The extent to which individual plants utilise nitrate and ammonium, the two principal nitrogen sources in the rhizosphere, is variable and many species require a balance between the two forms for optimal growth. The effects of nitrate and ammonium on gene expression, enzyme activity and metabolite composition have been documented extensively with the aim of understanding the way in which plant cells respond to the different forms of nitrogen, but ultimately the impact of these changes on the organisation and operation of the central metabolic network can only be addressed by analysing the fluxes supported by the network. Accordingly steady-state metabolic flux analysis was used to define the metabolic phenotype of a heterotrophic Arabidopsis thaliana cell culture grown in Murashige and Skoog and ammonium-free media, treatments that influenced growth and biomass composition. Fluxes through the central metabolic network were deduced from the redistribution of label into metabolic intermediates and end products observed when cells were labelled with [1-13C]-, [2-13C]- or [13C6]glucose, in tandem with 14C-measurements of the net accumulation of biomass. Analysis of the flux maps showed that: (i) flux through the oxidative pentose phosphate pathway varied independently of the reductant demand for biosynthesis, (ii) non-plastidic processes made a significant and variable contribution to the provision of reducing power for the plastid, and (iii) the inclusion of ammonium in the growth medium increased cell maintenance costs, in agreement with the futile cycling model of ammonium toxicity. These conclusions highlight the complexity of the metabolic response to a change in nitrogen nutrition.
Most plants acquire nitrogen in the form of nitrate and/or ammonium from the soil and an adequate nitrogen supply is a crucial factor in achieving high crop productivity (Kraiser et al., 2011; Xu et al., 2012). The availability of nitrate and ammonium is highly variable in unfertilised soils, and the nitrate:ammonium ratio can itself affect productivity (Marschner, 2012). In particular, growth on ammonium as the sole nitrogen source can be suboptimal, with the severity of the phenomenon depending on species, genotype and the concentration and composition of the rooting medium, and this gives rise to the concept of ammonium toxicity (Gerendás et al., 1997; Britto and Kronzucker, 2002; Hoffmann et al., 2007). Numerous reports have documented the deleterious effects of ammonium as the sole nitrogen source on the growth of Arabidopsis thaliana (Cao et al., 1993; Wilkinson and Crawford, 1993; M'rah Helali et al., 2010; Hachiya et al., 2012) and growth media for this plant typically contain roughly comparable amounts of nitrate and ammonium, both of which are taken up (Gazzarrini et al., 1999) and metabolised.
At a cellular level, the effects of nitrate and ammonium have been studied extensively through the analysis of transcripts (Wang et al., 2003; Scheible et al., 2004; Hoffmann et al., 2007; Gifford et al., 2008; Patterson et al., 2010), enzyme activities (Tschoep et al., 2009) and metabolite composition (Tschoep et al., 2009; Kusano et al., 2011). It has been established that a large fraction of the Arabidopsis genome responds to the form in which nitrogen is supplied to the plant (Vidal and Gutiérrez, 2008), and together with the observed changes in enzyme activities and metabolite levels, this points towards substantial rearrangements of metabolism in response to changes in nitrogen supply. However, the operation of the pathways of central carbon metabolism has been shown to be remarkably robust in many organisms, including plants (Rontein et al., 2002; Williams et al., 2008), and it is difficult to draw conclusions about the metabolic impact of a change in the nitrogen source without direct measurements of metabolic activity. In particular the implications of any putative metabolic reorganisation for substrate utilisation, and the supply of reductant and ATP for biosynthesis, can only be assessed from a knowledge of the metabolic fluxes supported by the network.
The critical importance of flux measurements for the analysis of metabolic networks at a systems level has been emphasised elsewhere (Stitt et al., 2010; Kruger and Ratcliffe, 2012) and currently steady-state metabolic flux analysis (MFA) provides the most powerful method for measuring multiple fluxes in the central metabolic network (Ratcliffe and Shachar-Hill, 2006; Libourel and Shachar-Hill, 2008; Allen et al., 2009; Kruger and Ratcliffe, 2009; Kruger et al., 2012; O'Grady et al., 2012). This isotope-based approach to flux analysis can be used to assess the metabolic impact of changes in nitrogen nutrition, and in earlier work steady-state MFA was used to compare the metabolic phenotypes of developing Brassica napus embryos grown on ammonium nitrate or a mixture of glutamine and alanine (Junker et al., 2007). This study showed a poor correlation between changes in enzyme activity and changes in flux, emphasising the need for direct measurements of metabolic activity if the metabolic impact of a change in nitrogen supply is to be properly identified.
