Gene expression patterns reveal tissue-specific signaling networks controlling programmed cell death and ABA- regulated maturation in developing barley seeds



This article is corrected by:

  1. Errata: Erratum Volume 47, Issue 6, 987, Article first published online: 30 August 2006

*(fax +49 39482 5000; e-mail


Gene expression patterns covering over 10 000 seed-expressed sequences were analyzed by macroarray technology in maternal tissue (mainly pericarp) and filial endosperm and embryo during barley seed development from anthesis until late maturation. Defined sets of genes showing distinct expression patterns characterized both tissue type and major developmental phases. The analysis focused on regulatory networks involved in programmed cell death (PCD) and abscisic acid (ABA)-mediated maturation. These processes were similar in the different tissues, but typically involved the expression of alternative members of a common gene family. The analysis of co-expressed gene sets and the identification of cis regulatory elements in orthologous rice gene ‘promoter’ regions suggest that PCD in the pericarp is mediated by distinct classes of proteases and is under the hormonal control of both jasmonic acid (JA) and ethylene via ethylene-responsive element binding protein (EREBP) transcription factors (TFs). On the other hand, PCD in endosperm apparently involves only the ethylene pathway, but employs distinct gene family members from those active in the pericarp, and a different set of proteases and TFs. JA biosynthetic genes are hardly activated. Accordingly, JA levels are high in the pericarp but low in the endosperm during middle and late developmental stages. Similarly, genes acting in the deduced ABA biosynthetic pathway and signaling network differ between endosperm and embryo. ABA in the endosperm appears to exert an influence over storage product synthesis via SNF1 kinase. In the embryo, ABA seems to influence the acquisition of desiccation tolerance via ABA response element binding factors, but the data also suggest the existence of an ABA-independent but interactive pathway acting via the dehydration-responsive element binding (DREB) 2A TF.


Barley seeds (strictly caryopses) are composed of three genetically distinct tissue types: the maternal pericarp and seed coat, the filial embryo, and the triploid endosperm. These tissues are all heterogeneous within themselves, and are subject to dramatic changes during development. Based primarily on the stage of endosperm development, three defined developmental phases (pre-storage, storage and desiccation) have been recognized. The most dramatic transcriptional and physiological changes occur during transition from the pre-storage to the storage phase (Sreenivasulu et al., 2004; Wobus et al., 2004). Preliminary transcriptomic analyses of Arabidopsis (Ruuska et al., 2002), rice (Zhu et al., 2003) and barley (Sreenivasulu et al., 2004) seeds have focused on various metabolic pathways, but have not provided much information to date regarding either the regulators of transcription or the genes involved in signaling networks.

Therefore, in this study, we analyzed the barley seed transcriptome throughout development in three different tissue preparations and focused our interest on transcriptional regulators/transcription factors (TFs), kinases and signaling networks involved in programmed cell death (PCD), ABA-regulated seed maturation and desiccation control.

PCD is an integral part of the developmental process in plants (Hoeberichts and Woltering, 2003; Kuriyama and Fukuda, 2002; Lam, 2004; Young and Gallie, 2000) and is probably effected by (caspase-like) proteases (Woltering, 2004). It has been studied most extensively in the starchy endosperm and the aleurone layer during seed germination (Fath et al., 2000; Young and Gallie, 2000). For example, in developing maize endosperm, a balance between ABA and ethylene appears to control the onset and progression of PCD (Young and Gallie, 2000; Young et al., 1997). Only a small number of investigations have been directed at PCD in maternal tissues such as the maize pedicel (Kladnik et al., 2004), the wheat nucellus (Dominguez et al., 2001) and the seed inner integument of oilseed rape (Wan et al., 2002). In barley, groups of expressed genes indicative of PCD have been identified in the maturing pericarp (Sreenivasulu et al., 2004). Here we extend this observation, and propose a regulatory network, involving proteases, jasmonic acid (JA), ethylene and several TFs controlling PCD during pericarp maturation.

The role of ABA in seed maturation is to initiate the synthesis of storage compounds and to determine the levels of both seed dormancy and desiccation tolerance (Koornneef et al., 2002; Seo and Koshiba, 2002). A complex set of signal transduction networks involving hormones, sugars and TFs controls the expression of a broad spectrum of response genes. Our present knowledge is largely based on genetic studies in Arabidopsis (Finkelstein et al., 2002), and there is a need to extend this to other species. This large-scale transcriptomic analysis of barley caryopsis development has allowed the definition of a number of regulatory networks, based on the assumption that a congruent expression pattern for a set of genes is indicative of functional relationships between them (the ‘guilty by association’ concept). The approach we have described defines tissue-specific putative regulatory networks involved in PCD, maturation processes and desiccation control during barley seed development.

Results and discussion

Distinct gene expression clusters characterize tissue types and major developmental stages

To define the genetic programs acting in specific barley caryopsis tissues during seed development, transcript profiles were assessed with a 12 K barley seed array (see Experimental procedures). In total, 365 397 hybridization signals were determined from 31 duplicated experiments, sampling maternal (0–16 DAF), endosperm (0–26 DAF) and embryo (12–26 DAF) tissue (DAF, days after flowering). Upon filtering, approximately 20% of the total gene set present on the macroarray (2384 clones) was retained as representing differentially expressed genes with respect to developmental time and/or tissue. The remaining approximately 80% of detectable genes was expressed across all three tissues either constitutively or at a level that was too low to allow reproducible detection. A similar proportion has emerged from a study of Arabidopsis seed development (Ruuska et al., 2002). Of the set of differentially expressed genes, 508 (21%) encode putative regulators such as TFs, hormone and signaling components. A K-mean clustering method was used to identify six major gene clusters with characteristic expression in maternal tissues, endosperm and embryo. Within each cluster, a self-organizing map (SOM) procedure then defined subclusters showing distinct temporal patterns during development. The cluster-specific expression patterns as related to tissue type and developmental stage are shown in Figure 1. Overall, it is apparent that both developmental stage and tissue type are associated with specific sets of differentially expressed genes showing distinct patterns of expression. To validate these conclusions, a random selection of clones from representative clusters were used as probes in Northern blotting experiments, and this analysis proved consistent with the primary output of the macroarray (Figure S1).

Figure 1.

