Reduced mitochondrial malate dehydrogenase activity has a strong effect on photorespiratory metabolism as revealed by 13C labelling

Highlight Using a custom-built chamber for 13C labelling, we show that reduced mitochondrial malate dehydrogenase activity affects photorespiratory metabolism.


Introduction
Mitochondrial malate dehydrogenase (mMDH) catalyses the interconversion of malate and oxaloacetate (OAA) in the tricarboxylic acid (TCA) cycle and its activity is important for redox control of the mitochondrial matrix, through which it may participate in regulation of the TCA cycle turnover (Nunes-Nesi et al., 2005;Tomaz et al., 2010). In addition, its activity is closely linked to malate/OAA exchange across the mitochondrial inner membrane. This process may be especially important for shuttling reductants between mitochondria and peroxisomes during photorespiration; in which equimolar amounts of NADH are produced during glycine oxidation in mitochondria and consumed in the reduction of hydroxypyruvate to glycerate in peroxisomes (Bauwe et al., 2010). Also, by interplay with malate/OAA shuttling between chloroplasts and cytosol -the so-called 'malate valve'-mMDH influences cellular redox balance (Scheibe, 2004). There are two isoforms of mMDH in Arabidopsis thaliana. Tomaz et al. (2010) found that the double mutant mmdh1mmdh2 had a dramatically reduced growth rate, which could be, at least partly, related to disturbances in photorespiratory metabolism. Although the single mutants, mmdh1 and mmdh2, both had a reduced mMDH activity (~60% and ~40% reductions, respectively), neither displayed an apparent growth phenotype when grown under a controlled environment in air. Nonetheless, a slight perturbation in the CO 2 response curve was recorded in mmdh1, suggesting that photorespiration might also be affected in these plants (Tomaz et al., 2010).
Metabolite analysis can provide good indications of an organism's physiological status at a given time under given conditions. However, such snapshots of metabolites' pool sizes can be complemented by metabolic fluxes and dynamics analyses in order to provide additional information and better decipher an organism's metabolic responses to environmental changes (Wiechert et al., 2007;Sweetlove et al., 2014). A well-established method for flux analysis in plants is feeding a 13 C-labelled substrate to a cell suspension culture until a steady state is reached (Wiechert, 2001;Wittmann and Heinzle, 2002;Sauer et al., 2004;Matsuda et al., 2007;Williams et al., 2008). 13 C is a favourable tracer as it is safe to handle, clean, stable, and compatible with mass spectrometry (MS) and nuclear magnetic resonance spectroscopy (NMR) techniques . However, for flux analysis in non-steady state systems, such as photosynthetic leaf tissue, the steady-state 13 C analysis provides little information and requires additional strategies (Sweetlove et al., 2014). Determining kinetics of labelled metabolites in non-steady state systems remains challenging, and several 13 CO 2 -based approaches have been proposed to address metabolic fluxes in planta (Huege et al., 2007;Arrivault et al., 2009;Szecowka et al., 2013;Heise et al., 2014;Ma et al., 2014). Another hurdle for 13 C analyses in plants lies in the lack of appropriate labelling chambers that allows control of the 13 CO 2 concentration and growth conditions, while facilitating rapid sampling and freezing of exposed leaves, without disturbing their environment. These features are essential for accurate determinations of pool sizes, their dynamics and robust conclusions regarding metabolic regulation. Furthermore, metabolite analysis is challenging, regardless of labelling, due to the sheer number of metabolites, the diversity of their chemical properties and the huge concentration differences (Fiehn, 2002;Weckwerth, 2011). Nonetheless, coverage can be improved by using multiple analytical techniques, for example, liquid-or gas chromatography coupled with MS (LC-and GC-MS) and NMR. Such combination analysis can substantially increase metabolite detection, thereby improving the depth of biological understanding (t' Kindt et al., 2009;Hiller et al., 2011;Kueger et al., 2012;Szecowka et al., 2013;Heise et al., 2014).
Here we report experiments using a specially constructed chamber for exposure of plants to 13 CO 2 under controlled conditions. A set of plants, consisting of both wild type and mmdh1, were subject to high and low CO 2 treatments, and sampled before and after 30, 60 and 120 min of 13 C labelling. Forty metabolites were detected with LC and GC-MS techniques to compare metabolic adjustments between wild type and mmdh1 plants under reduced or high photorespiratory conditions.
Growth phenotype was assessed from 7-week-old plants, which were placed in either ambient CO 2 (i.e. 380 ppm) or in low CO 2 (i.e. 150 ppm) conditions for 6 weeks.
Chemicals 13 CO 2 (25 l bottle; isotopic purity 90 atom % 13 C, <1.5 atom % 18 O) purchased from Spectra products (Littleport, Cambridgeshire, UK) and 12 CO 2 (20 l bottle) from AGA (Sweden). Atmospheric air was tapped from an internal supply. All metabolite standards and other chemicals were purchased from Sigma-Aldrich (Minneapolis, MN, USA), except sedoheptulose-7-phosphate, which was bought from Carbosynth Ltd (Berkshire, UK). All standards were purchased at the highest available purity.