Here, steady-state MFA is used to compare the metabolic phenotype of a heterotrophic Arabidopsis cell suspension grown in a conventional medium containing both nitrate and ammonium, with the phenotype of cells grown in an ammonium-free medium. This experimental system has been used in several applications of MFA (Williams et al., 2008, 2010, 2011; Masakapalli et al., 2010) and in a study that analysed the subcellular compartmentation of carbohydrate metabolism, it was found that the flux through the oxidative steps of the pentose phosphate pathway (oxPPP) was insufficient to meet the demand for NADPH for biosynthesis (Masakapalli et al., 2010). While the flux map showed that the shortfall in NADPH could be met by flux through an NADP+-dependent isocitrate dehydrogenase (ICDH), the need to do so was surprising and raises the possibility that the assimilation of nitrogen might be constrained by the supply of reducing power. To test this hypothesis, and to gain further insight into the influence of the nitrogen supply on heterotrophic metabolism, including the provision of reductant and ATP for biosynthesis, steady-state flux maps were constructed for cells grown on conventional and ammonium-free media. The results highlight the significant and variable contribution of non-plastidic processes to the provision of reducing power for the plastid; and they also suggest that the ammonium in the conventional growth medium has a deleterious effect on metabolic efficiency.
Growth and biosynthetic outputs of a heterotrophic Arabidopsis cell culture
Growth curves following subculture into fresh medium established that omitting ammonium from the normal MS medium increased the growth rate, whereas omitting nitrate prevented growth (Figure 1a). The latter observation is consistent with the commonly noted alleviation of ammonium toxicity by nitrate (Roosta and Schjoerring, 2008). Further characterisation of growth in the MS and ammonium-free (N40) media showed that: (i) N40 cells accumulated glucose from the growth medium more rapidly than MS cells (Figure 1b); (ii) net depletion of nitrate and ammonium from the medium occurred at the same rate in MS, while the rate of depletion of nitrate from the N40 medium was marginally greater than from MS (Figure 1c); and (iii) equivalent amounts of nitrate and ammonium were absorbed and metabolised after 5.5 days in MS (Figure 1d). The inability of the cells to grow on ammonium alone is consistent with the negative impact of ammonium as a sole nitrogen source on the growth of Arabidopsis plants (Cao et al., 1993; Wilkinson and Crawford, 1993; M'rah Helali et al., 2010; Hachiya et al., 2012) and so the subsequent analysis compared cells grown in MS and N40 media.
Radiorespirometric analysis showed that changing the growth medium from MS to N40 affected the pathways of carbohydrate oxidation (Figure 2). The ratios of 14CO2 released from positionally labelled glucose and gluconate are diagnostic of changes in the relative fluxes through particular pathways (Kruger et al., 2007) and the observed ratios suggested an increased flux through the oxPPP relative to metabolism via the tricarboxylic acid (TCA) cycle in the N40 cells. The C1/C6 ratio implied only a small increase in the oxPPP in the N40 cells (Figure 2a), but this was supported by the C1(gluconate)/C6 (glucose) ratio (Figure 2b), which provides a more sensitive measure of the contribution of the oxPPP to CO2 release (Garlick et al., 2002). N40 cells also showed a smaller C2/C1 ratio, suggesting either less recycling of hexose phosphate through the oxPPP or less oxidation via the TCA cycle; and an increase in the (C1–C6)/C3,4 ratio, suggesting an increase in flux through the oxPPP relative to entry into the TCA cycle via the oxidation of pyruvate. Overall these observations provided a qualitative indication of a change in the metabolic phenotype caused by the switch from MS to the ammonium-free medium.
Further support for this conclusion was obtained by incubating Arabidopsis cells with [U-14C]glucose and analysing the distribution of label in a fractionated extract after the pools of metabolic intermediates had reached an isotopic steady state (Table 1). The redistribution of the label was comparable to that reported previously for Arabidopsis cell cultures, and broadly as expected for actively growing heterotrophic plant cells (Kruger et al., 2007), but there were significant differences between the MS and N40 cells in almost every fraction. In particular N40 cells showed reduced labelling in CO2, amino acids and protein, and an increase in soluble sugars, cell wall material and organic acids. The increase in carbon conversion efficiency from 56 to 68%, and the changes in biosynthetic outputs, imply that switching from MS to an ammonium-free medium altered the metabolic fluxes in the pathways of central carbon metabolism.
Table 1. Influence of the nitrogen source on the metabolism of [U-14C]glucose by cell suspension cultures of Arabidopsis. Samples from 4.5-day-old cultures grown in MS or N40 media were incubated with 74 kBq [U-14C]glucose for 24 h prior to extraction in methanol/chloroform and subsequent metabolic fractionation. The radioactivity in each fraction is presented as a percentage of total 14C metabolised (the sum of radioactivity in CO2, ethanol, lipid, the methanol-soluble fraction excluding glucose and the methanol-insoluble fraction in each cell culture). Recovery of 14C is the sum of radioactivity in the terminal fractions obtained from each sample expressed as a percentage of total 14C metabolised. Each value is the mean ± SE from four independent cultures
Distribution of radioactivity in cultures grown in MS or N40 (% of total metabolised)
Indicates differences in labelling between MS and N40 cultures that are statistically different (P <0.05) in univariate comparisons following manova (Pillia's trace = 1.00, F-ratio = 1148, d.f. = 6,1, P = 0.023).
