The Plant Journal (2004) 37, 539±553

doi: 10.1046/j.1365-313X.2003.01981.x

Transcript pro®les and deduced changes of metabolic pathways in maternal and ®lial tissues of developing barley grains Nese Sreenivasulu, Lothar Altschmied, Volodymyr Radchuk, Sabine Gubatz, Ulrich Wobus and Winfriede Weschke Institut fuÈr P¯anzengenetik und Kulturp¯anzenforschung (IPK), Correnstrasse 3, D-06466 Gatersleben, Germany Received 25 April 2003; revised 22 October 2003; accepted 4 November 2003.  For correspondence (fax ‡49 39482 5138; e-mail [email protected]).

Summary Different aspects of barley grain development have been studied in detail, but a more global analysis of gene expression patterns is still missing. We have employed macro arrays, containing 1184 unique sequences from 1421 barley cDNA fragments, to study gene expression pro®les in maternal and ®lial tissues of developing barley caryopses from fertilization to early storage phase. Principle component analysis (PCA) de®ned distinct expression networks in the pre-storage (0, 2, and 4 days after ¯owering (DAF)) and early storage phase (10 and 12 DAF). During an intermediate phase (6 and 8 DAF), PCA visualizes a dramatic reprogramming of the transcriptional machinery. In maternal tissues, a large set of protein-mobilizing enzyme mRNAs, together with upregulated lipid-mobilizing enzyme and downregulated reactive oxygen species (ROS)-scavenging enzyme genes, suggests mobilization of stored compounds and programmed cell death (PCD). In the ®lial tissue fraction, a set of genes highly expressed during the pre-storage phase is involved in growth processes, including cell wall biosynthesis. The data suggest that the necessary UDPglucose is provided both by sucrose synthase (isoform 3) and an invertase-driven pathway. Further, major developmental changes in pathways producing energy are predicted. A bell-shaped expression pro®le with a peak during the intermediate phase is characteristic for genes associated with photosynthesis and ATP production. The photosynthesis-determined increase of ATP concentration could be a prerequisite for the initiation of grain ®lling, dominated by starch and storage protein synthesis. Storage product accumulation is accompanied by high transcriptional activity of genes involved in glycolysis and fermentation, as well as in the citric acid cycle. Keywords: barley, expression pro®ling, maternal and ®lial caryopsis tissues, coordinated development, grain-speci®c photosynthesis, initiation of grain ®lling.

Introduction The barley grain is a typical starch and protein storing sink organ, which when mature, is composed predominantly of the diploid embryo and the triploid endosperm. These ®lial organs are surrounded by maternal tissues. Nutrients are unloaded from the vascular bundles into the pericarp cells, which in turn, reach the underlying ®lial tissues in an apoplastic step (see Patrick and Of¯er, 2001). How and by which mechanisms maternal and ®lial tissues interact to drive and coordinate normal seed development are largely unknown. The major events of barley endosperm development have been described earlier, and are classi®ed into four main stages: syncytial stage, cellularized stage, differentiation stage, and maturation stage (Olsen et al., 1992). Based ß 2004 Blackwell Publishing Ltd

on growth characteristics, starch accumulation patterns, and metabolite pro®les, we described a slightly different staging scheme for whole caryopses in the barley variety `Barke' (Weschke et al., 2000, 2003). The ®rst phase is de®ned as pre-storage phase (0±5 days after ¯owering (DAF)) and is characterized mainly by cell division and absence of starch in the endosperm. During the following days, an initial accumulation of starch grains occurs in the endosperm (6±9 DAF), followed by a linear increase of storage product biosynthesis and deposition (10±20 DAF), which levels off thereafter (around 20 DAF). The accumulation of storage compounds (starch and storage proteins) is based mainly on imported sucrose, and it dominates endosperm development. Although the biochemical pathways 539

540 Nese Sreenivasulu et al. that produce these storage products are known, based on the measured activity of relatively few genes or enzymes (Bewley and Black, 1994; Kigel and Galili, 1995), further dissection of the networks of regulatory and metabolic pathways is required to obtain a more complete understanding of seed development, including storage compound accumulation. With the recent development of high-throughput technologies allowing the concomitant analysis of thousands of genes, proteins, and/or metabolites (see Fiehn et al., 2001), the analysis of complex networks governing developmental and metabolic processes has become possible for the ®rst time (for seeds, see Lee et al., 2002; Ruuska et al., 2002). To investigate regulatory networks operating in barley grains during the pre-storage and early storage phases, we started the analysis of expressed sequence tags (ESTs) (see Michalek et al., 2002; http://pgrc.ipk-gatersleben.de) and developed cDNA macro arrays for transcriptome analysis (Potokina et al., 2002; Sreenivasulu et al., 2002). Physiological studies, including enzyme activity and metabolite measurements, have been published (see Weschke et al., 2000, 2003), and detailed histological analyses of seed development including 3D-model building, as well as medium-scale in situ hybridization studies, are under way (Gubatz et al., in preparation). In this study, we used a cDNA array with 1421 sequences representing 1184 different genes (unigenes), according to our annotation procedure. Arrays were hybridized with probes derived from maternal and ®lial tissue preparations of barley caryopses. The probes represent seven time points during development from fertilization (0 DAF) to the early storage phase (12 DAF), at 2-day intervals. The data obtained verify principally known facts in more detail and point out unknown or insuf®ciently analyzed phenomena. Maternal tissues undergo degradation characterized by lipid and transient storage compound mobilization. Principle component analysis (PCA) of all expression data implies a dramatic stepwise re-programming of the transcriptional machinery between pre-storage and storage phases in a newly de®ned intermediate phase of the caryopsis development. Intermediate phase gene expression pro®les suggest an important role for photosynthetic oxygen production and ATP provision for subsequent storage processes. Filial tissue expression patterns are also indicative of changes in energy provision with respect to development and cellular compartmentation. Results and discussion Barley grain development is a highly complex process based on coordinated regulatory programs realized in three genetically different organs: the diploid maternal tissues, the ®lial triploid endosperm, and the ®lial diploid embryo. To analyze the underlying transcriptional programs, we

used hand-dissected maternal and ®lial tissue fractions, prepared from barley caryopses, between fertilization and maturation (0±12 DAF), at 2-day intervals (seven time points) to synthesize 33P-labeled cDNA probes. The probes were hybridized to arrays on nylon membranes (macro arrays), with 1421 cDNA fragments representing 1184 seed-expressed unique sequences. Hybridization experiments were repeated using independently grown plant material. Expression analysis results are presented and discussed in the following chapters. Principle component analysis visualizes massive re-programming of gene expression during the transition from pre-storage to storage phase Seed development is usually divided into three phases based on general features: ongoing cell division and morphogenesis (cell division or pre-storage phase), storage product accumulation (storage or maturation phase), and water loss (desiccation phase; Goldberg et al., 1989). To survey on transcriptional changes during development of barley grains, a macro array ®lter was hybridized with radio-labeled second-strand cDNA probes as described above. The complete data set from 28 experiments, comprising logarithmically scaled median array and gene-centered expression data (a total of 39 788 data points), was subjected to PCA. PCA calculates cumulative correlations and identi®es the maximal variances in a multidimensional data set (see Experimental procedures). A data plot with respect to the ®rst two principle components, representing almost two-thirds of the total variance (66.4%), shows that maternal and ®lial tissue preparations are clearly distinguishable from each other (Figure 1). Differences between the maternal and ®lial tissue samples are comparatively small during early development. As development progresses, the variances between the two tissue fractions increase. Furthermore, the expression pro®les of tissues during early (0, 2, and 4 DAF) and later (10±12 DAF) developmental stages form distinct groups for both maternal and ®lial samples. The distances between the two groups are larger for the ®lial fraction as shown in Figure 1, indicating more dramatic changes in the ®lial than in the maternal transcriptome. Likewise, the distances between the tissue samples represented by the PCA data points/groups 4 and 6, 6 and 8, as well as 8 and 10 DAF, allude to large transcriptional changes, i.e. a regulatory re-programming. Again, this process is much more distinct in the ®lial than in the maternal tissues. We interpret the PCA results as indicative of massive re-programming of gene expression during the transition from pre-storage to storage phase, especially in the ®lial tissues. Interestingly, fresh weight increases and starch accumulation of caryopses shows a small but characteristic transient plateau at the respective developmental stage (see Fig. 7 in Weschke et al., 2000). ß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

Expression patterns in developing barley grains

541

using all data points for the 337 cDNA fragments, i.e. maternal and ®lial tissues were considered together. Therefore, f- and m-panels, which occur side by side in Figure 2, show expression of the same genes. The number of genes

Figure 1. Developmental stages of the barley caryopsis as identi®ed by PCA. Tissue samples were clustered with respect to their gene expression status using PCA (for details, see Experimental procedures). Maternal (green) and ®lial tissues (yellow) fall in distinct groups, as well as tissue samples 0±4 and 10±12 DAF, identifying the pre-storage and the storage phases of caryopses development, respectively. Tissue samples 6 and 8 DAF (not associated with a group) indicate an intermediate stage during which the gene expression pro®les change dramatically. Two experimental series based on independently obtained plant material are shown by open and closed symbols. For a reduced data set containing only 337 differentially expressed genes, instead of all 1421 cDNA fragments on the array, the black cross-hairs de®ne the position of the groups of tissue samples, indicating that most of the variance is retained.

