• Open Access

A joint transcriptomic, proteomic and metabolic analysis of maize endosperm development and starch filling

Authors


* Correspondence (fax 33 (0)1 69 15 34 24; e-mail jean-louis.prioul@u-psud.fr)

Summary

The maize endosperm transcriptome was investigated through cDNA libraries developed at three characteristic stages: (i) lag phase [10 days after pollination (DAP)]; (ii) beginning of storage (14 DAP); and (iii) maximum starch accumulation rate (21 DAP). Expressed sequence tags for 711, 757 and 384 relevant clones, respectively, were obtained and checked manually. The proportion of sequences with no clear function decreased from 35% to 20%, and a large increase in storage protein sequences (i.e. 5% to 38%) was observed from stages (i) to (iii). The remaining major categories included metabolism (11%–13%), transcription–RNA processing–protein synthesis (13%–20%), protein destination (5%–9%), cellular communication (3%–9%) and cell rescue–defence (4%). Good agreement was generally found between category rank in the 10-DAP transcriptome and the recently reported 14-DAP proteome, except that kinases and proteins for RNA processing were not detected in the latter. In the metabolism category, the respiratory pathway transcripts represented the largest proportion (25%–37%), and showed a shift in favour of glycolysis at 21 DAP. At this stage, amino acid metabolism increased to 17%, whereas starch metabolism surprisingly decreased to 7%. A second experiment focused on carbohydrate metabolism by comparing gene expression at three levels (transcripts, proteins and enzyme activities) in relation to substrate or product from 10 to 40 DAP. Here, two distinct patterns were observed: invertases and hexoses were predominant at the beginning, whereas enzyme patterns in the starch pathway, at the three levels, anticipated and paralleled starch accumulation, suggesting that, in most cases, transcriptional control is responsible for the regulation of starch biosynthesis.

Introduction

The economic and nutritional value of the maize kernel is derived mostly from the endosperm, a starch-rich tissue that supports the embryo during germination. Thus, optimum kernel filling is one of the most highly regarded traits in maize breeding. Classical breeding approaches aimed at improving the quality and quantity of stored carbohydrates and proteins are currently complemented by the use of molecular markers and genomics, which promise to provide new and more efficient selection criteria. Quantitative trait locus approaches allow the identification of small genomic segments that account for the variation in quantitative traits. However, the identification of the candidate genes at the critical loci has so far limited this approach, for example as discussed in the context of starch metabolism (Prioul et al., 1999), and a knowledge of the interconnection of metabolic and developmental processes is lacking.

Cereal kernel development has been analysed thoroughly from cytological (Olsen, 2001), genetic and epigenetic (Chaudhury et al., 2001) points of view. Development can be divided into three partially overlapping stages: (i) the ‘lag phase’, occurring from fertilization to 12 days after pollination (DAP) and consisting of three substages (coenocytic, cellularization and differentiation) that result in rapid expansion of the endosperm; (ii) the ‘kernel filling’ stage, ranging from 12 to 40 DAP and corresponding to the conversion of imported sucrose and amino acids into starch and storage proteins which accumulate in the endosperm; and (iii) the ‘desiccation–maturation stage’, occurring from 40 to 70 DAP and leading to the mature kernel. The metabolic steps from sucrose import to the synthesis of amylose and amylopectin, the starch granule components, are well established, especially in the maize kernel, where a collection of mutants affecting filling have enabled the determination of all of the key steps (Schultz and Juvik, 2004). However, the interrelationships with developmental and energetic processes remain to be clarified. High-throughput genomic and post-genomic approaches are now providing new tools for a better understanding of the genetic and biochemical networks operating during kernel development. The first full-length cDNA libraries of maize endosperm (Lai et al., 2004) were constructed from early and middle mature stages (4–6 and 7–23 DAP), leading to the creation of a 5504 unigene set comprising 2911 cDNAs with an assigned function. However, the earliest microarray profiling on maize endosperm dealt with a comparison of RNA expression in the endosperm vs. immature ear or other organs (Cho et al., 2002; Fernandes et al., 2002). The same slide comprising 5000 expressed sequence tag (ESTs) was also used to analyse the effect of water deficit at 9 DAP (Yu and Setter, 2003). Lee et al. (2002), comparing 900 metabolism-targeted cDNAs during kernel development from 5 to 10 DAP and embryo development from 10 to 45 DAP, established coordinated gene expression in the tricarboxylic acid (TCA) cycle and glycolysis, which showed a large decrease after 20 DAP, and in fatty acid biosynthesis. Similar developmental studies performed in barley grain on 1421 cDNAs showed a strong relationship between the accumulation of storage product and the transcript level of respiratory metabolism genes (Sreenivasulu et al., 2004). On a larger scale, Zhu et al. (2003), starting from 21 000 rice cDNAs, focused on 491 genes belonging to three major pathways of nutrient partitioning, and identified 269 genes preferentially expressed during grain filling (0–21 DAP). Proteomics has also been used successfully to characterize gene product accumulation in grain from rice (Koller et al., 2002), barley (Finnie et al., 2002; Ostergaard et al., 2004), wheat (Vensel et al., 2005) and, within the European Union (EU)–Zeastar programme, maize (Méchin et al., 2004, 2007).