 Tissue- and development-specific clusters of gene expression profiles obtained from pericarp, endosperm and embryo tissue preparations during barley seed development.
A combination of K-mean and self-organizing map (SOM) clustering methods (see Experimental procedures) was applied to the differentially expressed set of 2384 genes. Major clusters (clusters 1–6) obtained by K-mean clustering, and subclusters obtained based on SOM analysis [subclusters 1(1), 1(2), 1(3) etc.] are indicated on the left by vertical bars. The number of genes (n) represented in each individual subcluster is indicated on the right. Expression values are given in logarithmically scaled (base 2) signal intensities: red, high expression; yellow, moderate expression; blue, low expression (see also colour scale at the top). Every horizontal row represents an individual gene. Developmental stages (days after flowering), assigned in 2-day intervals at the top, are represented in vertical columns.

Specific TF gene families are involved in the development of particular seed tissues

Extensive analysis of barley EST sequences extracted from 22 cDNA libraries has revealed about 1200 putative TFs (Zhang et al., 2004), of which 547 were present in the libraries made from developing seeds. Of these, 214 TFs (39%) were differentially expressed during seed development. According to the TRANSFAC and TIGR classifications, these TFs can be assigned to at least 20 families, among which the most prominent are zinc finger super family members (70 sequences), followed by AP2/EREBP family TFs (17 sequences), MADS box genes (16 sequences), and MYB TFs (12 sequences). Figure 2 illustrates the expression of the most abundant TFs in clusters 1(2) and 5(3), expressed specifically in pericarp and endosperm, respectively, and shows that members of certain TF families are significantly over-represented in particular clusters. A complete version of these data and analyses are given as Table S2 and Figure S2. In summary, zinc finger, MADS box and AP2/EREBP TFs dominate maternal tissue-specific subcluster 1(2), whereas MADS box genes are absent from endosperm subcluster 5(3), which includes, along with the zinc finger class, seven MYB TFs. This latter class is absent from the maternal subcluster 1(2). Other TFs [e.g. chromatin assembly factors of the bZIP class (nine genes), MCM family (four genes), HMG family (three genes) and WD-40 repeat family (two genes)] are expressed in all three developing tissues during early development (Figure S2).

Figure 2.

 Expression profiles of transcription factors preferentially expressed during pericarp [selected members from cluster 1(2), upper block] and endosperm [selected members from cluster 5(3), lower block] development.
The predicted name and family of putative transcription factor genes are indicated on the right. For further explanations, see legend to Figure 1. The complete set of transcription factor expression profiles is given in Figure S2.

Proteases, JA, ethylene and EREBP TFs are members of a putative network controlling PCD in the pericarp

Co-expression of genes is taken to indicate their involvement in the same functional pathway or network, as mentioned before. Here we present a comparison of putative regulatory networks of PCD between the maternal pericarp and the filial endosperm. The analysis is based on co-regulated gene clusters, and includes a search for TF-binding cis elements within the 5′ upstream regions of orthologous rice genes (as insufficient barley genomic sequence is available). This approach was validated by cross-species sequence comparisons between the two barley BAC clones 259I/6 (encoding granule-bound starch synthase) and 184G9 (encoding a globulin) and their rice orthologs, which demonstrated a high level of conservation not only between coding sequences but also between cis regulatory elements (data not shown). A similar conclusion has been drawn elsewhere using two different barley BACs (Dubcovsky et al., 2001).

Genes primarily expressed in maternal tissues.  During cereal seed development, the function of the maternal tissues changes from support for the filial tissue to its protection, via a process involving PCD. PCD events in seeds have been studied in some detail in the starchy endosperm and in the aleurone layers of germinating cereal grains (for recent reviews see Fath et al., 2000; Young and Gallie, 2000), but very little is known about PCD in maternal tissues with the exception of nucellar tissue (Bi et al., 2005; Chen and Foolad, 1997; Dominguez and Cejudo, 1998; Dominguez et al., 2001). PCD has been observed to dominate during the process of pericarp development, from 8 to 12 DAF (Sreenivasulu et al., 2004), a pattern confirmed in the present experiments, in which the expression of genes in subcluster 1(2) is largely restricted to maternal tissues (Figure 1). Of the 357 differentially regulated genes present in subcluster 1(2), at least 139 are associated with protein degradation (22 genes, mainly encoding proteases), lipid degradation (11 genes), transcriptional regulation (48 genes), signaling (34 genes) and hormonal pathways and their regulation (24 genes; Figure 3a; Tables S1, S2).

Figure 3.

 Identification and functional interpretation of the pericarp-specific genes from subcluster 1(2) related to programmed cell death (PCD).
(a) Expression profiles clustered using Pearson correlation. Based on the mean expression value of all cDNA fragments represented in this cluster, a green-shaded expression profile is depicted (right). After correlation-preserving expression profile alignment, standard deviation values shown in dark gray were calculated based on two independent experimental series. Vertical shading indicates developmental stages (yellow, pre-storage phase; light orange, intermediate stage; dark orange, storage phase). Selected enriched functional transcript groups possibly related to PCD are given as percentages of all genes in the subcluster and as absolute numbers in brackets. The individual functional categories selected for further detailed analysis are shown in bold; their corresponding expression profiles are shown in Figures 2 and 3(b).
(b) Hypothetical scheme of jasmonate and ethylene biosynthetic and signaling pathways related to protease gene control are shown in the centre and the expression profiles of selected genes found in cluster 1(2) to the left and right. Changes in transcript levels are indicated by colour scale (for further explanation, see legend to Figure 1). Abbreviations: 13-HPOT, 13S-hydroperoxy(9 Z,11 E,15 Z)-octadecatrienoic acid; OPDA, 12-oxophytodienoic acid; ACC, 1-aminocyclopropane-1-carboxylic acid; CTR, Raf-like protein kinase; ETR, ethylene receptor; EIN, ethylene-insensitive; ERF, ethylene response factor; EREBP, ethylene-responsive element binding protein. Levels of jasmonic acid and its precursor OPDA in the pericarp (black bars) and endosperm fractions (gray bars) during seed development are included on the left. Values represented on the y axis are pmol/g fresh weight. Two independent extractions and measurements were performed from two sets of independently grown plant material. Average values were plotted.
In the lower part, the presence of two motifs in the 1 kb upstream region of homologous rice ethylene/jasmonate response protease genes is indicated (for details, refer to Table S3).

