miRNA regulation in the early development of barley seed

Background During the early stages of seed development many genes are under dynamic regulation to ensure the proper differentiation and establishment of the tissue that will constitute the mature grain. To investigate how miRNA regulation contributes to this process in barley, a combination of small RNA and mRNA degradome analyses were used to identify miRNAs and their targets. Results Our analysis identified 84 known miRNAs and 7 new miRNAs together with 96 putative miRNA target genes regulated through a slicing mechanism in grain tissues during the first 15 days post anthesis. We also identified many potential miRNAs including several belonging to known miRNA families. Our data gave us evidence for an increase in miRNA-mediated regulation during the transition between pre-storage and storage phases. Potential miRNA targets were found in various signalling pathways including components of four phytohormone pathways (ABA, GA, auxin, ethylene) and the defence response to powdery mildew infection. Among the putative miRNA targets we identified were two essential genes controlling the GA response, a GA3oxidase1 and a homolog of the receptor GID1, and a homolog of the ACC oxidase which catalyses the last step of ethylene biosynthesis. We found that two MLA genes are potentially miRNA regulated, establishing a direct link between miRNAs and the R gene response. Conclusion Our dataset provides a useful source of information on miRNA regulation during the early development of cereal grains and our analysis suggests that miRNAs contribute to the control of development of the cereal grain, notably through the regulation of phytohormone response pathways.


Background
MicroRNAs (miRNAs) are a class of non-coding small RNAs (smRNAs) that act to reduce expression of target genes by interacting with their target mRNAs in a sequence-specific manner. Since their discovery it has become clear that miRNAs are an important component in the regulation of many genes in most eukaryotic cells. In plants, most currently validated miRNA targets code for transcription factor families with crucial developmental functions, including the control of root and shoot architecture, vegetative to reproductive phase transitions and leaf and flower morphogenesis [1,2]. miRNAs are processed from a primary miRNA transcript which folds to form an imperfect stem-loop. The pri-miRNA hairpin is recognised and processed to a smRNA duplex consisting of the miRNA and complementary miRNA* by a protein complex containing a DCL1-type RNase. The mature miRNA, which is typically 20-21 nt in length, is then incorporated into the RNA Induced Silencing Complex (RISC) to regulate one or more target genes in trans through a base pairing mechanism. Most plant miRNAs appear to trigger both mRNA cleavage (between the nucleotides matching the 10 th and 11 th position of the miRNA) and translational repression of their target genes [3]. Although these two mechanisms are additive, they can be dissociated when slicing activity is disabled by a mis-pairing in the central region between the miRNA and its target [4][5][6][7]. In plants, the high level of complementarity between the miRNAs and their targets suggests slicing is the predominant mode of action of miRNAs [7]. Alternatively, miRNAs can regulate their target indirectly through the production of trans-acting short interfering RNAs (tasiRNAs) [8,9]. tasiRNAs are synthesised from a noncoding mRNA that is processed to phased 21 nt smRNAs by a miRNA triggered process. Like miRNAs, tasiRNAs can regulate multiple target genes through a slicing mechanism.
The number of annotated miRNAs in miRBase has exponentially increased in the last decade [10]. The earliest group of miRNAs were identified in silico using algorithms to predict stem-loop precursors and targets present in the genome and/or EST databases [11][12][13][14][15]. Subsequent developments in high throughput sequencing made it possible to identify miRNAs based on sequencing of smRNA libraries in a wide range of species. Schreiber et al. [16] identified 100 miRNAs, including 44 new miRNAs, from barley leaves using short-read sequence data. A major challenge of sequencing based approaches is to identify the miRNAs amongst a smRNA population mostly composed of short-interfering RNAs (siRNAs) [17]. Distinguishing these two major smRNA classes relies principally on identifying their origin. An siRNA locus produces several overlapping siRNAs, whereas the pri-miRNA encoded by a MIR gene usually produces one miRNA from an imperfect RNA hairpin [18]. Additional criteria can also help classify a smRNA, such as its length and mode of action. Most miRNAs and tasiRNAs are 21 nt in length and posttranscriptionally regulate their target genes in trans, whereas the vast majority of the 24 nt smRNAs correspond to cis-acting siRNAs (casiRNAs) that regulate the transcription of their own locus of origin through a DNA methylation based mechanism. miRNA targets are often validated using a modified 5'RACE technique to detect the products of miRNAmediated cleavage [19]. For most currently annotated miRNA targets, cleavage has not been verified and therefore the function of the corresponding miRNA in vivo has not been established. Recently, techniques which combine 5'RACE and high throughput sequencing (Parallel Analysis of RNA Ends (PARE) and equivalent methods [20][21][22]), have been used to simultaneously validate all sliced miRNA targets in a given RNA extract. Such an approach has been successfully carried out in Arabidopsis, rice, soybean, grapevine, citrus and medicago [23][24][25][26][27][28]. However, identifying a miRNA regulation is dependent on examining the appropriate tissue and developmental stage. As miRNAs are predominantly post-transcriptional regulators [29,30], the impact of their regulation depends on the overlap of their spatiotemporal expression with that of their target genes [1,2]. miRNAs from the same family can potentially have different functions depending on their expression profile, as suggested for members of the miR169 and miR171 families that differentially accumulate in response to abiotic stress in rice [31,32].
Despite the growing knowledge of miRNA functions in plants, only the functions of highly conserved miR-NAs have been investigated in crop species. Perhaps the best characterized miRNAs in cereals are miR156 and miR172 which regulate SPL (Squamosa Promoterbinding protein-Like) and AP2-like genes, respectively. miR156 controls shoot branching in rice and maize [33][34][35] and miR172 regulates floral organ identity in rice, maize and barley [36][37][38][39][40][41]. In maize, miR172 accumulation is affected by miR156 and both miRNAs are involved in the regulation of the juvenile to adult phase transition [33]. In contrast to the highly conserved miRNAs, the majority of the newly discovered miRNAs are weakly expressed and only found in closely related species, suggesting that they have recently evolved and could contribute to determining speciesspecific traits.
Barley is the fourth most cultivated crop worldwide; its grains are used for both human consumption and livestock feed. From anthesis, it takes approximately 40 days to form a mature grain composed of 3 principal tissues: the embryo, the endosperm (starchy endosperm and aleurone layers), and the outside layer (seed coat and pericarp). The development of the grain can be divided in three principal stages based on morphological changes, metabolite accumulation and transcriptome analysis: pre-storage, storage (or maturation) and desiccation [42][43][44][45]. The pre-storage phase, which corresponds to the first 5 Days Post Anthesis (DPA), is characterized by extensive mitotic activity in both embryo and endosperm. The transition to the storage phase, roughly between 5 and 10 DPA, can be considered as an intermediate stage characterized by dramatic transcriptional changes in order to mobilize energy resources and initiate the differentiation of the tissues that will constitute the mature grain. Throughout the maturation phase, which lasts up to~25 DPA, aleurone and embryonic tissues acquire desiccation tolerance whereas the endosperm cells undergo endoreduplication and accumulate storage metabolites (mainly starch and proteins) [46].
In this study we investigated the miRNA-mediated gene regulation that takes place during the growth of the barley grain. Since the early stages of development play a key role in determining grain quality characteristics, we focused on the pre-storage and early storage phases (0-15 DPA). From analysis of smRNA and degradome libraries, 96 genes regulated by miRNAmediated cleavage were identified including transcription factors, kinases, oxidoreductases, hydrolases, transferases, receptors and transporters. Our data suggest that miRNAs contribute widely to the control of development of the cereal grain, notably through the regulation of phytohormone response pathways.