Identifies values that differ from MS cultures as assessed by Student's t-test (P <0.05).
Implementation of steady-state 13C-metabolic flux analysis
Steady-state 13C-MFA was used to deduce flux maps of the central metabolic pathways for cells growing on MS and N40 media. This was achieved using methods described elsewhere (Williams et al., 2008, 2010; Masakapalli et al., 2010; Kruger et al., 2012), but optimised as follows.
First, while the Arabidopsis cell culture provides a good experimental system for steady-state MFA, end products do not necessarily reach an isotopic steady state over the 5.5-day labelling period. The shortfall in the labelling was quantified (Table S1 in Supporting Information) and used to provide a scaling adjustment that allowed the labelling patterns of end products to be corrected, thus enabling them to be incorporated into the flux analysis (Table S2). Alternatively, where scaling adjustments could not be applied, proxy reactions were added to the metabolic network to compensate for the difference in labelling between metabolic intermediates and end products (Table S2).
Secondly, output fluxes from the network of central carbon metabolism were quantified (Tables S3–S5) using the 14C-labelling data in Table 1. This method provides a sensitive alternative to the more conventional bulk assay of biosynthetic outputs (Masakapalli et al., 2010; Kruger et al., 2012).
Thirdly, triplicate labelling experiments were conducted with each of [1-13C]-, [2-13C]- and [13C6]glucose on cells grown in MS and N40 media, and the redistribution of label into metabolic intermediates and end products was analysed exhaustively by NMR and gas chromatography-mass spectrometry (GC-MS) to generate positional and mass isotopomer measurements (Tables S6 and S7).
Fourth, metabolic fluxes were deduced from the stable isotope labelling measurements and the output fluxes using a compartmented model of central carbon metabolism (Table S8) that was similar to the one defined previously (Masakapalli et al., 2010). Differences in the labelling of the hexose moieties of cell wall and starch (Figure S1) justified the inclusion of cytosolic and plastidic glucose 6-phosphate pools in the model, and so the oxidative steps of the oxPPP were duplicated in the cytosol and plastids. The final model, which contained 32 free fluxes (Table S9), was obtained after extensive evaluation of additional reactions (Table S10), including amino acid recycling reactions (Table S11). One significant change in comparison with earlier models was the inclusion of unlabelled inputs of CO2 and methylene-tetrahydrofolate, both of which led to improved agreement between the model and the experimental measurements.
Finally, the best estimates of the fluxes were determined by the same iterative procedure as before (Masakapalli et al., 2010). A key feature of this procedure is the simultaneous fitting of all nine datasets within 13c-flux, and in a further refinement the flux solutions were subjected to a jackknife analysis (Figure 3). In this analysis, the effect of omitting one of the nine datasets from the fitting procedure was tested for each dataset to check for bias in the flux solutions. The results show that the flux solutions for the MS and N40 cells formed two discrete clusters, implying that the flux maps were different under the two culture conditions. Moreover, the difference between the flux maps for cells grown in MS and ammonium-free media was maintained irrespective of the dataset that was omitted, demonstrating the robustness of the fitting process and the estimates of the fluxes.
Flux maps of central carbon metabolism for cells grown in MS and ammonium-free media
The modelling procedure led to a set of fluxes (Tables 2 and S12) that provided a good explanation for the experimental data obtained for both MS and N40 cells (Figure S2, Data S1). Both models passed the χ2 test (Table S13) and the majority of the net fluxes were well defined. Comparison of the flux maps (Figure 4) for cells growing in the two media showed that growth in the ammonium-free medium led to: (i) increased flux through the plastidic oxPPP (pppp1), (ii) a switch from plastidic glycolysis to oxPPP, reflected in an increase in the pppp1/phex1 ratio, (iii) decreases in the glycolytic production of pyruvate (chex3, phex3), (iv) reduced entry into the TCA cycle (ana1, ana2, cmex), and (v) reduced TCA cycle fluxes. There was also a significant increase in the total flux through the oxPPP in the ammonium-free medium (Figure 5). Overall, these changes in the net fluxes of central carbon metabolism, together with the changes in biosynthetic outputs (Tables 1 and S3–S5), provide a comprehensive description of the metabolic consequences of the shift from MS to an ammonium-free medium.