Although metabolic pathways for starch and storage proteins during the accumulation period have been analysed thoroughly in a number of papers, the developmental aspects, especially at the transcript level, are less well documented and are limited to a few enzymes or proteins (Ober et al., 1991; Carlson et al., 2002). Changes in the flux pattern of central carbohydrate metabolism were analysed during kernel development, considering the kernel as a single compartment (Ettenhuber et al., 2005), but it is known that starch accumulation is polarized from distal layers to the base of the kernel (Brangeon et al., 1997). The relationship between enzyme activity and the corresponding mRNA and protein enzyme content has been examined for some key enzymes, such as adenosine diphosphate (ADP)-glucose pyrophosphorylase (AGPase) (Prioul et al., 1994) and pyruvate orthophosphate dikinase (PPDK) (Chastain et al., 2006), but, to our knowledge, no comparison of the developmental expression of the endosperm storage pathways has been made encompassing biochemistry to proteomic and transcriptomic measurements. Taking advantage of the earlier work of Méchin et al. (2004, 2007) on the endosperm proteome, the aim of this article is to link gene expression, enzyme content, enzymatic activities and some key enzyme localizations to metabolite variations, and to provide an overview of the regulation of carbohydrate metabolism during endosperm development in relation to general kernel metabolism.

Results

Changes in transcript profiling during development

An analysis of the functional distribution of endosperm cDNAs was performed for three kernel libraries of the F-2 genotype obtained at 10, 14 and 21 DAP. From the 1849 sequences analysed, 711 were obtained at 10 DAP, 757 at 14 DAP and 381 at 21 DAP. Each of the automated annotations was checked in order to verify the naming consistency and to ensure that the same classification of functional categories was used as for the proteomic data (Méchin et al., 2007).

The most striking variation in transcript abundance was observed for storage protein transcripts, which changed from 5% at 10 DAP to 32% and 38% at 14 and 21 DAP, respectively (Figure S1, see ‘Supporting information’). Excluding storage protein transcripts and so-called function not yet clearly identified (NYC) and non-identified (NI) transcripts (35% to 20%, from 10 to 21 DAP), the largest individual category was metabolism, with the relative proportion of transcripts increasing from 21% to 28% during development (Figure 1). The protein synthesis (including RNA processing) and transcriptional function categories together constituted a major group, the abundance of which remained approximately constant at about one-third of the total (32%, 27% and 29% of transcripts for 10, 14 and 21 DAP, respectively). The protein destination category (12%–14% of transcripts) mainly consisted of degradation proteins associated with the ubiquitin–proteasome complex, which represented 53% of the transcripts in this category at 14 DAP. The cellular communication/signal transduction-related transcripts were at 15%–16% at 10 and 14 DAP, and later dropped to 7%, but this variation hid a large increase in the proportion of kinases, from 18% at 10 DAP to 34% and 42% at 14 and 21 DAP, respectively (Table S1, see ‘Supporting information’). Cell rescue–defence/cell death and ageing represented a significant but almost constant category, amounting to 6%–9%, with transcripts commonly involved either in defence or detoxification reactions. The distribution of the most numerous categories was highly significantly different between the three stages (χ2 = 163.3, d.f. = 21, P < 0.001). The largest divergence occurred between the 10-DAP and 21-DAP stages (main functional categories were protein synthesis and cellular communication at 10 DAP vs. metabolism and protein destination at 21 DAP), whereas the 14-DAP stage was intermediate, with a high proportion of metabolism and cellular communication categories.

Figure 1.

Distribution into functional categories of endosperm transcripts from maize sampled at 10, 14 and 21 days after pollination (DAP). Non-identified (NI), not yet clearly identified (NYC) and storage protein transcripts were omitted (see Figure S1 and Table S1 for the complete set of transcripts). Inset: proteome distribution in the same categories at 14 DAP (from Méchin et al., 2007).

The distribution of metabolism transcripts showed significant variation over the time course considered (χ2 = 34.5, d.f. = 21, P < 0.05). The energetic metabolism obtained by summing together glycolysis, the TCA cycle, C4 photosynthesis (C4P) (PPDK), the hexose monophosphate (HMP) pathway, fermentation and energy accounted for 32%, 25% and 37% at 10, 14 and 21 DAP, respectively (Figure 2); TCA cycle and fermentation transcripts were not detected at 21 DAP. The amino acid metabolism category increased from 11% at 10 DAP to 16%–17% at 14 and 21 DAP. C4P transcripts appeared at 21 DAP, all belonging to the same enzyme: PPDK. Starch metabolism transcripts tended to decrease at 21 DAP, but the cumulative proportion of both starch and carbohydrate metabolism was approximately constant, accounting for 21%, 15% and 23% at 10, 14 and 21 DAP, respectively. The nucleotide metabolism transcripts decreased slightly over the time course, but their proportion was always higher than 10%. Secondary metabolism compounds were also well represented, and showed large variations from a 12% minimum to a 23% peak at 14 DAP.

Figure 2.

Distribution into functional subcategories of metabolism-related transcripts from maize endosperm sampled at 10, 14 and 21 days after pollination (DAP). Inset: proteome distribution in the same subcategories at 14 DAP (data from Méchin et al., 2007). C4p, C4 photosynthesis; HMP, hexose monophosphate; TCA, tricarboxylic acid.

