–11 14

15

Ho CL, Saito K (2001) Molecular biology of the plastidic phosphorylated serine biosynthetic 16

pathway in Arabidopsis thaliana. Amino acids 20: 243-259 17

18

Holzhütter HG (2004) The principle of flux minimization and its application to estimate 19

stationary fluxes in metabolic networks. Eur J Biochem 271: 2905–2922 20

21

Iyer VV, Sriram G, Fulton DB, Zhou R, Westgate ME, Shanks, JV (2008) Metabolic flux 22

maps comparing the effect of temperature on protein and oil biosynthesis in developing 23

soybean cotyledons. Plant Cell Environ 31: 506-517 24

25

Joshi M, Seidel-Morgenstern A, Kremling A (2006a) Exploiting the bootstrap method for 26

quantifying parameter confidence intervals in dynamical systems. Metab Eng 8: 447–455 27

28

Joshi V, Laubengayer KM, Schauer N, Fernie AR, Jander G (2006b) Two Arabidopsis 29

threonine aldolases are nonredundant and compete with threonine deaminase for a 30

common substrate pool. Plant Cell 18: 3564-3575 31

32

Junker BH, Lonien J, Heady LE, Rogers A, Schwender J (2007) Parallel determination of 33

enzyme activities and in vivo fluxes in Brassica napus embryos grown on organic or 34



30

inorganic nitrogen source. Phytochemistry 68: 2232-2242 1

2

Kauffman KJ, Prakash P, Edwards JS (2003) Advances in flux balance analysis. Curr 3

Opin Biotechnol 14:491–496 4

5

Kelly EB, McMillen DP, Buono MJ (1990) Constructing confidence intervals using the 6

bootstrap: an application to a multi-product cost function. Rev Econom Stat 72: 339-344 7

8

Kim H, Salem N (1993) Separation of lipid classes by solid phase extraction. J Lipid Res 12: 9

2285-2289 10

11

Kleijn RJ, van Winden WA, van Gulik WM, Heijnen JJ (2005) Revisiting the 13C-label 12

distribution of the non-oxidative branch of the pentose phosphate pathway based upon 13

kinetic and genetic evidence. FEBS J 272: 4970-4982 14

15

Krook J, Vreugdenhil D, Dijkema C, van der Plas LHW (1999) Sucrose and starch 16

metabolism in carrot (Daucus carota L.) cell suspensions analysed by 13C-labelling: 17

indications for a cytosol and a plastid-localized oxidative pentose phosphate pathway. J 18

Exp Bot 49: 1917-1924 19

20

Kruger NJ, Huddleston JE, Le Lay P, Brown ND, Ratcliffe RG (2007a) Network flux 21

analysis: impact of 13C-substrates on metabolism in Arabidopsis thaliana cell suspension 22

cultures. Phytochemistry 68: 2176-2188 23

24

Kruger NJ, Le Lay P, Ratcliffe RG (2007b) Vacuolar compartmentation complicates the 25

steady-state analysis of glucose metabolism and forces reappraisal of sucrose cycling in 26

plants. Phytochemistry 68: 2189-2196 27

28

Kruger NJ, Ratcliffe RG (2008) Metabolic organization in plants: a challenge for the 29

metabolic engineer. In HJ Bohnert, H Nguyen, NG Lewis, eds, Bioengineering and 30

Molecular Biology of Plant Pathways, Vol.1. Elsevier, Amsterdam, pp 1-27 31

32

Kruger NJ, Ratcliffe RG (2009) Insights into plant metabolic networks from steady-state 33

metabolic flux analysis. Biochimie 91: 697-702 34



31

1

Kruger NJ, Ratcliffe RG, Roscher A (2003) Quantitative approaches for analyzing fluxes 2

through plant metabolic networks using NMR and stable isotope labeling. Phytochem 3

Rev 2: 17-30 4

5

Kruger NJ, Troncoso-Ponce MA, Ratcliffe RG (2008) 1H-NMR metabolite fingerprinting 6

and metabolomic analysis of perchloric acid extracts from plant tissues. Nat Protocols 3: 7

