Combined Noninvasive Imaging and Modeling Approaches Reveal Metabolic Compartmentation in the Barley Endosperm W OA

The starchy endosperm of cereals is a priori taken as a metabolically uniform tissue. By applying a noninvasive assay based on 13 C/ 1 H-magnetic resonance imaging (MRI) to barley ( Hordeum vulgare ) grains, we uncovered metabolic compartmentation in the endosperm. 13 C-Suc feeding during grain ﬁlling showed that the primary site of Ala synthesis was the central region of the endosperm, the part of the caryopsis experiencing the highest level of hypoxia. Region-speciﬁc metabolism in the endosperm was characterized by ﬂux balance analysis (FBA) and metabolite proﬁling. FBA predicts that in the central region of the endosperm, the tricarboxylic acid cycle shifts to a noncyclic mode, accompanied by elevated glycolytic ﬂux and the accumulation of Ala. The metabolic compartmentation within the endosperm is advantageous for the grain’s carbon and energy economy, with a prominent role being played by Ala aminotransferase. An investigation of caryopses with a genetically perturbed tissue pattern demonstrated that Ala accumulation is a consequence of oxygen status, rather than being either tissue speciﬁc or dependent on the supply of Suc. Hence the 13 C-Ala gradient can be used as an in vivo marker for hypoxia. The combination of MRI and metabolic modeling offers opportunities for the noninvasive analysis of metabolic compartmentation in plants.


INTRODUCTION
Cereal grains form the basis of most of the human diet and most animal feed. Among grains, barley (Hordeum vulgare) is widely cultivated in all temperate regions and ranks fourth in worldwide production. Because of its importance in agriculture, the barley genome is currently being sequenced (Schulte et al., 2009; http:// barleygenome.org), and the development of the barley caryopsis, or grain, has been studied in some detail (Sreenivasulu et al., 2010). Apart from the transfer cells and aleurone layer, the barley endosperm appears quite homogeneous. However, in vivo tracking during grain filling has established regional variation in the supply of Suc to the endosperm (Melkus et al., 2011), clear developmental gradients (Sabelli and Larkins, 2009), and uneven levels of hypoxia within the caryopsis (Rolletschek et al., 2004). These localized differences in the internal environment of the endosperm suggest a compartmentation of metabolic activity, although this is difficult to detect directly without access to markers for distinct biochemical events.
Alanine aminotransferase (Ala-AT; EC 2.6.1.2) catalyzes the interconversion of pyruvate to Ala and is regarded as a critical branch point separating aerobic from anaerobic metabolism (Good and Crosby, 1989;Miyashita et al., 2007;Miyashita and Good, 2008;Rocha et al., 2010). The heterologous expression of barley Ala-AT produces a marked improvement in the nitrogen use efficiency of canola (Brassica napus; Good et al., 2007) and rice (Oryza sativa; Beatty et al., 2009). It was argued that this has revolutionary importance for future crop production (Ridley, 2009). In the barley caryopsis, the gene encoding Ala-AT becomes more highly expressed as the assimilate storage phase proceeds (Sreenivasulu et al., 2006), corresponding well with the rising level of internal hypoxia (Rolletschek et al., 2004). Its expression is higher in the endosperm than in the pericarp (Sreenivasulu et al., 2006), both in barley and in rice grains (Kikuchi et al., 1999). The in vivo function of this enzyme appears to be to catalyze the synthesis of Ala, but the reversibility of this reaction complicates the picture because its direction is dependent on the local oxygen status and the relative concentrations of Ala and pyruvate present. Ala-AT activity has long been used in the medical field as a marker for certain diseases (Salaspuro, 1989;Nyblom et al., 2004) and inflammatory responses (Gelderblom et al., 2008) and can be detected by noninvasive methods.
Proton nuclear magnetic resonance ( 1 H-NMR) spectroscopy allows for the noninvasive in vivo detection, quantification, and mapping of certain metabolites in plants (Ratcliffe et al., 2001;Kö ckenberger et al., 2004). For example, major assimilates, such as Suc and Gln, can be detected noninvasively in seeds of pea (Pisum sativum) using localized 1 H-NMR spectroscopy and chemical shift imaging (Melkus et al., 2009). The advantage of using the signal from 1 H instead of other nuclei, such as 13 C or 15 N, is its higher sensitivity, but its utility is limited to measuring steady state levels. By contrast, 13 C NMR can be combined with the feeding of 13 C labeled substrates to track their localization along with their metabolic products. By combining NMR spectroscopy and imaging, it is possible to acquire metabolic and spatial information of the 13 C enriched molecule and its products in the same experiment. To further improve the sensitivity of detection of the 13 C nucleus, different inverse detection schemes were developed for preclinical animal and human applications (Bax et al., 1983, Rothman et al., 1985. These pulse sequences provide different strategies to edit and detect the protons connected to the 13 C nuclei and reach therefore in general higher sensitivity than direct 13 C detection (Novotny et al., 1990;Heidenreich et al., 1998;de Graaf et al., 2003).
Here, we applied the gradient enhanced heteronuclear multiple quantum coherence (geHMQC; Hurd and John, 1991) sequence to enable the detection of 13 C-Ala, derived from a supply of 13 C-Suc to a barley plant. Noninvasive metabolite imaging was able to demonstrate that 13 C-Ala synthesis is restricted to the innermost, most hypoxic region of the endosperm. Biochemical and flux balance analysis (FBA) have also been applied to interpret this spatial arrangement, producing a spatially resolved metabolic model of the starchy endosperm in barley.