Chamber construction
The chamber was built in a glove-box style ( Supplementary Fig. S1, available at JXB online). The enclosure housing the plants was custom-built by Rexonic AB (Piteå, Sweden) from 8 mm thick Plexiglas (volume 0.12 m 3 ). It was mounted on a metal scaffold, above a tabletop surface, leaving space for a freezing container and providing an ergonomically comfortable working height. The scaffold also supported the light source; a high-pressure metal-halide lamp providing a light intensity of 180 μE m −2 s −1 at plant level. A pair of integrated rubber gloves enabled sampling during treatment. An extraction port with a membrane in the chamber floor allowed fast transfer of samples into a container with liquid nitrogen. The extraction port was strategically placed to minimize risks of shading the plants when working in the chamber. An expansion vessel compensated for the reduction in air volume resulting from hands working in the chamber. A one-way restrictor valve allowed the release of internal pressure during treatment. The gas system included three lines (coupled to the chamber via a single merged inlet port). One line was for nitrogen (humidified before entering the chamber) to flush CO 2 from the chamber prior to treatments. One supplied 13 CO 2 via a regulator fitted with a magnetic valve, controlled by a custom-made computer interface, after mixing with CO 2 -free air (both 12 CO 2 and 13 CO 2 gas could be used in ranges from 100 to 10 000 µl l −1 ). The other was the air supply, tapped from the laboratory's gas lines and controlled by a high-pressure regulator and a rotameter (Platon NG, Type FNGVB211A, Roxspur Measurement & Control Ltd, Sheffield, UK). CO 2 was removed from the air by two aqueous CO 2 scrubbers in series followed by a limestone cartridge.
The cooling system consisted of a heat exchange package through which water was pumped continuously by a F12-MA thermoregulated water bath (Julabo GmbH, Seelbach, Germany). Two fans, controlled by the computer interface, were placed behind the heat exchanger. The fans were switched on if the temperature exceeded the set target value. An additional fan for air circulation was kept running when the chamber was in use.
An Engine K30 FR CO 2 sensor (SenseAir, Delsbo, Sweden) was used to monitor CO 2 levels and a Sensirion SHT75 dew point sensor (Sensirion, Staefa ZH, Switzerland) was used to monitor temperature and relative humidity. Both sensors were connected to an Arduino UNO microcontroller (http://arduino.cc/), to enable communication with the computer interface. The computer interface was programmed using MATLAB ver. 8.1 (MathWorks, Natick, MA, USA). The data collected by the sensors were displayed by the interface in real-time plots ( 12 CO 2 , 13 CO2, temperature and RH) and data were automatically recorded as *.txt files.

Plant treatments
The temperature was kept at 22 °C in the labelling chamber, and the relative humidity at 80-85%. The 13 CO 2 gas pressure was 1 bar and the air flow rate to the CO 2 scrubber was 2 l min −1 for the high CO 2 treatment and 5 l min −1 for the low CO 2 treatment. Target values for the high and low CO 2 treatments were 1 000 and 150 µl l −1 , respectively. All experiments were initiated in the middle of the photoperiod.

Sampling
Fully expanded leaves were cut from 6-week-old Arabidopsis plants in the labelling chamber, transferred to a 20 ml scintillator tube that was loosely capped (making it possible for the liquid nitrogen to come in contact with the sampled leaf) and dropped into liquid nitrogen through the extraction port (the whole sampling procedure took under 10 s). Four biological replicates (i.e. independent plants) per genotype were sampled at four time points: just before treatment and after 30, 60 and 120 min treatment.