Table 2. Flux estimates for Arabidopsis cell suspension cultures grown in MS and N40 media. The fluxes are presented as best estimates of flux ± SD as determined by the EstimateStat tool in 13c-flux. The 95% confidence limits of the free fluxes were determined by non-linearised statistics and are presented as (minimum, maximum mentioned in italic). All fluxes are expressed on a molar basis relative to a glucose uptake of 1 ± 0.05. A normalised exchange flux for which SD takes the estimated value outside the permissible [0,1] range is considered indeterminate (Ind)
Validation of the flux maps was achieved by using them to predict the steady-state outcome of the radiorespirometric analysis using positionally labelled glucose (Figure 2a,c,d). These measurements were not used in the flux analysis, and since the release of 14CO2 from the positionally labelled glucose was at or approaching isotopic steady state at 48 h they can be used to provide an independent check on the validity of the flux maps. In fact the predicted pattern of CO2 release from specific carbon atoms of glucose is in very good agreement with the experimentally determined ratios of 14CO2 release from the cultures (Figure 6). Note that the largest of the very small discrepancies in Figure 6 are consistent with the observation that the steady-state release of 14C from C2 was furthest from an isotopic steady state (Figure 2c). Further validation was provided by comparison of the rates of nitrogen assimilation predicted from the flux maps with those estimated from measurements of nitrogen depletion from the growth media (Figure 1c). The rates of nitrogen uptake by the cell cultures calculated between 2 and 6 days after subculture were 1.75 ± 0.135 and 0.78 ± 0.036 μmol h−1 g−1 fresh weight (mean ± SE) for cells in MS and N40 medium, respectively, and these values are in close agreement with the rates of assimilation deduced from the flux maps (Table 3). Thus on the basis of these independent comparisons the flux maps provide an accurate description of the metabolic phenotypes under the two growth conditions.
Table 3. Growth characteristics of Arabidopsis cultures grown in MS and N40 media. All values were determined from the flux maps defined by the network output constraints and the flux estimates presented in Tables 2, S5 and S12. Carbon conversion efficiency is the fraction of metabolised carbon accumulating in end products other than CO2
Glc, glucose; FW, fresh weight.
Carbon conversion efficiency
0.55 ± 0.002
0.69 ± 0.007
Conversion to biomass (C atom/mol Glc)
3.30 ± 0.011
4.14 ± 0.041
N assimilation (N atom/mol Glc)
0.34 ± 0.013
0.11 ± 0.007
Rate of C conversion to biomass (μmol g FW−1 h−1)
2.53 ± 0.009
3.95 ± 0.039
Rate of N assimilation (μmol g FW−1 h−1)
1.55 ± 0.058
0.65 ± 0.038
Rate of nitrate assimilation (μmol g FW−1 h−1)
0.77 ± 0.029
0.65 ± 0.038
C/N ratio of biomass
9.84 ± 0.363
36.92 ± 1.857
Defining the flux response to a change in nitrogen supply
Omitting ammonium from the growth medium of the heterotrophic Arabidopsis cell culture caused substantial changes in the biosynthetic output of the system (Table 1), reflecting the expected impact of the nitrogen source on metabolic composition. For example, the difference in the labelling of amino acids and organic acids between the MS and N40 media mirrors numerous reports of higher organic acid and lower amino acid levels in nitrate-grown plants relative to ammonium-grown plants (Chaillou et al., 1991; Pasqualini et al., 2001; Hachiya et al., 2012). Similarly the reduction in CO2 labelling in the ammonium-free medium, and the concomitant increase in the carbon conversion efficiency, is consistent with the observation that ammonium as the sole nitrogen source often leads to an increase in respiration (Britto et al., 2001; Escobar et al., 2006; Hachiya et al., 2010). These parallels support the use of the Arabidopsis cell culture as a system for investigating the impact of the nitrogen source on the organisation of central carbon metabolism in heterotrophic tissues, although it should be noted that the typical nutrient concentrations in cell culture media are greater than those that would normally be encountered by plants under natural conditions.
The changes in biosynthetic output (Table 1) implied changes in the flux distribution supported by the central pathways, and the impact of the growth medium on the flux distribution was quantified by steady-state 13C-MFA (Figure 4, Tables 2 and S12). In principle constraints-based flux balance analysis (Sweetlove and Ratcliffe, 2011) using the biosynthetic outputs as constraints and an appropriate optimisation function could have been used to predict some of the changes in the flux distribution (Williams et al., 2010; Hay and Schwender, 2011), but isotope-based MFA provides a more complete picture of the metabolic response. The flux map for the cells grown in MS medium was broadly similar to that reported in the earlier analysis of the subcellular compartmentation of carbohydrate oxidation in the same system (Masakapalli et al., 2010), and the reliability of the flux measurements was sufficient to detect significant differences in multiple fluxes and flux ratios for cells grown on the ammonium-free medium. These differences occurred mainly in the oxPPP and TCA cycle, with implications for the provision of reductant and ATP for biosynthesis. The predicted growth characteristics of the cell cultures derived from these flux maps are summarised in Table 3 and reinforce the conclusion that the metabolic phenotype of the cells is strongly dependent on the composition of the growth medium.