Experimental Design for Localizing Ala Synthesis in the Caryopsis
To localize Ala synthesis, we first needed to determine the best precursor form and route to deliver the 13 C label. Analysis of caryopsis sugar composition indicates a characteristic developmental shift from mainly hexoses to mainly Suc (see Supplemental Figure 1 online). Ala was the most prominent free amino acid present, at ;30% of the total pool. The Ala present in the caryopsis included a portion delivered through the vascular strands in the crease vein and a portion synthesized within the caryopsis. The latter was detected by means of feeding 13 C-Suc to the intact caryopsis via the stem ( Figure 1A; for details, see Methods). The extent and kinetics of the uptake of the labeled Suc and its subsequent synthesis to Ala was determined by applying mass spectrometry. The proportion of 13 C-Suc present rose significantly within 12 h of feeding ( Figure 1B), and the accumulation of labeled Ala followed a similar time course. The presence of labeled Glc and Fru remained low.
For noninvasive studies, we applied 13 C/ 1 H-magnetic resonance imaging (MRI) to visualize the anatomy of the ear (see Supplemental Movie 1 online) and the allocation of 13 C-Suc (see Supplemental Movie 2 online). Dynamic MRI demonstrates that Suc passes via crease vein and nucellar projection toward the endosperm and spreads centrifugally within endosperm (for more details, see Melkus et al., 2011).
In addition, we optimized the detection method to allow us to distinguish 13 C-Ala from 13 C-Suc and other molecules. The noninvasive localization of de novo Ala synthesis was achieved using the geHMQC editing scheme combined with chemical shift imaging for spatial localization (see Supplemental Figure 2A online). Sufficient water suppression was achieved with a frequency selective water presaturation module (VAPOR suppression;Tká c et al., 1999). To improve the signal-to-noise ratio (SNR), the 13 C nuclei were irradiated with a Malcolm Levitt decoupling pulse scheme, while the 1 H signal was acquired (see Methods). Supplemental Figures 2B and 2C show spectra acquired with the geHMQC from 100 mM solutions of 13 C-Suc and 13 C-Ala. The resonances of the protons from the 13 C-Suc are visible in the spectral region between 3.4 and 4.2 ppm, and the anomeric hydrogen on the pyranose ring is detectable at 5.4 ppm (see Supplemental Figure 2B online). The protons connected to 13 C-Ala resonate at 1.48 ppm (methyl group) and 3.78 ppm (methin group) (see Supplemental Figure 2C online). The methyl resonance from Ala is therefore well separated from the Suc signals in the spectrum, allowing the separation of the two (A) Schematic representation of 13 C-Suc feeding to the barley ear; the cage indicates the placement of the NMR coil used for metabolite imaging.
(B) The ratio of 13 C-to-12 C metabolites in grain extracts during feeding, as measured by LC-MS. (C) A slice of a three-dimensional reconstruction of an ear showing the pericarp (p), endosperm (en), and nucellar projection (np) of an intact caryopsis, as measured by NMR (see Supplemental Movie 1 online). (D) The scheme indicates the direction of Suc flow within the endosperm (see Supplemental Movie 2 online). The three-dimensional model of caryopsis is based on micro-computer tomography, and flow direction is based on NMR. metabolites also in the in vivo spectra. Figure 2A shows a 1 H reference image of the barley caryopsis, where the locations of two voxels from the geHMQC sequence are indicated. Typical geHMQC spectra are shown in Figures 2D and 2E. The spectrum from the vascular region ( Figure 2D) shows intense signal from the delivered 13 C-Suc, while signals from other 13 C metabolites are not visible. The central region of the endosperm features a major 13 C-Ala resonance ( Figure 2E). This signal is well separated from resonances of 13 C-Suc. We conclude that the 13 C-Ala was derived from the incoming 13 C-Suc.

The Primary Site of Ala Synthesis Is the Central Endosperm
To determine where Ala is synthesized in the endosperm, we used seeds fed with 13 C-Suc and determined the localization of labeled Suc and Ala. The metabolic Suc image was calculated from the Suc resonances, avoiding the spectral region around 3.78 ppm to prevent signal contamination arising from the Ala methin group (see Supplemental Figure 2B online). The distribution of Ala was computed from the Ala methyl resonance at 1.48 ppm (see Supplemental Figure 2C online). Two sites of high Suc levels were detected, one at the crease vein/nucellar projection and a second, larger one within the central part of the endosperm ( Figure 2B). Ala also accumulated in this latter region, but not in the pericarp or elsewhere in the peripheral region of the endosperm ( Figure 2C). Over the course of the feeding experiment, the Ala signal gained in intensity within the central part of the endosperm, forming steep gradients toward the periphery. Thus, the major site of Ala synthesis clearly lay within the central part of the endosperm, a conclusion borne out by a comparison of Ala accumulation between the inner and outer regions of the endosperm (see below).