LC-MS analysis
Analysis was done by combined ultra-high-performance liquid chromatography-electrospray ionization-triple quadrupole-tandem mass spectrometry (UHPLC-ESI-QqQ-MS/MS) in multiple-reaction-monitoring (MRM) mode. An Agilent 6490 UHPLC chromatograph equipped with a Waters Acquity UPLC BEH Amide1.7 µm, 2.1 × 50 mm column (Waters Corporation, Milford, USA) coupled to a QqQ-MS/MS (Agilent Technologies, Atlanta, GA, USA) was used. The washing solution, for the auto sampler syringe and injection needle, was isopropanol:water (1:1, v/v). The mobile phase consisted of 85% B (acetonitrile:10 mM aqueous ammonium formate, v/v) for 0.5 min followed by linear gradients from 85 to 70% from 0.5 to 5.5 min then 70 to 10% B from 5.5 to 8 min, followed by 85% B for equilibration from 8 to 15 min. The balance (mobile phase A) consisted solely of 10 mM aqueous ammonium formate. The flow rate was 1.6 l min −1 during equilibration and 0.250 l min −1 during the chromatographic runs. The column was heated to 60°C, and injection volumes were 2 μl. The mass spectrometer was operated in negative ESI mode with gas temperature 210°C; gas flow 11 l min −1 ; nebulizer pressure 60 psi; sheath gas temperature 200°C; sheath gas flow 8 l min −1 ; capillary voltage 3 000 V (neg.); nozzle voltage 0 V; iFunnel high pressure RF 90 V; iFunnel low pressure RF 60 V. All MRM transitions were run in negative mode: dwell time 50 s; fragmentor voltage 380 V; cell acceleration voltage 5 V. Every sample was injected twice to reduce the number of MRM transitions per analysis. For a list of MRM transitions see Supplementary Table  S1. Data were normalized with respect to sample fresh weights and processed using MassHunter Qualitative Analysis and Quantitative Analysis (QqQ; Agilent Technologies, Atlanta, GA, USA) and Excel (Microsoft, Redmond, Washington, USA) software.

GC-MS analysis
The GC-MS analysis followed the GC-TOF-MS procedure published by Gullberg et al., (2004). Electron inpact (EI) was used for ionization. Quality control samples and a n-alkane series (C 8 -C 40 ) were included in each analysis ((Schauer et al., 2005). The derivatized samples (1 μl) were injected into a split/splitless injector in splitless mode, by an CTC PAL systems auto sampler (with a 10 μl syringe), into an Agilent technologies 7890A GC system (Agilent Technologies, Atlanta, GA, USA) equipped with a 30 m×0.250 mm diameter fused silica capillary column with a bonded 0.25 μm Durabond DB-5MSUI stationary phase (part no. 122-5222UI, Agilent J&W GC columns). The settings were: injector temperature, 260°C; front inlet septum purge flow rate, 3 ml min −1 ; gas flow rate, 1 ml min −1 ; column temp 70°C for 2 min, then increased by 20°C min −1 to 320°C (held for 8 min). The column effluent was introduced into the ion source of a Pegasus HT GC, high-throughput TOF-MS (LECO Corp., St Joseph, MI, USA), with: transfer line temperature, 270°C; ion source temperature, 200°C; detector voltage, 1520 V; electron impact electron beam, −70 V; ionization current, 2.0 mA. 20 spectra s −1 were recorded with a 50-800 m/z mass range, and 290 s solvent delay.
The raw data were converted from SMP-format to NetCDFformat using ChromaTOF software. Peak detection and peak area calculations of both labelled and unlabelled fragments (selected fragments listed in Supplementary Table S2) were performed using Frag_calc, in-house software programmed in MATLAB ver. 8.1 (MathWorks, Natick, MA, USA). Frag_calc required a text file as input, containing unique names, ion channels and retention time windows of the metabolites to be analysed. Data were normalized with respect to internal standards according to (Redestig et al., 2009). Unlabelled metabolites were identified by comparing their retention indices and mass spectra with entries in commercial and in-house mass spectra libraries using NIST MS Search 2.0 (National Institute of Standards and Technology, 2001). In-house software, 13C_est, was used to correct for natural abundance of 13 C and isotope contributions from TMS-groups, and to calculate percentages of 13 C incorporation for each identified metabolite.

Statistical analysis
Multivariate analysis was performed using SIMCA 13.0 software (Umetrics, Umeå, Sweden). All variables were log 10 -transformed, meancentred, and scaled to unit variance before further analysis. Principal component analysis (PCA) was used to overview the data, e.g. observe trends/clusters and detect outliers. Orthogonal projection to latent structures (OPLS) analysis, a supervised technique, was used to connect information regarding two-block variables (X and Y) (Trygg and Wold, 2002) and OPLS-Discriminant (OPLS-DA) analysis was used for modelling maximum class separation (Trygg and Wold, 2002;Bylesjö et al., 2006;Trygg et al., 2006). For all models R2X(cum) is the cumulative modelled variation in X, while R2Y(cum) is the cumulative modelled variation in Y. The range of these parameters is 0-1, where 1 indicates a perfect fit. Q2 is the estimated predictive ability of the model (−1 to 1).

