Two sRNA libraries in which M. truncatula seedlings were treated with drought stress (DS) and without stress, control (CK) constructed from shoots of seedlings grown under the two treatment-regimes were sequenced by an Illumina-Solexa sequencer. High-throughput sequencing of CK and DS libraries led to generation of 6,808,877 and 6,874,742 primary reads, respectively. There were 6,420,234 clean reads (2,685,754 unique sequences) for CK and 6,505,965 clean reads (2,937,417 unique sequences) for DS after initial processing (Table 1). The length distribution of reads showed that the majority of the reads was 20-25 nt in size, of which the class of 24 nt was the most abundant group followed by the group of 21 nt class (Figure 1a).

The common/specific sequences were analyzed between CK and DS libraries for the total and unique sequences. There were 19.14% and 44.95% specific sequences in the DS library for total and unique sequences, respectively.

After initial processing, the high-quality small RNA reads were mapped to the M. truncatula genome sequence (Mt3.0) using SOAP 54 . The number of total and unique sequences that matched genome was 4,875,034/1,713,480 and 4,819,359/1,867,375 in the two libraries, respectively (Table 1).

The known miRNAs in M. truncatula were identified by comparing with the up-dated miRNAs database (miRBase 17, released in April 26, 2011) 55 . We identified 283 and 293 known miRNAs in the two libraries, respectively (Table 1). Sixty-six members belonging to 14 conserved miRNA families were obtained after removing those miRNAs whose expression levels are too low to be analysed for differential expression. Of these families, the most abundant two reads were miR156 and miR166. Among the identified miRNAs, miR399 contained the most members, including miR399a-i and miR399k-q (Figure 1b).

In addition to the conserved miRNAs, we also identified 137 non-conserved miRNAs that belong to 44 miRNA families (Additional file 1) after removing miRNAs with expression levels that were too low to be analysed for differential expression. Among the 137non-conserved miRNAs, some were legume-specific. The most abundant miRNA was miR2086 followed by miR2111 in the control (CK) library. In the drought stress (DS) library, miR2111 was the most abundant, followed by miR1507. In addition, miR2630 was found to have the most members in the two libraries (miR2630a-y).

The formation of stable hairpin structure is one of the essential features for identification of new miRNAs 56 . To identify new miRNAs and new members of known miRNA families, we pooled the reads obtained from the two libraries, and identified the miRNAs by folding the sequences of potential miRNA precursors using the mfold web server 57 . We obtained 238 potential miRNA candidates (Table 2, Additional file 2). In addition to the hairpin structure, detection of miRNA*s is another proof that has been widely used to confirm the new miRNAs 58 . Among the 238 potential miRNA candidates, we found 29 miRNAs with complementary miRNA*s (Table 2, Figure 2, Additional file 3), indicating that these candidate miRNAs are likely to be new miRNAs or new members of known miRNA families in M. truncatula. The counts of several miRNA*s of these miRNAs were low (Table 2). This may make the annotation of these new miRNAs questionable. However, no requirement on the counts of several miRNA*s was given in the most recent criteria for annotation of plant miRNAs 58 . Therefore, these miRNAs were still considered to be new miRNAs or new members of known miRNAs. A similar identification of new miRNAs has also been reported by others in the literature. For instance, candidate miRNAs with only one miRNA* have been identified as new miRNAs in rice 59 .

Twenty-two out of 29 new miRNAs/new members of known miRNA families had length of 21 nt, while remaining miRNAs had length of 22 nt. The majority of the miRNAs started with a 5' uridine, a hallmark of miRNAs 60 . The minimum free energy (MFE) for hairpin structure of miRNA precursors was lower than -30 kcal/mol (Table 2). This feature is in agreement with other reported values in the literature 61 . The length of new miRNAs' precursors ranged from 86 nt to 315 nt (Table 2). These values are comparable to those reported for M. truncatula in the literature 42 51 52 62 63 .