Carbohydrate and respiratory metabolism

As starch is valuable and forms the largest component of the mature endosperm, the genes related to carbohydrate metabolism were examined in more detail. Starch and soluble carbohydrate metabolism transcripts represented 15%–23% of the sequenced transcripts (Figure 2) and 15% of the processed proteins (Figure 2, inset), whereas the proportions of respiration–energy metabolism transcripts and proteins were 25%–39% and 47%, respectively. Placing the transcriptomic and proteomic information on a metabolic scheme for carbohydrate and respiration (Figure 3) showed that 41 of the 55 enzymes involved were detected either as transcripts (33) or proteins (34), or both (26). Time courses of transcript and protein were in agreement, and transcript levels increased in anticipation of protein abundance. Enzymes of the two respiratory pathways (i.e. glycolysis + TCA cycle and HMP pathways) were represented. For carbohydrate metabolism, all the enzymes of the starch pathway, from sucrose cleavage to amylose and amylopectin synthesis, were found. The cell wall metabolism enzymes uridine diphosphate (UDP)-glucose dehydrogenase and UDP-glucosyl transferase were also present. Sucrose-phosphate synthase was detected as a transcript at an early stage (10 DAP). This expression may be related to invertase, which is active at this stage in all three forms: cell wall, vacuolar and cytosolic (see below). This probably indicates a re-synthesis of sucrose to provide a substrate for sucrose synthase (Susy), which is a mandatory step in starch and cellulose synthesis. Similarly, HMP enzymes, such as phosphogluconate dehydrogenase and transketolase, were detected at an early stage, which is consistent with the involvement of this pathway in the synthesis of the nucleic acid components necessary for cell development. By contrast, the presence of transcripts for glycolysis extended to 14 and 21 DAP, whereas TCA enzyme transcripts were notably absent at 21 DAP, concurrent with the decreasing trend noted at the protein level. This raises the question of the origin of adenosine triphosphate (ATP) needed for starch and storage protein accumulation, and may be related to the hypoxic conditions in the endosperm at this stage, as discussed later. PPDK (C4P metabolism) was detected in the proteome data at 14 DAP, but only at 21 DAP in the transcript data, a result that may reflect low transcript abundance. PPDK reversibly converts pyruvate and ATP into phosphoenolpyruvate and pyrophosphate. This typical enzyme of the C4 cycle in leaves has been reported in cereal grains for a long time (Aoyagi and Bassham, 1984).

Figure 3.

Time course of transcript and protein at 10, 14 and 21 days after pollination (DAP) for the respiration and starch metabolism pathways. T, transcript data: black/white squares, presence/absence of transcript in the cDNA libraries; black to grey squares, amount of transcripts by quantitative polymerase chain reaction (Q-PCR). P, proteomic data from two-dimensional gels: green, black and red squares, low, medium and high protein quantities, respectively; black square, proteins identified at 14 DAP but not reliable for quantification. 1cw, cell wall invertase Incw2; 1n, neutral invertase; 2, sucrose synthase; 3, hexokinase; 4, glucose-6-phosphate isomerase; 5, UDP-glucose pyrophosphorylase; 6, UDP-glucose dehydrogenase; 7 and 7′, phosphoglucomutase; 8 and 8′, AGPase; 9 and 9′, granule-bound starch synthase II-1 (GBSSII-1); 10, starch branching enzyme (SBE); 11, isoamylase; 12, pullulanase; 13, sucrose phosphate synthase; 14, sucrose phosphatase; 15, fructose-1,6-bisphosphatase; 16, fructokinase-II; 16′, diphosphate-fructose-6-phosphate-1-phosphotransferase; 17, fructose bisphosphate aldolase; 18, triosephosphate isomerase; 19, phosphoglycerate dehydrogenase; 20, phosphoglycerate kinase-3; 21, phosphoglycerate mutase (bis); 22, phosphoglycerate mutase; 23, enolase; 24, pyruvate kinase; 25, pyruvate orthophosphate dikinase; 26 A, pyruvate dehydrogenase; 26 B, dihydrolipoamide S-acetyltransferase; 27, acetyl-CoA carboxylase; 28, phosphoenolpyruvate carboxylase; 29, citrate synthase; 30, aconitate hydratase = citrate hydrolyase; 31, isocitrate dehydrogenase; 31′, NADP-specific isocitrate dehydrogenase; 32, oxoglutarate dehydrogenase; 33, dihydrolipoamide dehydrogenase precursor; 34, dihydrolipoamide S-succinyltransferase; 35, succinyl-CoA ligase (synthetase); 36, succinate dehydrogenase; 37, fumarase; 38, malate dehydrogenase; 39, glucose-6-phosphate 1-dehydrogenase; 40, phosphogluconate dehydrogenase-6; 41, ribose-5-phosphate epimerase; 42, ribulose phosphate 3-epimerase; 43, transketolase; 44, transaldolase; 45, starch phosphorylase; 46, endo-1,4-β-glucanase; 47, pyruvate decarboxylase; 48, alcohol dehydrogenase; A, adenylate translocator; B, glucose-6-phosphate translocator; OAA, oxaloacetate.

Starch metabolism: compared expression at the transcriptomic, proteomic and enzyme activity levels

The activity of starch metabolic enzymes was measured at 10, 14, 21, 30 and 40 DAP in parallel with the changes in the substrate or product concentrations. Data covering protein accumulation over the same period of time are available from Méchin et al. (2007). Proteins were identified on 14-DAP two-dimensional gels and quantified at the other stages by comparison with the 14-DAP stage, which corresponds to the end of the lag phase and the start of starch accumulation. Therefore, some enzymes accumulating poorly at 14 DAP [such as granule-bound starch synthase (GBSS) and debranching enzyme (DBE)] are missing from the quantitative proteomic data set. In addition, some enzyme monomers that have high relative molecular mass (Mr) and/or basic isoelectric point (pI), which are poorly resolved, were excluded from the quantification. Quantitative reverse transcriptase-polymerase chain reaction (RT-PCR) was used to assess gene expression.