Results and discussion
The early development of the seed is marked by largescale transcriptional changes, especially during the transitional phase. In order to correlate those changes with variation in miRNA abundance, we made smRNA and mRNA-degradome libraries from the whole caryopsis at three consecutive developmental stages: (A) from 1 to 5 DPA (early pre-storage), (B) from 6 to 10 DPA (late pre-storage or transition phase), and (C) from 11 to 15 DPA (early storage). An overview of our analysis is presented Figure 1. We first used the smRNA libraries to detect known miRNAs and to identify new miRNAs based on the presence of their precursor in cDNA databases. We then used the degradome libraries to identify potential endonuclease cleavage sites in EST sequences and selected those that could result from slicing by a sequenced smRNA. The smRNAs associated with a cleavage site in the degradome data are designated as potential miRNAs (pot-miRNAs).

Diversity of the small-RNA population in early grain development
Approximately equal numbers of sequence reads (20 million) were generated from each of the smRNA libraries (Table 1, Additional file 1). The size distribution in the smRNA datasets was similar to previous reports with about 44 % 24 nt sequences that are likely to consist predominantly of casiRNAs and 7 % 21 nt smRNAs that will include the bulk of the miRNAs ( Figure 2). The datasets showed a decrease in the percentage of 24 nt smRNAs and an increase in the percentage of 21 nt smRNAs from stages A to C, which correlates with data from developing rice grain samples from 1-5 DPA and 6-10 DPA [47]. If unique signatures are considered, both 21nt and 24 nt smRNA diversity increased from stage A to B, suggesting a higher smRNA complexity during the reprogramming phase of grain development. As the grain matures further (sample C), the number of unique 21 nt signatures decreases while the 24 nt increase ( Figure 2). The continuing increase in 24 nt smRNA diversity with development may reflect an increase in heterochromatin formation as cells become   [48,49].