Provision of reductant for plastidic biosynthesis
The incorporation of the nitrogen from nitrate into amino acids is a major sink for reducing power, and an analysis of pea root plastids concluded that the process could be limited by the capacity of the oxPPP to meet the demand for NADPH (Bowsher et al., 2007). Specifically, it was observed that the rate of glucose 6-phosphate-dependent nitrite reduction by isolated plastids was lower in the presence of substrates for glutamate synthase, and that the rate of glutamate synthesis decreased in the presence of nitrite. The implied restriction on plastidic metabolism arising from a shortage of reducing power has led to the suggestion that manipulation of flux through the oxPPP in roots could enhance nitrogen use efficiency (Foyer et al., 2011). This strategy assumes that plastidic reductive biosynthesis is largely dependent on plastidic sources of reductant, but this is not necessarily self-evident. For example a radiorespirometric analysis of 6-phosphogluconate dehydrogenase-deficient genotypes of maize indicated that the cytosolic and plastidic oxPPP cooperated in the provision of reductant for plastidic biosynthesis, most probably through plausible, but as yet unidentified, metabolite shuttles (Averill et al., 1998). This result suggests that the availability of reductant needs to be considered at a network level, and that the contribution of the plastidic steps should be viewed in the context of the metabolic processes occurring in other subcellular compartments.
The coenzyme balance across the network of central carbon metabolism can be analysed using the net fluxes deduced from steady-state MFA. In Arabidopsis cells grown in MS it has already been shown that the NADPH requirement for biosynthesis cannot be met by either the flux through the plastidic oxPPP or by the combined flux through the cytosolic and plastidic oxPPP (Masakapalli et al., 2010). Thus reductive biosynthesis in the plastid depends on non-plastidic sources of reductant, and the relevance of this conclusion to nitrogen assimilation was explored here by analysing the effect of a change in the nitrate and ammonium composition of the growth medium on the coenzyme balance of the central metabolic network.
Analysis of the flux maps reveals that the NADH and NADPH requirements for biomass production were greater in the MS medium than in the N40 medium (Tables 4 and S14). Most of the biosynthetic demand for reductant in the Arabidopsis cells is associated with the assimilation of nitrogen (Tables 4 and S14, italicised entries), with the conversion of nitrate to ammonium alone accounting for 47 and 74% of the total biosynthetic requirement for NADPH and NADH, respectively, in the MS cells. These figures increased to 61 and 90%, respectively, in the N40 cells, reflecting the use of nitrate as the sole nitrogen source in the growth medium. Omitting ammonium from the growth medium decreased the total assimilation of nitrogen into amino acids and protein (Table 1) and led to a much higher C:N ratio for N40 biomass (Table 3). Interestingly the cells were unable to respond to the absence of ammonium in the growth medium by increasing the assimilation of nitrate. It is possible that this reflects the limitation in the supply of reductant observed earlier in isolated plastids (Bowsher et al., 2007), although the availability of extra-plastidic sources of reductant in an intact cell makes this less likely. However, it is probably unrealistic to seek a simple explanation for the change in nitrogen assimilation in terms of reductant supply since the metabolic response to the change in growth medium revealed in the flux maps was complex.
Table 4. Influence of the nitrogen source on coenzyme turnover during metabolism in heterotrophic Arabidopsis cell suspension cultures. Coenzyme requirements for biosynthesis, expressed as mean values ± SE, were determined from the network output constraints (Table S5) for cell cultures grown in MS and N40 media. Italicised values are the requirements directly attributable to the reduction of nitrate to ammonium. Generation of coenzymes by the metabolic network was determined from net fluxes between intermediates in the flux solutions (Table S12). Separate estimates assume conversion of isocitrate to 2-oxoglutarate via NAD- or NADP- isocitrate dehydrogenase (ICDH). When isocitrate is metabolised by NAD-dependent isocitrate dehydrogenase, then virtually all of the NADPH is generated by the oxPPP
Absolute flux (μmol h−1 g−1 fresh weight)
Requirement for biomass production
Generated by network including
Values for ATP generation: based on production via substrate level phosphorylation.
Values for ATP generation: calculated assuming that the NADH not required for biosynthesis generates ATP. All fluxes are expressed as molar values per unit fresh weight. Molar flux estimates relative to glucose uptake are presented in Table S14 which provides further details of these calculations.
First, as well as incorporating less nitrogen into amino acids and protein (Table 1), N40 cells assimilated less nitrate than the MS cells (Table 3), showing that the cells were unable to maintain the same rate of nitrate reduction, let alone compensate for the loss of ammonium by increasing the assimilation of nitrate. Thus the presence of ammonium favoured the assimilation of nitrate, and this is certainly not explicable in terms of the supply of reducing power for nitrate reduction. Secondly, although the assimilation of nitrate decreased in the absence of ammonium, the total flux through the cytosolic and plastidic oxPPP increased in the N40 cells (Figure 5), such that the entire biosynthetic requirement for NADPH could be met by the oxPPP (Table 4). Again this is an unexpected result that emphasises the way individual metabolic events reflect the response of the entire network to the change in the nitrogen supply. Finally, the potential supply of reductant from the activity of NADP-ICDH was lower in the N40 cells, reflecting reduced fluxes in the TCA cycle (Figure 4), and this was partially offset by provision of reductant elsewhere in the network. The overall conclusion is that while the flux maps show that cytosolic processes make a significant and variable contribution to the provision of reducing power for the plastid, it is the network as a whole that coordinates the metabolic response to the change in nitrogen supply. In particular it is notable that flux through the oxPPP varied independently of the demand for NADPH for reductive biosynthesis.