Two main patterns were observed (Figure 4). Firstly, the three invertase activities (cell wall, cytosolic and vacuolar) reached a maximum at 21 DAP, declining thereafter. The concentrations of their substrate (sucrose) and products (glucose and fructose) varied accordingly. Secondly, all the enzymes of the starch pathway had low or undetectable activity before 14 DAP. Activity progressively increased to reach a maximum at 30 DAP and then declined. This time course is consistent with the starch and amylose accumulation rates, which reached a maximum at 30 DAP and declined thereafter. Similarly, ADP-glucose, the AGPase substrate, had a concentration which varied with the enzyme activity. Glucose-6-phosphate and fructose-6-phosphate, the main primary substrates of glycolysis, declined after 21 DAP, which is in agreement with the relative decrease observed at the proteome level for glycolytic enzymes during the late grain-filling period (Méchin et al., 2007). The relatively constant level of UDP-glucose may be the result of two biosynthetic pathways occurring successively: from glucose-6-phosphate or fructose-6-phosphate through the sequential action of three enzymes, namely phospho-glucose isomerase, phosphoglucomutase and UDP-glucose pyrophosphorylase, during the first 20 DAP, and through Susy during the starch accumulation period. The high glucose-6-phosphate/fructose-6-phosphate ratio during the early phase is indicative of a phospho-glucose-fructose isomerase functioning near its thermodynamic equilibrium. The difference in the time course for amylose and starch suggests an imbalance in the amylose/amylopectin ratio during development.

Figure 4.

Comparison of the time course of transcriptomic and proteomic data with the corresponding biochemical products and activity of the enzymes of the starch pathway from 10 to 40 days after pollination (DAP). Sucrose, glucose, hexoses, phosphorylated sugars, nucleotide sugars and starch concentration were expressed as milligrams per gram of endosperm fresh weight. Amylose was expressed as a percentage of starch; starch was expressed as a percentage of fresh weight. Enzyme activities were expressed as nanomoles per second per gram of product per endosperm fresh weight, except for granule-bound starch synthase (GBSS) and soluble starch synthase (SSS), analysed by zymograms. Protein quantification was performed by image analysis of two-dimensional gels, and was expressed in arbitrary units (Méchin et al., 2007). The star and the arrow indicate the approximate position of the endosperm-specific Sh2 subunit of AGPase. The mRNA content analysed by reverse transcriptase-polymerase chain reaction (RT-PCR) was expressed as a percentage of total RNA.

Time courses for transcript accumulation generally showed parallel variations to the corresponding enzyme activity, and two main types of profile were discernible. When several transcripts could be quantified for a given enzyme, they exhibited similar [Susy, AGPase, starch branching enzyme (SBE)] or slightly different [soluble starch synthase (SSS)] time courses, but one transcript was always predominant over the others. In most cases, the maximum transcript levels were reached earlier than enzyme activity or protein accumulation. Few data were available for protein accumulation, but here again the time course was consistent with the activity pattern.

In situ localization of carbohydrate and starch metabolism enzymes

Enzyme activities were directly located on tissue sections from frozen kernel samples using enzyme-specific substrates and optimum pH, i.e. ADP-glucose and pyrophosphate for AGPase, UDP and sucrose at neutral pH for Susy, and sucrose at acid pH for acid invertases. As both cell wall and vacuolar invertases have an acidic optimum pH, they could only be distinguished on thin sections after resin embedding. With such preparations, most of the labelling occurred in the basal endosperm tissue layer up to 21 DAP and earlier in the pericarp, but, as some background labelling was also observed in the control endosperm and vascular tissue (data not shown), localization was checked using specific cell wall invertase antibodies (Figure 5, top panel).

Figure 5.

Top panel: immunolocalization of cell wall invertase (Incw) at 21 days after pollination (DAP) using the VECTOR® red labelling system: (a) non-immune serum control; (b) INCW-specific antibodies. Bottom panel: in situ activity of AGPase (c) and sucrose synthase (d) in kernels at 14, 21 and 40 DAP. Activity was visualized with blue staining (nitroblue tetrazolium reduction).

The tissue localization of AGPase (Figure 5c) and Susy (Figure 5d) activity showed large developmental changes: labelling was mainly located in the maternal tissues, in the nucellus and then in the pericarp up to 14 DAP. Afterwards, it increased massively, firstly in the outer endosperm (21 DAP) and secondly in the basal zone (40 DAP). This polarization of activity is consistent with the well-established progression of starch accumulation in the kernel, which is also shown by the yellow colour of the endosperm in cross-sections representing flint regions (Figure 5). Starch accumulates from the distal part of the endosperm towards the basal zone where sucrose enters from maternal tissues. Thus, the localization of the activity marked the active starch-synthesizing zones. The highly similar pictures for AGPase and Susy are fully consistent with the cooperative contribution of both enzymes to starch synthesis.