Results
It has been reported that when grown under ambient air conditions, the two single mMDH mutant plants, i.e. mmdh1 and mmdh2, did not exhibit apparent growth phenotype (Tomaz et al., 2010). Still, a slight perturbation in the CO 2 response curve was observed, suggesting that photorespiration might be affected. Here, wild-type and mmdh1 plants were grown under short day conditions and in CO 2 controlled environments for 6 weeks. Under 380 ppm CO 2 , the rosettes of mmdh1 were only slightly smaller than the ones from wild-type plants while a drastic growth difference was noticeable when plants were grown under 150 ppm CO 2 (Fig. 1A). In addition, the respective shoot biomass fresh weight was quantified. The growth ratio (GR) between wild type and mmdh1 was estimated at nearly 50% under 380 ppm CO 2 whereas mmdh1 produced only 14% of the wild-type shoot biomass under low CO 2 conditions (Fig. 1B). Interestingly, the difference in shoot biomass production between ambient air and low CO 2 conditions was 68% for wild-type plants whereas it reached more than 90% in the mmdh1 mutant plants. Together, this phenotypical quantification clearly showed that a reduced amount of mMDH affects plant growth, particularly under high photorespiratory conditions. This therefore prompted us to investigate in more detail the metabolic adjustments in mmdh1, particularly under high and low photorespiratory conditions.

Validation of the labelling chamber
A labelling chamber, designed as a glove box, was constructed with the capacity to house eight fully-grown Arabidopsis plants. The chamber enabled simultaneous treatment of four biological replicates of two genotypes, but also handling and sampling plants during treatment (Fig. 2). A connected gas system provided either 12 CO 2 or 13 CO 2 at controlled concentrations. Samples were collected within the chamber, and directly snap frozen by pushing the sample tubes through an extraction port in the chamber floor into a liquid nitrogen bath. Sampling did not disturb the experimental environment.
The chamber's ability to provide reproducible conditions and uniformly exposed plants to the 13 C tracer was tested in four preliminary experiments by four sets of wild-type Arabidopsis plants. The CO 2 concentration was kept at 1 000 µl l −1 in two of these experiments and at 150 µl l −1 in the other two (hereafter referred to as 'high' and 'low' CO 2 , respectively), in both cases once with 12 CO 2 and once with 13 CO 2 . Fully grown leaves were sampled before treatment and after 30, 60 and 120 min of each treatment, then analysed by GC-MS. Data acquired from the experiments were compared by OPLS , using length of treatment as the Y-variable and metabolite abundances as the X-variables. A t o [1]/ t[1] score plot including all four experiments showed the samples based on the variation in X depending on Y, the orthogonal variation was the CO 2 concentration ( Supplementary Fig. S2). The results showed that the plants responded to the treatments and validated the chamber's suitability for exposing plants to varied CO 2 concentrations. The reproducibility between experiments was also evaluated by OPLS, by comparing two experiments with the same CO 2 concentration. The score plot from the model of the low CO 2 treatment showed no separation between experiments (Fig. 3A), but consistent separation of time points (Fig. 3B). As a final validation of the robustness of the chamber an OPLS model of data from each low CO 2 experiment was used to predict the metabolic profile of the samples from the other low CO 2 experiment. The responses proved to be highly linear, with R2-values of 0.95 and 0.61, showing that the chamber provided highly reproducible conditions ( Supplementary Fig. S3A, B).
To scrutinize the 13 C incorporation we measured the abundance and 13 C labelling of sucrose under high and low CO 2 conditions. Sucrose was chosen since it is a stable product from carbon fixation. Sucrose production was expected to be reduced in low CO 2 treatment compared to high CO 2 treatment. Indeed, the relative abundance of sucrose decreased during the 2-h low CO 2 treatment, but remained stable in the high CO 2 treatment (Fig. 3C). Furthermore, sucrose was more rapidly labelled in high CO 2 treatment compared to low (Fig. 3D). Hence, we concluded that the 13 CO 2 labelling of plants in the chamber was robust under both high and low CO 2 treatments.