Previously identified miRNAs present in barley grain
We found 84 smRNA signatures that were identical (in sequence and length) to at least one previously identified plant miRNA, representing 47 miRNA families ( Figure 3, Additional file 2). Of these, 11   Total RPM (A+B+C) Absent in hvu-miRBase but detected in barley leaves [16] Never detected in barley Figure 3 Known miRNA families found in the barley seed. miRNA families were grouped according to their conservation level across the plant kingdoms (based on the farthest species from barley in which at least one member has been found). For each miRNA family, the number of members (or unique sequences) identified in barley is presented in brackets and the total number sequences from the three libraries is presented in reads per million (RPM). In black are the families for which at least one member has been classified in miRBase, in black stripes are the miRNA families not classified in miRBase but previously detected in barley leaves [16] and in red are the known miRNA families absent in barley leaves.
previously classified as hvu-miRNA in miRBase and 32 were previously reported in barley leaves but not classified as hvu-miRNA in miRBase [10,16]. We found 4 miRNA families (hvu-miR894, hvu-miR158, hvu-miR161, hvu-miR391) that were not observed in barley leaf and so may be seed specific [16] (Figure 3, Additional file 2). As previously observed by Colaiacovo et al. [50], the vast majority of the miRNAs are 21 nt in length (Additional file 2). The specificity of each family was determined according to the farthest species (from barley) in which at least one member has been found. We note that the highly conserved families are not necessarily highly expressed in barley seed, examples are miR894 and miR408 which accumulate at less than 10 RPM. Conversely, miR5071, miR5048 and miR5067 which have only been identified in barley, are expressed at over 100 RPM in the seed. Overall only~0.01% of the unique 21 nt signatures correspond to known miRNAs.
To determine whether the number of cloned sequences in the libraries reflects the relative abundance of a smRNA in planta, the accumulation of three known miRNA families (hvu-miR164, hvu-miR168 and hvu-miR390) was monitored during seed development ( Figure 4). For all three families, the abundance of the mature miRNA detected by northern blot between the three development stages followed the same trend as the numbers of reads in the libraries. Therefore, the relative expression of each smRNA between the 3 samples (A, B and C) can be directly inferred from the numbers of sequence reads.

New miRNAs identified based on the presence of their precursor
A miRNA can potentially evolve as a result of the transcription of one of the many inverted repeats present in the genome if the resulting hairpin structure has the features to be recognised and processed by a DCL protein.
In the absence of a barley genome sequence, sequence information is restricted to EST databases. For this analysis we used the HarvEST database which contains over 50,000 unigenes and searched for miRNA precursors corresponding to smRNA sequences in our database ( Figure 1). A putative precursor (pri-miRNA) was found for 15 smRNA sequences. Eight of these were for miRNAs present in miRBase including the three highly conserved miRNAs hvu-miR159, hvu-miR171a and hvu-miR168a for which a miRNA* was also present ( Table 2, rows 1-8; Additional file 3). There were also putative pri-miRNAs for seven smRNA sequences not present in miRBase (Table 2, rows 9-15). Two of these have sequences closely related to known miRNA families and were therefore annotated hvu-miR5071b and hvu-miR1120b. Hvu-miR1120b (21 nt) is a short version of hvu-miR1120 (24 nt) with 3 nucleotides missing at the 5' end; both are predicted to originate from the same pri-miRNA. Since hvu-miR1120 was only predicted in silico [51] and hasn't been detected in barley leaves, it may not exist in planta. We temporarily annotated the other five smRNAs as new miRNAs starting from hvu-miR6001 as they show the expected features of a miRNA other than the presence of a miRNA*. The lack of miRNA* sequences may reflect the low abundance of these miRNAs.