Discussion

Transcriptome and proteome

Large-scale analyses of gene expression can provide information on the regulation of gene networks underlying metabolic and cellular processes. Transcript abundance primarily reflects the transcriptional control of gene expression and has been shown in many instances to be poorly related to protein abundance (Gygi et al., 1999; Watson et al., 2003). Proteins are the actual cellular effectors. Their abundance depends on the balance between translation and degradation which, in turn, can be influenced by various co- and post-translational processes. Comparison of the distribution of transcripts and proteins in the different functional categories at a given developmental stage can shed light on the relative importance of these processes. Integration of the temporal dimension may help to reveal sets of co-regulated gene products and to decipher the respective roles of transcriptional and (co- and post-) translational control on elements of the networks.

To date, few investigations have combined transcriptomic and proteomic data. The most recent reports have focused on small numbers of proteins in special situations, such as seed protein processing (Higashi et al., 2006) or microspore-derived embryo development (Joosen et al., 2007). The transcript analysis in this article deals with several hundred clones from cDNA libraries at three critical stages of endosperm development. As expected, the major changes in the transcriptome between the three stages involve the increased proportion of storage protein transcripts, but also the decrease in NYC and NI transcripts, suggesting that a large number of endosperm-specific functions are yet to be deciphered at the early stage. A proteome map consisting of 632 proteins has been established previously for the 14-DAP endosperm of the F-2 genotype in the same EU–Zeastar programme as the transcriptome analyses (Méchin et al., 2004), with the protein identifications updated recently (Méchin et al., 2007). The distribution of the functional categories – excluding storage proteins and NYC and NI transcripts – of the 10-DAP transcriptome is closer to that of the 14-DAP proteome, revealing a lag phase between gene transcription and protein accumulation. The proportion of sequences in the metabolism category is higher in the proteome data than in either transcriptome data set (40% vs. 21%–28%). This observation, associated with the high proportion of isoforms for a given function generally observed in proteome data, suggests that a number of post-translational modifications are possible for these accumulating proteins.

The combined protein synthesis and RNA processing category is a major constituent of the transcripts throughout the time course. RNA processing proteins are absent in the proteome analysed at 14 DAP, probably indicating high turnover with low accumulation. The RNA processing transcript proportion increases at 14 and 21 DAP, whereas, at 10 DAP, the high proportion of protein synthesis-related transcripts (87 of 105) appears to precede the accumulation of proteins of this category at the early to mid developmental stages (Méchin et al., 2007). These observations may be related to the increasing translation and protein synthesis activity needed to support zein production, beginning at about 14 DAP (Méchin et al., 2007). The protein destination category is also a major category over the time course, with the proportion of transcripts devoted to degradation increasing up to 14 DAP and decreasing afterwards (44%, 53% and 33%, respectively). This trend appears to be consistent with the maximum accumulation of proteins involved in degradation processes at the early stages (Méchin et al., 2007). The relatively high proportion of expressed genes devoted to degradation over the time course is indicative of important protein turnover during endosperm development, with a maximum at 14 DAP. Although it is not surprising to see this transition from the lag phase to the accumulation phase when the set of enzymes must be renewed to enable new functions, it is less expected during the phase in which the starch accumulation rate is constant. Of the transcripts coding for protein-folding genes, another subcategory of protein destination, it is tempting to associate the progressive and large accumulation of one specific protein, disulphide isomerase (gene index identification for protein: 1709619), up to 30 DAP with folding of the most abundant protein, i.e. zein. Transcription-related transcripts are the most abundant at 21 DAP, which may be related to the preparation of the maturation phase, which progresses after 30 DAP. Transcripts for defence (antifungal and antimicrobial proteins, etc.) and detoxification enzymes against reactive oxygen species (ROS) and xenobiotics are always present, and the abundance of proteins identified in this category at 14 DAP reaches a maximum during the earliest stages and diminishes afterwards. The presence of an anti-ROS protein at this early stage can be explained by the large oxygen availability, as opposed to the hypoxic situation which prevails after 14 DAP (Rolletschek et al., 2005; Méchin et al., 2007). However, superoxide dismutase (SOD) transcripts have been reported at 28 DAP, and their expression is enhanced by xenobiotics (Mylona et al., 2007). It has been suggested that this expression may be related to seed maturation. This would be consistent with the presence of metallothioneins and glutathione S-transferase transcripts at 21 DAP, which are also associated with this process (Table S1, see ‘Supporting information’). The communication/signal transduction functional category is represented more by transcripts (12%–15%) than by proteins (approximately 1%) at 14 DAP. A probable explanation is that these transcripts, which mostly code for kinases and transduction chain elements, are not detected on two-dimensional gels because of their low abundance.

We investigated the metabolism category in more detail in order to examine the relationships between energy metabolism and storage compound synthesis. The decrease in TCA cycle and mitochondrial transcripts to their disappearance at 21 DAP, and the appearance of the PPDK transcript, are consistent with the observations at the protein level, although the variation is more dramatic and contrasting for transcripts than for proteins. These data can be interpreted by the near absence of oxygen at these late stages, which impedes functioning of the mitochondrial electron transport chain because of the deficit in the final electron acceptor. In this context, starch synthesis, which does not need oxygen, is favoured. In addition, the increase in PPDK may act on the protein/starch balance at the end of starch filling, as discussed by Méchin et al. (2007). The largest diversity of starch and amino acid metabolism transcripts is observed at the beginning of the starch accumulation phase, preparing for the synthesis of the numerous enzymes needed for further storage product synthesis. The large increase in carbohydrate transcripts at 21 DAP corresponds to UDP-glucose metabolism enzymes (conversion into UDP-galactose or UDP-glucuronate) and to glucosidase, which is probably linked to cell wall formation. Accordingly, a relatively high UDP-glucose level is maintained at later stages. With regard to the proteome, especially at the mid stage, the proportion of transcripts for secondary compounds and nucleotide metabolism is surprisingly high compared with those for carbohydrate metabolism, suggesting that the contribution of these two categories of metabolism may need to be re-evaluated.