Metabolite detection and calculation of 13 C incorporation
To detect 13 C incorporation in metabolites early in carbon fixation, such as hexose phosphates (fructose-6-phosphate, glucose-6-phosphate and glucose-1-phosphate) and UDPglucose, the LC-MS system was operated in negative MRM mode. The pseudomolecular ion of each compound, [M-H −1 ], was set as the precursor ion and the phosphate group as the product ion ([PO 3 ] − m/z 79 or [PO 4 ] − m/z 96). For each labelled metabolite there are (n+1) possible isotopomers, were n is the number of carbons in the metabolite (conceptual illustration Supplementary Fig. S4A). Hence, (n+1) precursor ions were detected for every labelled metabolite. In unlabelled plant material 12 additional sugar phosphates were detected, but their abundance was too low for estimation of 13 C incorporation.
The majority of the metabolites were identified by GC-TOF-MS after methoxyamine and trimethylsilyl (TMS) derivatization. To calculate the 13 C incorporation correctly the contribution from the derivatization and the number of carbons in the fragment must be known. For example, the most abundant fragment for the two carbon metabolite glycine is m/z 276. This fragment contains both carbons and three TMS-groups where one of the TMS-groups has lost a methyl group. Once this relation has been established the (n+1) formula can be used to decide the number of isotopomers to be detected. Thus, for 13 C-labelled glycine m/z 276 to 278 must be monitored to be able to calculate the 13 C incoporation correctly (conceptual illustration Supplementary Fig.  S4B). Hence, the number of isotopomers required to cover the 13 C incorporation increases with the number of carbons of the metabolite and the total number of isotopomers of the analysis increases with the number of metabolites analysed.
Metabolite abundance and 13 C incorporation were calculated by a targeted approach, were a list of all metabolite fragments were processed by in-house scripts. In the same process the natural occurrence of 13 C, ~1.1% of all C carbons (Smith, 1972), and the contribution from the TMS-groups (for the GC-MS data), was subtracted. This was done by sequential isotope compensation from an unlabelled reference spectrum (Fig. 4) for every isotope, the relative standard deviation (RSD) of the unlabelled control, and the contributions from the natural abundance of 13 C and the TMS-groups ( Supplementary Fig.  S5). The visualization plots were an important part of the quality evaluation of the data.

Comparison of metabolic phenotypes of wild type and mmdh1
Metabolic profiles of wild type and mmdh1 under different CO 2 concentrations, using 13 CO 2 , were acquired by GC-and LC-MS analyses and first explored by PCA. Score plots for samples of both genotypes exposed to high and low CO 2 showed a clear separation between treatments, but no evident differences between genotypes (Fig. 5A). A second PCA model based solely on data from samples (wild type and mmdh1) exposed to high CO 2 showed no clear differences between genotypes, or changes over time (Fig. 5B). However, a PCA model based on metabolic profiles of the two genotypes at low CO 2 showed a progressive separation of the genotypes during the time course (Fig. 5C). This was further confirmed by OPLS-discriminant analysis (OPLS-DA) , which showed significant correlations between shifts in abundance of some metabolites and one or the other of the genotypes. Under low CO 2 , the mutant displayed a time course separation in the first orthogonal component (vertical axis of the OPLS-DA score plot; Supplementary Fig. S6C). Collectively, these results clearly indicated that metabolic differences between the mmdh1 mutant and wild type were most pronounced under low CO 2 concentrations, while the two genotypes responded very similarly under high CO 2 . More specifically, after 2 h treatment in low CO 2 treatment the levels of sucrose, two organic acids (pyruvate and glycerate) and several amino acids (asparagine, aspartate, alanine and serine) had decreased in both genotypes (Table 1, Fig. 6, Supplementary Tables S3-6). Interestingly, levels of isoleucine, valine, tryptophan, phenylalanine and tyrosine decreased in wild-type plants, while levels of glutamate and glutamine decreased in the mmdh1 mutant. In addition, in mmdh1 plants, pools of several metabolites (succinate, α-ketoglutarate, lysine, glycine, and the three aromatic amino acids: tyrosine, tryptophan and phenylalanine) were more abundant under low CO 2 than under high CO 2 , a relation that was not found in the wild type counterparts (Fig. 6). Thus, due to the increase in glycine and decrease in serine content in low CO 2 , the glycine/serine ratio was much higher in mmdh1 than in wild-type plants ( Supplementary Fig. S7). In high CO 2 the glycine/serine ratio was low in both mutant and wild type whereas the ratio in ambient CO 2 was intermediate between high and low CO 2 in the mutant. In wild type under limiting CO 2 the glycine/serine ratio initially increased during the first hour but then returned to a low ration after 2 h, suggesting an adaptive mechanism.
Moreover, the 13 C incorporation estimations showed that the metabolites directly produced from and closely associated with the Calvin-Benson cycle, were rapidly labelled. This was exemplified by sucrose and maltose, as well as metabolites associated with photorespiration, such as glycine, Green bars represent an average of four samples treated in high CO 2 conditions and orange bars represents an average of four samples treated in low CO 2 conditions. Relative abundance of sucrose under high and low CO 2 conditions was compared by a Student's t-test and significance was marked as ** for a P<0.01. (D) 13 C incorporation in sucrose, in leaves of wild-type Arabidopsis plants exposed to high (green circles) and low (orange triangles) CO 2 treatment over 2 h. Data were normalized to fresh weight (mg) and technical variation (t-score of internal standards).
Glycolate was also labelled albeit more weakly than the other analysed photorespiratory intermediates (in both genotypes at every time point and under both CO 2 treatments).
Generally, 13 C-incorporation patterns were similar in mutant and wild-type samples under both high and low CO 2 conditions (Table 1). Apart from glycine, most of the amino acids were more strongly labelled under high than low CO 2 conditions. Several amino acids were rapidly labelled, including: aspartate and its biosynthetic derivatives isoleucine and lysine; the aromatic amino acid phenylalanine, for which the biosynthetic intermediate shikimate also showed substantial labelling; and both alanine and valine, which are linked to pyruvate metabolism. Unfortunately, the abundance of pyruvate was too low for a reliable detection of 13 C incorporation. At most time points glycine was more strongly labelled in mmdh1 than in wild-type plants, particularly under low CO 2 .
Intriguingly, glutamate and glutamine were slightly labelled under high CO 2 , but hardly at all under low CO 2 (Table 1). This is interesting as they are key intermediates in refixation of the NH 4 + released in mitochondria during photorespiratory conversion of glycine to serine. Several organic acids (e.g. fumarate and malate) showed an intermediate labelling rate. Other metabolites, including citrate and α-ketoglutarate, showed a very low labelling rate. Finally, uniquely among the monitored metabolites, substantial incorporation of 13 C was detected for succinate in mmdh1, but not in wild-type plants.