Growth curves, with lag phases as phenotypic indications of underlying discontinuous developmental processes, have also been described in pea (Wang and Hedley, 1993) and Arabidopsis seeds (Baud et al., 2002). To reveal which genes are primarily responsible for the variance between the tissue samples, the complete data set was reduced to cDNA fragments, which showed differential expression across the experiments. These were selected based on two criteria: signal intensity, which had to exceed either three arbitrary units, or three times the average background, and ratio between minimal and maximal signal for a cDNA fragment across all experiments, which had to exceed the value of 10. For log2-transformed data, it would mean that the difference between the minimal and the maximal normalized signal intensity would have to exceed 3.3219. The reduced data set of 337 cDNA fragments was again subjected to PCA and this led to similar results (Figure 1, red versus black lattice). Thus, most of the variance of the original data set is retained, implying that the 337 differentially regulated genes are major components of the stepwise developmental changes. Identification of developmental stage-specific maternal and filial gene expression clusters Signal intensities (re¯ecting relative mRNA levels) of the 337 differentially expressed cDNA fragments were subjected to cluster analysis. Sixteen groups of cDNA fragments, showing distinct expression patterns during maternal and ®lial tissue development (Figure 2), were created by k-mean clustering. Clusters were calculated ß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

Figure 2. Clusters of co-expressed genes. Three hundred and thirty-seven differentially expressed genes were grouped into 16 clusters using the k-mean algorithm. Clusters are identi®ed by numbers (top of each panel), of which the ®rst digit denotes a group of clusters with qualitatively similar expression behavior and the second digit denotes the cluster within the group. The expression behavior of maternal (m) and ®lial tissues (f) is shown side by side. Red and blue symbols identify the mean expression value of all cDNA fragments in that cluster for two independent experimental series; gray shaded areas demarcate extreme expression values. Expression values are given as logarithmically scaled (base 2) array- and gene-centered signal intensities so that one unit on the vertical axis represents an expression ratio of a factor of 2. Vertical shading indicates developmental stages of the barley caryopsis: yellow, pre-storage phase 0±4 DAF; orange, intermediate stages 6 and 8 DAF; and brown, early storage phase 10±12 DAF.

542 Nese Sreenivasulu et al. Table 1 cDNA clones preferentially expressed in the maternal tissues Cluster-id

EST-id

Putative function

Assignment

EST cluster

Regulation of cellularization 1_1 HY05H16 1_1 HY02D05 1_1 HY02F10 1_1 HY01E17 1_1 HY07I05 1_1 HY01A07 1_1 HY04M20 1_1 HY07G15

DNA J protein homolog 2 DNA J protein homolog 60 S ribosomal protein l40 Histidyl-tRNA synthetase Actin de-polymerizing factor Peptidyl-prolyl cis-trans isomerase Peptidyl-prolyl cis-trans isomerase Peptidyl-prolyl cis-trans isomerase, chloroplastic

Potential Secure Secure Potential Potential Secure Secure Secure

± cl-134 ct-198 cl-153 ct-218 ± cl-735 ct-866 cl-3 ct-3 cl-3 ct-3 ±

Carbohydrate metabolism 1_1 HY01A03 1_1 HY03H04 1_1 HY07C03 1_1 HK04H22 1_1 HY06N18 1_2 HvSTP 1_2 CWINV 1_3 HY05O10

b-amylase b-amylase Glucan endo-1,3-b-glucosidase Glucokinase 14-3-3-like protein A Hexose transporter Cell wall invertase Fructokinase

Secure Secure Secure Secure Secure Potential Potential Secure

cl-1 ct-1 cl-1 ct-1 cl-1775 ct-1934 cl-1626 ct-1781 cl-68 ct-89 ± ± cl-512 c-t63

Energy production 1_1 HY09J04 1_1 HY03N13 1_2 HY05B22 1_2 HY05B22

ATP synthase beta chain, mitochondrial Cytochrome P450 71b2 NADP-dependent oxidoreductase P1 NADP-dependent oxidoreductase P1

Secure Potential Potential Potential

cl-894 ct-13 cl-1821 ct-198 cl-56 ct-68 cl-56 ct-68

Hormonal regulated genes 1_2 HY05P23 1_3 HY09I02 1_2 HK03J06

Estradiol 17 b-dehydrogenase 4 Gibberellin-regulated protein 3 Brassinosteroid-regulated protein

Potential Potential Secure

cl-69 ct-734 cl-891 ct-127 cl-1124 ct-1264

Lipid degradation 1_1 HY03C16 1_2 HY02E15 1_2 HY03J19 1_2 HY07L03 1_3 HY10O23 1_3 HK04H17

Acyl-CoA-binding protein Lipoxygenase 1 Hormone-sensitive lipase Non-specific lipid-transfer protein 4 Non-specific lipid-transfer protein 5 Non-specific lipid-transfer protein 4.1

Secure Secure ± Secure Potential Secure

cl-28 ct-368 cl-123 ct-184 cl-327 ct-432 cl-752 ct-883 cl-554 ct-674 cl-1658 ct-1816

Protein degradation 1_1 HY05M15 1_1 HY09E10 1_2 HK03G06 1_2 HK03G06 1_2 HY07K19 1_2 HW01G04 1_2 HY05P07 1_3 HK04D21 1_3 HY10O06

Ubiquitin-conjugating enzyme Cathepsin b-like cysteine proteinase precursor Cysteine proteinase 1 Cysteine proteinase 1 Cysteine proteinase 1 Cysteine proteinase 1 Vacuolar processing enzyme Serine carboxypeptidase Aspartic proteinase

Secure Secure Secure Secure Secure Secure Potential Potential Secure

cl-58 ct-626 cl-55 ct-98 ± ± cl-75 ct-881 cl-75 ct-881 cl-1871 ct-23 ± cl-966 ct-113

Metabolism of various compounds 1_3 HY05K19 Cobalamin-independent methionine synthase isozyme 1_1 HY08E09 Myo-inositol-1-phosphate synthase 1_1 HY05F22 Myo-inositol-1-phosphate synthase

Secure Secure Secure

cl-148 ct-212 cl-164 ct-1796 cl-29 ct-65

Non-classified genes 1_2 HY08I05 1_3 HY09N04 1_1 HY06P18 1_1 HY05E15 1_1 HY03E16 1_1 HW01N21 1_1 HY08B10

Secure Secure Potential Potential Potential Secure Secure

cl-829 ct-964 cl-8 ct-934 cl-69 ct-819 ± cl-254 ct-342 cl-1897 ct-256 cl-688 ct-817 cn-935

Tonoplast intrinsic protein Argonaute protein CAJ1 protein Dihydroflavonol-4-reductase Sorbitol dehydrogenase Dehydrin cor410 Plasma membrane intrinsic protein 1c

ß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

Expression patterns in developing barley grains

543

Table 1 continued Cluster-id 1_2 1_3 Non-annotated 1_1 1_1 1_1 1_1 1_1 1_1 1_1 1_1 1_1 1_1 1_1 1_1 1_1 1_1 1_2 1_2 1_2 1_2 1_2 1_2 1_3 1_3 1_3 1_3 1_3 1_3 1_3 1_3 1_3