Starch synthesis: towards a synthetic view from transcripts, proteins and enzyme activities

The core reactions involved in the starch synthetic pathway are well known, especially in maize kernels, thanks to a series of mutants marking each step. Individual analyses of transcript and protein expression have been made by classical techniques, and post-translational modifications, possibly affecting in vivo activity, have been demonstrated for some enzymes, such as redox modification in AGPase and phosphorylation in SBE (Tetlow et al., 2004). The coordination of individual reactions has been tested genetically by assessing the effect of single mutations on the rest of the pathway. Taking advantage of the triploid structure of the endosperm, the dosage effect of mutant alleles of Sh1, Bt2, Ae1 (encoding SBEIIb) and Su1 (encoding DBE) on the content of soluble sugars and starch formation has been examined for various enzyme activities (Singletary et al., 1997). These workers found a direct relationship between the amount of starch produced and gene dosage, and observed, for all three dose mutants, except su1, large pleiotropic effects on the various enzyme activities, associated with an elevated level of soluble sugars. Similarly, sh2, bt2, ae1, du1 and waxy mutations resulted in both an increased transcription rate of the non-mutated genes of the starch pathway and an effect on zein synthesis, which suggested a coordinated transcriptional control of the genes involved in storage compound biosynthesis in maize endosperm (Giroux et al., 1994). In rice, the study of 269 genes belonging to different metabolic pathways that were coordinately up-regulated during grain filling, as shown by the monitoring of 21 000 microarrayed genes, has recently provided some clues on the possible mechanisms involved in coordinated expression (Zhu et al., 2003). Indeed, analysis of available promoter sequences indicated a significant over-representation of an AACA element in these grain-filling genes when compared with other genes, suggesting that this motif may play a role in the coordinated expression of various pathways during grain filling. A group of transcription factors potentially interacting with this element was identified.

The present approach in maize, which considers the simultaneous developmental changes at key steps of starch biosynthesis at the level of transcript and protein accumulation and enzyme activity, in relation to substrate and/or product concentration, provides information on the integrated functioning of the pathway during endosperm development and its relationship to the energy metabolism examined at the transcript and protein level.

Two main patterns are observed for carbohydrates: (i) hexoses present an early maximum followed by a decline; and (ii) starch, ADP-glucose and sucrose present a low level before 14 DAP and a progressive increase up to 30 or 40 DAP. The enzymes involved in the cleavage/synthesis or accumulation of these compounds display consistent activity patterns. Initially high invertase activities decline thereafter, whereas the activities of the core enzymes of the starch pathway (Susy, AGPase, GBSS, SSS and SBE) progressively increase from a low initial level to a maximum at 30–35 DAP, in parallel with the ADP-glucose content. These two distinct patterns may be interpreted with regard to the contribution of the metabolites to development, either as signalling molecules or as trophic components, during the lag and accumulation phases.

Invertases play a prominent role during the early phase, producing hexoses from imported sucrose, as inferred from the analysis of the miniature phenotype produced by the mn1 mutation. This mutant is characterized by a deficiency in both soluble and cell wall invertase activities, although genetic and molecular analyses show that the mutation only affects the cell wall gene Incw2. This gene is specifically expressed in the basal endosperm tissue cell layer, where carbohydrate transfer from the maternal placento-chalazal zone to seminal tissue takes place (Cheng et al., 1996). As cell wall and vacuolar activities vary in parallel in the mn1 mutant, a coordinated control of both isoforms has been suggested. We confirm that the cell wall form is the most active invertase in the endosperm, as noted previously (Cheng et al., 1996; Qin et al., 2004). Four genes encode cell wall invertase (Taliercio et al., 1999; Kim et al., 2000), but only Incw2 and, to a much lesser extent, Incw1 are expressed in the kernel. Accordingly, the main transcript detected is Incw2, with maximum expression appearing earlier than activity. However, the activity is relatively high at 10 DAP, but the transcript level is still low, suggesting that, at the early stage, few transcripts are required, and that, afterwards, the cell wall invertase inhibitor recently reported in maize kernels hinders activity (Bate et al., 2004). Invertase activity appears to be critical for early kernel development (Cheng and Chourey, 1999) because hexoses act as a positive signal on cell division (Vilhar et al., 2002) and α-tubulin transcript expression (Datta and Chourey, 2001). Accordingly, α-tubulin transcripts are most numerous at 14 DAP and proteins are mostly expressed in the lag phase (Méchin et al., 2007).

Susy and AGPase are two multimeric enzymes encoded by small multigene families with specific tissue expression. Sh1 for Susy and Bt2 and Sh2 for AGPase are the main genes expressed in the endosperm. Transcripts for two other Susy genes, Sus1 and Sus3 (now called Sus2), have been reported in the endosperm (Carlson et al., 2002). Indeed, very low expression of these two genes compared with Sh1 was found, but quantitative proteome data were only available for Sus1, showing maximum expression at 30 DAP, consistent with the enzyme activity pattern. The major transcripts of Sh1 peak at 21 DAP, which is earlier than for the enzyme activity, consistent with a predominant role for transcriptional control of the expression of Susy in the endosperm, as also reported in developing leaves (Nguyen-Quoc et al., 1990; Nguyen-Quoc, 1991).