Ala Gradients Reflect the Local Oxygen State of the Endosperm
The oxygen level within the developing caryopsis has been examined in detail by Rolletschek et al. (2004). They showed that the central region of the endosperm is highly hypoxic (falling to <5 mM oxygen), while in the photosynthetically active portion of the pericarp, the oxygen concentration is, if anything, oversaturated under lit conditions ( Figure 3B). In the absence of light, the conditions in the central region of the endosperm tend toward stronger hypoxia (<2 mM oxygen), unlike in the periphery, where hypoxia does not occur ( Figure 3C). Ala synthesis is therefore largely confined to the most hypoxic region of the endosperm.
The gradient in Ala also colocalized with pattern of cell expansion (see Supplemental Figure 3A online). The cell size is maximal within the central hypoxic region and decreases toward the periphery. We did not detect any structural barriers between these regions, but rather multiple plasmodesmata connecting cells within the entire starchy endosperm (see Supplemental Figures 3B and 3C online).

Steady State Metabolite Levels Differ in Inner versus Outer Regions of Starchy Endosperm
To determine whether other metabolites showed a similar gradient to Ala, we next compared the inner and outer endosperm by liquid chromatography-mass spectrometry (LC-MS). The  NMR-based metabolite imaging indicated some spatial heterogeneity within the endosperm for sugar to amino acid conversion, and this conclusion was supported by the LC-MS-based comparison of steady state metabolite levels between the endosperm's inner and outer regions ( Figure 3). These data demonstrated that the intermediate products of both Suc breakdown (i.e., hexoses and hexose phosphates) and glycolysis persisted at higher levels in the peripheral region of the endosperm. Likewise, several tricarboxylic acid cycle-related organic acids, with the exception of fumarate, were distributed in this manner.
Most of the free amino acids were evenly distributed throughout the endosperm, with the exception of Ala and Gly, accumulating in the inner regions, and homocysteine (and also 5-methyltetrahydrofolate), which concentrated in the peripheral region. Despite remarkable differences in oxygen influx (corresponding to the respiratory oxygen consumption), those intermediates related to energy metabolism (except NADP) maintained comparable steady state levels across the endosperm. The picture is therefore one where metabolite levels and presumably metabolic pathway activity differs substantially across the starchy endosperm.

The Response of Ala Metabolism to Variation in the Supply of Oxygen
We next tested whether Ala synthesis changed with different oxygen levels, contrasting intact ears subjected to either normal (21 kPa) or elevated (42 kPa) levels of oxygen. The level of Ala declined as the oxygen supply was increased ( Figure 4, top panel), as did the levels of phosphoenolpyruvate, pyruvate, Glu, and Asp. By contrast, those of oxoglutarate and oxaloacetate (acceptor molecules for the amino group in the transamination reactions catalyzed by Ala-AT and aspartate aminotransferase [Asp-AT]) remained unchanged and ammonia increased (see Supplemental Table 1 online). There was no perceptible shift in the fermentation products lactate and ethanol. Ala-AT activity, as estimated by its peak catalytic activity (V max ) in tissue extracts, was also unaffected by the increased supply of oxygen ( Figure 4), nor likewise were Asp-AT (related to Ala metabolism) and the fermentation enzymes pyruvate decarboxylase, alcohol dehydrogenase, and lactate dehydrogenase. The oxygen-dependent shifts in metabolite pools indicate altered activities in the pathways related to Ala metabolism. None of the genes encoding enzymes above was transcriptionally induced, as assayed by macroarray analysis (V. Radchuk, personal communication). It appeared that the regulation of Ala metabolism is rather independent of transcriptional/translational control.
To elucidate the in vivo enzymatic activities, we then attempted a modeling approach (FBA) based on the barley caryopsis metabolic network established by Grafahrend-Belau et al. (2009a). Ala-AT function was simulated using the synthesis/degradation of Ala in response to changes in the oxygen supply ( Figure 4, bottom panel). This predicted that Ala accumulates under hypoxic conditions, producing an Ala gradient opposite to that of oxygen. As the supply of oxygen was increased, the accumulation of Ala falls as a result of a reduced flux of both the cytosolic and mitochondrial isoforms of Ala-AT, both of which direct Ala synthesis. At an oxygen level above 8 mmol g dry weight (DW) 21 h 21 , Ala was no longer accumulated due to the deactivation of mAla-AT and the reverse action of cAla-AT, which converts Ala back to pyruvate.

Simulation of Compartment-Specific Metabolism in the Caryopsis
Based on the above FBA, a simulation was then performed to model metabolic flux distributions appearing in central versus peripheral endosperm ( Figure 5; see Supplemental Data Set 2 online). What follows is a description of the tissue-specific flux maps, with the major changes to metabolic behavior being highlighted.