C labelling and detection techniques for metabolite analysis
The labelling chamber, developed and applied in this study, offers unique possibilities to expose plants to 13 CO 2 , with satisfactory numbers of biological replicates. The chamber Fig. 4. Principle of 13 C estimations. The 13 C incorporation was calculated by sequential isotope compensation from a reference spectra (unlabelled).
was easy to handle as a set of plants could be exposed to 13 CO 2 at a desired concentration, ranging from 100 to 10 000 µl l −1 , under carefully controlled and recorded conditions. Nonetheless, the chamber had one main limitation: its large volume would prevent rapid atmospheric changes, and hence high-resolution kinetic analysis of metabolic responses. It is however unlikely that this would strongly affect metabolites other than those with a fast turnover.
The aim during development of the analytical methodology was to combine a fast sample preparation with quick LC-and GC-MS methods enabling high-throughput analysis and providing good coverage of plant primary carbon metabolism. Thus, the same extraction protocol was used for both MS analyses (apart from the derivatization step for the GC analysis), which inevitably prevented detection of some metabolites that would have required a more elaborate extraction. This resulted in a relatively low coverage of metabolites involved in early stages of carbon assimilation, but not for metabolites acting further downstream in the plants' primary metabolism. For the GC-MS analysis the major challenge in label detection was to find the most representable fragment for each metabolite, e.g. containing the highest possible number of carbons without interference from other compounds, preferably with a high signal-to-noise ratio. The use of customized in-house scripts dramatically reduced the time required for data processing and aided both interpretation of the results and quality control.