Identification of potential miRNA targets using the degradome libraries
The 84 known and 7 new miRNAs identified in this study account for only 1 % of the unique 21 nt signatures in the smRNA sequence dataset, suggesting that  these analyses did not identify all the miRNAs present. An alternative approach to identify the presence of a miRNA is to detect its post-transcriptional regulatory activity on a target gene. In plants, most miRNAs characterised to date show slicing activity on their target, hence degradome analysis was carried out using the Parallel Analysis of RNA Ends (PARE) technique [21], constructing libraries from samples A, B and C as used for the smRNA libraries ( Figure 1). We reasoned that having smRNA and degradome libraries from the same sets of samples would allow us to follow the miRNA regulation of target genes and increase the likelihood of detecting a cleavage that occurs at a particular developmental stage. Approximately 30 million sequence tags corresponding to cleaved 5' ends of mRNAs were obtained from each of the degradome libraries (Table 3, Additional file 1). After trimming of adapter sequences, most sequences were of the expected size of 20 or 21 nt. To simplify analysis, the 21 nt sequences had the 3' nucleotide trimmed and were then pooled with the 20 nt sequences giving 8.86 million unique sequence tags ( Table 3). The number of unique sequences was higher in sample B which also had the greatest diversity of 21 nt smRNAs ( Figure 2). This suggests that there is a larger diversity of transcripts regulated by miRNA cleavage between 6 and 10 DPA.
The degradome sequences were mapped to the Har-vEST dataset. The total reads mapping to an EST were used to establish a threshold which was calculated using the average number of reads of all degradome signatures matching the EST plus two standard deviations. Sequences that were more abundant than the threshold were considered to be degradome peaks ( Figure 5). The degradome peak was then used to define a Target Signature Sequence (TSS, Figures 1 and 5) which extended 16 nt in each direction from the 5' end of the degradome sequence that identified the peak. The TSSs were then compared to the known miRNAs, the new miRNAs and to the 19-23 nt smRNAs filtered against repeat elements ( Figure 1; see Methods for detail). The three degradome libraries were analysed separately. This identified 1126 ESTs with at least one TSS that was predicted to be a binding site for a known, new or one of the 19-23 nt siRNAs. We refer to the 19-23 nt siRNAs that match a TSS as potential miRNAs (pot-miRNAs) from here on. As there is no precursor information for the pot-miRNAs they could include tasiRNAs and other siRNAs as well as genuine miRNAs. The majority of the ESTs The pri-miRNAs were identified in the harvEST dataset according to their ability to form a hairpin-like secondary structure that could generate one of the smRNAs present in the libraries. For each miRNA, its length and number of reads (in RPM) in each library (A, B, C) are indicated. The column "Star" indicates when a corresponding miRNA* was found in the libraries. we identified had multiple degradome peaks. As the degradome libraries were made by reverse transcription from the polyA + tail, a miRNA target would be expected to have a peak corresponding to the miRNA cleavage site, together with degradation products from downstream of the cleavage site. This prediction was observed for 17 out of 18 conserved targets of known miRNAs (data not shown). Based on this observation, we selected 96 ESTs for which the first TSS was predicted to be targeted by at least one miRNA (known, new or potential) that perfectly aligned with the predicted cleavage site (offset =0) (Additional file 4). These 96 ESTs included 21 with only one TSS and were targeted by a total of 1013 miRNAs (Additional file 4). Most of the TSSs were predicted to be targeted by multiple miRNAs; some were aligned to the predicted cleavage site, while others aligned at positions without a corresponding degradome product (offset of +/− 1 to 3). This last group of miRNAs, which do not appear to cleave the mRNA may be present in different tissues to the target mRNA. Another explanation could be that the precise position of the miRNA binding site on the target mRNA is critical for efficient cleavage by RISC due to structural constraints. As the presence of multiple miRNAs raises some doubt about the validity of the mRNA target site, they were assigned to three groups ( Figure 5). Category I included the ESTs only targeted by miRNAs with a perfect offset. Category II contains ESTs targeted by a majority of miRNAs with a perfect offset. Category-III contains the ESTs where the aligning miRNAs with a perfect offset were in the minority. Among the 96 potential miRNA targets, we found 17 targets of known miRNAs and three targets of new miRNAs (hvu-miR6005, hvu-miR6001, hvu-miR5071b) ( Table 4, Additional file 4). The cleavage of three targets of known miRNAs was verified by RLM-5'RACE (Additional file 5). The pot-miRNAs identified in this analysis included many homologs of known miRNA families that varied in sequence and length to previously identified sequences (e.g. 71 miR156 homologs and 24 miR168 homologs). In the absence of complete genomic sequence data it is not possible to determine whether these represent genuine additional family members or errors from library construction and sequencing. The presence of large numbers of these alternate length sequences for some miRNA families suggests that there may be differential processing of the pri-miRNAs or later processing of the terminal nucleotides.
The degradome analysis revealed that most miRNAs target only one EST with a smaller group targeting 2 to 4 different ESTs, which are usually members of the same gene family, (e.g. ARFs, CBFs and SPLs; Additional file 4). This is in contrast to an average of 10 ESTs bioinformatically predicted as targets for each miRNA (data not shown). One obvious reason for not detecting target mRNA cleavage is that expression of the miRNA and mRNA may not overlap. Published microarray data [45,52] shows that a majority of the genes predicted to be regulated by a slicing mechanism are expressed during seed development (data not shown) however this does not preclude the non-overlapping expression of miRNA and target in the same cell types. In addition we found four sliced targets in our dataset that were predicted to be regulated through a translational repression  Total 96 The targets correspond to barley ESTs potentially regulated by miRNA mediated cleavage. The categories refer to a level of confidence based on the distribution of the smRNAs predicted to bind the target site (see Results and Discussion). The known miRNAs are those identified in previous studies, the new miRNAs are smRNAs for which we identified a pri-miRNA and the pot-miRNAs refer to the remaining set of 19-23 nt smRNAs matching to a TSS ( Figure 1). Note that targets of known and new miRNAs may also be targeted by pot-miRNAs.
mechanism only (Additional file 4). Our observations suggest that there are inadequacies in the algorithms currently used to predict miRNA targets, hence experimental verification is required to confirm these predictions in planta.