Metabolism in the Central Hypoxic Endosperm
The metabolic flux distribution in the central endosperm reflects the outcome of limited respiration, the activation of fermentation, and the stimulation of glycolysis (fermentative Pasteur effect). The TCA cycle shifts to a noncyclic mode due to inactive succinate dehydrogenase (SDH). The activation of cAla-ATA and mAla-ATA results in the accumulation of Ala, the use of pyruvate, and the production of oxoglutarate, which can act as a cosubstrate for glutamate dehydrogenase (GDH) in the mitochondria. As the mitochondrial electron transport chain is limited, GDH represents the main source of mitochondrial NAD + regeneration, providing the redox equivalents required for the TCA cycle. In the cytosol, oxoglutarate reacts with aspartate to form Glu and oxaloacetate through the action of cAsp-AT. Cytosolic malate dehydrogenase (MDH), which catalyzes the reaction from oxaloacetate to malate, is the main source of cytosolic NAD + regeneration. Cytosolic NAD + needs to be constantly regenerated to permit glycolysis, the primary source of energy under hypoxia. In the cytosol, the cytosolic MDH-mediated regeneration of NAD + is boosted by the import of oxaloacetate synthesized in the mitochondria by mitochondrial MDH. In addition to Ala accumulation, the reverse action of fumarase acting on malate results in the accumulation of fumarate. There is a minor accumulation of succinate via TCA cycle reactions, but the Gaba shunt is not involved. Neither the Gaba shunt nor GAD are active, and there is no accumulation of Gaba.

Metabolism in the Peripheral Aerobic Endosperm
Metabolism in the peripheral endosperm is characterized mainly by elevated TCA cycle activity, with little or no fermentation, and decline in glycolytic flux. Thereby, the TCA pathway follows the common cyclic mode. Neither Ala, fumarate, nor succinate accumulates in this part of the endosperm. Ala serves either as a precursor for protein storage synthesis or is converted to pyruvate by cAla-AT. mAla-AT is no longer active. The induction of the anaplerotic reaction of phosphoenolpyruvate carboxylase (PEPC) results in the allocation of carbon skeletons required for storage product synthesis. In the cytosol, the flux via MDH is strongly reduced, reflecting a much reduced requirement for NAD + regeneration, while in the mitochondria, the main source of NAD + regeneration is represented by oxidative phosphorylation, with GDH acting in the direction of NADH synthesis. (A) Steady state metabolite levels and enzyme activites (V max ) in endosperms developing under ambient and elevated levels of oxygen availability. Stars indicate statistically significant (P < 0.05) differences according to a t test (n = 6). ADH, alcohol dehydrogenase; GabaT, g-aminobutyric acid aminotransferase; LDH, lactate dehydrogenase; PDC, pyruvate decarboxylase; PK, pyruvate kinase; SSA, succinic semialdehyde; n.d., not determined. Ethanol and SSA were undetectable in seed extracts. Metabolite and enzyme data are also given in Supplemental Table 1  The FBA also suggests other compartment-specific features. For example, while there is no PEPC activity in the central endosperm, this enzyme is active in the peripheral endosperm ( Figure 5). Localization of PEPC in peripheral caryopsis tissues (Macnicol and Jacobsen, 1992;Gonzá lez et al., 1998) is in aggreement with this prediction. Our data can also be taken to suggest that the V max values of the various enzymes (see Supplemental Table 1 online) are much higher than is necessary to satisfy the endosperm's in vivo requirements (see Supplemental Data Set 2 online), as has also been observed in embryos of oilseed rape (Junker et al., 2007).

Suc and Ala Distribution in Plants Engineered to Have Altered Suc Allocation
We next examined whether the Ala distribution is altered in plants with altered Suc distribution. The gene Jekyll controls cell differentiation and cell death within the nucellar projection of caryopsis  and in so doing also affects Suc release from the pericarp into the endosperm. When Jekyll was downregulated by means of RNA interference, the release rate of Suc was greatly reduced (Melkus et al., 2011). A strong repression of Jekyll (by ;90%) resulted in a switch from cell death to calluslike growth in the nucellar projection, which expanded and eventually replaced the starchy endosperm ( Figure 6A). As it is enclosed by a pericarp/cuticle, this callus-like tissue readily becomes hypoxic, with the level of oxygen in the central part of these caryopses (as measured by a microsensor) falling to 1.5 6 0.6 mM. The application of 13 C/ 1 H-MRI analysis on these caryopses showed that the distribution of Suc was altered, while that of Ala was left unchanged ( Figure 6B). This observation was taken to show that Ala synthesis is related to the hypoxic status of the tissue, rather than being dependent on either tissue type or the pattern of Suc supply.

C/ 1 H-NMR and the in Vivo Imaging of Plant Metabolism
The use of NMR to visualize and quantify metabolites in planta remains challenging. The present approach has relied on a spectral editing scheme: the geHMQC sequence used the 1 J CHcoupling to detect the protons attached to 13 C nuclei, while at the same time other magnetization was dephased by pulsed magnetic field gradients (Hurd and John, 1991;Tse et al., 1996). This method allowed the monitoring of 13 C-Ala accumulation with a high level of sensitivity, which in turn demonstrated the localized synthesis of Ala from Suc fed to the plant. The approach provides several major advantages over available alternatives. First, the assay is noninvasive, avoiding the artifacts arising from sample preparation; second, apart from complications associated with feeding the plant with 13 C-labeled Suc, the outcomes are faithful to the in vivo situation, in terms of the presence of endogenous gradients in the concentrations of metabolites and oxygen, and the compartmentation structure of the caryopsis remains undisturbed; third, the spatial resolution of the 13 C-Ala mapping is sufficient to allow the detection of tissue-specific differences in metabolism; fourth, the assay times (;20 min) offer a means to observe the kinetics of 13 C-Ala uptake, synthesis, and degradation; fifth, the parallel tracking of anatomy and 13 C-Ala means that Ala accumulation can be related to localized structures and the developmental changes that occur within them; and finally, individual caryopses can be subjected to post hoc analyses (e.g., MS-based detection of other metabolites). There are a number of possible applications for this NMR approach in the general area of studying plant metabolism in vivo. For example, one can equally provide 13 C-Ala (instead of 13 C-Suc or other substrates) to study the routes through which Ala is taken up and distributed within the plant, becomes degraded, or is incorporated into proteins.