Protease genes.  Proteases are heavily implicated in PCD (Woltering, 2004), and the protease genes grouped into subcluster 1(2) mostly fall into the C1 class and γ-VPE (vacuolar processing enzyme)-type cysteine proteases, along with cathepsin-related and subtilisin-like serine proteases (Figure 3b). In Arabidopsis, most of the latter form a gene family, the function of whose members remains largely uncharacterized. Some of these sequences are thought to play a specialized role in regulating senescence (Beers et al., 2004). Interestingly, the two subtilisin-like serine proteases (SAS1 and SAS2) exhibit caspase-like activity and have been shown to be associated with pathogen-induced PCD in Avena sativa (Coffeen and Wolpert, 2004). The presence of S8-type subtilisin-like serine protease transcripts (CA026376, CA023452, CA022715, CA026139) and an S14-type ATP-dependent ClP protease mRNA (BU975774) in the degenerating pericarp (Figure 3b) provides a further indication that this class of protease may be of as much importance for PCD as the cysteine and cathepsin proteases described previously (Sreenivasulu et al., 2004). An ortholog (sequence CA023673) of the integument-specific subtilisin-like protease SCS1 of soybean (Batchelor et al., 2000) was expressed both in the maternal tissue and the endosperm fraction [subcluster 2(1)], peaking at around 8–10 DAF (Table S3). Its presence in the endosperm fraction is probably a contamination artefact resulting from adherence of the integument. A C1 class cysteine protease (CA021721), expressed in late (12–16 DAF) pericarp, shares a high level of homology with two Brassica napus proteinases [BnSAG12 (AF0898848), a leaf senescence-associated cysteine proteinase, and BnCysP1 (AF448505), a cysteine protease involved in seed inner integument PCD] (Wan et al., 2002). Similarly, the other C1-type cysteine proteases in the cluster, CA020367 and CA023736, are related to Cys-EP from castor bean, which has been shown to be involved in nucellar PCD (Greenwood et al., 2005) and RD21, which is active during the senescence of Arabidopsis leaves (Yamada et al., 2001). Other cysteine proteases (CA022934, CA023714, CA023325 and CA028339) are caspase-related γ-VPEs, which have been implicated in the process of cell death in Arabidopsis seed coat cell layers (Nakaune et al., 2005). Finally, aspartic proteinase transcripts showing a high degree of similarity to PCD-linked animal cathepsin  D (CA021538, CA021489) and cathepsin  H (CA028182; aleurain) are expressed in degenerating pericarp tissue. Overall the indications are that the pericarp-expressed proteases of cluster 1 are involved in PCD processes.

TF genes.  Having defined the set of putative PCD-related proteases, we determined whether the TFs congruently expressed with the respective genes in subcluster 1(2) act in a functional context. This subcluster contains nine AP2/EREBP TFs, 18 zinc finger family members and nine MADS box genes (Figure 2). Rice orthologs of the 21 protease genes were determined (Figure 3b), and their 5′ upstream regions were scanned for promoter cis elements. This resulted in the identification of one (6 times found) and two to four (7 times found) CAACA/CACCTG elements in 13 regions, associated mainly with S8-type subtilisin-like serine protease and cysteine protease genes (Table S3). Such elements are recognized by AP2/EREBP TFs (RAV1, Kagaya et al., 1999). Four upstream regions contain either a GCCGCC or an ERF (ethylene response factor) motif, to which other members of the AP2/EREBP family can bind (Fujimoto et al., 2000). The AP2/EREBP clone CA028270 (Figure 3b) is most closely related to tobacco ERF2 (Ohta et al., 2000; 69% amino acid identity) and Catharanthus ORCA3 (van der Fits and Memelink, 2001; 60% identity at the nucleotide level), both of which are ethylene- and JA-responsive. Ethylene enhances PCD in cereal endosperm (Young and Gallie, 2000), while ACC oxidase, a subcluster 1(2) member and the key gene in the ethylene biosynthetic pathway that is expressed throughout development in maternal tissues, is induced by JA (Emery and Reid, 1996).

JA and ethylene pathways related to PCD.  Little is known concerning the role of JA in either seed development or PCD. Subcluster 1(2) genes that are upregulated in the degenerating maternal tissue include transcripts involved in both JA and ethylene biosynthesis. Their gene products are assumed to increase levels of jasmonate and ethylene in the pathways depicted in Figure 3(b). JA biosynthesis begins with a GDSL-motif lipase (CA020570, BQ469565, CA024107) and other lipase gene family members (CA020586, AL511679, CA020727, BU980997), together with lipid transfer proteins [CA025922, CA022749, CA021042, CA027579, CA022845, BU982975, see Table S2, cluster 1(2)], which are abundantly expressed during late pericarp development. Their precise functions remain unknown. The co-expression of lipoxygenase  2 genes (CA027764, CA024178, CA020483, CA022863, CA025433, BU974281) obviously reflects their participation in the supply of 13-hydroxy-9,11,15-octadecatrienoic acid (13-HPOT), which is involved in PCD (Bachmann et al., 2002; Maccarrone et al., 2001). Taken together, these observations suggest that lipolysis and the JA biosynthetic pathway are initiated in the senescing pericarp tissue. Also co-expressed are allene oxide synthase 1 and 2 genes (CA021313, AL511064), which have been shown to require 13-HPOT as substrate (Maucher et al., 2000). As a result of the activity of allene oxide cyclase (AOC), 12-oxo-phytodienoic acid (OPDA) is formed from highly unstable epoxide. The increase in AOC expression during late pericarp development correlates with increased OPDA levels, as confirmed by metabolite measurements (Figure 3b). CA021001 shares significant homology with tomato/Arabidopsis OPR3 (OPDA reductase 3) and is up-regulated in the pericarp. The OPR3 gene product specifically recognizes cis-(+)-OPDA (Schaller et al., 2000), which is transformed into JA by sequential β-oxidation. Thus the elevated levels of JA in the maturing pericarp (Figure 3b), also observed in soybean seeds (Simpson and Gardner, 1995), emphasize its importance in mediating PCD in pericarp tissue. JA has also been shown to promote leaf senescence (He et al., 2002), a comparable process. Although little is known concerning the key regulators of JA signaling networks in seeds, the TFs ERF2 (CA028270) and MYC2 (CA029559), co-expressed in the degenerating pericarp, may be involved, given that they act to integrate JA signaling networks in defense responses (Lorenzo and Solano, 2005). Furthermore, COI1 expression can be identified in the degenerating pericarp. The fact that this gene encodes an F-box protein (BU968442) involved in JA signaling (Xie et al., 1998) underlines further that JA very probably plays an important role in protein degradation in the pericarp.