mmdh1 has photorespiratory perturbations resulting from a redox imbalance
The responses to high CO 2 treatment were very similar between wild-type and mmdh1 plants, both with respect to metabolite pools and incorporation of 13 C (Fig. 6). Also, when high and low CO 2 were compared, the differences observed were in most cases similar between wild type and mutant including decreased pools of hexoses and sucrose, and some amino acids (aspartate, asparagine and alanine). However, under low CO 2 treatment both abundance and labelling of some metabolites significantly differed between the two genotypes. Interestingly, most of the observed differences were closely associated with photorespiration. The glycine/serine ratio increased in mutant as compared to wild-type plants and the effect was more pronounced at low CO 2 as compared to ambient air and high CO 2 ( Supplementary Fig. S7). The increase in glycine/serine ratio could indicate that in mmdh1 the limitation in OAA to malate conversion directly influences the glycine to serine conversion although the increased ratio could also reflect an adjustment to maintain the flux through the GDC. However, the significant reduced growth of mutant plants in strong photorespiratory conditions and the effects on glutamate/glutamine/α-ketoglutarate (see below) support a direct limitation in the reaction. A reduced capacity to shuttle NADH produced in glycine decarboxylation from the mitochondria out to the peroxisomes is likely to result in an increased NADH/NAD + ratio in the mitochondrial matrix. This could in turn inhibit the glycine decarboxylase complex, which is inhibited by NADH with a K i of 15µM (Bykova et al., 2014). Furthermore, the reductions in glutamate and glutamine pools together with the increase in α-ketoglutarate are most likely related to the reduced rates of ammonium production, from mitochondrial glycine oxidation, which would limit its re-fixation via the GS/GOGAT  Fig. 5. PCA score plots for mmdh1 and wild type leaves, under: (A) high (dark green) and low (orange) CO 2 treatment (PCA model: components=5, n=64, X-variables=37, R2X(cum)=0.731, Q2=0.455); (B) high CO 2 treatment (PCA model: components=4, n=32, X-variables=37, R2X=0.623, Q2=0.187); (C) low CO 2 treatment (PCA model: components=4, n=32, X-variables=37, R2X=0.701, Q2=0.402). The colour coding indicates sampling times: control (blue), and after 30 min (green), 60 min (red) and 120 min (yellow) treatment. Data were normalized to fresh weight (mg) and technical variation (t-score of internal standards).
system. An important role of glutamate dehydrogenase (GDH) in this unbalanced ratio between α-ketoglutarate and glutamate has been discarded for two main reasons: (i) in the present scheme, NADH would be more available than NAD + , which thus would not support the catabolism of glutamate, and (ii) although it has been proposed that GDH could play a role in ammonium re-assimilation by mitochondria, the high K m of this enzyme for NH 4 + (in the range of a few mM) would not particularly favour the reductive amination of α-ketoglutarate. Additionally, an increase in the mitochondrial NADH/NAD + ratio can contribute to inhibition of key reactions in the TCA cycle, particularly steps catalysed by the pyruvate dehydrogenase complex and isocitrate dehydrogenase (Bykova et al., 2005), thereby limiting the turnover of the cycle. Consequently, a light-dependent limitation in TCA cycle turnover is generally observed in the light even though different mechanisms may be operating in high and low CO 2 conditions. A partial TCA cycle in the light has previously been described by several investigations using different approaches (Sweetlove et al., 2010;Tcherkez et al., 2009). An alteration in mitochondrial photorespiratory reactions may also affect the chloroplasts, as limitation of glycine decarboxylation in barley reportedly affects the chloroplast redox state (Igamberdiev et al., 2001). The 'malate-valve' probably plays an important role in linking metabolic processes between the cellular compartments (Scheibe, 2004), although additional systems seem to be active (Hebbelmann et al., 2012).
However, the incorporation of 13 C into several metabolites differed between our experiments and this report. For example, alanine and aspartate were very poorly labelled after 6 h in the light in Brassica in air or low CO 2 whereas we observed high labelling of these two metabolites already within our 2 h experiment, both in high and low CO 2 (Fig 6). The contrasting results can be due to differences between species and different experimental setups, including use of detached or attached leaves etc. The interactions between carbon and nitrogen metabolism is complex and additional detailed studies with different species and experimental systems will be needed to resolve this issue. Interestingly, a higher incorporation of 13 C label in succinate was observed in mmdh1 plants compared to wild-type plants, at every time point, under both high and low CO 2 (Table 1, Fig. 6B). Furthermore, the labelling was much higher than in citrate and α-ketoglutarate, and in low CO 2 the succinate pool was also much larger in mmdh1 plants than in wild-type plants. A possible explanation for these observations, illustrated in Fig. 6B, is that in mmdh1 OAA to malate conversion is limited, which impairs malate/OAA exchange. In this situation cytosolic PEP carboxylase can fix H 13 CO 3 − to form OAA, which can be reduced to malate by cytosolic MDH. This labelled malate could then be taken up by mitochondria and converted to succinate, which accumulate under these conditions. Therefore, it is tempting to propose a mechanism whereby succinate formation from malate via fumarate at the expense of FADH 2 oxidation could perhaps relieve some of the limitation imposed on glycine oxidation by a high matrix redox state. Accordingly, it has been shown that an increase of α-ketoglutarate, a metabolite at the junction between the TCA cycle and nitrogen assimilation, can lead to activation of succinate dehydrogenase (Saito and Matsuda, 2010).