Ala Synthesis Is an in Vivo Marker for Hypoxia in Plants
Ala is the characteristic and major free amino acid accumulated when plants are suffering from hypoxia, as shown here for barley ( Figure 2C) and as previously shown for Zea mays, Medicago truncatula, and Arabidopsis thaliana (Good and Muench, 1993;Muench and Good, 1994;Ricoult et al., 2005Ricoult et al., , 2006Miyashita et al., 2007). Ala synthesis relies on the reversible action of Ala-AT, with the direction of catalysis being able to flip rapidly, depending on the local conditions in planta Beatty et al., 2009). Because Ala synthesis is responsive to environmental variation (Wallace et al., 1984;Limami et al., 2008), its localized accumulation can be considered as a marker for the metabolic state. We have demonstrated here that 13 C-Ala synthesis occurs most readily in the central part of the endosperm, which is the region of the endosperm in which the conditions are the most hypoxic ( Figures 2C, 3B, and 3C). When the starchy endosperm is replaced by the nucellar projection, as occurs when Jekyll is suppressed, the level of 13 C-Ala synthesis is retained in the most hypoxic region of the caryopsis, even though the pattern of Suc delivery has been greatly altered. Thus, it appears that the accumulation of Ala acts as a marker for localized hypoxia, at least in the developing seed, mirroring the situation in animal cells, where Ala synthesis is widely used for the diagnosis of hypoxiarelated disorders (Nyblom et al., 2004). The 13 C-Ala imaging described here has the potential to identify which specific tissue is the most susceptible to hypoxic constraint. As the assay is an in vivo one, it represents an alternative to oxygen-sensitive microsensors, which inevitably cause a degree of tissue damage during penetration and in some situations are not usable at all (due to particular tissue features). We suggest that the relationship drawn here between Ala accumulation and hypoxia in the endosperm can quite possibly be extended to diverse plant tissues, such as the roots/stem phloem, in which both elevated Ala synthesis/accumulation

of 14
The Plant Cell and reduced oxygen levels have been observed (Good and Crosby, 1989).

Ala-AT Is a Key for Metabolic Adjustments in Central Endosperm
In Arabidopsis cell suspension cultures, the raising of oxygen availability increases flux throughout the metabolic network, including Ala-AT (Williams et al., 2008). In barley endosperm, not only the metabolic flux responds to local oxygen availability but also the direction of enzyme catalysis (Ala-AT) can turn around. This adds another level of complexity to the regulation of in vivo metabolism. The FBA model of the barley caryopsis predicts high Ala synthesis rates via Ala-AT in central endosperm that in turn inhibits pyruvate accumulation (Figure 4), a metabolite which is increasingly produced via glycolysis under hypoxia. In the mitochondria, the oxoglutarate produced by Ala-AT can be used by oxoglutarate dehydrogenase and SuccCoA ligase to produce ATP, which would compensate in part for the reduced supply of respiratory ATP. This may help to maintain a high level of ATP and other energy-rich intermediates in inner endosperm regions ( Figure 3). In our model, subsequent reactions via both GDH and MDH provide the major source of NAD + regeneration (which represents a critical step under O 2 deficiency). Thus, achieving a higher flux through Ala-AT represents an advantageous alternative under conditions where the operation of the usual TCA cycle is inhibited. Similar conclusions were drawn from studies on photosynthetically active Arabidopsis tissues (Lee et al., 2008) and metabolic changes in response to waterlogging in Lotus japonicus (Rocha et al., 2010). By removing pyruvate, the activation of ethanol/lactate fermentation is inhibited ( Figure 4A), and the stimulation of alternative oxidase activity can be avoided (Oliver et al., 2008). Ala synthesis under endogenous oxygen limitation is regarded as a means to maintain glycolysis and energy provision. The caryopsis can improve its energy efficiency/homeostasis and thereby enhance productivity and performance (De Block and Van Lijsebettens, 2011). While root-specific isoforms of Ala-AT and other homologs expressed in vegetative tissues respond to oxygen supply (Good and Crosby, 1989;Good and Muench, 1993;Miyashita et al., 2007), the endosperm-specific one does not (V. Radchuk, personal communication). A plausible interpretation is that in the endosperm, posttranslational regulation of Ala-AT activity is of particular significance. This represents a rational strategy given that the high energy costs of shifting protein abundance via transcription/translation (Piques et al., 2009).
The flux maps were derived from a specific constraints-based analysis and have provided several striking and/or unexpected predictions of the consequences of hypoxia, which naturally will need to be confirmed experimentally. First, the FBA suggests that SDH is inactive under hypoxic conditions. SDH is involved in both the TCA cycle and the respiratory chain and has been shown to become inactivated under various conditions, including O 2 deficiency (reviewed in Sweetlove et al., 2010). Lowering the activity of SDH will reduce the flux through the TCA cycle (Araú jo et al., 2011), producing a compensatory rise in the glycolytic flux (the latter being part of the pseudohypoxic response seen in animal tumor cells; King et al., 2006). The picture is therefore one where glycolysis is upregulated in the central region of the barley endosperm, where the supply of sugar is more than adequate ( Figure 2B), while that of oxygen is limited (Figures 3B and 3C).
Second, the FBA predicts that barley caryopses, unlike roots (Armstrong et al., 1994;Bailey-Serres and Voesenek, 2008;Miyashita and Good, 2008), neither accumulate Gaba nor invoke the Gaba shunt pathway. Our data demonstrated that the Gaba-T enzyme activity was marginally induced by hypoxia ( Figure 4A), while the level of SSA (its reaction product) rose in the interior (hypoxic) region of the endosperm (Figure 3). Nevertheless, the Gaba-T V max remained very low (see Supplemental Table 1 online), and the concentration of Gaba was not affected by variation in the supply of oxygen ( Figure 4A). Thus, it is hardly likely that the Gaba shunt can carry any significant flux in the endosperm. This might represent an endosperm-specific adaptation to the situation where oxygen deficiency is long-lasting rather than transient (Rolletschek et al., 2004). Third, the suggestion is that in the interior (hypoxic) region of the endosperm, the assimilation of ammonium operates via GDH rather than GS/GOGAT. GDH is among the genes upregulated by hypoxia (Bailey-Serres and Voesenek, 2008;Branco-Price et al., 2008), and its product delivers a more energy efficient assimilation than GS/GOGAT provides (Qiu et al., 2009). In the central hypoxic zone of the endosperm, ammonia is used to generate Glu via GDH ( Figure  5B), while in its more peripheral (oxygenated) regions, GDH does not assimilate ammonia, but rather releases it. The measured levels of Glu, Ala, and ammonia (Figure 3; see Supplemental Table 1 online) are in line with this prediction.