EST-id

Putative function

Assignment

EST cluster

HY01C15 HY08D04

Glutathione S-transferase 1 Catalase isozyme 2

Potential Secure

cl-2 ct-5 cl-279 ct-367

Dihydrolipoamide dehydrogenase, mitochondrial (garden pea) Myosin heavy chain (amoeba) Potassium channel protein 1 (fruit fly) Hypothetical 59.6-kDa protein (baker's yeast) b2 Protein (carrot) Calcium-binding mitochondrial protein (fruit fly) Blue copper protein precursor (garden pea) No BLAST hit Zinc finger protein 90 (mouse) D-ribose-binding protein precursor (Bacillus) S-adenosylmethionine (Haemophilus) Heat shock factor protein 3 (chicken) DNA-directed RNA polymerase (Pseudomesenteroides) Large tegument protein (herpesvirus type 1) Legumin b (garden pea) Insulin receptor substrate-2 (human) Keratin, ultra high-sulfur matrix protein (human) Excinuclease abc subunit (Salmonella) Excinuclease abc subunit (Salmonella) Mucoropepsin precursor (Rhizomucor) Trimethylamine oxidase (Escherichia coli) Chaperone protein DNA K (Myxococcus) Hypothetical 137.2-kDa protein (fission yeast) Anther-specific proline-rich protein (Arabidopsis) Zinc finger protein 40 (mouse) Hypothetical 48.5-kDa protein (Caenorhabditis elegans) Branched-chain amino acid aminotransferase (Methanococcus) Endo-1,4-b-glucanase 4 (Bacillus) Ribonucleoprotein (fruit fly)

None None None None None None None None None None None None None None None None None None None None None None None None None None None None None

± ± cl-39 ct-79 ± cl-1844 ct-23 ± cl-641 ct-768 cl-781 ct-915 cl-18 ct-248 cl-63 ct-757 cl-95 ct-141 ± cl-772 ct-96 ± cl-133 ct-196 cl-269 ct-357 cl-269 ct-357 cl-3 ct-389 cl-3 ct-389 cl-285 ct-374 ± cl-424 ct-536 cl-674 ct-83 cl-785 ct-919 ± cl-477 ct-591 cl-668 ct-797 ± cl-564 ct-684

genes HK04B02 HY05N14 HY10M15 HY05D22 HY06P19 HY05G05 HY06G03 HY07P23 HY02H12 HY06E13 HY09N11 HY08A18 HY07O10 HY05H20 HY02D04 HY03B06 HY10J06 HY03G16 HY03G16 HY03D09 HK05C07 HK01E14 HY06M01 HY08B02 HK03G15 HY04P23 HY08J12 HY08P04 HY05D10

All clones of clusters 1_1, 1_2, and 1_3 are listed. The expression of these cDNAs is at least twofold higher in maternal than in filial tissues. The top hits from a BLASTX2 search against the protein databases SwissProt and PIR (June 2002) are given together with our classification of this assignment (see Experimental procedures). EST cluster numbers describe the affiliation of every single clone to a specific cluster (cl), contig (ct), or consensus sequence (cn), and reflect parts of the STACK_PACK analysis results obtained by using about 6500 50 - and 30 sequences only from caryopses-specific clones, which can also be obtained from the Web server at the IPK (http://pgrc.ipk-gatersleben.de/ seeds/).

within a cluster is always indicated in the m-part. Based only on expression pattern, these 16 clusters can be further arranged into seven cluster groups (see legend of Figure 2 for cluster designation). Cluster group 1 is characterized by the fact that expression of genes is twofold (on average) higher in maternal tissues than in ®lial tissues. In contrast to group 1, genes of cluster groups 2±5 are expressed at higher levels in the ®lial tissues. Similar relative mRNA expression levels in both tissue types principally characterize group 6 genes, whereas group 7 genes are outstanding with respect to the obvious differences between the replicate experiments (see Figure 2). Most likely, expression of group 7 genes was strongly in¯uenced by changes of unknown environmental factors, which remained uncontrolled within the growth chambers and did not in¯uence majority of genes in all ß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

other clusters. Cluster members and their functional annotation are provided in Figure 5; Tables 1 and 2; and Tables S1±S4.1 Most of the clusters are characterized by functional groups of genes (Figure 3). Whereas hormonal regulated genes and genes speci®c for degradation processes are exclusively expressed in the maternal tissues (clusters 1_2 and 1_3), genes involved in carbohydrate metabolism, including starch synthesis, are predominantly expressed during 6±12 DAF and typically found in the ®lial-speci®c cluster 4. Cluster 5 is dominated by storage protein and proteinase inhibitor genes (Figure 3) most highly 1 To con®rm the conclusions of the paper, the supplemental data can be accessed through the authors website (http://pgrc.ipk-gatersleben.de/ seeds/).

544 Nese Sreenivasulu et al. Table 2 Genes expressed in ®lial tissues during early storage phase (10±12 DAF) Cluster-id

EST-id

Putative function

Assignment

EST cluster

Regulatory genes 5 HY06E05 5 HY06M23 5 HY05K12 5 HY04P15 5 HY08M21

MYB-related protein DNA excision repair protein haywire Nuclear polyadenylated RNA-binding protein Tubulin 3 b-chain Glycyl-tRNA synthetase

Potential Potential Potential Secure Secure

± cl-645 ct-773 ± cl-141 ct-25 ±

Energy production 5 HY06M18 5 HY05G12 5 HY05G12 5 HY02L08

Inorganic pyrophosphatase 5'-AMP-activated protein kinase, b-2 subunit 5'-AMP-activated protein kinase, b-2 subunit Plasma membrane ATPase 4

Potential Potential Potential Secure

cl-637 ct-764 ± ± cl-789 ct-923

Inhibitors of proteolytic degradation 5 HY09N22 Trypsin inhibitor cme precursor 5 HY02P19 Alpha-amylase inhibitor bmai-1 5 HY06K18 Alpha-amylase/trypsin inhibitor cmb 5 HY08P07 Alpha-amylase inhibitor bdai-I 5 HY03M02 Alpha-hordothionin 5 HY04J03 Trypsin/factor xiia inhibitor precursor

Secure Secure Secure Secure Secure Potential

cl-259 ct-347 cl-196 ct-265 cl-314 ct-416 cl-25 ct-6 cl-122 ct-183 cl-439 ct-552

Storage protein accumulation 5 HY06G01 5 HY06A05 5 HY07A07 5 HY06E20 5 HY10K04 5 HY08A03

B1-hordein B3-hordein B3-hordein Gamma-hordein 3 Glutenin, high-molecular-weight subunit Glutenin, high-molecular-weight subunit

Potential Potential Secure Secure Secure Potential

cl-468 ct-581 cl-468 ct-582 cl-468 ct-581 cl-632 ct-759 cl-63 ct-17 cl-65 ct-11

Non-annotated 5 5 5 5 5 5 5 5

Ornithine decarboxylase antizyme, long isoform (zebrafish) Gamma-interferon-inducible thiol reductase precursor (human) Periodic tryptophan protein 1 homolog (human) Adenovirus type 2 late 100-kDa protein (human) Adenovirus type 2 late 100-kDa protein (human) Prolamin pprol 17 (rice) Large proline-rich protein bat2 (human) Puroindoline-b precursor (wheat)

None None None None None None None None

cl-25 ct-6 cl-353 ct-461 ± ± ± cl-179 ct-247 ± cl-83 ct-965

genes HY01D10 HY03N12 HY04M15 HY04F23 HY04F23 HY02H10 HY04G19 HY08I07

The table contains all members of cluster group 5. For further explanations, see legend of Table 1.