miRNA regulation during seed development
The early development of the grain is controlled by a complex interaction of signalling and gene regulation networks to allow the proper expansion and specialisation of the different tissues that will constitute the mature grain. Based on our combined analysis of smRNA and degradome data we identified 96 genes likely to be miRNA regulated (by cleavage) during the first 15 DPA of seed development. Using annotated sequences from barley, wheat, rice and Arabidopsis, we found significant homology to an annotated gene for 77 of the miRNA target genes. Based on sequence homology these genes are predicted to encode a wide range of protein functions, including transcription factors, kinases, oxidoreductases, hydrolases, transferases, receptors and transporters (Additional file 4). We performed an ontology analysis of these targets and compared it to a set of over 8000 ESTs previously detected in the seed and annotated by Sreenivasulu et al. (Figure 6) [45]. Enrichment of GO terms was declared statistically significant if they met the criteria of P < = 0.01 using a hypergeometric one-tailed test with correction for multiple testing (Benjamini-Hochberg). This analysis shows that in the barley grain, miRNAs target a significantly higher percentage of genes annotated in the hormone signalling pathways, RNA cellular processes (which includes transcription factors) and energy mobilization categories. Using our data, the variation of mature miRNA abundance was compared to that of the cognate degradation products across the three stages of grain development. The detection of mRNA cleavage products indicates that the expression domains of the miRNA and target gene are at least partially overlapping. The more the miRNA and its target are expressed, the more degradation products should be generated. The following paragraphs highlight what we think are the more interesting data based on the function of the targets. Since it is impossible to distinguish which one of the miRNAs (if not all) are present and functional in the same tissue as the target, all miRNAs with zero offset to the cleavage site were considered. The number of distinct miRNAs and the sum of their reads for each library is summarised in Table 5. As noted above, related miRNAs tend to have similar expression profiles and thus the sum of their reads is a good indication of their individual expression patterns.

Cell differentiation
Perhaps the best known function of the miRNA pathway is to control the cell fate through the regulation of transcription factor coding genes. We validated 11 conserved targets of six known miRNA families which code for transcription factors known to control key steps in plant development: miR156-SPL (2 genes), miR159-Myb, miR164-NAC, miR167-ARF (potentially 3 genes sharing the same degradome peak), miR169-CBF (3  genes) and miR172-AP2like (Table 5). We also found evidence for miRNA regulation of the DOF (DNA binding with one finger) plant specific transcription factor family. Its expression seems to be restricted to the early development of the grain since degradation products were observed only during stage A (Table 5). DOFs are plant specific transcription factors known to play a critical role in growth and development [53]. In maize and finger millet, DOF proteins are thought to be involved in carbon metabolism and the accumulation of storage proteins [54,55]. In rice, RPBF (rice prolamin box binding factor) which contains a DOF domain, was shown to be involved in the regulation of endosperm expressed genes [56].