Implications of the Existence of Metabolic Compartmentation for Seed Filling
Our data demonstrate that the Suc transported to the caryopsis moves via the well-oxygenated nucellar projection toward the central endosperm (see Supplemental Movie 2 online), where an appreciable level of hypoxia prevails. In central endosperm, a significant level of localized Ala-AT activity is established, and this synthesizes Ala ( Figure 2C) to avoid pyruvate accumulation/ fermentation and to provide additional ATP. Eventually the nitrogen use efficiency is also improved due to the reutilization of ammonia, which is released by deterioration of adjacent nucellar tissues . Multiple plasmodesmata interconnect the cells (see Supplemental Figure 3 online), so that the energy efficient movement of Ala along a concentration gradient toward the more peripheral endosperm cells is enabled. Since the peripheral cell is not in a hypoxic state, Ala-AT can act to reconvert Ala back to pyruvate ( Figure 4B).
The proposed metabolic compartmentation of the endosperm is also mirrored by the structure of the tissue, since the central part of the endosperm is composed largely of large cells, whereas those at the periphery are small (see Supplemental Figure 3A online). Although possibly coincidental, this pattern of cell size corresponds to the known promoting effect of hypoxia on cell expansion/elongation (Armstrong et al., 1994;Bailey-Serres and Voesenek, 2008).
Hypoxia is commonly assumed to stunt vegetative growth and compromise cell metabolism (Armstrong et al., 1994;Bailey-Serres and Voesenek, 2008), but it now appears to be somewhat beneficial in certain regions of the caryopsis. Constraints are not uniformly experienced throughout a caryopsis, as a result of the architecture of the organ and the nonuniform availability of Suc/oxygen. These imbalances, as demonstrated by this experimental and modeling approach, can be balanced by local metabolic adjustments. We suggest that the metabolic compartmentation of the starchy endosperm provides a mechanism to ensure metabolic flexibility and eventually contributes to the high carbon conversion efficiency shown by the cereal starchy endosperm (Alonso et al., 2011).

Plant Growth and Stable Isotope Labeling
Barley plants (Hordeum vulgare cv Barke and Golden Promise) and transgenic Jekyll downregulated barley plants were grown under standard greenhouse conditions at 188C with 16 h of light and a relative air humidity of 60%. Determination of developmental stages for developing barley caryopses and tissue isolations were performed as described by Radchuk et al. (2006). Stable isotope labeling was performed using previously used 13 C feeding procedures (Singh and Jenner, 1983;Ekman et al., 2008). Stems were cut 5 cm below the ear and placed in nutrient solution containing quarter-strength Murashige and Skoog medium, 10 mM Gln, 10 mM Asn, 2 mM MES buffer, pH 6.0, and 100 mM UL-13 C 12 -Suc (Omicron Biochemicals).