In addition to genes involved in JA biosynthesis and signaling, an extensive set of ethylene signal transduction pathway genes is activated during pericarp degeneration, including ethylene receptors (ETR3, AL511633), CTR1 (CA029560, BU971644, BU973230, CA024753), a MAP kinase (BU968393), EIN2 (CA026029, CA020555) and ERF2 TF (CA028270; Figure 3b). Genetic analysis of tomato orthologs expressed during fruit ripening suggests that ETR interacts with CTR1, and that signal transduction is effected via a MAP kinase cascade to EIN2. EIN2 is thought to be central to the control of several hormone signal transduction pathways, and can activate the TF ERF (an EREPB family member; Adams-Phillips et al., 2004). The ethylene biosynthesis and signaling pathway depicted in Figure 3(b) is based both on barley gene expression data and on comparative data derived from fruit ripening studies of tomato (Adams-Phillips et al., 2004), a process bearing some similarity to cereal pericarp development.

We suggest that both JA and ethylene are involved in regulating maternal seed tissue PCD, and that the barley ERF2 TF (CA028270) acts as the signal integrator and inducer of PCD-related proteases, as a result of the ERF-binding motifs GCCGCC and CAACA/CACCTG present in their promoters. Whether other EREBP factors are also involved remains unclear. In addition, we note that the two-component response regulator ARR, which acts downstream of ETR in ethylene signaling independently of CTR1 (Hass et al., 2004), is also expressed in degenerating pericarp tissue (CA024680; Table S2). In summary, both hormones may trigger PCD either via the EREBP TFs (Figure 3b) and/or by directly inducing PCD-involved proteases. The latter possibility is suggested by the observation that JA induces a subtilisin-like serine protease in several Arabidopsis tissues (Golldack et al., 2003).

The possible role of MADS box TFs, receptor kinases and calcium signaling in pericarp degeneration

Seven of the MADS box TFs (CA023749/CA021244, CA023514, CA024629, CA024291, AL511037, AL506475, BU981579) in the pericarp-specific subcluster 1(2) are homologous both to the AGL (agamous-like) family of Arabidopsis (Ma et al., 1991) and to the MIKC-type ZmMADS2 protein of maize (Figure 2). Whether their co-expression leads to heterodimer formation, as occurs with the petunia MADS-box FBP protein members (Immink et al., 2002), requires further investigation. MIKC maize protein is expressed during anther dehiscence, which is a PCD process (Schreiber et al., 2004). Two sequences (CA021403, BU967322) are homologous with a tomato RIN MADS-box TF involved in the regulation of floral development and fruit ripening. This gene is required for the initiation of climacteric respiration and associated ethylene biosynthesis, and functions upstream of ethylene in the regulatory cascade. It is thought to represent a global developmental regulator of ripening (Vrebalov et al., 2002). The preferential expression of agamous-like MADS box TFs in barley maternal seed tissue after fertilization is similar to the situation in Arabidopsis (de Folter et al., 2004), and this suggests an additional function for this gene (apart from its established role in flower development) in the development of the maternal seed parts.

Subcluster 1(2) also contains a large number of receptor kinase sequences (CA022559, CA026732, CA020740, BQ459175, AL509203, CA021329, CB884149, CA023468, CA028738), as well as calmodulins (BU972080, CA027509, BQ458375, CA022282, CA023915, CA026655, BQ458716, CA022383, CA022507) and a calcium-dependent protein kinase (BU977255) [Figure S3, cluster 1(2)]. It is noteworthy that various non-ripening tomato mutants (NR, RIN and NOR) lack calcium-dependent protein kinase  1 expression (Leclercq et al., 2005), and that EIN2 is implicated as a calcium transporter (Adams-Phillips et al., 2004). However, the functional relevance of the gene expression patterns during pericarp degradation in relation to calcium has yet to be elucidated.

PCD-related processes in the starchy endosperm are characterized by selective proteolysis

PCD in developing barley endosperm progresses in a gradient manner from central to peripheral regions (L. Borisjuk , IPK, Gatersleben, personal communication). As a consequence of PCD, the starchy endosperm successively becomes a dead tissue during the latter part of its development, although it still retains cellular structures with remnants of nuclei, RNA and ribosomes. This contrasts with the pericarp, where cell structures are almost completely re-absorbed. Genes highly expressed during the later phase of endosperm development are grouped in subcluster 5(3). The protease genes in this subcluster are metalloproteases (Table S3) and components of the proteasome complex, such as the 20S (AL510971)/26S proteasome (BU975809, BU968712), ubiquitin and its conjugates (BU973122, BU988115, BU981092, BU987993, BU970892, BU972422, BU990445) and ClpAP protease (BU986436; Figure 4). This complex acts as a threonine protease. Serine and aspartic protease transcripts, characteristically present in the pericarp [subclusters 1(2) and 2(1)], are absent from the endosperm. An F-box protein gene, ORE9 (BQ464059), is specifically expressed in the endosperm along with the ubiquitin/26S proteasome complex. Its Arabidopsis ortholog prolongs the onset of senescence (Woo et al., 2001). The ubiquitin/26S proteasome pathway confers a high selectivity to proteolytic processes (see review by Moon et al., 2004), which is of particular importance in the starchy endosperm, a tissue in which protein synthesis, storage product accumulation and degradation occur simultaneously, and in which starch accumulation is linked to PCD (Young and Gallie, 2000). Some fine-tuning may be conferred by a set of proteinase inhibitors (Solomon et al., 1999) [23 entries, see Table S2, cluster 5(3)], likewise absent from the pericarp.

Figure 4.

 Identification and functional interpretation of the endosperm-specific programmed cell death (PCD)-related gene subcluster 5(3).
On the left, a hypothetical scheme of ethylene biosynthetic and signaling pathways related to protease gene control and the inhibitive role of ABA in endosperm PCD is illustrated based on literature data (Bethke et al., 1999), whereas the expression profiles of selected genes found in cluster 5(3) are shown on the right. A total of 25 + 15 homologous rice genes were defined and analyzed for the presence of ERF/EREPB-binding cis elements (left). For details, refer to Table S3. Changes in transcript levels (right) are indicated by the colour scale (for further explanation, see legend to Figure 1). For abbreviations, see legend to Figure 3(b).