expressed 10±12 DAF (Figure 2). Gene expression patterns different in clusters 4 and 5 are indicative of independent control of starch and protein biosynthesis. The same conclusion was reached for Arabidopsis seeds based also on expression pro®ling (Ruuska et al., 2002) and for legume seeds based on biochemical analyses (Golombek et al., 2001). Figure 3 also highlights the association of cluster 3 genes with energy production (see below for detailed explanations) and the large percentage of genes with unknown functions (about 20±25% in every cluster) expressed in both maternal and ®lial tissues. Maternal gene expression is dominated by sets of genes involved in transient storage, reserve mobilization, and tissue degradation Thirty-nine genes preferentially expressed in the maternal tissue fraction are most highly active during early (0±2 DAF)

development (Figure 2, m-part of cluster 1_1, gray-shaded pro®le, notice log scale). Roughly, one-third exhibits no signi®cant homology to SwissProt data base entries (Figure 3a, red block of cluster 1_1); others are involved in regulation of cellularization (8), energy production (2), and carbohydrate metabolism (5; Figure 3a, black blocks; Table 1). The other half of the maternal genes discussed (n ˆ 40) was found to be upregulated speci®cally in later stages of development (Figure 2, m-part of clusters 1_2 and 1_3). Functional annotation revealed their involvement in hormonal-induced processes, lipid mobilization, and protein degradation (Figure 3a, yellow blocks; Table 1), but again about one-third of the genes (n ˆ 24) have unknown functions (Figure 3a, red block of clusters 1_2 and 1_3). Furthermore, genes encoding reactive oxygen species (ROS)scavenging enzymes are downregulated, e.g. catalases and different types of peroxidases (Figure 4). The observed ß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

Expression patterns in developing barley grains

545

(a) Functional groups

Clusters 1_2, 1_3

1_1 Regulation of cellularization Energy production Carbohydrate metabolism Metabolism of various compounds Hormonal regulated genes Lipid degradation Protein degradation Unclassified genes Unknown genes 0

25

0

25

50 %

(b) Functional groups

Cluster groups 2

3

4

5

Regulation of cellularization Protein degradation Lipid degradation Energy production Carbohydrate metabolism Amino acid metabolism Storage protein accumulation Inhibitors of protein degradation Nucleotide metabolism Genes coupled to cofactors and vitamins Transport Metabolism of various compounds Unclassified genes Unknown genes 0

25

0

25

0

25

0

50

25 %

Figure 3. Composition of clusters speci®c for the maternal (a) and the ®lial tissue preparations (b) of the barley caryopsis. The composition of clusters or (cluster) groups is shown with respect to functional annotation of their cDNA members. Functional groups dominating the respective cluster are shown in red; functional groups speci®cally expressed in maternal tissues are shown in yellow.

expression patterns suggest the involvement of programmed cell death (PCD) in later development of the maternal seed part. PCD has been described for cereal endosperm (Young and Gallie, 1999) and especially for aleurone (for review, see Fath et al., 2000), but not for maternal tissues. Gibberellins (GAs) are involved in the initiation of PCD in the aleurone of germinating cereal seeds (Fath et al., 2001). It is therefore interesting that the EST of a putative GAß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

regulated protein (HY09I02; Table 1) with homology to proteins from Oryza sativa (AC084282, score 149) and Arabidopsis (AAC27845, score 102) is expressed in maternal tissues. Embryo-produced GAs (Appleford and Lenton, 1997; Wang, 1997) are transferred to the aleurone layer during early germination, and they ®rst induce lipid breakdown. Genes involved in the two described pathways for lipolysis (Feussner et al., 1997) are among the maternally expressed genes upregulated later in development: a

546 Nese Sreenivasulu et al. lipase, which releases fatty acids from the glycerol backbone, and a lipoxygenase (see Table 1, lipid metabolism), which catalyzes oxygenation of storage and/or membrane lipids (Feussner et al., 1995). Increased transcript levels of an acyl-CoA-binding protein (Table 1) suggest signi®cant concentrations of acyl-CoA, the product of b-oxidation of fatty acyl chains. In addition, data mining of 10 040 ESTs produced from a maternal-speci®c cDNA library (H. Zhang, personal communication) has identi®ed the complete set of enzymes necessary for the b-oxidation pathway, together with the peroxisomal enzymes of the glyoxylate cycle and a highly expressed malate dehydrogenase. The high abundance of lipoxygenase ESTs out of the maternal-speci®c cDNA library supports the idea that lipoxygenases, rather than lipases, initiate b-oxidation, at least in senescing leaf tissues (Feussner et al., 1997). b-oxidation requires molecular O2 (probably provided by seed photosynthesis as discussed below) as electron acceptor and releases H2O2, a ROS. ROS play a key role in aleurone PCD (Bethke and Jones, 2001; Fath et al., 2001). H2O2 as well as other molecules of ROS are detoxi®ed by ROS-scavenging enzymes like ascorbate peroxidase (APX), catalase (CAT), and superoxide dismutase. Genes for these enzymes are also active in maternal caryopses tissues during earlier developmental stages. Later stages are characterized by decreasing transcript levels of CAT1, glutathione peroxidase, a cytosolic APX, and a peroxidase (Figure 4). A similar downregulation of ROS-scavenging enzyme RNAs was found in aleurone layers and GA-treated barley aleurone protoplasts (Bethke et al., 2001; Fath et al., 2001), indicating the decreasing ability of the cells to detoxify ROS, such as H2O2. The downregulation of transcript levels of ROS-scavenging enzymes parallels the disintegration of the pericarp tissue observed from 8 DAF onwards (S. Gubatz, personal communication) and therefore, may be connected to PCD of the pericarp cells as speculated by Bethke et al. (2001) for GA-treated aleurone cells and protoplasts. Beside ROS, proteases also play a role in PCD in different plant species and organs (for review, see Beers, 1997). For instance, cysteine (Bethke et al., 1998) and aspartic proteases were shown to play important roles in PCD of aleurone layer during barley seed maturation (ToÈrmaÈkangas et al., 1994). A cysteine proteinase (SmCP), active in degenerating nucellus tissue of Solanum melongena, was suggested to function in mobilizing protein reserves for the developing embryo (Xu and Chye, 1999). Furthermore, Dominguez and Cejudo (1998) described the expression of a germination-related carboxypeptidase in the wheat nucellus, and an aspartic protease, called nucellin, was detected in barley nucellus (Chen and Foolad, 1997). According to our array analysis, two isoforms of cysteine±proteinase, one aspartic protease, a serine carboxy peptidase, and a cathepsin-like cysteine proteinase are activated at the transcript

level in maternal tissues (cf. Table 1, protein degradation). This large set of protein-mobilizing enzyme mRNAs, together with upregulated transcription of lipid-mobilizing genes and downregulated transcription of ROS-scavenging genes, supports the suggestion that barley maternal caryopses tissues mobilize stored compounds and undergo PCD. During the developmental period analyzed (0±12 DAF), both accumulation and degradation of storage products take place. Interestingly, storage protein genes, as well as a-amylase and proteinase inhibitor genes, exhibit similar expression patterns in both the ®lial and the maternal caryopsis fraction (see clusters 4_1, 4_3, and 5 in Figure 2). However, clear differences exist regarding the level of expression (Figure 2, notice logarithmic scale). Speci®c cell layers of the maternal tissue, such as the outer and the nucellar epidermis, as well as parts of the integuments, persist until late in development. Western blot analyses indicated that these cell layers accumulate storage proteins. Whether these reserves are mobilized and the manner in which they are mobilized remain to be determined. Both invertase- and sucrose synthase-dependent pathways of sucrose cleavage are probably involved in cell wall biosynthesis during early development of the filial tissues The most characteristic process during early development of ®lial tissues is cell division. Accordingly, more than 25% of the genes in cluster group 2 are involved in cell division and elongation processes (Figure 3b; Table S1). Cell division requires intensive cell wall biosynthesis, a process dependent on UDP-glucose. UDP-glucose can be produced directly by sucrose synthase or by an invertase-dependent pathway via glucose-1-phosphate (G-1-P; see Figure 7). Our data suggest that both processes are active. As expected for young tissues, different invertase isoforms are expressed during the pre-storage phase of the caryopsis (Figure 4; Weschke et al., 2003). Their expression correlates with high amounts of hexoses within the ®lial fraction and the expression of speci®c hexose transporters, especially HvSTP1 (Weschke et al., 2003). High relative mRNA levels during early development have also been detected for hexokinases (HKs), HK1 and HK2, as well as for glucokinase (GK; Figure 7). Thus, supposing that the mRNA levels roughly correlate with the activities of the corresponding enzymes, high amounts of hexose-6-phosphates necessary for glycolytic processes should be present in the cells of the ®lial tissue. G-6-P can be converted to G-1-P by phosphopglucomutase (PGM; exhibiting suf®cient relative transcript levels also during early development; Figure 7) and further metabolized preferentially by UDP-glucose pyrophosphorylase (UGPase) to UDP-glucose, whereas conversion by ADP-glucose pyrophosphorß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

Expression patterns in developing barley grains

547

120

signal intensity (%)