1001-1012 8

9

Kruger NJ, von Schaewen A (2003) The oxidative pentose phosphate pathway: structure 10

and organisation. Curr Opin Plant Biol 6: 236-246 11

12

Lehmann M, Schwarzländer M, Obata T, Sirikantaramas S, Burow M, Olsen CE, 13

Tohge T, Fricker MD, Møller BL, Fernie AR, Sweetlove, LJ and Laxa M (2009) The 14

metabolic response of Arabidopsis roots to oxidative stress is distinct from that of 15

heterotrophic cells in culture and highlights a complex relationship between the levels of 16

transcripts, metabolites and flux. Mol Plant 2: 390-406 17

18

Libourel IGL and Shachar-Hill Y (2008) Metabolic flux analysis in plants: from intelligent 19

design to rational engineering. Annu Rev Plant Biol 59: 625-650 20

21

Lunn JE (2007) Compartmentation in plant metabolism. J Exp Bot 58: 35-47 22

23

Malone JG, Mittova V, Ratcliffe RG, Kruger NJ (2006) The response of carbohydrate 24

metabolism in potato tubers to low temperature. Plant Cell Physiol 47: 1309-1322 25

26

Miyashita Y, Dolferus R, Ismond KP, Good AG (2007) Alanine aminotransferase catalyses 27

the breakdown of alanine after hypoxia in Arabidopsis thaliana. Plant J 49: 92-102 28

29

Namba A (2004) Simulation studies on bootstrap empirical likelihood tests. Communications 30

in Statistics – Simulation and Computation 33: 99-108 31

32

Nielsen J, Villadsen J, Lidén G (2003) Bioreaction Engineering Principles, Ed 2. Springer, 33

New York, pp 124-142 34



32

1

Poolman MG, Miguet L, Sweetlove LJ, Fell DA (2009) A genome-scale metabolic model 2

of Arabidopsis thaliana and some of its properties. Plant Physiol 10.1104/pp.109.141267 3

4

Ratcliffe RG, Shachar-Hill Y (2005) Revealing metabolic phenotypes in plants: inputs from 5

NMR analysis. Biol Rev Camb Philos Soc 80: 27-43 6

7

Ratcliffe RG, Shachar-Hill Y (2006) Measuring multiple fluxes through plant metabolic 8

networks. Plant J 45: 490-511 9

10

Rontein D, Dieuaide-Noubhani M, Duforc EJ, Raymond P, Rolin D (2002) The metabolic 11

architecture of plant cells. J Biol Chem 277: 43948-43960 12

13

Roscher A, Kruger NJ, Ratcliffe RG (2000) Strategies for metabolic flux analysis in plants 14

using isotope labelling. J. Biotechnol 77: 81-102 15

16

Scharte J, Schön H, Tjaden Z, Weis E, von Schaewen A (2009) Isoenzyme replacement of 17

glucose-6-phosphate dehydrogenase in the cytosol improves stress tolerance in plants. 18

Proc Natl Acad Sci 106: 8061–8066 19

20

Schuster S, Pfeiffer T, Fell DA (2008) Is maximization of molar yield in metabolic 21

networks favoured by evolution? J Theor Biol 252: 497-504 22

23

Schwender J, Ohlrogge JB, Shachar-Hill Y (2003) A flux model of glycolysis and the 24

oxidative pentosephosphate pathway in developing Brassica napus embryos. J Biol Chem 25

278: 29442-29453 26

27

Schwender J, Shachar-Hill Y, Ohlrogge JB (2006) Mitochondrial metabolism in 28

developing embryos of Brassica napus. J Biol Chem 281: 34040-34047 29

30

Schwender J (2008) Metabolic flux analysis as a tool in metabolic engineering of plants. 31

Curr Opin Biotechnol 19: 131-137 32

33

Scott P, Kruger NJ (1995) Influence of elevated fructose-2,6-bisphosphate levels on starch 34