Signal transduction components related to ethylene are present in both the pericarp and endosperm, but as different gene family members. This holds also for ethylene receptor ETR3 (BU967042), histidine kinase receptors (BU969609, BU990435), CTR1 (BU971280), EIN3 (BQ460618) and several EREBP-related TFs (BQ464182, BU972517, BU972019), which are all specifically expressed during starchy endosperm PCD [Figure 4 and Figure S3, subcluster 5(3)]. In Arabidopsis, the F-box proteases EBF1 and EBF2 are involved in the ubiquitinization-dependent post-translational regulation of EIN3, a trigger of ethylene-responsive gene expression (Potuschak et al., 2003). The barley F-box protease orthologs BU972329, BU71788 and BQ464059, and the EIN3 ortholog BQ460618, are expressed late in barley endosperm development, suggesting that they play a similar role in barley. Based on the transcription profiles of ethylene-related genes, we postulate a role for ethylene in starchy endosperm PCD, as in the pericarp. However, in contrast to the model for the pericarp, JA signaling genes are not expressed to a significant degree in the endosperm. Instead, ABA biosynthetic genes are active during middle to late endosperm development [Figure S3, subcluster 5(3)]. ABA is known to prevent the PCD of aleurone cells in barley (Bethke et al., 1999). In the growing embryo, no ethylene biosynthesis genes were active, whereas the expression levels of ABA biosynthetic genes were elevated. Tissue-specific transcription profiles of ethylene biosynthesis and signaling genes in developing maize endosperm and embryo (Gallie and Young, 2004) are largely similar to those in barley, but there do appear to be differences in the expression of ethylene receptor genes in the embryo between these species.

Gene expression profiles reveal differences in ABA biosynthesis and deduced signaling networks between endosperm and embryo during seed maturation

ABA is the primary mediator of seed maturation. It induces both the synthesis of storage products as well as desiccation and dormancy processes (Koornneef et al., 2002; Seo and Koshiba, 2002). Although the general role of ABA in cereal seed maturation is well understood, a comparison between the regulatory networks in the endosperm and embryo is lacking. We have explored the endosperm and embryo transcriptome during maturation, and have identified putative regulators driving the synthesis of storage compounds in the endosperm and the acquisition of desiccation tolerance in the embryo. Special emphasis was given to expression patterns of genes involved in ABA biosynthesis and signaling, co-regulated TFs and putative ABA-responsive genes. The latter were identified by searching for ABA-responsive elements in the upstream regions of rice orthologs.

Approximately 950 genes out of the approximately 12 000 present on the array are expressed during late endosperm development. This is the largest gene group specifically expressed in an individual tissue during a particular stage of seed development. Gene annotation identifies about 18% of these as being involved in regulatory processes. Prominent TFs, expressed specifically in the endosperm during the main storage phase, group to subcluster 5(3) and belong mostly to the zinc finger super family (21 bZIP, DOF and C3H/C3HC4 genes), along with seven MYB and five NAM members (Figure 2). Genes associated with ABA, gibberellic acid (GA) , auxin and ethylene are also expressed during late endosperm development [ Figure 5, subclusters 5(1) and 5(3)], with ABA being the most important of these hormones. However, the interpretation of cluster 5 gene expression patterns is complicated by the probable presence of aleurone tissue in the starchy endosperm fraction. Prominent TFs expressed preferentially in the embryo during maturation belong to the AP2-type DREB family and the bZIP-type ABA response element binding factors.

Figure 5.

 Groups of genes preferentially expressed in endosperm [subclusters 5(1), 5(2), 5(3)] and embryo [subclusters 6(1), 6(2)] during maturation.
Expression profiles on the right were determined as described in the legend to Figure 3(a). Functional transcript groups are given as percentages of all genes in a given subcluster and as absolute numbers in brackets. The individual functional categories selected for further detailed analysis are shown in bold; their corresponding expression profiles are depicted in Figures 6 and 7.

ABA in developing seeds is supplied either by the maternal tissues or is synthesized de novo in the embryo (Frey et al., 2004; Karssen et al., 1983). ABA biosynthesis and signaling genes are well studied in non-seed tissues responding to abiotic stress, whereas little is known of the equivalent processes in seeds (Nambara and Marion-Poll, 2003). In plastids of green tissues, biosynthesis is initiated by the epoxidation of zeaxanthin, and the resulting xanthoxin is imported into the cytosol, where a two-step conversion occurs from xanthoxin to ABA via ABA-aldehyde (Schwartz et al., 2003). Aldehyde oxidase (AO) catalyzes the conversion of ABA-aldehyde to ABA (Seo and Koshiba, 2002). At least two members of an AO gene family are expressed during barley seed development. HvAO1 mRNA (CB883703) is found exclusively in developing endosperm beginning at 10 DAF (Figure 6), whereas HvAO2 mRNA (BU988574) is seen mainly in the embryo preparation, although also, but at a much lower level, in the late endosperm fraction (Figure 7). Moreover, Cnx1 [BU984387; cluster 6(2)] encoding the key enzyme for molybdenum cofactor (MoCo) biosynthesis and allowing AO activity, is unexpectedly present at much higher levels in the embryo than in the endosperm. We conclude that ABA biosynthesis occurs independently in developing embryo and endosperm and is enabled by distinct AO gene family members.

Figure 6.

 Expression pattern of key regulators related to maturation events in the endosperm.
A hypothetical scheme of ABA signaling pathways and its relation to starch biosynthesis is shown, together with relevant barley accession numbers and expression profiles of selected genes (right; for explanation, see legend to Figure 3b). Analysis of co-expressed gene sets of endosperm-specific clusters of barley and identification of ABRE cis regulatory elements in the respective rice gene ‘promoter’ regions suggest that MYB, zinc finger, SNF1, RGA1 and PKABA1 are ABA-responsive (see also Table S4), and that ABA-mediated regulation seems to exert control on starch biosynthesis via SNF1, MYB and MYC factors. ABA also seems to inhibit pre-mature germination via RGA1, PKABA1 and ABI3.

Figure 7.

 Expression pattern of key regulators related to maturation events in the embryo.
Putative response genes of the putative ABA signaling pathway in the embryo defined by the presence of ABRE motifs in the 5′ upstream regions of rice genes homologous to genes of clusters 6(1) and 6(2) were used and checked for the occurrence of ABRE motifs. ABA-responsive genes identified from cluster 6(1) are indicated along with functional classes on the left (see also Table S5). Regulators controlling desiccation tolerance during embryo maturation via ABA-dependent pathways from members of cluster 6(2) are shown on the right.