The use of 13 C unveils novel metabolic dynamics and regulation
The present study takes advantage of the potential to use 13 C labelling to get additional information about plant metabolism. While analysis of pool sizes could identify a metabolic block in the glycine to serine conversion the labelling data gives further information about effects on the turnover of the TCA cycle. In addition, we found that sugars (more specifically hexoses/hexose phosphates and sucrose) rapidly became highly labelled following exposure to the tracer. Two other sugars, trehalose and maltose, had similar pool sizes and no further information about them would have been obtained using a conventional metabolomics approach. However, 13 C labelling revealed that maltose had similar labelling kinetics to the other sugars, contrary to trehalose, which remained poorly labelled throughout the 13 C treatment. This indicates that maltose is somehow a direct product of primary carbon incorporation (Fig. 6A). A similar observation, obtained by a different experimental system, has recently been reported by Szecowka et al. (2013). The cited authors proposed that the dramatic increase in labelled maltose is unlikely to originate from transient starch degradation, but rather from de novo biosynthesis of maltose in light. This is consistent with observations of very rapid 14 CO 2 labelling of maltose in spinach chloroplasts in earlier studies (Allen et al., 2003;Bauwe et al., 2010). Another interesting result is the accumulation of several amino acids, including aromatic and branched-chain amino acids, in the mmdh1 mutant under low CO 2 . Surprisingly, however, the 13 C incorporation in these metabolites was not correspondingly high, suggesting that their accumulation was due to transfers between pools rather than de novo biosynthesis (Fig. 6A). A similar reasoning may explain the low 13 C incorporation in glutamine, glutamate and α-ketoglutarate (Fig. 6B). Indeed, the altered photorespiratory cycle in mmdh1 certainly leads to an imbalanced ammonium refixation in the chloroplast. Furthermore, since glutamate, glutamine and α-ketoglutarate share the same carbon backbone, these metabolites probably only exchange carbon skeletons among their existing pools while some of the nitrogen accumulates in glycine. This recycling of backbone skeletons would limit the 13 C incorporation into these three metabolites.
An obvious limitation in the present experimental setup lies in the fact that the CO 2 concentration was changed simultaneously with the start of the 13 C labelling. Consequently, as metabolite pools do change during the labelling period, metabolic fluxes cannot be properly calculated. Future prospects for the 13 C estimations would therefore be to combine them with a flux model to estimate fluxes for the defined metabolic network. Recently an elegant method was published were non-stationary 13 C flux analysis was used to monitor flux in Arabidopsis rosettes (Ma et al., 2014). Furthermore, the experimental setup described in the present study has a limitation in time resolution, however, this may not be a problem when metabolic reactions downstream of the Calvin-Benson cycle are in focus. The report by Ma et al. (2014) showed that the metabolites involved in this cycle were saturated with label within ~15 min whereas downstream metabolites were poorly labelled. Thus a longer labelling period would be needed to get reliable data. Another future prospect would be to include aspects of cell compartmentalization, which is necessary to fully elucidate metabolic fluxes and metabolism. The labelling results provide information about the rate of conversion of specific metabolites, which in turn can give hints on the existence of several subcellular pools. For example, the rather low labelling rate of malate, fumarate and succinate can be explained by the size and location of these pools in plant cells. The biggest fractions of these organic acids are located in the vacuole, where malate and fumarate can form temporary carbon sinks for photosynthate Keerberg et al., 2011). Small, organellar pools with a rapid turnover will therefore be masked by bigger and less metabolically active pools. For such studies non-aqueous fractionation (Heise et al., 2014) and protoplast fractionation (Keerberg et al., 2011) procedures could be used in order to follow up leads from labelling experiments.

Supplementary data
Supplementary data are available at JXB online. Fig. S1. Labelling chamber. Fig. S2. PCA score plot of Arabidopsis wild-type leaves from two experiments, one treated in high CO 2 and one in low CO 2 . Fig. S3. Y-predictive plots for chamber validation. Fig. S4. Conceptual diagrams of 13 C detection in (A) LCand (B) GC-MS. Fig. S5. Script output of 13 C visualization plots, here data for sucrose under high CO 2 in mmdh1 and wild-type samples. Fig. S6. Biological validation of the system and methodology by OPLS-DA of metabolic profiles of mmdh1 and wildtype samples at high (A, B) and low CO 2 (C, D). Fig. S7. Glycine/serine ratios observed in wild type and mmdh1 samples under (A) high, (B) ambient and (C) low CO 2 conditions. Fig. S8. Percentage 13 C incorporated in hexose phosphates at high CO 2 and low CO 2 in mmdh1 and wild-type samples. Table S1. All MRM transitions and instrumental setup for LC-QqQ-MS analysis. Table S2. List of metabolites analysed by GC-MS. Table S3. Metabolite abundance under high and low CO 2 treatment. Table S4. Student's t-test of the analysed metabolites in mmdh1. Table S5. Student's t-test of the analysed metabolites in wild type. Table S6. Student's t-test for the analysed metabolites.