ATP supply and ammonium toxicity
The ATP requirement for biosynthesis, whether expressed in absolute terms (Table 3) or relative to glucose utilisation (Table S14), was greater for MS cells, even though the N40 cells grew more rapidly (Figure 1a). In fact potential ATP production by the central metabolic network greatly exceeded the biosynthetic requirement, to the extent that substrate-level phosphorylation alone would have been sufficient to meet the requirement in MS cells. Only 10–13% of the potential ATP production was used for the net synthesis of biomass in the MS cells, a figure in agreement with earlier measurements on the same cell culture (Masakapalli et al., 2010), increasing to 12–16% in the N40 cells (Table 4).
Ammonium-dependent stimulation of respiration has been shown to arise from up-regulation of the cytochrome pathway, rather than changes in coupling mediated by the alternative oxidase, in Arabidopsis shoots (Hachiya et al., 2010). Moreover nitrate, which suppresses expression of the alternative oxidase in Arabidopsis (Escobar et al., 2006; Hachiya et al., 2010), was present at the same level in the MS and N40 media supporting the inference of greater potential ATP production in the MS cells (Table 4). In fact comparison of the energy potentially available for ATP production with the biosynthetic requirement reveals a striking difference between the MS and N40 cells: the surplus ATP production in cells grown on MS and N40 was 44.6 and 31.4 μmol h−1 g−1 fresh weight, respectively, using NAD-ICDH to convert isocitrate to 2-oxoglutarate, or 34.0 and 23.7 μmol h−1 g−1 fresh weight, respectively, using NADPH-ICDH (Table 4). Thus potential ATP consumption by processes other than net biosynthesis was over 40% higher in the MS cells than the N40 cells. This difference, which correlated with the increased carbon conversion efficiency of the N40 cells, implies a higher maintenance cost for cells grown in the presence of ammonium. An ammonium-dependent increase in cell maintenance cost is an inevitable consequence of the futile cycling model of ammonium toxicity (Britto et al., 2001; Britto and Kronzucker, 2002) and the maintenance costs revealed in the flux maps provide strong support for this model. However, since it is well known that the extent of futile ammonium cycling depends on the relative concentrations of potassium and ammonium (Szczerba et al., 2008; Balkos et al., 2010), it can be anticipated that this contribution to cellular maintenance costs will vary according to the nutritional conditions.
All materials, including specifically labelled isotopomers of 13C- and 14C-glucose, were purchased as before (Masakapalli et al., 2010).
Cell suspensions of Arabidopsis thaliana (L.) Heynh. (ecotype Landsberg erecta) (May and Leaver, 1993) were maintained at 22°C in MS medium, supplemented with 166.6 mm glucose, 0.5 mg L−1 1-naphthaleneacetic acid, and 0.05 mg L−1 kinetin, in 250 ml Erlenmeyer flasks sealed with a double layer of aluminium foil on an orbital shaker running at 150 r.p.m. under a 16-h light, 8-h dark cycle (Williams et al., 2008). Heterotrophic cell suspensions were produced by subculturing 6 ml of a 7-day-old light-grown cell suspension into 36 ml of fresh growth medium in 100 ml Erlenmeyer flasks, followed by incubation in the dark at 22°C. Two growth media were used: normal MS medium, which contained 18.8 mm KNO3 and 20.6 mm NH4NO3; and an ammonium-free medium (N40), in which the nitrogen in the MS medium was replaced by 40 mm KNO3. Subculturing into fresh media is followed by a lag phase in which the reprogramming of gene expression includes a response to the change in nitrogen content of the medium. The transcriptomic response to a change in nitrogen status is typically complete within hours (Scheible et al., 2004) and since the growth of the cell culture is negligible over this period, the acclimation process has no impact on the redistribution of label. Cells were expected to approach an isotopic steady state after 5.5 days (Williams et al., 2008) and the extent to which this was achieved was determined by growth on 20% [13C6]glucose. Three different 13C-labelling strategies were used for MFA: 99% [1-13C]glucose; 99% [2-13C]glucose; and a mixture of 20% [13C6]glucose and 80% unlabelled glucose (Sigma, http://www.sigmaaldrich.com/). Cells were grown in the dark at 22°C until they reached the putative isotopic steady state at 5.5 days. It has previously been shown that the use of non-tracer amounts of [13C]glucose does not perturb the flux distribution in the core metabolic network (Kruger et al., 2007).
Extraction procedures after stable isotope labelling
Soluble metabolites were extracted from lyophilised cells using perchloric acid (Kruger et al., 2007, 2008), and then fractionated into neutral, acidic and basic components using ion-exchange chromatography (Kruger et al., 2007). Protein was extracted from half of the insoluble pellet using 6 m urea/2 m thiourea, and then precipitated with ice-cold 15% (v/v) trichloroacetic acid prior to hydrolysis in 6 m HCl for 24 h at 100°C to release the component amino acids. Starch was extracted from the other half of the insoluble pellet by digestion with 20 U α-amylase (from Bacillus amyloliquefaciens; Roche, http://www.roche.com/) and 5 U amyloglucosidase (from Aspergillus niger; Roche) for 16 h at 37°C (Allen et al., 2007). After solubilisation of the starch, the remaining cell wall material was hydrolysed using 2 m trifluoroacetic acid (TFA) for 1 h at 120°C. The supernatant was vacuum dried, redissolved in methanol and vacuum dried again to remove TFA.