Energy mobilization
The early development of the seed is associated with an elevated metabolic activity limited by energetic resources. Photosynthesis related genes are mainly expressed during the first 5 DPA within the pericarp tissue [43]. Four of the potential miRNA targets (encoding a ferredoxin, a chlorophyll a/b binding protein, a carbonic anhydrase and a ribonuclear protein) are likely to be involved in chloroplast function. An EST coding for a PGlcT (Plastidic GlucoseTranslocator) homolog is also cleaved by a pot-miRNA during the early development of the grain. PGlcT is involved in the export of stored starch into the cytoplasm at night [57]. The level of PGlcT degradation products in our dataset increases during grain development (Table 5) which correlates with a previous observation in rice that expression of a PGlcT homolog gene increases in the endosperm during the first 15 DPA [58].
Auxin concentration together with other local factors, contributes to cell differentiation and specification of cell fate [63,64] and is known to be involved in embryo patterning [65]. In Arabidopsis, the auxin signal is tightly linked to the miRNA pathway, with four conserved miRNA families (miR160, miR167, miR390 and miR393) regulating the auxin receptor TIR1 (Transport Inhibitor Response1) and different subgroups of ARF (Auxin Response Factor) genes [9,[66][67][68][69][70][71]. We identified a TIR1 homolog and 7 ARF genes potentially regulated by miR-NAs and tasiRNAs during seed development (Table 5). Our data shows that in the barley grain the regulation of TIR1 and potentially 3 ARF genes (the same degradome peak matches three distinct ESTs) by the miR393 and miR167 families is conserved. We noticed that hvu-miR167a and d, which are the highest expressed members in this family, show a reciprocal accumulation pattern which could suggest they are expressed in different tissues where they differentially regulate the same target genes (Additional file 4). We also identified smRNAs homologous to the tasiR-ARFs which regulate four ARFs (ARF 4/5/6/7). The accumulation of these smRNAs correlates with hvu-miR390 which gradually decreases in abundance from stage A to C (Table 5, Additional file 2), suggesting that, as in Arabidopsis, the production of the tasiR-ARFs requires miR390-mediated cleavage.
The antagonistic role of GA and ABA in the control of the switch between dormancy and germination is a well known mechanism; however the function of these hormones during the early stages of seed development remains unclear. Our data suggest that there is miRNA regulation of ABA and GA signalling during the early stages of grain development, with degradome analysis identifying two ABA-Insensitive homolog genes (ABI3 and ABI8), a GA3oxidase1 and a homolog of the GA receptor GID1 (Gibberellin Insensitive Dwarf1) as targets of miRNAs or pot-miRNAs (Table 5). ABI3 is cleaved with a perfect offset by 2 members of the grass specific miRNA family hvu-miR516. Degradation products of ABI3 only accumulate during stage C whereas the corresponding miRNAs are expressed earlier. In contrast, the cleavage of GID1 mostly occurs during early stages while the cognate pot-miRNAs accumulate in later stages. This suggests that in both cases the miRNAs could act to prevent leakage in target gene expression, ensuring that GID1 function is restricted to early stages and ABI3 to later stages. These data support the current belief that GA is required during early embryogenesis but its function is repressed in later phases when a higher ABA/GA balance is needed for the proper maturation of the grain. This also correlates with the observation that the late presence of GA may inhibit embryogenic cell differentiation [72].
Our data also suggest that there is miRNA regulation of ethylene responses with an ACC oxidase homolog cleaved by a pot-miRNA during the early maturation phase. Along with ABA, ethylene is thought to play a major role in the development of the endosperm by affecting grain filling and the timing of programmed cell death (PCD) [44,46].

Defence response
Plants recognize many pathogens through the action of a diverse family of R genes, whose protein products are necessary for the direct or indirect recognition of pathogen avirulence (avr) proteins in order to initiate the defence response. In addition to their role in defence responses, R genes may be involved in the regulation of developmental processes in Arabidopsis and rice [73,74].
In barley, the R gene MLA10 acts as a receptor of fungal infection by recognising avirulence proteins and confers resistance against the powdery mildew fungus [75,76]. In wheat the expression of several miRNAs is responsive to powdery mildew infection, suggesting that the miRNA pathway could be involved in triggering the defence response [77,78]. Our degradome analysis indicates that two HvMLA genes homologous to the rice MLA1 and MLA10 genes are cleaved by miRNAs or pot-miRNAs (Table 5).