100 80 60 40 20 0 0

2

4

6

8

10

12 14 DAF

Catalase 1 (HY01K15) Catalase 2 (HY03C15) Glutathione peroxidase (HY02F15) L-Ascorbat peroxidase (HY01M23) Peroxidase (HK01O22)

Figure 4. Expression of ROS-scavenging enzymes in the maternal fraction of barley caryopses 0±12 DAF. For all genes, normalized signal intensities for each time point were averaged for the two experimental series and expressed as percentages of the maximal intensity.

ylase (AGPase) is typical for the storage phase (Figure 7). This latter reaction needs ATP, whereas UGPase produces pyrophosphate (PPi). According to Stitt (1998), high PPi levels are characteristic for cytoplasm-rich young cells. This observation is in favor of the preferential conversion of G-1P into UDP-glucose in young ®lial barley tissues. Based on relative transcript levels (Figure 7), we assume that the high 120

signal intensity (%)

100 80 60 40 20 0 0

2

4

6

8

10

12 14 DAF

Phosphoribulokinase (HY09O01) Rubisco small subunit (HY07K23) Rubisco large subunit (HY06P01) Chlorophyll a/b binding protein (HY02C13) Oxygen-evolving enhancer protein 2 (HY05I05) Oxygen-evolving protein 1 (HY05D21) Photosystem II 22 kD protein (HY02P17) Plastocyanin (HY07N21)

Figure 5. Transcript levels of genes related to photosynthesis in ®lial tissues of barley caryopses 0±12 DAF. Normalized signal intensities of the respective cDNA fragments were treated as described in the legend of Figure 4.

ß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

Figure 6. Transcript levels of genes related to ATP-synthesis and ATPtransport in ®lial tissues of barley caryopses 0±12 DAF. Normalized signal intensities of the respective cDNA fragments were treated as described in the legend of Figure 4. The expression pro®le of the triosephosphate translocator was determined by Northern blotting and quanti®ed as described by Weschke et al. (2000).

amounts of UDP-glucose necessary for cell wall biosynthesis are not only synthesized from G-1-P produced by the invertase-dependent pathway, but also from SUS3, the only sucrose synthase isoform represented by higher relative levels of mRNA during early development of the ®lial tissues (Figure 7). SUS3 at the sequence level is most homologous to the maize Sh1/Sus1 gene. The Sus1encoded SS1 maize isoform has been suggested to play a dominant role in providing substrate for cellulose biosynthesis (Chourey et al., 1998). The respective rice isoform is RSuS2 (gene RSs1; Wang et al., 1999). Biochemical work on cotton ®ber cellulose biosynthesis had already led to the hypothesis of direct metabolic channeling of carbon from sucrose to glucan polymers by a sucrose synthase/glucan synthase complex located at the plasma membrane (Amor et al., 1995). Whether the invertase and sucrose synthase pathways to UDP-glucose in barley seeds are spatially separated is unknown. Surprisingly, one invertase isoform (CWINV3) is preferentially expressed during the storage phase. Exclusively at this developmental stage, a previously unknown HK3 is active at the transcript level (Figure 7). By Northern blotting, CWINV3 mRNA was localized in the embryo/

548 Nese Sreenivasulu et al. scutellum (data not shown). Respective data for HK3 are still missing. The intermediate phase of seed development is characterized by a transient burst of gene expression related to photosynthesis and energy production Plant seeds often contain green tissue involved in photosynthesis. In barley grains, a green tissue layer of maternal origin, called the chlorenchyma, surrounds the seed sac. Its chloroplast-containing cells most likely express the genes forming cluster 3_2 (Figure 2). The expression pro®le of the 16 involved genes is bell-shaped in the ®lial tissue preparation (cluster 3_2f in Figures 2 and 5), with a broad peak within the intermediate phase of seed development (6± 8 DAF) de®ned by PCA (Figure 1). Genes of that cluster are involved in photosynthetic reactions, oxygen evolution, and energy production (Table S3; Figure 5). The appearance of these transcripts in the ®lial fraction (in spite of the maternal origin of the chlorenchyma tissue) is because of outer integument and pericarp disintegration processes during the pre-storage phase, which cause adherence of chlorenchyma tissue either to the maternal fraction (0±2 DAF), to both maternal and ®lial fraction (4 DAF), or exclusively to the ®lial fraction (6±12 DAF) during tissue preparation for mRNA isolation. Thus, the assignment of expression pro®les of cluster 3_2 genes to either the maternal or the ®lial part of the grain is partially uncertain. In developing Arabidopsis seeds, photosynthesis-related genes also show a bell-shaped expression pro®le, with a peak after the starch and before the protein accumulation period, but coinciding with maximal oil accumulation (Ruuska et al., 2002). The authors discuss their results mainly with respect to the capture of carbon released during fatty acid biosynthesis. The major function of photosynthesis is CO2 ®xation. However, in canola green embryos, the ®xation rate is low (less than 1/10 of the rate in leaves), although signi®cant electron transport was detected (Asokanthan et al., 1997). Studies in cotyledons of developing grain legumes led Rolletschek et al. (2002, 2003) to postulate that a more important function of seed photosynthesis is to provide oxygen for the ®lial tissues, which are hypoxic during early development. In barley caryopses, the situation is different from that in cotyledons, as the chlorenchyma is a maternal tissue separated from the ®lial starchy endosperm by cuticules. However, whereas CO2 cannot pass through this barrier (Nutbeam and Duffus, 1978), O2 produced in the chlorenchyma is able to pass (Duffus and Cochrane, 1993; Nutbeam and Duffus, 1978), thus increasing the oxygen content of the ®lial seed part. Therefore, photosynthetic activity of the barley chlorenchyma changes the hypoxic state of young ®lial tissues (H. Rolletschek, L. Borisjuk, personal communications) to an aerobic one during the

intermediate phase, and as a consequence, oxidative phosphorylation and ATP production are favored. In accordance with this scenario, genes associated with energy production and transport are upregulated during the intermediate phase, as for instance, the 6-kDa subunit of the mitochondrial ATP-synthase, the mitochondrial ATP/ ADP translocator, the plastidic ATP-synthase g-chain, and the plastidic triose phosphate translocator (Figure 6). The expression pro®les of the genes in Figure 6, which peak at the beginning of the storage phase, suggest an important role in energy provision for storage product biosynthesis and accumulation processes. The supposed increase in ATP concentration may provide a prerequisite and signal for cells to enter the storage phase as speculated for developing legume embryos (Rolletschek et al., 2003). Changing pathways of energy production during endosperm/embryo development as deduced from gene expression patterns Expression pro®les of genes involved in energy production pathways in the ®lial tissues (Figure 7) suggest striking differences between developmental stages. Enzymes depicted were selected from the complete set of expression analysis data based on two criteria: (i) signal intensities had to exceed either three arbitrary units or three times the average background; (ii) expression behavior had to be principally the same in experiments 1 and 2. Average values of signal intensities, as well as standard deviations between the two experiments, can be found at http:// pgrc.ipk-gatersleben.de/seeds/. During the pre-storage phase, high levels of G-6-P are expected based on invertase, as well as GK and HK expression pro®les (see above). Suf®ciently high early relative mRNA levels are also seen for both the cytosolic isoform of glucose-6-phosphate dehydrogenase (G6PDH) and the two subunits of pyrophosphate-dependent phosphofructokinase (PFP; Figure 7). The respective enzymes connect the hexose phosphate pool to the oxidative pentose phosphate (OPP) pathway and to glycolysis, respectively. Cytosolic G6PDH and 6-phosphogluconate dehydrogenase catalyze oxidative reactions of the OPP pathway. Their mRNA expression is higher during early than during later development (Figure 7), and correlates with high demands for RNA, DNA, and proteins in the ®lial part of the seed. Fructose-6-phosphate (F-6-P) phosphorylation (F-6-P to F1,6-P) is highly regulated and catalyzed by PFP, as mentioned above, and a second isoform, the ATP-dependent phosphofructokinase (PFK). However, because of missing information on plant-speci®c PFK sequences, we were unable to identify a putative PFK sequence even within our collection of more than 40 000 barley seed-speci®c ESTs. High relative mRNA levels of the PFP a-subunit during the pre-storage phase are contrasted by relatively ß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