For AGPase, the situation is even more complex as, in addition to the small and large subunits encoded by the Bt2 and Sh2 genes expressed in the endosperm cytosol, another subunit couple, Agp1 and Agp2, constitutes an AGPase expressed in both the endosperm and embryo amyloplasts (Denyer et al., 1996), but mainly in the embryo (Giroux and Hannah, 1994). Transcripts for each AGPase subunit were identified for the embryo forms, but no endosperm Bt2 transcript was detected in our cDNA library. The RT-PCR results show that Sh2 is the most highly expressed overall. A similar predominant expression of only one AGPase transcript (OsAPL3, a Sh2-like gene) in the middle phase of seed development has been reported from comparisons of the expression profiles of six cDNAs encoding the two small subunits and four large subunits in rice. Furthermore, OsAPL3 has been shown to be accumulated significantly in response to increasing levels of sucrose and abscisic acid (Akihiro et al., 2005). The present sucrose time course is consistent with a similar control of Sh2 expression in maize endosperm. Classical Northern blot analyses with Bt2 and Sh2 probes of the time course have shown previously that relative variations of both transcripts coincide, but slightly precede AGPase protein and activity levels (Prioul et al., 1994), a result consistent with the present observations. It has also been noted, as in this study, that the maximum activity coincides with the maximum rate of starch accumulation. The low level of the AGPase protein spot assigned to the large subunit of the embryo (plastid) form at 10 DAP, in the absence of any measurable activity, may originate from a contamination of the endosperm by maternal tissues, which are difficult to separate at this stage. Accordingly, up to 14 DAP, the AGPase transcripts and peptides detected at low levels are mostly located in the pericarp and nucellus (Brangeon et al., 1997). Afterwards, the contribution of the plastid form to the total AGPase in the endosperm has been estimated to be approximately 10% (Denyer et al., 1996). In any case, the major form becomes the typical endosperm AGPase from 14 DAP onwards. Our in situ measurements of AGPase activity show that it is concentrated in the endosperm, first in the distal part and then towards the base. At 40 DAP, starch synthesis is completed in the upper endosperm, as shown by the yellow zone marking flint starch where the enzyme is no longer active. The pattern is very similar for Susy, as is the time course for both enzyme transcripts, supporting the cooperative functioning of both enzymes and suggesting a similar transcriptional control.

Three groups of enzymes are involved in starch synthesis [the starch synthases (GBSS and SSS), SBEs and, finally, DBEs hydrolysing branched linkages], yielding the highly ordered and clustered structure of the amylopectin polymer. Pullulanase, another DBE group, is more likely to be involved in starch degradation. Of the starch synthases, the granule-bound form (GBSSI) is dedicated to the synthesis of amylose, which consists of long linear chains with a few branches. The very low transcript and protein levels at 10 and 14 DAP increase further as amylose accumulation increases, which may be indicative of a coarse transcriptional control, although some post-translational modification cannot be excluded because of some discrepancies between the amlyose accumulation rate and both transcript and protein levels. The SSSs are involved in the elongation of amylopectin chains. Four SSS isoforms (I, IIa, IIb, III = DU1) have been described, but most of the activity has been reported to rely on SSSI and SSSIII products (Cao et al., 1999). However, of the three transcripts characterized here, SSSIIa is the most expressed overall, which is consistent with our proteomic data. Thus, the contribution of SSSIIa to global activity in the endosperm needs to be re-evaluated. Three SBEs (I, IIa and IIb) have been described (Ball and Morell, 2003); the transcripts assigned to SBEII are expressed three times more highly than SBEI, whereas SBEIIa is only weakly expressed, mainly in the early phase. The activity and transcripts nearly disappear at 40 DAP, suggesting rapid inactivation or degradation of the enzyme.

The present functional approach strongly suggests that the primary control of enzyme activity leading to starch biosynthesis takes place in a coordinated manner at the transcriptional level. The explanation may be found by searching for common specific motifs in gene promoters when complete sequences become available, as performed in rice (Zhu et al., 2003). However, in the case of cell wall invertase, AGPase and SSS for example, the joint analysis of the transcript, protein (when available) and activity level suggests a potential control at the protein level. The various controls [post-transcriptional modifications or feedback loops, hexose signalling, as shown for Incw2 and kernel cell division (Cheng and Chourey, 1999; Vilhar et al., 2002), and protein–protein interactions producing metabolic channelling (Hennen-Bierwagen et al., 2008)] remain to be elucidated.

Experimental procedures

Growth conditions for field experiments

The maize flint line F-2 was first grown in Bergamo (Italy) from April to July 2001 to provide material for the cDNA endosperm libraries from 2 to 28 DAP. A second season in Clermont-Ferrand (France) from April to August 2002 was dedicated to parallel analysis of kernel development at five sampling dates from 10 to 40 DAP by transcriptomic, proteomic and biochemical methods. For this purpose, endosperm sets at the five sampling dates were ground in liquid nitrogen, and aliquots were used for the various experiments.

Transcriptomics procedures

RNA extractions were performed using a phenol–chloroform protocol (Baudo et al., 2006). Traces of genomic DNA were eliminated by a deoxyribonuclease (DNase) treatment. RNA was quantified spectrophotometrically, and ribonuclease (RNase) contaminations were tested by incubating the RNA for 5 min at 37 °C. The quality of the RNA was then checked by gel electrophoresis. Unamplified cDNA was used to construct the various libraries in the vector pBluescript II SK+ using standard procedures. All libraries were directionally cloned between the EcoRI and XhoI restriction enzyme recognition sites.