Further investigation of OsMLA10-like and GA3oxidase1 regulation
The impact of regulation by a miRNA depends on the relative spatio-temporal accumulation of the miRNA and the target mRNA. In this study, we focused on investigating miRNA regulation at three consecutive stages of grain development. Since the tissues that will constitute the mature grain are not formed during the early stages, we used the whole caryopsis to be able to compare the abundance of the mature miRNAs and the degradation of their targets between the samples. Consequently, further investigation of the function of a miRNA cleavage identified in our analysis requires further assessment of the tissue specificity of both miRNA and target mRNA expression.
To illustrate this, the regulation of two category-I targets, OsMLA10-like and GA3oxidase1, and their associated miRNAs was investigated. The abundance of both miRNAs and targets were quantified in embryo, endosperm and pericarp tissues dissected from the caryopsis at stage C (Figure 7). For OsMLA10-like in these tissues, the corresponding miRNAs (which belong to the miR5071 family) are mostly detected in the embryo and pericarp whereas OsMLA10-like expression is higher in the endosperm. The degradation products detected during earlier stages suggest that OsMLA10-like transcription was initially higher and that the function of the miRNA was to inhibit its expression in the embryo and pericarp. In contrast, the pot-miRNAs targeting the GA3oxidase1 gene predominantly accumulate in the endosperm where the target is also actively transcribed. According to the degradome data, the GA3ox1 probably starts to be expressed during stage B when the first cleavage products can be detected. The role of the miRNA(s) may be to modulate level of GA3ox1 transcripts and consequently prevent excess GA accumulation in the endosperm. These two examples highlight the complexity of multilayer gene regulations and the requirement for complementary studies in order to analyse how, where and when a gene is regulated.

Conclusion
The data we have generated provides a comprehensive source of information about the timing of miRNA regulation during grain development. Regulation by miRNAs peaks during the transition phase (5-10 DPA) which correlates with the timing of a major change in transcript profiles. The 96 potential miRNA target genes we identified are predicted to be involved in various functions including photosynthesis, carbohydrate translocation, phytohormone signalling, cell differentiation and defence response. Our data suggest an upstream function of the miRNAs in coordinating tissue specification and energy mobilization to ensure proper growth and development of the grain.
As increasing amounts of genome sequence data become available our data can be re-examined to identify more miRNA precursors and refine the predictions of which genes are under miRNA regulation. The analysis of the biological roles of miRNAs in cereals currently depends on transgenic approaches; however the identification of miRNA resistant target mRNAs that give rise to altered phenotypes in rice and barley [34,35,38] suggests that the use of high-throughput methods to identify sequence changes leading to miRNA-resistant targets will allow assessment of the roles of other miRNAs.

Biological material
Barley (Hordeum vulgare) plants were grown in naturally lit phytotron glasshouses with air temperature set at 17°C/9°C day/night cycle. The plants were grown from October to December when the time of anthesis was determined for each head based on the dissection of the middle spikelet. Immature grains were harvested from the middle six rows of the head at 1 to 15 Days Post Anthesis (DPA). Total RNA was extracted from 100 mg of seeds from each DPA (which correspond to~50 seeds at 1 DPA and 2 seed at 15 DPA) using the following method. Whole caryopsis was ground in a mortar using liquid nitrogen.