33

mobilization in transgenic tobacco leaves in the dark. Plant Physiol 108: 1569-1577 1

2

Selivanov VA, Meshalkina LE, Solovjeva ON, Kuchel PW, Ramos-Montoya A, 3

Kochetov GA, Lee PWN, Cascante M (2005) Rapid simulation and analysis of 4

isotopomer distributions using constraints based on enzyme mechanisms: an example 5

from HT29 cancer cells. Bioinformatics 21: 3558-3564 6

7

Sriram G, Fulton DB, Iyer VV, Peterson JM, Zhou R, Westgate ME, Spalding MH, 8

Shanks JV (2004) Quantification of compartmented metabolic fluxes in developing 9

soybean embryos by employing biosynthetically directed fractional 13C labeling, two 10

dimensional [13C, 1H] nuclear magnetic resonance, and comprehensive isotopomer 11

balancing. Plant Physiol 136: 3043-3057 12

13

Sriram G, Fulton DB, Shanks JV (2007) Flux quantification in central carbon metabolism 14

of Catharanthus roseus hairy roots by 13C labeling and comprehensive bondomer 15

balancing. Phytochemistry 68: 2243-2257 16

17

Sweetlove LJ, Fell D, Fernie AR (2008) Getting to grips with the plant metabolic network. 18

Biochem J 409: 27-41 19

20

Wakao S, Andre C, Benning C (2008) Functional analyses of cytosolic glucose-6-21

phosphate dehydrogenases and their contribution to seed oil accumulation in Arabidopsis. 22

Plant Physiol 146: 277-288 23

24

Wiback SJ, Famili I, Greenberg HJ, Palsson BØ (2004) Monte Carlo sampling can be 25

used to determine the size and shape of the steady-state flux space. J Theor Biol 228: 26

437-447 27

28

Wiechert W, Mollney M, Petersen S, de Graaf AA (2001) A universal framework for 13C 29

metabolic flux analysis. Metab Eng 3: 265-283 30

31

Williams TCR, Miguet L, Masakapalli SK, Kruger NJ, Sweetlove LJ, Ratcliffe RG 32

(2008) Metabolic network fluxes in heterotrophic Arabidopsis cells: stability of the flux 33

distribution under different oxygenation conditions. Plant Physiol 148: 704-718 34



34

FIGURE LEGENDS 1

2

Figure 1. Alternative metabolic models for the subcellular compartmentation of the PPP. A, 3

Oxidative and non-oxidative reactions of the PPP confined to the plastid; B, oxidative steps 4

of the PPP in both the cytosol and plastid, with the non-oxidative steps in the plastid; and C, 5

the complete pathway in both compartments. Net and exchange fluxes are shown with uni- 6

and bi-directional arrows respectively, and the direction of a unidirectional arrow corresponds 7

to a positive flux in the output of the model. Flux names are given in italics and the pentose 8

phosphates are grouped into a single indistinguishable pool in all three models. 9

Abbreviations: Ery-4-P, erythrose 4-phosphate; Fru-6-P, fructose 6-phosphate; Glu-6-P, 10

glucose 6-phosphate; Rib-5-P, ribose 5-phosphate; Rbu-5-P, ribulose 5-phosphate; Sed-7-P, 11

sedoheptulose 7-phosphate; TA-C3, the three-carbon substituted enzyme intermediate in the 12

reaction catalyzed by transaldolase; TK-C2, the two-carbon substituted enzyme intermediate 13

in the reaction catalyzed by transketolase; Xlu-5-P, xylulose 5-phosphate. 14

15

Figure 2. Metabolic model of central carbon metabolism with the oxidative steps of the 16

plastidic oxidative PPP duplicated in the cytosol. Output fluxes used to constrain the model 17

are shown with dotted black arrows; net and exchange fluxes deduced from the 13C-label 18

redistribution are shown with solid black uni- and bi-directional arrows, respectively. The 19