NMR Experiments
The NMR experiments were performed on a 17.6 T wide bore (89 mm) superconducting magnet (Bruker BioSpin) with a maximum gradient strength of 1 mT m 21 using a custom built double resonant 13 C/ 1 H-NMR coil (inner diameter 5 mm).
For metabolic 13 C imaging, the geHMQC pulse sequence with phase encoding gradients for spatial localization was used (see Supplemental Figure 2A online). An additional frequency selective water saturation scheme (VAPOR; Tká c et al., 1999) was built into the sequence for efficient suppressing of the water signal. The geHMQC sequence consisted of a slice 908 selective excitation pulse on the protons, followed by phase encoding gradients in the remaining two spatial dimensions. After t = 1/(2 1 J CH ), a 908 hard pulse was applied on 13 C channel, which creates multiple quantum coherences (MQCs). The following 1808 pulse on the protons interchanges the MQCs and refocuses the chemical shift evolution. The last 908 pulse on 13 C converts MQCs back to detectable magnetization. The MQC gradients G1, G2, and G3 were used in a ratio of 2:2:1 to select a coherence pathway and to suppress signal from protons, which are connected to 12 C. This MQC gradient combination allows refocusing 50% of the 1 H magnetization coupled to 13 C spins. The spoiler gradients (Sp) were used to dephase magnetization, which results from imperfections of the 1808 pulse. A total of 2048 complex points were acquired at a bandwidth of 20 kHz with 13 C decoupling (Malcolm Levitt decoupling), which suppresses the 1 J CH coupling and improves the SNR. Further experiment parameters are as follows: TR = 1.5 s, TE = 8.2 ms, t = 2.9 ms, number of scans = 800 (hanning-weighted k-space acquisition scheme), FOV = 8 3 8 mm 2 , spatial resolution 0.4 3 0.4 mm 2 , slice thickness = 2.5 mm, MQC gradient strength (G1: G2: G3) = (0.4: 0.4: -0.2) T/m, MQC gradient duration = 1 ms, water suppression bandwidth = 500 Hz, duration for one experiment = 20 min.
Image reconstruction and analysis was done with in-house written software (Java and MATLAB 7.9.0; The MathWorks). The data were filtered with an exponential function (time constant = 3.3 ms) in the time domain to improve the SNR. For the dynamic uptake study of 13 C-Suc, the data for every frame were averaged from seven consecutive experiments. The data from these frames were linear interpolated to create additional frames for a smoother video (see Supplemental Movie 2 online).

Experimental Treatment of Samples Used for Biochemical Analysis
To study the effect of distinct light and oxygen regimes on metabolism of developing caryopses (12 to 16 d after fertilization), intact ears were kept in either light (600 mE) or darkness (enclosed with black paper) for 6 h. In addition, some dark-incubated ears were aerated with gas mixtures containing either reduced (2.1 kPa) or elevated oxygen levels (42 kPa; balanced by N 2 via a multigas controller). After 6 h, caryopses were quickly removed and immediately frozen in liquid nitrogen. To analyze metabolite pattern in inner versus outer regions of endosperm, freshly harvested caryopses were snap-frozen in liquid N 2 and freeze-dried. Subsequently, the outer pericarp layers were peeled off and the endosperm was separated into inner and outer parts by use of a razor blade.
For analysis of other metabolites, frozen caryopses were grinded as above and extracted with chloroform/methanol/water (1:1:3) according to Soga et al. (2002). The analysis of soluble sugars was performed by ion chromatography using pulsed amperometric detection (Rolletschek et al., 2005). Ammonia was measured by ion chromatography coupled to conductivity detection (ICS-3000; Dionex); separation was done using the IonPac CS12A (4 3 250 mm) in the isocratic mode with 10 mM H 2 SO 4 .
The remaining metabolites were measured by LC-MS using three different methods. Ion chromatography (ICS-3000) coupled to an electrospray ionization triple quadrupole mass spectrometer (API 4000) in negative mode was used for compounds listed in Supplemental Table 2 online. The separation was performed on an IonSwift MAX-100 (1 3 250 mm; Dionex) with a constant column temperature of 408C and a column 10 of 14 The Plant Cell flow of 150 mL/min. The suppressor was set to a value of 38 mV. As sodium hydroxide eluent, we used the following gradient: t = 0 min (5 mM); t = 10 min (5 mM); t = 16 min (12 mM); t = 28 min (25 mM); t =32 min (100 mM); t = 38 min (100 mM); t = 42 min (5 mM); t = 56 min (5 mM). The other methods were performed with the coupling of LC (Ultimate 3000) and the same mass spectrometer. The hydrophilic interaction chromatography method with an aminopropyl column (Phenomenex Luna NH2; 250 mm 3 2 mm, particle size 5 mm) is based on Bajad et al. (2006). To obtain maximum separation of different compounds, the column was used in combination with the mass spectrometer in negative and positive mode and varying eluent gradients. The LC eluents are as follows: solvent A, 20 mM ammonium acetate + 20 mM ammonium hydroxide in 95:5 water:acetonitrile, pH 9.45; solvent B, acetonitrile. The gradients in the negative mode to determine the metabolites (listed in Supplemental Table 3 online) are as follows: t = 0, 85% B; t = 15 min, 0% B; t = 38 min, 0% B; t = 40 min, 85% B; t = 50 min, 85% B. The gradients in the positive mode to determine the metabolites (listed in Supplemental Table 4 online) are as follows: t = 0, 85% B; t = 15 min, 0% B; t = 28 min, 0% B; t = 30 min, 85% B; t = 40 min, 85% B. Data were normalized to plant milligrams of fresh weight and to an internal standard added during extraction ( 13 C-succinate; Cambridge Isotope Labs).