Nuclear magnetic resonance spectroscopy
The 1H and 13C NMR spectra of extracts and growth media were recorded at 20°C on a Varian Unity Inova 600 spectrometer (Agilent, http://www.home.agilent.com) and then processed using nuts (Acorn NMR, http://www.acornnmr.com/) software as before (Williams et al., 2008, 2010; Masakapalli et al., 2010). The 13C NMR spectra of samples derived from [13C6]glucose feeding experiments were recorded with nuclear Overhauser enhancement using a 6 sec relaxation delay; whereas the 13C NMR spectra of samples derived from [1-13C]- or [2-13C]glucose feeding experiments were recorded without nuclear Overhauser enhancement using a 19 sec relaxation delay. The 1H-decoupled 14N NMR spectra were recorded at 43.35 MHz using a 10 mm broadband probe, a 90° pulse angle, a spectral width of 459 ppm, WALTZ-16 1H decoupling during the acquisition and a 4 sec relaxation delay. Relative measurement errors for 13C NMR peak areas, i.e. standard deviations divided by measurement averages, were obtained from the previously established empirical relationship (Williams et al., 2008), and these values were increased by 0.025 to provide flexibility in fitting peaks with high signal-to-noise ratios.
Gas chromatography-MS analysis
The GC-MS analysis of derivatised amino acids, organic acids and carbohydrates was performed on an Agilent 7890A GC coupled to an Agilent 5975C quadrupole MS (http://www.agilent.com/) using similar procedures to those described previously (Williams et al., 2010). Amino acids and organic acids were analysed as tert-butyldimethylsilyl (TBDMS) derivatives; carbohydrates were treated with methoxyamine hydrochloride (Sigma) and then derivatised with N-methyl-N-(trimethylsilyl)-trifluoroacetamide (MSTFA). Mass spectra (m/z 146–600) in scan mode were acquired at either 3.19 or 4.38 scans sec−1; and single ion monitoring of particular ions was done with a dwell time of between 10 and 30 ms. MetAlign was used for baseline correction (Lommen, 2009) and peaks were assigned using AMDIS32 (National Institute of Standards and Technology, http://www.nist.gov/), an extensive library of trimethyl silyl-derivatised plant metabolites (Kopka et al., 2005), and libraries of TBDMS-derivatised amino acids and organic acids generated in our laboratory and elsewhere. Mass isotopomer abundances for 13C-labelled metabolites were determined as before (Williams et al., 2010).
The release of 14CO2 from a 4.5-day-old cell suspension metabolising positionally labelled substrate was monitored as before (Harrison and Kruger, 2008; Masakapalli et al., 2010). Experiments were conducted by incubating four replicate flasks for each culture condition with 3.7 kBq of the positionally-labelled [14C]substrate. The radiolabelled compounds were added to the culture in 0.1 ml volumes at a specific activity of: [1-14C]glucose, 2.04 TBq mol−1; [2-14C]glucose, 2.04 TBq mol−1; [3,4-14C]glucose, 1.11 TBq mol−1; [6-14C]glucose, 2.07 TBq mol−1; or [1-14C]gluconate, 2.47 GBq mol−1. Radioactivity was determined in a Beckman LS 6500 scintillation counter using Wallac OptiPhase ‘HiSafe’ 3 scintillation cocktail (Fisher Scientific, http://www.fisher.co.uk/).
Quantification of biosynthetic outputs by radiolabelling
Fluxes to biosynthetic products were obtained from the redistribution of label following metabolism of [U-14C]glucose (Masakapalli et al., 2010), but with the following modifications to the procedure described previously. First, as well as recovering [14C]ethanol from the incubation medium by distillation, excreted organic acids and protein were also recovered. Organic acids were obtained by fractionating 1 ml of cell culture filtrate with an anion-exchange column. Protein was precipitated from the remaining cell culture filtrate by the addition of BSA (2 mg ml−1) followed by 2% deoxycholate and 15% v/v trichloroacetic acid. Standing on ice overnight resulted in a clear protein pellet which was separated by centrifugation at 13 000 g for 15 min. The pellet was washed with 1 ml of acetone and centrifuged at 13 000 g for 15 min. The acetone washing was repeated twice and the radioactivity in the protein pellet was measured. Secondly, GC-MS spectra of unlabelled fractions corresponding to soluble extracts, protein hydrolysates and cell culture filtrates were used to measure the relative proportions of individual amino acids, organic acids and sugars. Appropriate standard curves were used to obtain the relative proportions of the metabolites. In a refinement of this approach, the labelling of metabolites obtained from a steady-state [13C6]glucose labelling experiments was used to deduce the actual carbon atom abundances that were generated due to metabolism. This ensured there was no influence of pre-existing pools on the relative abundance of de novo synthesised biomass.