direction of a unidirectional arrow corresponds to a positive flux in the output of the model 20

and the dotted grey boxes indicate that the enclosed metabolites were considered as a single 21

pool. Standard abbreviations are used for the amino acids. Other abbreviations: C3-P, three-22

carbon phosphate ester pool; cit, citrate; CO2in, internal carbon dioxide; Fru-6-P, fructose 6-23

phosphate; fum, fumarate; Glc-6-P, glucose 6-phosphate; αKG, 2-oxoglutarate; OAA, 24

oxaloacetate. 25

26

Figure 3. Representation of the oxidative PPP for steady-state MFA. A, Conventional, and B, 27

alternative stoichiometries for the non-oxidative reactions of the PPP reflecting the substrate 28

specificity of transketolase (TK). C, Formulation of the non-oxidative reactions of the PPP in 29

terms of the two- and three-carbon substituted enzyme intermediates formed during the ping-30

pong mechanisms of TK and transaldolase (TA). D, Steady-state analysis of the fluxes 31

through the reactions catalyzed by TK for a simulated set of 13C-labeling data. Isotopomer 32

abundances were first deduced for a flux distribution corresponding to the conventional 33

network in A, and then these abundances were used to predict the flux distribution through a 34



35

metabolic network that permitted flux through both the conventional (A) and alternative (B) 1

pathways. The 100 solutions each correspond to a different proportion of the three exchange 2

reactions catalyzed by TK: (6) Xlu-5-P + Rib-5-P ↔ Tri-3-P + Sed-7-P; (○) Xlu-5-P + Ery-3

4-P ↔ Tri-3-P + Fru-6-P; (□) Sed-7-P + Ery-4-P ↔ Rib-5-P + Fru-6-P. E, The result of the 4

same exercise performed with the network (C) defined using half reactions for TK: (○) Xlu-5

5-P ↔ Tri-3-P + Enz-C2; (□) Fru-6-P ↔ Ery-4-P + Enz-C2; (6) Sed-7-P ↔ Rib-5-P + Enz-6

C2. Using half-reactions removes the ambiguity apparent in (D) and leads to a unique 7

solution for the flux distribution (E). All flux solutions were obtained using the metabolic 8

models defined in Supplemental File S3. 9

10

Figure 4. Iterative procedure for determining metabolic fluxes from steady-state stable 11

isotope labeling experiments. 12

13

Figure 5. Procedure for determining the global best fit for the model and the optimal flux 14

values. 15

16

Figure 6. Establishing the threshold for the best fit flux solutions used to define the subset of 17

Monte Carlo flux space for calculating the global best fit (Fig. 5). A, Histogram showing the 18

frequency distribution for the residua of the feasible solutions. The inset shows an expansion 19

of the region covering the lowest residua with an arrow to indicate the threshold for inclusion 20

in the subsequent analysis. B, Distribution of scores for the first component following 21

principal component analysis of the flux solutions from 1000 Monte Carlo simulations. 22

Feasible flux solutions with residua <3000 (●) form a distinct cluster that is well resolved 23

from those with residua >3000 (○). The latter have a more scattered distribution which is 24

similar to that obtained from random combinations of fluxes that produce infeasible flux 25

maps (�). 26

27

Figure 7. Flux maps corresponding to three models for the subcellular compartmentation of 28

the oxidative PPP. A, Oxidative and non-oxidative reactions of the PPP confined to the 29

plastid; B, oxidative steps of the PPP in both the cytosol and plastid, with the non-oxidative 30

steps in the plastid; and C, the complete pathway in both compartments. The width of the 31

arrows is proportional to the net molar flux for each step as listed in Table II. 32