To identify drought-responsive miRNAs, the normalized expression of miRNAs in the two libraries (control, CK and drought stress, DS) was compared. Based on the results of high-throughput sequencing, we selected those miRNAs with changes in expression levels being greater than 1.5-fold in response to drought treatment (Figure 3a and 3b, Additional file 4) to validate the expression patterns by real-time quantitative PCR. As shown in Figure 3b and 3c, expression patterns of drought-responsive miRNAs from high-throughput sequencing and RT-qPCR exhibited similar results. Results from both the methods showed that 22 members in 4 miRNA families, i.e., miR399, miR2089, miR2111 and miR2118, were up-regulated in response to the drought stress. Conversely, 10 members belonging to 6 miRNA families, i.e., miR164, miR169, miR171, miR396, miR398 and miR1510, were down-regulated in response to drought stress (Figure 3b and 3c). These miRNAs have been reported to be involved in diverse cellular processes in plants 14 17 22 52 . The known target genes of these miRNAs and their function annotations were summarized in Table 3. For those miRNAs whose targets are not known, we predicted their targets using the psRNATarget http://bioinfo3.noble.org/psRNATarget/ and the srna-tools http://srna-tools.cmp.uea.ac.uk/plant/ 64 , and the results are given in the Table 4. Among these miRNAs, several miRNAs have been reported to be involved in abiotic stresses. For example, miR399 and miR2111 have been reported to be up-regulated by phosphate starvation 23 65 , while miR169 with target of CCAAT Binding Factor is down-regulated in response to drought stress 14 17 .

We further examined whether the 29 new miRNAs/new members of known miRNA families were responsive to drought stress using the same method as the known miRNAs. The miRNAs with changes in expression levels being greater than 1.5-fold and p-values less than 0.05 are presented in Table 4. Results from high-throughput sequencing and RT-qPCR showed that, among the 14 new miRNAs, 5 miRNAs were responsive to drought stress with miR5274b and miR5558 being up-regulated, while miR5554a-c being down-regulated. In addition, among the 15 new members of known miRNA families, 3 miRNAs were responsive to drought stress with miR2111u, v being up-regulated and miR4414a being down-regulated (Figure 3b and 3c, Table 4).

High-throughput sequencing has been used to study miRNAs at whole genome level in several plant species, including Arabidopsis 39 40 , rice 43 , wheat 44 , soybean 45 , cotton 46 , grapevine 53 , Medicago 42 51 52 , Brachypodium 20 , tomato 41 , populus 47 , citrus 48 , peanut 49 , Porphyra 50 . However, most of the studies on high-throughput sequencing miRNAs have focused on miRNAs under non-stressed, normal growth conditions, and only a few studies have compared miRNAs under control conditions to those under conditions of abiotic stresses by high-throughput sequencing plant miRNAs 20 43 44 . For instance, Sunkar et al. (2008) reported a total of 714,202 reads in rice seedlings from three independent libraries (control, drought and salt stress libraries) 43 . Twenty-eight and 12 miRNAs in response to cold stress in Brachypodium 20 44 and heat stress in wheat 44 have been identified by high-throughput sequencing, respectively. In the present study, we constructed two sRNA libraries, non-stressed, control and drought stressed in legume model plant, M. truncatula and identified known and new miRNAs that were responsive to drought stress.

The successful application of high-throughput sequencing technology to systemically identify plant miRNAs has greatly advanced our knowledge on the functions of miRNAs in plants in recent years. There have been several reports on the identification of miRNAs in M. truncatula in the literature. For instance, 26,656 and 844,110 sRNA reads were reported in two recent studies from M. truncatula by Jagadeeswaran et al. (2009) and Lelandais-Briere et al. (2009), respectively, with 97,028 unique sequences being obtained by the latter studies 51 52 . However, in both cases, sRNAs were sequenced using Roche 454 sequencer, and the total reads and unique sequences obtained from these studies are much less than we reported in the present study (Table 1). In a study by Szittya et al. (2008), the authors obtained 3,948,871 reads and 1,563,959 unique sequences from two sRNAs libraries of M. truncatula using the same high-throughput sequencing technology (Illumina-Solexa) 42 . In contrast to these reported findings, we got a total of 13,683,619 reads and 4,878,445 unique sequences from two libraries of M. truncatula. Therefore, our results contain more reads than those previously reported for miRNAs in M. truncatula in the literature. Moreover, the high-throughput sequencing technology used in the present study allows us to identify the miRNAs with low abundance, thus accounting for the greater unique sequences found in our studies. In addition to the greater amounts of miRNAs sequencing data, the use of most updated miRBase also contributes to our identification of more known miRNAs in M. truncatula. The available database for M. truncatula contains more miRNAs than those for other plant species. For example, in the database, there are 19 miRNAs and 41 miRNAs for Brachypodium and wheat, respectively, while there are 375 miRNAs in the database for M. truncatula (miRBase 17). In addition, to the best of our knowledge, we are the first one to use the M. truncatula genome sequence Mt3.0 to analyze miRNAs in this species. The more genomic information in Mt3.0 than Mt2.0 may also contribute to predicting more new miRNAs in M. truncatula.