Osakabe et al. (2005) have recently demonstrated the participation of Arabidopsis RPK1 (receptor protein kinase) in the early perception of ABA. Its barley ortholog BU986612 is expressed during both endosperm and embryo development (Figure 6). In addition, distinct gene family members of ABA signaling genes are expressed during late endosperm and embryo development. While ABA-insensitive (ABI) protein phosphatases (AL511807, BU974458, BU973214, CA022711, BU975906, CA028969), calcium-dependent protein kinases (BU990065, BU990390, AL511267, BU987382) and MAP kinase (BU989059), all known to participate in ABA signaling, are expressed during late endosperm development, other protein phosphatases (BU990528, BU973761, BU984765) and a distinct MAP kinase (CA022676) are preferentially expressed in late embryo development (Figure 7). ABRE-binding factors (ABFs) HvABF1 (BU984330) and HvABF2 (BU984330), which activate ABA-dependent gene expression (Finkelstein et al., 2002), are co-expressed in both endosperm and embryo tissues during the maturation phase. Our data suggest overall that the ABA signaling networks in the endosperm and embryo are different from one another.

During endosperm maturation, ABA is involved in the control of starch synthesis probably via SNF1 kinase

Nambara and Marion-Poll (2003) have established the existence of cross-talk between ABA and sugar signaling during endosperm development, which may be mediated via SNF1-related kinase (Halford and Paul, 2003). This kinase in turn may regulate key starch biosynthesis genes, such as sucrose synthase and AGPase (Halford and Paul, 2003; Tiessen et al., 2003). The barley data are consistent with this model, as barley SNF1 kinase (BU986153, BU973086) is co-expressed with bZIP ABI5 TF (BU971616, BU970111) and MYB TF (BQ460434) in subcluster 4(2) (Figure 6, Table S2). Furthermore, the rice ‘promoters’ of SNF1, MYB TF AL510872 and the zinc finger TF BU971291 all possess ABRE elements (Table S4). All these TFs are known to be transcriptional activators responding to ABA signaling and able to bind to MYB and MYC motifs (Abe et al., 2003). In a screen for cis elements in the upstream regions of rice starch metabolism genes, which are orthologs of barley cluster group 5 genes, only MYC, MYB and GA response motifs (pyrimidine boxes and GARE elements) were found (Table S4). In contrast, hordein promoters possess a prolamin box in addition to the MYC, MYB and GA response motifs, and it is the former that mainly determines expression (Mena et al., 1998). We conclude that, during endosperm maturation, ABA is apparently involved in the modulation of starch synthesis via SNF1 kinase and a set of TFs.

ABRE motifs suggest that an ABA-regulated gene network imposes control on lipid storage via oleosin gene expression

In a search for genes regulated by ABA through ABRE binding factors, 1 kb of the 5′ upstream regions of rice orthologs of cluster 5, 6(1) and 6(3) genes were queried for the presence of ABRE motifs. Whereas the major endosperm-specific cluster 5 contained only 12 such genes (of 115) (Table S4), 20 of 56 embryo-specific genes in cluster 6(1) and 16 of 63 genes in cluster 6(2) possess at least one ABRE motif (Tables S5 and S6). Interestingly, ABRE elements were found in the upstream regions of rice orthologs of the steroleosins BQ464496, BQ464153, BU986762 and oleosin BU969988, but not in those of the lipid biosynthesis genes ketoacyl-ACP synthase  II, mono- and diacylglycerol acyltransferase, epoxide hydrolase and lysophospholipase  2. In barley, lipids are stored almost exclusively in the embryo, and storage is often associated with the synthesis of oleosins. As oleosins are generally restricted to the oil bodies of seeds undergoing dehydration during maturation, where they serve as stabilizers at low water potential (Buchanan et al., 2000), their control by ABA is consistent with its function during desiccation. Other genes characterized by at least one ABRE motif in the orthologous rice promoters are storage globulin genes (BU990108, BU987019), genes controlling redox potential (BU985104, BU986709) and five genes of unknown function (Figure 7, Table S5).

DREB TFs are probably involved in the acquisition of desiccation tolerance during late embryo development either dependently or independently of ABA

At least 20 of 56 genes from cluster 6(1) feature a DRE motif, which binds DREB TFs, in their rice orthologous upstream region. As at least some DREB TFs are ABA-responsive, both ABFs and DREB TFs are likely to act within a maturation-controlling regulatory network. Most notably, among the genes carrying DRE elements in their ‘promoters’ are aldehyde oxidase (BU988574) and the lipid biosynthetic genes mono- and diacylglycerol acyltransferase (BU975092, BU987629), as well as a redox potential gene (BQ464170; see Table S5).

Subcluster 6(2) contains 65 genes expressed almost exclusively during embryo development (Figure 1), at a time when desiccation is initiated by declining water content in the endosperm, but where the embryo remains largely unaffected by this stress due to its acquisition of desiccation tolerance. It is therefore reasonable to assume a functional relationship between genes in the cluster with respect to desiccation [Figure 5, subcluster 6(2)]. The most prominent groups within this subcluster are stress-responsive genes (18%), which include late-embryogenesis-abundant (LEA) proteins (BU989580, BU987165, BQ464073, BU985128, BU987905, BQ464035, BU989436) and LEA class III dehydrins (BU990280, BQ464373, BU989205), and genes of unassigned function (30%; Figure 5 and Table S6). Involvement of barley LEA proteins in desiccation tolerance has been clearly demonstrated in transgenic experiments (Bahieldin et al., 2005). These proteins are thought to induce changes in protein conformation and thereby prevent damage to protein aggregates during water stress (Goyal et al., 2005).

Within cluster 6(2), TFs and other regulatory genes are co-expressed with LEA transcripts (Figure 5). Among the TFs in this cluster are three putative DREB gene sequences (AP2/EREBP family members), belonging to the DREB2 subfamily. BU984499 (HvDREB2A) shares 96% identity with DREB2 of Hordeum brevisubulatum (AAU29412) and 71% identity with rice OsDREB2A (Dubouzet et al., 2003; AN02487), whereas BQ469825/BU986453 sequences (HvDREB2B) show 82% identity with OsDREB2 (AAP70033). The full-length sequences of HvDREB2A and HvDREB2B contain open reading frames of 278 and 283 amino acids, respectively, with gene products of predicted molecular masses of 30.3 and 30.5 kDa.