33



36

Figure 8. Validation of Arabidopsis flux map by radiorespirometric analysis. The ratios of 1

the specific activity of 14CO2 released from cell cultures supplied with [14C]glucose 2

specifically labeled in the carbon positions indicated was compared with those predicted from 3

the flux maps. The latter were derived from the specific abundance of label in CO2 4

determined by simulations in CumoNet using different glucose isotopomers as the input 5

substrate. Experimental data are shown as the mean ± SD of ratios from four replicate 6

measurements for each of two independent cell cultures (white bars). The predicted ratios 7

were determined from the global best-fit solutions (see Table II, also Supplemental Table S5) 8

for the networks in which the PPP is confined to the plastid (Model A, light gray), the 9

oxidative steps of the PPP are located in both the cytosol and plastids (Model B, dark gray), 10

or the entire pathway is present in both compartments (Model C, black). There is no 11

significant difference in the experimentally determined ratios for the two separately grown 12

cultures as assessed by repeat measures ANOVA, and all the predicted ratios are within the 13

95% prediction intervals of the ratios determined experimentally. 14

15

Figure 9. Assessment of goodness-of-fit of isotopomer data to the global best fit solution of 16

the metabolic network defined in Figure 2. Data from cell cultures labeled separately with [1-17 13C]glucose, [2-13C]glucose and [U-13C6]glucose were used in these evaluations. A. 18

Comparison of predicted and measured isotopomer abundances for all metabolites analyzed. 19

B. Comparison of the predicted relative fractional abundance of isotopomers of aspartate with 20

measured values, before and after scaling of subgroups of 13C peaks with multiplet structure. 21

Shading is conserved for individual isotopomers in the pie charts for a given labeled substrate 22

23

Figure 10. Statistical analysis of flux estimates obtained with different steady-state labeling 24

strategies. The standard deviations of optimum flux estimates were derived from the global 25

best fit solution of the metabolic network defined in Fig. 2 using the EstimateStat routine in 26

13C-FLUX. Isotopomer measurements obtained from analysis of extracts of cell cultures 27

labeled separately with either [1-13C]glucose (1-13C), [2-13C]glucose (2-13C) or [U-28 13C6]glucose (U-13C) were used independently or in combination. Fluxes were expressed 29

relative to the rate of glucose consumption (set to unity) and standard deviations are coded 30

from black (low; flux well defined) to white (high; flux poorly defined) as indicated by the 31

logarithmic greyscale at the top of the figure. The bottom row contains an analysis of data 32

combined from all three feeding strategies (1/2/U-13C) in which the standard deviation of all 33

isotopomer measurements was reduced to 5%. 34



37

Table I. Metabolism of [U-14C]glucose by cell suspension cultures of Arabidopsis 1

Samples (5 ml, 0.618 ± 0.017 g fresh weight) from 6 d old cultures grown in glucose at 2

natural abundance 13C were supplemented with 185 kBq [U-14C]glucose (yielding a specific 3

activity of 308 ± 15 MBq mol-1) and incubated for 24 h prior to extraction in methanol and 4

subsequent metabolic fractionation. Total 14C uptake is the sum of radioactivity in CO2, 5

ethanol and the chloroform soluble, aqueous methanol soluble and methanol insoluble 6

fractions in each cell culture. Each value is the mean ± SE of the distribution of radioactivity 7

in four cell cultures. 8

9

Metabolic fraction Radioactivity in specified fraction (dpm) CO2 1,076,468 ± 35,772 Chloroform soluble 63,439 ± 5,420 Neutral lipids 41,242 ± 2,217 Free fatty acids 1,083 ± 225 Neutral P-lipids 5,046 ± 722 Acidic P-lipids 21,597 ± 2,921 Methanol soluble 1,021,858 ± 26,185 Organic acids/P-esters 109,533 ± 553 Amino acids 386,305 ± 13,201 Asp/Glu 31,938 ± 2,563 Others 334,737 ± 20,786 Sugars 389,986 ± 17,377 Sucrose 251,596 ± 16,562 Glucose 58,680 ± 5,384 Fructose 37,060 ± 2,298 Others 3,943 ± 623 Methanol insoluble 591,272 ± 53,215 Starch 72,424 ± 11,971 Protein 116,647 ± 10,699 Cell wall 103,763 ± 13,417 Residue 77,010 ± 21,924 Ethanol 18,500 ± 2,115 Total 14C uptake 2,771,536 ± 67,703 Total 14C recovered 2,361,265 ± 25,419 Recovery (%) 84.2 ± 2.1