Expression patterns in developing barley grains

549

Figure 7. Pathways of energy production and starch accumulation in ®lial caryopses tissues. Normalized signal intensities of cDNA clones with homology to genes encoding pathway enzymes (see Experimental procedures) were selected from the entirety of expression data and treated as described in the legend of Figure 4. The enlarged expression pro®le of G6PDH located at the upper right corner illustrates scaling of the x- and y-axis used for every single bar diagram shown in the ®gure. For explanation of vertical shading, see legend of Figure 2. Designation to a speci®c cell compartment (cyt, cytosol; plast, plastid; mito, mitochondria) is based on the annotation and functional classi®cation of the ESTs as described in Experimental procedures.

low levels of the b-subunit (Figure 7). Later in development, the relative mRNA amount of the a-subunit decreases, whereas the b-subunit values increase, probably leading to high and low a/b-subunit ratios during pre-storage and storage phase, respectively. According to Plaxton and coworkers (Plaxton, 1996; Theodorou et al., 1992), an increase in PFP activity and its sensitivity to the activator fructose2,6-bisphosphate is associated with an increase of the a/bsubunit ratio. Therefore, a relatively high PFP activity can be expected during early development. Surprisingly, fructose bisphosphate aldolase (FBA) and both isoforms of glyceraldehyde-3-phosphate dehydrogenase (GPD) regulating the metabolization of F-1,6-P are expressed at very low relative mRNA levels during early development, indicating a reduced activity of the upper part of the glycolytic pathway. As at the same time, G6PD is upregulated (see above), the OPP pathway may be preß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

ferred. As discussed by Neuhaus and Emes (2000), this alternate route for glucose metabolism would bypass the highly regulated phosphofructokinase step of glycolysis by production of triosephosphate and generation of NADPH necessary for biosynthetic reactions in the cytosol. Later steps in glycolysis are characterized by genes suf®ciently (phosphoglycerate kinase, phosphoglycerate mutase, enolase 2, pyruvate kinase 1) or highly expressed (enolase 1, pyruvate kinase 2) during early development. On the contrary, the gene encoding the plastidic pyruvate kinase 3 is exclusively expressed during later development. Hence, isoforms 2 (cytosolic) and 3 (plastidic) of pyruvate kinase seem to be highly speci®c for the pre-storage and the storage phase, respectively. The result parallels ®ndings of Ruuska et al. (2002) for developing Arabidopsis seeds. In general, many plastidial isoenzymes are strongly upregulated during storage phase, for instance, both AGPase

550 Nese Sreenivasulu et al. subunits and starch synthase, GAPDH, pyruvate kinase, and HK, indicating an increase in the plastidial metabolism during storage product accumulation. During the storage phase, photosynthetic activity is downregulated at the transcript level. Instead, genes encoding enzymes of the citric acid cycle and one related transporter (malate translocator) are upregulated together with transcripts of the fermentation pathway (Figure 7). Several isoforms of alcohol dehydrogenase transcripts are already found, whereas a D-lactate dehydrogenase transcript is nearly missing early during development (Figure 7). Only sequences of two isoforms of the latter enzyme catalyzing the pyruvate-to-lactate fermentation process were identi®ed in the set of about 40 000 seedspeci®c ESTs, one of them expressed in the maternal and the other one in the ®lial seed part (H. Zhang, personal communication), but practically only during the storage phase (Figure 7). Thus, the direction of the fermentation process apparently differs during development of the ®lial tissues. Ethanol might be produced during pre-storage as well as storage phases, whereas lactate production possibly takes place only during later developmental stages. In summary, gene expression pro®les suggest various changes in energy provision with respect to development and cell compartment. Intensive biochemical work is necessary to prove or disprove the indications provided here. Concluding remarks In the present study, macro array expression pro®ling was applied to gain insight into transcriptional networks active in maternal and ®lial tissues of the barley grain during early to mid-stages of development (0±12 DAF). We derived at the following general conclusions: (i) while seed development is usually divided in three phases characterized by cell division, maturation, and desiccation, our study identi®ed an intermediate phase separating cell division from accumulation processes. During that time span, massive transcriptional re-programming takes place as indicated by PCA; (ii) especially characteristic for the intermediate phase is a transient burst of gene expression related to photosynthesis and energy production. We infer from this and unpublished studies (H. Rolletschek and L. Borisjuk, personal communications) that oxygen, produced in maternal chlorophyll-containing cell layers, is able to enter the endosperm and it drives oxidative phosphorylation. The high amounts of ATP produced are a prerequisite for storage product accumulation; (iii) pro®les of genes expressed in maternal tissues, mainly in pericarp, suggest degradation of membrane lipids by b-oxidation and mobilization of storage compounds during mid-term development (>8 DAF). We note parallels to aleurone PCD and germination-related mobilization processes; (iv) a comparison of expression pro®les of genes involved in the various energy-producing

pathways revealed striking changes with respect to developmental time and involved cellular compartments. However, only detailed cell- and compartment-speci®c analysis of synthesis and fate of RNA, protein, and metabolites can provide the necessary information to fully understand seed development and its regulation. Experimental procedures Plant material and tissue preparation Hordeum vulgare cv. Barke, a two-rowed spring barley, was cultivated in growth chambers at 20/188C under a 16-h light/8-h dark cycle until seed set. The developmental stage of a caryopsis was determined from the mid-region of the ear as described by Weschke et al. (2000). Young developing seeds were harvested from that region at 0, 2, 4, 6, 8, 10, and 12 DAF. The developing caryopsis consists of the maternal tissues, i.e. mostly pericarp and the ®lial embryo/endosperm. The maternal and ®lial tissue fractions were isolated by manual dissection as described by Sreenivasulu et al. (2002).

Macro array preparation, hybridization, and evaluation After determination of ESTs, a total of 1412 cDNA clones were selected from cDNA libraries of developing caryopses (1235 clones), etiolated seedlings (73 clones), and roots (104 clones). Clones from etiolated seedlings and roots were chosen to increase the number of genes and gene isoforms related to carbohydrate metabolism and stress response. The number of unique genes was estimated using cluster analysis with StackPack (Christoffels et al., 1999; Miller et al., 1999) of a larger set of ESTs (Michalek et al., 2002). All sequences except a few of minor quality were deposited in the European Molecular Biology Laboratory (EMBL) sequence database and can be obtained together with their expression data from our WWW-server (http://pgrc.ipk-gatersleben.de/seeds/). Ampli®cation of cDNA fragments, spotting on Nylon membranes, and preparation of 33P-labeled cDNA probes, as well as the hybridization procedure, have been described previously by Sreenivasulu et al. (2002). After quantitative removal of the probe, each array was used up to ®ve times. Probe removal was achieved by washing the arrays for 15 min in 0.1 SSC, 0.1% SDS (heated to 1008C immediately before use), followed by an alkaline treatment (0.4 M NaOH at 658C for 15 min) and a successive neutralization of the array in 0.2 M Tris±HCl (pH 7.5) at room temperature. After hybridization, the radioactive signals on a cDNA array were detected using a phosphoimager (Fuji BAS, 2000, Fuji Photo Film Co., Ltd, Tokyo, Japan), and the intensities of individual spots and the corresponding local backgrounds were determined (ArrayVision, Imaging Research, St Catharine's, Ont., Canada). These data were exported to a spreadsheet program for further processing. Local background was subtracted from spot intensities, and the signal intensities of duplicated spots for each cDNA fragment were averaged. To allow the comparison of signal intensities between experiments, the median of the logarithmically scaled (log2) intensity distribution for each experiment was set to zero (median centering of arrays; Eisen et al., 1998). Before clustering of expression data, the median of the logarithmically scaled, arraycentered values for each gene across all experiments was set to zero (median centering of genes; Eisen et al., 1998). Median ß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

Expression patterns in developing barley grains centering of arrays and genes was repeated four more times. This transformation yields a data set, which emphasizes different expressions irrespective of absolute signal intensities. The resulting logarithmic relative signal intensities were used for all clustering algorithms in J-EXPRESS (Dysvik and Jonassen, 2001) usually employing Euclidian distances measures, i.e. for k-mean clustering and PCA. The logarithmic, array-centered data were backtransformed to non-logarithmic values leading to normalized signal intensities comparable to those of the previous normalization procedure (Sreenivasulu et al., 2002), but improving the comparison of experiments with somewhat higher background values. Normalized, non-logarithmic signal intensities (see above) were used to exclude cDNA fragments with low intensities across all experiments from a more detailed analysis. Furthermore, signal intensities transformed back to non-log scale were used to calculate the relative signal intensities presented in Figures 4±7.