Metabolic modelling was performed using 13c-flux (version 20050329; Wiechert et al., 2001; Masakapalli et al., 2010; Kruger et al., 2012), but with the following modifications to the procedure described previously. First, non-linear confidence intervals were determined for free fluxes using a grid search method (Antoniewicz et al., 2006; Williams et al., 2011). Secondly, two methods were used to correct for the impact of pre-existing unlabelled pools with slow turnover rates on the analysis of the isotopic steady state. These unlabelled pools can include proteins, carbohydrates and cell wall components and their contribution to the observed labelling patterns was included either by adding proxy reactions to the metabolic network (Lonien and Schwender, 2009; Williams et al., 2010; Kruger et al., 2012) or by implementing a scaling adjustment based on a 20% [13C6]glucose labelling experiment.
Adding proxy reactions to the network (Tables S2 and S8) added a free flux to the metabolic model for each unlabelled pool. These extra free fluxes had only a minor effect on the fitting procedure, since the model was still over-determined, with the number of isotopomer measurements (typically more than 2000) invariably much greater than the number of free fluxes (typically 30–60), and preliminary simulations established that it was possible to analyse a version of the Arabidopsis model network containing as many as 101 free fluxes. However when handling large numbers of free fluxes, the proxy free fluxes were initially constrained during the Monte Carlo analysis of the model and only set as free in the final optimisation to find the best fit. This procedure avoided generating an unacceptably high proportion of infeasible flux solutions during the Monte Carlo analysis. Also, since 13c-flux does not permit an input flux to be constrained below 0.0001, proxy reactions carrying negligible fluxes were removed from the network to minimise the total number of free fluxes.
To reduce the reliance on proxy fluxes, mass isotopomer measurements were also corrected for pre-existing pools of unlabelled metabolites using a scaling adjustment (Table S2). In the first step, the average fractional 13C abundance (F) of a precursor metabolite (S) that had reached an isotopic steady state, and a product (P) derived from it, were compared to define a correction factor (Δ):
where 0.0113, the natural abundance of 13C, is the expected value of F for an unlabelled metabolite. A value of 0 for Δ implies that P and S are in the same isotopic steady state; a value of 1 implies that P derives entirely from a pre-existing unlabelled pool. The contribution of the unlabelled pool to the observed mass isotopomers was estimated by multiplying the fractional abundance of the mass isotopomers for unlabelled P by Δ. The resulting values were subtracted from the measured values to give a set of corrected mass isotopomer measurements for P that were rescaled to sum to 1. Corrections deduced from replicate [13C6]glucose labelling experiments were applied to measurements obtained in experiments with other positionally labelled substrates that were conducted in parallel on the same batch of cells.
The scaling adjustment for P depends on identifying a measurable pool of S that reaches an isotopic steady state in a 20% [13C6]glucose labelling experiment. For cell wall and starch, it was assumed that the putative precursors were labelled to 20% since the precursors would have been derived more or less directly from the hexose pool. However, the average steady-state 13C abundance of metabolites produced by reactions involving CO2 and/or methylene-tetrahydrofolate could not be assumed to reach this isotopic abundance because of the potential contribution of atmospheric CO2 in their biosynthesis. Thus, for proteins, soluble pools of amino acids were used to scale the labelling of the amino acids in protein hydrolysates, but when these pools were too small for accurate measurements, a detectable precursor to the amino acid was used as a reporter molecule instead (Table S2).
The mean ± SE of ratios of independent unpaired samples were determined as described by Motulsky (1995). Flux maps were compared by principal components analysis following scaling to unit variance of mean-centred flux estimates (Simca-P 11.5, Umetrics). Other differences between cell cultures were assessed using Student's t-test (two-tailed, unequal variance) and manova (SPSS Statistics 18, IBM). The arcsine [arcsin (y/100)0.5] of percentage values and the logarithm of ratios were used in these analyses (Wardlaw, 1985). Homogeneity of variance of the dependent variable(s) was confirmed by Levene's test. The outputs of multivariate analyses were assessed by Pillai's trace (Field, 2009). Differences between data that were not normally distributed as assessed by the Shapiro–Wilk test were compared using the non-parametric Mann–Whitney test. Statistical comparisons for which P <0.05 are considered significant, and effects with a point-biserial (Pearson) correlation coefficient (r) > 0.5 are defined as large (Cohen, 1992).
SKM acknowledges financial support from the University of Oxford (Clarendon Scholarship), Exeter College (Mr Krishna Pathak Scholarship) and a UK Overseas Research Student award. We thank W. Wiechert (Forschungszentrum Jülich GmbH) for permission to use 13c-flux and P. Spelluci (Fachbereich Mathematik, Technische Universität Darmstadt, Germany) for developing the Donlp2 algorithm.