Determination of Ethanol
Samples were ground with mortar and pestle, and 1000 mL of methanol (100%, MS grade) was added. After shaking for 1 h at 108C, samples were cleaned using Vivaclear filters and filled into gas chromatography vials. Ethanol concentration in the extracts was measured by a gas chromatograph (model 2014; Shimadzu) equipped with a flame ionization detector. The column was Zebron ZB-1 with dimensions 3 mm, 60-m length, and 0.32-mm inner diameter. The following analysis parameters were used: injector temperature, 2008C; carrier gas, He; linear velocity, 35 cm/s; split ratio, 30; oven temperature program, 508C hold for 5 min, increase to 708C within 2min, followed by temperature rise to 2408C at 508C/min; detector temperature, 2508C. Limit of detection was determined as ;1 nmol ethanol/g fresh weight (based on additions of known standards).

FBA
FBA was applied to a recently established model of primary metabolism in the barley caryopsis (Grafahrend-Belau et al., 2009a). To elucidate the role of Ala accumulation during hypoxia, the model was extended by reactions involved in hypoxic metabolism (i.e., Ala export, mitochondrial Ala-AT, export of TCA cycle intermediates, and Gaba export). Assuming that efficient use of energy is the main objective of hypoxic metabolism, the minimization of ATP production per flux unit was used as the objective function for the hypoxic growth simulations. The objective was computed as a two-step optimization process, where the first step is to minimize ATP production and the second step is to minimize the overall intracellular flux. The biological assumption underlying the objective function is that the cells grow while (1) using the minimal amount of energy, resulting in energy conservation (first optimization step), and (2) achieving an efficient channeling of metabolites (second optimization step). The computation was performed by fixing the growth rate at 1.8 h 21 , a value representative for the central endosperm, and by adding the objective value of the first optimization as an additional constraint to the second optimization. With respect to the aerobic growth simulations, the maximization of biomass per flux unit was used as the objective function (Grafahrend-Belau et al., 2009a).
All simulations were performed by constraining the uptake rate for Suc between 0 and 8 mmol g 21 DW h 21 , a value experimentally determined for the developing grain of barley (Felker et al., 1984). Due to the lack of barley-specific data, the uptake rate for Asp and Glu was constrained between 0 and 4 mmol g 21 DW h 21 , based on experimental reports of soybean (Glycine max) seeds (Iyer et al., 2008;Allen et al., 2009,). In addition, Ala was allowed to be taken up under aerobic growth conditions by constraining the exchange flux between 0 and 6 mmol g 21 DW h 21 ; the latter value corresponds to the maximal hypoxic Ala accumulation rate. In accordance with the experimental results, the accumulation of the classical fermentation products (ethanol and lactate) was inhibited (ethanol) or restricted to a minimal value (lactate, 0.4 mmol g 21 DW h 21 ) under hypoxic growth conditions. To simulate growth under different O 2 supply, the O 2 uptake rate was fixed at representative values for each tissue (central endosperm, 3 mmol g 21 DW h 21 ; peripheral endosperm/pericarp, 10 mmol g 21 DW h 21 ). The latter value corresponds to respiration rates measured for whole caryopses (ranging between 10 and 13 mmol oxygen g 21 DW h 21 at the stage analyzed here; Tschiersch et al., 2011). The same biomass composition for both the central and the peripheral endosperm was used for modeling because starch is the major storage product all over the endosperm tissue at the stage analyzed here ) and accumulation of lipid and protein in aleurone occurs later in development (Neuberger et al., 2008).
To elucidate the in silico function of Ala-AT in response to varying O 2 supply, simulations were performed with the Suc uptake rate fixed at 8 mmol g DW 21 h 21 , the O 2 uptake rate varying from anaerobic to aerobic conditions (1 to 10 mmol g DW 21 h 21 ) and the maximization of biomass per flux unit chosen as the objective function. The remaining exchange fluxes were constrained as described above. Simulations as well as visualization of the resulting flux maps were performed using FBA-SimVis (Grafahrend-Belau et al., 2009b).

Generation of Oxygen Maps, Microsensor Analysis, and Micro-Computer Tomography
Oxygen maps are based on O 2 microsensors studies (Rolletschek et al., 2004). Spatially highly resolved O 2 concentration profiles were derived from microsensor insertions along various transects across the caryopsis and drawn as isolines of O 2 concentration. The oxygen influx was calculated from the gradient in oxygen concentration across the corresponding region of endosperm. Oxygen measurements in Jekyll plants were done according to Rolletschek et al. (2004).
Structural models of barley caryopsis (12 d after flowering) were acquired by applying the manufacturer's protocols and a Skyscan1172 instrument.

Supplemental Data
The following materials are available in the online version of this article.
Supplemental Figure 1. Characteristics of the Barley Starchy Endosperm during Grain Filling.
Supplemental Figure 2. The geHMQC Pulse Sequence and in Vitro Spectra.