EST sequencing

Clones from each library were subjected to a single-pass sequence using the SK primer. The number of clones sequenced for each library was 10 × 96 clones for 10, 14 and 21 DAP; however, the number of readable sequences with a minimum quality threshold score of Phred 10 over a minimum length of 100 bp was 783, 805, and 408, respectively. Only clones meeting this minimum quality were taken forward for further analysis. Sequences were submitted to the European Molecular Biology Laboratory (EMBL) nucleotide sequence database [EMBL Acc: FM180579 to FM196445 for cDNA (15867)].

Real-time RT-PCR

Real-time RT-PCR experiments were performed using an ABI Prism 7000 Sequence Detection System (Applied Biosystems, Courtabeuf, France). Total RNA (10 µg per sample) was reverse transcribed with a polyadenylated-mRNA control. PCRs were prepared by adding 10 µL of reaction mix (containing 1 µL of diluted RT reactions and 100 pmol of gene-specific primer pair) (Table S2, see ‘Supporting information’) to 10 µL of 2 × QuantiTect-SYBR-Green-PCR master mix (Qiagen, Courtabeuf, France). The amplification was performed in a MicroAmp-optical 96-well reaction plate (Applied Biosystems). The PCR parameters were as follows: 15 min at 95 °C, 45 cycles of 95 °C for 30 s, 60 °C for 60 s and 72 °C for 60 s, with a final incremental 1 °C increase up to 95 °C. The purity of the products was analysed by agarose gel electrophoresis. Fluorescence readings were analysed by associated software. Human β-globulin amplification over a range of human genomic DNA concentrations was used as standard for quantification (Roche Applied Science, Meylan, France). The results were adjusted for the efficiency of reverse transcription and PCR amplification using the TRAF1 spike in each sample. Primers were designed from the sequence of clones, which were checked by re-sequencing. The annotation of these clones, detailed in Table S2, was updated during manuscript preparation. Primer specificity was also re-checked in silico.

Carbohydrate metabolism

Invertases and Susy

Soluble and insoluble acid invertase activities were measured, after extract desalting, by an enzyme coupling assay (Thévenot et al., 2005). The cell wall invertase activity was measured from the washed resuspended pellet, a method more effective than the commonly used 1 m NaCl extraction (Qin et al., 2004). Susy and neutral invertase activities were measured in the same extract (Thévenot et al., 2005), except that N-2-hydroxyethylpiperazine-N′-2-ethanesulphonic acid (Hepes)–NaOH buffer was replaced by 50 mm sodium phosphate buffer (pH 7) to avoid potential neutral invertase inhibition by Hepes–NaOH buffer. Sucrose, glucose and fructose were assayed from the boiled extract by the same coupled enzyme assay as used for enzyme activities.

ADP-glucose pyrophosphorylase and starch-synthesizing enzymes

The method used was derived from Sweetlove et al. (1996) with in gel assays for SSS and GBSS essentially as described. Native gel electrophoresis was performed using precast tris(hydroxymethyl) aminomethane (Tris)–glycine gels supplied by Invitrogen Ltd. (Cergy-Pontoise, France); staining of the gel for SSS activity was performed according to Kossmann et al. (1999). SBE activities were measured as described in Smith (1988).

Starch

Total starch content and percentage amylose were determined by assaying starch samples solubilized by treatment with perchloric acid (Hovenkamp-Hermelink et al., 1988). Sodium dodecylsulphate-polyacrylamide gel electrophoresis (SDS-PAGE) was carried out using ‘Novex NuPAGE’ precast gels and methods supplied by Invitrogen Ltd.

Activated metabolites

A freshly harvested endosperm sample (or aliquot of endosperm powder, stored at –80 °C, up to 200 mg) was suspended in 0.5 mL of 1.41 m perchloric acid. After centrifugation, the supernatant was neutralized with 5 m K2CO3. The supernatant collected after precipitation was used for metabolite measurements. Glucose-6-phosphate, glucose-1-phosphate and fructose-6-phosphate were assayed spectrophotometrically using a coupled enzyme assay. UDP-glucose and ADP-glucose were separated by high-performance liquid chromatography on a DIONEX column, and identified by high-performance anion exchange chromatography with pulsed amperometric detection.

In situ localization

Invertase, Susy and AGPase activities were evaluated on thin kernel sections (10 µm) from a cryomicrotome, using the same coupling enzyme principle as for in vitro measurement. The protocol was adapted from Sergeeva and Vreugdenhil (2002). Frozen kernels were fixed (paraformaldehyde 2% in phosphate buffer) to keep the enzymes in tissue, and washed several times to remove all soluble compounds. Blue staining was performed by nitroblue tetrazolium reduction after coupling enzyme reactions terminated by NADH production. Immunolocalization was performed on 2-µm sections, embedded in LR white medium, with a red fluorescent labelling system from Vector® (Paris, France), using two antibodies specific to cell wall invertase and vacuolar invertase, respectively (Kim et al., 2000).

Acknowledgements

The financial support of the European Union (EU), 6th PCR, through the Zeastar project (QLK3-2000-00020), The Cell Factory section, is gratefully acknowledged.

Ancillary