Annotation and functional classification of genes represented by ESTs For annotation and functional classi®cation, all ESTs were compared with the SwissProt database (Bairoch and Apweiler, 2000) using BLASTX2 (Altschul et al., 1997). Relevant information was extracted from the BLASTX2 results using a custom made Perl script. ESTs were sorted into three categories based on the score and the length of the aligned sequence segment with the top database hit. Two straight lines that separate the three categories were de®ned on a scatter plot of score versus aligned length by manual annotation of approximately 700 sequences. These lines run through a common point de®ned by the minimal alignment length of 12 amino acids (aa) and a corresponding score of 27 bits. `Secure' and `potential' annotations are separated by a line with the slope of 1.36 bits per aa. `Potential' and `unannotated' sequences are separated at 0.62 bits per aa. All remaining cDNA clones on our array were categorized automatically using these criteria. In case 50 - and 30 -end sequences were available, the higher annotation category was assigned to the cDNA clone. For functional classi®cation of the genes represented by ESTs, cDNA clones with `secure' and `potential' annotations were used to extract EC numbers of the corresponding proteins from the description line of the SwissProt database hit. With help of the EC numbers, the positions of encoded proteins within biochemical pathways were identi®ed using the KEGG database (Kanahisa et al., 2002).

Reliability of the expression analysis results Probes were prepared from maternal and ®lial fractions at 2-day intervals from 0 to 12 DAF, i.e. 14 probes in total, and sequentially hybridized to the described macro array. The whole set of experiments was repeated once, using independently grown seed material. To verify the expression analysis results, 15 Northern blot analyses were carried out as described by Sreenivasulu et al. (2002) and Potokina et al. (2002). Evaluation of expression data con®rmed reproducibility of the procedure. Nearly identical results were obtained in the ®rst and second experimental series labeled by blue and red color, respectively, of lines and dots in Figure 2. Only cluster group 7 pro®les differ substantially between the two experiments (14 out of 337 differentially expressed genes (4.2%)). In cluster 7_2, this difference is essentially caused by four of the seven cluster members, coding for different types of heat shock proteins. The upregulation of stress genes only in that material used for the ®rst experimental ß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

551

series (blue symbols in Figure 2) might indicate some uncontrolled changes in the environmental conditions. Hybridization signal intensities of different cDNA fragments (ESTs) coding for the same gene and for identical fragments, represented two times on the array but ampli®ed independently, were in most of the cases in good accordance between the two independent sets of experiments (see, for instance, Table 1). However, cDNA fragments representing members of big gene families, for instance, the sucrose synthases, sometimes gave similar but not identical expression pro®les (notice, for instance, localization of sucrose synthase 2 expression pro®les in cluster 4_1, as well as cluster 4_2; Table S4). Whereas only three sucrose synthase isoforms could be identi®ed based on the EST collection used to select the 1421 unigenes described here, data mining of more than 40 000 seed-speci®c ESTs clearly identi®ed ®ve members of the gene family. Two of them are highly homologous to each other, but encode different isoforms (SUS 2.1 and 2.2; N. Sreenivasulu, unpublished). Whether the sucrose synthase 2 expression pro®les found in either cluster 4_1 or 4_2 belong to one or the other sucrose synthase 2 isoform can be decided only by full-length sequencing of all the sucrose synthase cDNA fragments spotted on the macro array. The complete data set from this cDNA array analysis together with the list of all clones represented on the macro array, as well as all comparisons of array data with Northern blot results, will be available on our web site (http://pgrc.ipk-gatersleben.de/seeds/).

Principle component analysis The entirety of 39 788 normalized, log-transformed array and gene-centered expression values from 28 hybridization experiments, with an array containing 1421 cDNA fragments, was analyzed by PCA (J-EXPRESS; Dysvik and Jonassen 2001). Basically, these 28 experiments de®ne 28 vectors each with 1421 components, represented by the expression values for all cDNA fragments in the respective tissue samples. The coordinate system of this 1421-dimensional space is transformed during PCA in such a way that the new origin is a centre of gravity de®ned by the end points of the 28 vectors. Furthermore, the ®rst axis is placed in the direction of the largest variance component between the end points, the second orthogonal axis in direction of the second largest variance, and so forth. As 48.5% of the total variance is represented along the ®rst axis, 17.9% of the variance along the second axis, and 7.3% along the third, the dimensionality of the system can be drastically reduced. Coordinates on the ®rst two axes, which together represent almost two-thirds of the total variance (66.4%), are plotted in Figure 1. Distances between experiments give an impression of the differences between the analyzed tissue samples.

Northern blot analyses Northern blot analyses were carried out as described by Sreenivasulu et al. (2002).

Acknowledgements We thank Hangning Zhang for STACK_PACK analysis of the barley EST collection, for discussions, and a lot of helpful comments. We are grateful to Angela Stegmann, Gabi Einert, Uta Siebert, and Elsa Fessel for excellent technical assistance. The comments of two anonymous reviewers helped to improve the manuscript considerably. This work was supported by a BMBF Grant (GABI-SEED, FKZ 0312282).

552 Nese Sreenivasulu et al. Supplementary Material The following material is available from http://www.blackwell publishing.com/products/journals/suppmat/TPJ/TPJ1981/TPJ1981sm. htm Table S1 Genes preferentially expressed in the ®lial tissues during the pre-storage phase Table S2 Genes preferentially expressed in the ®lial tissues during late pre-storage/intermediate phase Table S3 Genes of cluster group 4, containing candidates involved in carbohydrate metabolism, including starch synthesis, predominantly expressed during 6±12 DAF in the ®lial tissues Table S4 Genes of cluster group 6, upregulated mainly in the ®lial tissues during storage phase Table S5 Genes of cluster group 7 showing remarkable differences in gene expression between experiment 1 and experiment 2

References Altschul, S.F., Madden, T.L., Schaffer, A.A., Zhang, Z., Miller, W. and Lipman, D.J. (1997) Gapped BLAST and PSI-BLAST: a new generation of protein database search programs. Nucl. Acids Res. 25, 3389±3402. Amor, Y., Haigler, C.H., Johnson, S., Wainscott, M. and Delmer, D.P. (1995) A membrane-associated form of sucrose synthase and its potential role in synthesis of cellulose and callose in plants. Proc. Natl. Acad. Sci. USA, 92, 9353±9357. Appleford, N.E.J. and Lenton, J.R. (1997) Hormonal regulation of alpha-amylase gene expression in germinating wheat (Triticum aestivum) grains. Physiol. Plant. 100, 534±542. Asokanthan, P., Johnson, R.W., Grif®th, M. and Krol, M. (1997) The photosynthetic potential of canola embryos. Physiol. Plant. 101, 353±360. Bairoch, A. and Apweiler, R. (2000) The SWISS-Prot protein sequence and its supplement TrEMBL in 2000. Nucl. Acids Res. 28, 45±48. Baud, S., Boutin, J.-P., Miquel, M., Lepiniec, L. and Rochat, C. (2002) An integrated overview of seed development in Arabidopsis thaliana ecotype Ws. Plant Physiol. Biochem. 40, 151±160. Beers, E.P. (1997) Programmed cell death during plant growth and development. Cell Death Differ. 4, 649±661. Bethke, P.C. and Jones, R.L. (2001) Cell death of barley aleurone protoplasts is mediated by reactive oxygen species. Plant J. 25, 19±29. Bethke, P.C., Swanson, S.J., Hillmer, S. and Jones, R.L. (1998) From storage compartment to lytic organelle: the metamorphosis of the aleurone protein storage vacuole. Ann. Bot. 82, 399±412. Bethke, P.C., Fath, A. and Jones, R.L. (2001) Regulation of viability and cell death by hormones in cereal aleurone. J. Plant Phys. 158, 429±438. Bewley, J.D. and Black, M. (1994) Seeds ± Physiology of Development and Germination, 2nd edn. New York, London: Plenum Press, 445 pp. Chen, F. and Foolad, M.R. (1997) Molecular organization of a gene in barley which encodes a protein similar to aspartic protease and its speci®c expression in nucellar cells during. Plant Mol. Biol. 35, 821±831. Chourey, P.S., Taliercio, E.W., Carlson, S.J. and Ruan, Y.-L. (1998) Genetic evidence that the two isozymes of sucrose synthase present in developing maize endosperm are critical, one for cell wall integrity and the other for starch biosynthesis. Mol. Gen. Genet. 259, 88±96.