Construction and analysis of smRNA libraries
To determine the smRNA populations present in samples A, B and C, 60 μg of total RNAs were used to prepare libraries for Illumina sequencing (www.geneworks.com.au). A custom script was used to trim reads of 3' adapter sequences and then to pool identical reads to create a non-redundant sequences (=signatures) list. Sequences that were over 50% homopolymer or dinucleotide repeat were filtered out. Read counts for each sequence were normalised to reads per million of total sequenced (RPM). The diversity of the signatures present in the three libraries had a high number of singletons, over 80% (Table 1). To facilitate the analysis and build a level of confidence from the sequences cloned in the libraries, signatures were only considered that were 18 to 25 nucleotides in length and with a minimum expression of 1 RPM, representing 137,614 smRNAs. For identification of previously known miRNAs, signatures were checked for an exact match (in sequence and length) to a known miRNA present in miRBase [10] or recently in barley leaves [16]. To identify new miRNAs, the signatures of 19 to 23 nucleotides in length were kept and aligned to the HarvEST unigene set (release-21, http://www.harvestweb.org/) using SOAP [79]. Signatures with no more than 20 matching unigenes and with no match to a smear of overlapping smRNA sequences were kept. For near identical sequences aligning to the same location (for example, differing in length by a base or being offset by a base) only the sequence with the highest read count was retained. The miRNA precursors were searched by extracting the unigene sequence surrounding the aligned smRNAs and testing their potential to form a hairpin secondary structure using Vienna RNALFold (www.tbi.univie.ac.at/~ivo/RNA/). The smRNA signature was required to have no more than 4 mismatches against the complementary sequence in the hairpin structure and no more than 2 bulges. Considering that MIR genes may originate from the evolution of an inverted repeat element that initially can produce endogenous 24 nt-siRNAs [80,81], we kept the precursors sharing less than 70% homology within the pre-miRNA region to repeat elements in the Plant Repeat Databases [82]. smRNA which could be found in a stem of a potential miRNA precursor-like hairpin structure in the folded sequences were marked as new miRNAs. For the downstream analysis of miRNA target genes, the remaining 19-23 nt signatures with no more than 90% homology to a repeat element were kept (Plant Repeat Databases [82], Figure 1).
As for the small RNA analysis, reads were trimmed, reduced to a non-redundant set, filtered for repetitive sequence and their read counts were normalised in read per million (RPM). For subsequent analysis, sequences of 21 nt in length were trimmed back to 20 nt, then only sequences of 20 nt in length and having at least 1 RPM were retained. Kanga (http://code.google.com/p/ biokanga/) was used to align the 20 nt signatures to the HarvEST unigene sequences (release-21, http://www. harvest-web.org/). No mismatches were allowed in the alignment. For each matching EST the number of aligned degradome sequences at each position along the EST was investigated to identify signature peaks ( Figure 5). Positions for which the number of aligned sequences exceeded the mean plus two standard deviations for a sample along an EST were retained. From each of these retained signature peaks, a 32 nt sequence was extracted from the EST, centred around the 5' end of the aligned signature, to constitute the Target Signature Sequence (TSS). To identify the smRNAs that could potentially bind to a TSS we used psRNAtarget (http:// plantgrn.noble.org/psRNATarget). We ran the known miRNAs, new miRNAs and 19-23 nt smRNAs against the TSS with a maximum expectation of 5 and an hspsize (which is the length of the region used to score the complementarity between the miRNA and its target) equal to the length of the smRNA (to ensure that the entire sequence of the smRNA is considered by the scoring algorithm). We indicated the offset between the predicted cleavage site of the smRNA (position 10-11nt of the smRNA) and the detected cleavage site (center of the TSS, Figure 5), using the formula: offset = detected cleavage position on the EST -predicted cleavage position on the EST. For each TSS, we kept all smRNAs in a −3/+3 offset window. Since the smRNAs with a common 3' end binding position on the EST share the same predicted cleavage site we considered them as one group and categorized the targets according to the offset distribution of these smRNA groups (Table 4). We categorized the targets as follows: category-I; the ESTs targeted by a unique smRNA group with a perfect offset (considered as unique when the smRNA group represented more than 97% of the total number of smRNAs predicted to bind to the TSS), category-II; the ESTs targeted by a majority of smRNAs with a perfect offset and category-III; the remaining ESTs targeted by a minority of smRNAs with a perfect offset.

Quantitative RT-PCR
RT-PCR reactions were performed as previously described [40]. In brief, first-strand cDNA was synthesized using oligo (dT) primers and Super-Script III reverse transcriptase (Invitrogen). PCR reactions were performed on an AB 7900 HT Fast Real-Time PCR System (Applied Biosystems). 1.0 μL of 1:10 diluted template cDNA was used in a 10 μL reaction. The amplification program was: 1 cycle of 15" at 95°C, 35 cycles 15" at 95°C, 30" at 60°C, 30" at 72°C, and then followed by a thermal denaturing step. All primers pairs of the tested genes showed a similar amplification efficiency to the one used for the ACTIN gene which was used as reference. Relative transcript levels of biosynthesis were calculated with the ΔΔCt method (Applied Biosystems). Forward and Reverse primers: HvDCL1a F (AGAAGCCTTGACTGCTG CAT) and R (ATCAATTTCGCCCTCCTCTT); HvDCL1b F (GCCCCAAAAGTGCTATCTGA) and R (GCCCCGA CATCTCCTTTAGT); HvDCL1c F (CGGCAGAAACAATT GATGAG) and R (CAAAGCTTCCTGTTGCACTG); GA3ox1 F (GCACTACCGCCACTTCTCTG) and R (CTCTCGGTGA GGTTGTGCTC); OsMLA10-like F (ATAAGATACGTCGT CTGTCCATG) and R (TCCAACACCCGCAGAGCATG).

RLM-5' RACE
RNA ligase-mediated 5′ rapid amplification of cDNA ends (RLM 5′-RACE) was performed using the GeneRacer kit (Invitrogen). The manufacturer's protocol for 5'end analysis was followed with the exception of the 5' de-capping step. In brief, total RNA was isolated from whole caryopsis tissues at