10



38

Table II. Influence of metabolic network structure on determination of flux estimates 1

Fluxes were determined by fitting isotopomer abundances and biosynthetic output estimates 2

to a family of models differing in the extent of pentose phosphate pathway (PPP) 3

compartmentation (Figs 1, 2). Each model contained a complete plastidic PPP pathway. Free 4

fluxes varied during the fitting procedure are indicated in bold. Net fluxes and normalized 5

hyperbolic transformed exchange fluxes are the global best fit estimate ± SD as determined 6

by EstimateStat and are calculated relative to the rate of glucose uptake (set to unity). A 7

normalized exchange flux for which SD takes the estimated value outside the permissible 8

[0,1] range is considered indeterminate (Ind). 9

10

Flux name Flux (relative to glucose uptake)

No cytosolic PPP Cytosolic oxidative PPP

steps only Complete cytosolic PPP

Net flux chex1 0.317 ± 0.062 0.191 ± 0.083 0.665 ± 0.074 chex2 0.241 ± 0.062 0.115 ± 0.083 0.365 ± 0.187 chex3 0.778 ± 0.016 0.786 ± 0.021 0.894 ± 0.040 phex1 0.049 ± 0.067 0.209 ± 0.101 -0.160 ± 0.090 phex2 0.305 ± 0.063 0.442 ± 0.088 0.228 ± 0.187 phex3 0.072 ± 0.016 0.075 ± 0.020 0.000 ± 0.040 cppp1 - 0.372 ± 0.028 0.266 ± 0.019 cppp2a - - -0.224 ± 0.200 cppp2b - - -0.112 ± 0.100 cppp2c - - -0.112 ± 0.100 cppp3a - - -0.112 ± 0.100 cppp3b - - -0.112 ± 0.100 pppp1 0.405 ± 0.018 0.000 ± 0.039 0.000 ± 0.027 pppp2a 0.138 ± 0.006 0.127 ± 0.009 0.204 ± 0.102 pppp2b 0.118 ± 0.006 0.107 ± 0.009 0.184 ± 0.102 pppp2c 0.256 ± 0.012 0.234 ± 0.018 0.388 ± 0.204 pppp3a 0.138 ± 0.006 0.127 ± 0.009 0.204 ± 0.102 pppp3b 0.138 ± 0.006 0.127 ± 0.009 0.204 ± 0.102 tca1 0.650 ± 0.006 0.661 ± 0.009 0.696 ± 0.009 tca2 0.650 ± 0.006 0.661 ± 0.009 0.696 ± 0.009 tca3 0.624 ± 0.006 0.635 ± 0.009 0.670 ± 0.009 tca4 0.572 ± 0.006 0.583 ± 0.009 0.618 ± 0.009 tca5 0.573 ± 0.006 0.584 ± 0.009 0.619 ± 0.009 ana1 0.241 ± 0.004 0.241 ± 0.004 0.243 ± 0.005 ana2 0.039 ± 0.003 0.040 ± 0.004 0.039 ± 0.004 ana3 0.026 ± 0.001 0.026 ± 0.001 0.029 ± 0.003 cmex 0.610 ± 0.007 0.621 ± 0.009 0.657 ± 0.009



39

cpex 0.112 ± 0.015 0.109 ± 0.019 0.181 ± 0.037 gpt 0.496 ± 0.062 0.251 ± 0.088 -0.118 ± 0.082 tpt/ppt 0.540 ± 0.124 0.801 ± 0.171 0.523 ± 0.471 xpt - - 0.603 ± 0.309