DREB2 TFs are activated during dehydration stress (Liu et al., 1998), and in Arabidopsis (Sakuma et al., 2002; Yamaguchi-Shinozaki and Shinozaki, 1994) and rice (Dubouzet et al., 2003) bind to the dehydration-response element DRE with its core motif G/ACCGAC. HvDREB2A and HvDREB2B are expressed not in response to external stress conditions, but rather during the course of normal seed development and maturation (Table S6), where they may be implicated as regulators of seed desiccation. To gain support for this hypothesis, the promoter regions of the rice orthologs of subcluster 6(2) genes were searched for DRE elements, and at least one DRE element was found in 29 out of the 63 relevant genes (Table S6). Major LEA proteins, including class III type dehydrins, and a substantial number of unknown genes in the cluster show one or more DRE motifs in their orthologous rice promoter regions. The promoter of OsDREB2A itself possesses three separate ACCGAC motifs, while OsDREB2B has one GCCGAC motif (Table S6).

Because ABA plays a role in the activation of DREB genes (Narusaka et al., 2003), the 5′ upstream regions of OsDREB2A and OsDREB2B were screened for the presence of ABRE motifs. While there are no such motifs in the OsDREB2A promoter, two (CACGTGGC and CACGTGTC) are present in the promoter region of OsDREB2B (Figure 7 and Table S6). As ABRE and DRE can act interdependently (Narusaka et al., 2003), we suggest that while DREB2B activation is under the influence of ABA via ABRE motifs, DREB2A expression is independent of ABA. In addition, DREB2A and DREB2B may support their own expression, either by cross-activation and/or via self-regulation imposed by the DRE elements in their promoters.

Other genes in subcluster 6(2) are apparently regulated by ABA and DREB TFs, as nine upstream orthologous rice regions bear both ABRE and DRE elements (Table S6). At least one of these genes, the functionally uncharacterized embryo-specific barley gene PM19 (BU985212, BU987552), with two DRE motifs and two ABRE sequences, has been experimentally shown to be activated by ABA application (Ranford et al., 2002). In summary, we conclude that embryo-specific gene expression patterns delineate a regulatory network for the acquisition of desiccation tolerance, which is controlled by ABA-mediated signal transduction via ABFs, and in an apparently ABA-independent manner via the transcription factor DREB 2A.

Concluding remarks

The data we present provide the most comprehensive publicly accessible transcriptomic analysis of cereal seed development. It is based on a set of nearly 12 000 genes derived from cDNA libraries of developing barley seeds. We have analyzed separately three tissue types at eight (embryo), nine (pericarp) and 14 (endosperm) time points between fertilization and the late maturation stage, and have provided an extensive insight into tissue-specific gene expression patterns. A number of well-defined sets of genes characterize gene expression both in a tissue- and in a developmental stage-specific manner (Figure 1), and reflect tissue-specific physiological processes. A process such as PCD is executed by different sets of proteases in different seed tissues (pericarp and endosperm). PCD appears to be under the control of ethylene and JA signal transduction pathways in the pericarp, and of ethylene and ABA in the endosperm ( Figure 8). It is of special interest that several steps in the suggested ethylene pathway are mediated by the expression of tissue-specific gene family members.

Figure 8.

 Model scheme representing putative key regulators controlling PCD and storage processes in distinct tissues of developing barley seed.
Time-series transcriptome analysis, selected metabolite profiling and analysis of orthologous rice gene promoters for known cis elements reveal key regulatory components putatively controlling PCD and pathways of storage product accumulation. Only a fraction of the candidate regulatory genes identified in this study are shown here. The complete list of regulatory genes (TFs, kinases and hormone-related genes) are included in Figures S2, S3 and S4 respectively.

A further significant outcome of this study is the identification of various key regulators of maturation pathways in endosperm and embryo. Based on identical gene expression patterns and regulatory cis elements defined in orthologous rice gene upstream regions, we suggest that ABA exerts control over starch biosynthetic genes in the endosperm via SNF1 kinase and several TFs, whereas, in embryo, ABA influences a different gene network including oleosin and storage globulin genes (Figure 8). Desiccation-related genes in the embryo appear to be controlled in two different but probably interacting ways: ABA-dependent via ABA-responsive element binding factors including DREB 2B, and ABA-independent via the TF DREB 2A.

This analysis of tissue-specific gene expression in developing barley seeds underlines the value of genome-wide transcriptome analysis for the elucidation of regulatory networks, although many of the conclusions drawn will require experimental validation. More importantly, the tissue-specific results stress the need to bring transcriptome studies generally to the tissue and cell-type level, because a whole seed is a too complex an organ to analyze in toto.

Experimental procedures

Plant material

Barley plants were grown in controlled environment conditions under a 16 h light/19°C and 8 h dark/14°C regime during their vegetative phase. The time of fertilization and developmental stage of caryopses were determined according to the method described by Weschke et al. (2000). Developing seeds from the middle part of the spike were harvested at 2-day intervals from 0 to 26 days after flowering (DAF), i.e. from fertilization to late maturation. Maternal tissue including pericarp, for simplicity mostly called ‘pericarp’, and filial ‘endosperm’, as well as ‘embryo’ fractions, were dissected under a microscope and quick-frozen in liquid nitrogen. The pericarp fraction was isolated from 0 to 16 DAF samples, the embryo fraction from 12 to 26 DAF and the remaining fraction representing the major proportion of endosperm from 0 to 26 DAF. The tissue fraction samples were pooled from at least ten caryopses for the purposes of homogeneity. Experiments were replicated once with independently grown material. Whereas embryo and pericarp preparations were largely free of contamination by endosperm cells, early endosperm fractions (0 to 4 DAF), in particular, included maternal-related tissues such as nucellar epidermis and nucellus tissue. Furthermore, photosynthesizing chlorenchyma cells tend to adhere to transition-phase endosperm [for further details, see Sreenivasulu et al. (2002) and].

Identification and functional categorization of a 12 K unigene set from developing seed cDNA libraries

A total of 40 752 ESTs were generated from four independent cDNA libraries sampled from 0 to 7 DAF maternal tissue, 0–7 DAF filial tissue, 8–15 DAF caryopses and 16–25 DAF caryopses. Details of cDNA library construction, quality check parameters of ESTs and functional categorization based on Munich Information Center for Protein Sequences (MIPS) functional catalogues have been described elsewhere (Zhang et al., 2004). Sequence and clustering information, functional annotation, TOP-BLAST X results and predicted open reading frames for all ESTs can be accessed at The 40 752 sequences were clustered into 7682 contigs and 2962 singletons using the program StackPACK version 2.1.1. For optimal annotation, functional assignment of the resulting 10 644 sequences was based on homology-based search results generated from seven different databases. Four of these [MapMan, MIPS, Swiss-Prot and DBORA (Lange et al., 2005)] were used for the annotation of cellular and metabolic pathway genes. TF and kinase genes were assigned with the TRANSFAC and Plants P databases. tBLASTX searches were performed using predicted open reading frame sequences of TFs identified in the Arabidopsis and rice genome sequences. Plants T and ARAMEMNON databases were used to identify transporter genes. The annotation results were manually curated and structured, according to MapMan vocabulary.