Christoffels, A., Miller, R. and Hide, W. (1999) STACK_PACK and STACK (sequence tag alignment and consensus knowledgebase): a novel comprehensive, hierarchical EST clustering and consensus generation and analysis system providing unique insights into the human genome. Am. J. Hum. Genet. 65, 477. Dominguez, F. and Cejudo, F.J. (1998) Germination-related genes encoding proteolytic enzymes are expressed in the nucellus of developing wheat grains. Plant J. 15, 569±574. Duffus, C.M. and Cochrane, M.P. (1993) Formation of the barley grain morphology, physiology and biochemistry. In Barley: Chemistry and Technology (MacGregor, A.W. and Bhatty, R.S., eds). St Paul. Minnesota, USA: American Association of Cereal Chemists, Inc., pp. 31±72. Dysvik, B. and Jonassen, I. (2001) J-Express: exploring gene expression data using Java. Bioinformatics, 17, 369±370. Eisen, M.B., Spellman, P.T., Brown, P.O. and Botstein, D. (1998) Cluster analysis and display of genome-wide expression patterns. Proc. Natl. Acad. Sci. USA, 95, 14863±14868. Fath, A., Bethke, P., Lonsdale, J., Meza-Romero, R. and Jones, R. (2000) Programmed cell death in cereal aleurone. Plant Mol. Biol. 44, 255±266. Fath, A., Bethke, P.C. and Jones, R.L. (2001) Enzymes that scavange reactive oxygen species are down-regulated prior to gibberellic acid-induced programmed cell death in barley aleurone. Plant Physiol. 126, 156±166. Feussner, I., Wasternack, C., Kindl, H. and Kuhn, H. (1995) Lipoxygenase catalysed oxygenation of storage lipids is implicated in lipid mobilization during germination. Proc. Natl. Acad. Sci. USA, 92, 11849±11853. Feussner, I., KuÈhn, H. and Wasternack, C. (1997) Do speci®c linoleate 13-lipoxygenases initiate b-oxidation? FEBS Lett. 406, 1±5. Fiehn, O., Kloska, S. and Altmann, T. (2001) Integrated studies on plant biology using multiparallel techniques. Curr. Opin. Biotechnol. 12, 82±86. Goldberg, R.B., Barker, S.J. and Perez-Grau, L. (1989) Regulation of gene expression during plant embryogenesis. Cell, 56, 149±160. Golombek, S., Rolletschek, H., Wobus, U. and Weber, H. (2001) Control of storage protein accumulation during legume seed development. J. Plant Physiol. 158, 457±464. Kanahisa, M., Goto, S., Kawashima, S. and Nakaya, A. (2002) The KEGG database at GenomeNet. Nucl. Acids Res. 30, 42±46. Kigel, J. and Galili, G. (1995) Seed Development and Germination. New York, Basel, Hong Kong: Marcel Dekker, Inc., 847 pp. Lee, J.-M., Williams, M.E., Scott, V.T. and Rafalski, J.A. (2002) DNA array pro®ling of gene expression changes during maize embryo development. Funct. Integr. Genomics, 2, 13±27. Michalek, W., Weschke, W., Pleissner, K.-P. and Graner, A. (2002) EST analysis in barley de®nes a unigene set comprising 4000 genes. Theor. Appl. Genet. 104, 97±103. Miller, R.T., Christoffels, A., Gopalakrishnan, C., Burke, J., Ptitsyn, A.A., Broveak, T.R. and Hide, W.A. (1999) A comprehensive approach to clustering of expressed human sequences: the sequence tag alignment and consensus knowledge base. Genome Res. 9, 1143±1155. Neuhaus, H.E. and Emes, M. (2000) Non-photosynthetic metabolism in plastids. Annu. Rev. Plant Physiol. Plant Mol. Biol. 51, 111±140. Nutbeam, A.R. and Duffus, C.M. (1978) Oxygen exchange in the pericarp green layer of immature cereal grains. Plant Physiol. 62, 360±362. Olsen, O.-A., Potter, R.H. and Kalla, R. (1992) Histo-differentiation and molecular biology of developing cereal endosperm. Seed Sci. Res. 2, 117±131.

ß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

Expression patterns in developing barley grains Patrick, J.W. and Of¯er, C.E. (2001) Compartmentation of transport and transfer events in developing seeds. J. Exp. Bot. 52, 551±564. Plaxton, W.C. (1996) The organization and regulation of plant glycolysis. Annu. Rev. Plant Physiol. Plant Mol. Biol. 47, 185±214. Potokina, E., Sreenivasulu, N., Altschmied, L., Michalek, W. and Graner, A. (2002) Differential gene expression during seed germination in barley (Hordeum vulgare L.). Funct. Integr. Genomics, 2, 28±39. Rolletschek, H., Borisjuk, L., Koschorreck, M., Wobus, U. and Weber, H. (2002) Legume embryos develop in a hypoxic environment. J. Exp. Bot. 53, 1099±1107. Rolletschek, H., Weber, H. and Borisjuk, L. (2003) Energy status and its control on embryogenesis of legumes: embryo photosynthesis contributes to oxygen supply and is coupled to biosynthetic ¯uxes. Plant Phys. 132, 1196±1206. Ruuska, S.A., Girke, T., Benning, C. and Ohlrogge, J.B. (2002) Contrapunctual networks of gene expression during Arabidopsis seed ®lling. Plant Cell, 14, 1191±1206. Sreenivasulu, N., Altschmied, L., Panitz, R., HaÈhnel, U., Michalek, W., Weschke, W. and Wobus, U. (2002) Identi®cation of genes speci®cally expressed in maternal and ®lial tissues of barley caryopses. A cDNA array analysis. Mol. Genet. Genomics, 266, 758±767. Stitt, M. (1998) Pyrophosphate as an energy donor in the cytosol of plant cells: an enigmatic alternative to ATP. Bot. Acta, 111, 167±175. Theodorou, M., Cornel, F.A., Duff, S.M.G. and Plaxton, W.B. (1992) Phosphate starvation-inducible synthesis of the alpha-subunit of the pyrophosphate-dependent phosphofructokinase in black mustard suspension cells. J. Biol. Chem. 267, 21901±21905.

ß Blackwell Publishing Ltd, The Plant Journal, (2004), 37, 539±553

553

ToÈrmaÈkangas, K., Kervinen, J., OÈstman, A. and Teeri, T. (1994) Tissue-speci®c localization of aspartic proteinase in developing and germinating barley grains. Planta, 195, 116±125. Wang, M. (1997) The role of abscisic acid in the regulation of barley grain germination. Seed Sci. Technol. 25, 67±74. Wang, T.L. and Hedley, C.L. (1993) Genetic and developmental analysis of the seed. In Peas: Genetics, Molecular Biology and Biotechnology (Casey, R. and Davies, D.R., eds). Norwich: CAB International. Wang, A.-Y., Kao, M.-H., Yang, W.-H., Sayion, Y., Liu, L.-F., Lee, P.-D. and Su, J.-C. (1999) Differentially and developmentally regulated expression of three rice sucrose synthase genes. Plant Cell Physiol. 40, 800±807. Weschke, W., Panitz, R., Sauer, N., Wang, Q., Neubohn, B., Weber, H. and Wobus, U. (2000) Sucrose transport into barley seeds: molecular characterization of two transporters and implications for seed development and starch accumulation. Plant J. 21, 455±467. Weschke, W., Panitz, R., Gubatz, S., Wang, Q., Radchuk, R., Weber, H. and Wobus, U. (2003) The role of invertases and hexose transporters in controlling sugar ratios in maternal and ®lial tissues of barley caryopses during early development. Plant J. 33, 395±411. Xu, F.-X. and Chye, M.-L. (1999) Expression of cysteine proteinase during developmental events associated with programmed cell death in brinjal. Plant J. 17, 321±327. Young, T.E. and Gallie, D.R. (1999) Analysis of programmed cell death in wheat endosperm reveals differences in endosperm development between cereals. Plant Mol. Biol. 39, 915±926.