Normalized exchange flux chex1 0.990 ± 0.102 0.990 ± 0.108 0.423 ± 0.066 chex2 0.682 ± 0.055 0.667 ± 0.053 0.803 ± 0.073 phex1 0.476 ± 0.040 0.475 ± 0.054 0.148 ± 0.103 cppp2a - - 0.990 ± 0.276 cppp2b - - 0.320 ± 0.082 cppp2c - - Ind cppp3a - - Ind cppp3b - - 0.000 ± 0.194 pppp2a Ind Ind Ind pppp2b 0.990 ± 0.141 0.990 ± 0.166 0.990 ± 0.588 pppp2c 0.990 ± 0.120 0.990 ± 0.115 0.990 ± 0.502 pppp3a Ind Ind Ind pppp3b 0.268 ± 0.039 0.299 ± 0.059 0.252 ± 0.123 tca5 0.719 ± 0.043 0.719 ± 0.043 0.730 ± 0.044 gpt 0.970 ± 0.051 0.695 ± 0.067 0.462 ± 0.096 tpt/ppt 0.227 ± 0.140 0.441 ± 0.115 0.000 ± 1.424 xpt - - 0.309 ± 0.362 1

2



40

Table III. Coenzyme requirements for the biosynthetic outputs from the central metabolic 1

network in heterotrophic Arabidopsis cells. 2

The coenzyme requirements for biosynthesis were determined from the fluxes in Table S4 3

using the reaction stoichiometries in Table S6, and are presented as mean ± SE based on the 4

estimates of label partitioning following metabolism of [U-14C]glucose in Table 1. The 5

coenzyme requirements for lipid assume synthesis of triacylglycerol containing only oleoyl 6

(C18:0) side chains. Costs for amino acid production are based on the assimilation of equal 7

amounts of NO3- and NH4

+. The ATP requirement for protein production is calculated from 8

the cost of synthesising proteins from the component amino acids, and is equivalent to 4 ATP 9

per amino acid residue. Accumulating organic acids are withdrawn directly from intermediate 10

pools in the network and incur no additional cost. Generation of coenzymes by the central 11

metabolic network is determined from the net fluxes of the internal interconversions (i.e. 12

excluding output steps) of the model defined in Figure 2 using conventional reaction 13

stoichiometries and ignoring the possible contribution of inorganic pyrophosphate as a 14

phosphoryl donor. NADH and NADPH yield in italics assume conversion of isocitrate to 2-15

oxoglutarate by NADP+-dependent isocitrate dehydrogenase. ATP generation is attributed to 16

substrate level phosphorylation, and FADH2 production is assumed to be equivalent to 0.6x 17

NADH. Values for coenzyme generation are the mean ± SE of estimates obtained from the 18

best-fit flux solutions of 501 Monte Carlo simulations; values in parentheses are determined 19

from the global best-fit flux solution of the network in which the oxidative steps of the PPP 20

are located in both the cytosol and plastids as detailed in Table S5. 21

22 Product Coenzyme requirement (molar flux relative to glucose uptake)

ATP NADH NADPH Sucrose 0.076 ± 0.0039 - - Starch 0.042 ± 0.0070 - - Cell wall 0.079 ± 0.0112 -0.039 ± 0.0045 - Lipid 0.066 ± 0.0045 0.068 ± 0.0045 0.066 ± 0.0045 Ethanol - 0.020 ± 0.0023 - Amino acids 0.726 ± 0.0090 0.144 ± 0.0028 1.207 ± 0.0141 Protein 0.268 ± 0.0285 - - Total requirement 1.257 ± 0.0332 0.193 ± 0.0073 1.273 ± 0.0148

Generated by the central metabolic network

0.999 ± 0.0011 3.964 3.330

±±

0.0024 0.0021

0.771 1.407

±±

0.0022 0.0018

(1.001) (4.055) (3.420)

(0.770) (1.405)

23























Running Head: 2 Steady-state metabolic flux analysis in ... · 3 feature of plant cell metabolism. Ideally, steady-state metabolic flux analysis (MFA), an 4 emerging tool for creating

Documents