12 K barley seed cDNA macroarray design

The 12 K barley seed array was created using the unigene set of 10 644 clones described above. A further 1143 clones representing alternative members of large gene families were included as controls. The resulting 11 787 clones were arrayed into a 96-plate format using a TECAN robot (Genesis workstation  100). All 11 787 EST clones were re-sequenced from the 3′ direction to confirm their sequence identity. PCR amplifications, purification of amplified PCR products and spotting of equal amounts of DNA on 11 × 22 cm nylon membranes was carried out at German Resource Center for Genome Research (RZPD) (Berlin, Germany). The clones were spotted in duplicate, and 1461 spot positions were left free of DNA to be used for background subtraction and normalization of hybridization signals. Spotting was performed in blocks of 48 × 24, where each block contained 25 spots arranged in a 5 × 5 pattern, representing 12 genes spotted in duplicate along with a central empty guiding spot.

Content of the 12 K seed array

A stringent StackPACK clustering analysis showed that 73% of the sequences on the array overlapped with those represented on the 22 K barley Affymetrix gene chip (Close et al., 2004), leaving a set of 3088 unique seed ESTs. In all, annotation succeeded in assigning function to 6585 genes (Figure S6), including 2177 metabolism-related genes, 547 TFs, 462 signaling kinases and 196 hormone metabolism genes. The remaining 5201 genes are unassigned due to lack of specified annotations or poor similarities to characterized genes from other species. In summary, the 12 K barley (Hordeum vulgare) seed array represents the largest collection of sequences from developing barley seeds.

Probe preparation, hybridization and processing of cDNA arrays

RNA samples were prepared from dissected tissues of developing seeds according to GENTRA kit instructions (Biozyme, Germany), and subjected to DNAase digestion (Qiagen, Milden, Germany), and an additional purification step using the RNeasy kit (Qiagen). Purified RNA (35 μg) was used for the synthesis of first-strand cDNA. 33P-labeled second-strand cDNA was obtained by a random priming reaction (Amersham, USA). Synthesis of 33P-labeled cDNA probes, hybridization and processing of cDNA arrays were carried out as described by Sreenivasulu et al. (2002). The hybridized cDNA arrays were exposed to phosphor image processing screens for 6 h, which were then scanned at 100 μm resolution using a Fuji BAS 3000 phosphor scanner (Tokyo, Japan).

Data analysis

The Array Vision Image analysis program (St. Catharines, Ontario, Canada) was used for spot detection and signal intensity measurements. Local background was subtracted from the intensity of each target spot, and a mean intensity was generated from the duplicate spots of each clone. Data were normalized by median centering of the arrays using an R-based script as described by Sreenivasulu et al. (2004), and transformed to a logarithmic scale. Two filtering criteria were employed. The first removed weak and non-reproducible signals, the second applied a standard deviation method (1.0) according to Genesis (Sturn et al., 2002) to predict differentially expressed genes showing significant differences of expression between different tissues during the developmental periods analyzed. As a result, 2384 were retained. In order to extract the fundamental patterns of gene expression inherent in pericarp, endosperm and embryo development of barley caryopses, a combination of K-means clustering and self-organizing maps (SOM) provided by the Genesis software was applied (Sturn et al., 2002). First, tissue-specific major patterns were identified by K-means. These patterns were subsequently taken as starting points for a more sensitive search for co-expressed genes by using SOM clustering. A more detailed description is given in Note S1.

The expression patterns obtained from two independent sets of experiments were in nearly perfect accordance with one another (Table S2), and largely matched a more restricted data set published previously (Sreenivasulu et al., 2002, 2004). The entire set of differentially expressed gene expression profiles, along with functional annotation and corresponding sequences, is searchable at and is also provided as Table S2.

The multi-dimensional scaling (MDS) approach described by Strickert et al. (2005) was applied for the full 365 397 hybridization signals generated in 31 experiments with the 12 K seed array. The MDS data plot (Figure S5) indicates that, with the exception of the intermediate stages, a low degree of variation, i.e. a high degree of reproducibility, characterized the two independent sets of experiments. Further, the MDS data output clearly separates the tissues and supports the developmental stages established from a much smaller expression data set (Sreenivasulu et al., 2004). An intermediate (transition) stage had been defined in cultivar Barke, featuring a massive re-programming of gene expression especially in the endosperm (Sreenivasulu et al., 2004). The present data set provides even stronger evidence for this intermediate stage in endosperm development occurring between 6 and 10 DAF in cultivar Brenda, but is less conclusive with respect to the development of maternal tissues (Figure S5). The transition-stage differences between the two data sets (open and filled symbols for maternal tissue and endosperm of Figure S1) are most probably a consequence of rapid transcriptional changes during this stage.

Promoter analysis

Consensus barley sequences of the tissue- and developmental- specific cluster members were extracted and used for a BLAST search of the rice genomic sequence via the TIGR database ( From the ATG start site, 5′ upstream 1 kb ‘promoter’ regions were extracted and queried for the presence of known cis elements, as listed in Tables S3–S5.

Quantitative analysis of jasmonates and octadecanoids

Tissues of developing seed (pericarp and endosperm fractions) obtained from two independently grown sets of plants were extracted with 80% v/v methanol and prepared for GC-MS analysis for the quantification of endogenous JA, as described by Stenzel et al. (2003). For quantitative analysis of OPDA, we followed the method described by Mueller and Brodschelm (1994).


We thank Claus Wasternack for his valuable comments, which helped to improve the manuscript. Uwe Scholz and Heiko Meihe were instrumental in creating the expression database. Börn Usadel, Oliver Timm and Svenja Meyer helped in functional gene annotation. We are grateful to Angela Stegmann, Gabriele Einert and Elsa Fessel for their excellent technical assistance. This work was supported by the Federal Ministry of Education and Research (BMBF; GABI SEED II grant).