Plant miRNAs have a strong propensity to target genes associated with development, particularly those genes encoding transcription factors and F-box proteins 25 . In the present study, we found that the known or predicted targets of miR164, miR169, miR171, miR396, miR1510 and miR5558 were either transcription factors or F-box proteins (Table 3, 4). Under drought stress, an increase in root/shoot ratio was found in M. truncatula (data not shown). miR164 has been reported to regulate root development by a homeostatic mechanism to clear NAC1 mRNA, leading to down-regulation of auxin signals. It has been shown that decreases in NAC1 mRNA levels due to inducible expression of miR164, thus resulting in reduction in lateral root emergence in Arabidopsis 9 . Therefore, it is conceivable that the suppression of miR164 expression may contribute to the increase in root/shoot ratio under drought stress.

Several studies have revealed that miR169 is responsive to abiotic stresses such as drought, cold and salinity in different species 14 15 17 22 33 . Under drought stress, miR169 exhibited different expression patterns among different species. For example, in Arabidopsis, miR169 is down-regulated by drought stress through an ABA-dependent pathway, resulting in accumulation of NFYA5 with high affinity and sequence specificity for the CCAAT box, which is crucial for the expression of a number of drought-responsive genes 14 . In contrast to Arabidopsis, miR169g in rice is up-regulated by drought 17 18 . In rice, DREs (dehydration-responsive element) as the upstream of MIR169g are regulated by cold and drought stress, thus leading to up-regulation of miR169g 17 . Trindade et al. (2010) reported that the expression of miR169 in leaves of M. truncatula is not responsive to drought stress 15 . In contrast, our results from both the high-throughput sequencing and RT-qPCR showed that miR169 was down-regulated under drought stress in M. truncatula (Figure 3b and 3c, Additional file 4). This discrepancy may result from the difference in degree of drought stress exposed to plant materials in the two studies. For instance, in the present study, the drought stressed samples were collected after exposing of plants to drought for varying period (6, 8, 10 and 12 d after withholding water), thus our samples included plant materials suffering from a wide range of drought stress. This type of stress has been used in study of molecular response of plants to drought stress 66 67 . In the studies of Trindade et al. (2010), water stress was exposed by suppressing water supply with relative water content in leaves being approx. 50% and 30%, respectively 15 . In the present paper, the drought stressed leaves with relative water content of 87.4%, 78.8%, 75.2% and 68.3% were used for analysis of miRNAs. Therefore, our samples were mainly those suffering from mild and moderate water stress compared to those of Trindale et al. (2010). If miR169 is responsive to mild and/or early drought stress exclusively, the responsiveness of miR169 may not be detected by the more severe drought stress used by the authors 15 . In addition to withholding water supply, several studies on effect of drought on miRNAs also treated plants with PEG or mannitol for varying period 17 33 . It is difficult to compare the natural drought stress with PEG-induced and/or mannitol osmotic stress as the two treatments may differ in induction of water stress in terms of rapidity and severity.

Accumulation of reactive oxygen species (ROS) is a common phenomenon in response of plants to abiotic stress. The accumulated ROS damage nucleic acid, oxidize proteins and cause lipid peroxidation 68 69 . Superoxide dismutases (SODs) detoxify superoxide radicals. The targets of miR398 are two Cu/Zn superoxide dismutases (cytosolic CSD1 and chloroplastic CSD2), and miR398 expression was reported to be down-regulated transcriptionally by oxidative stresses 27 . Oxidative stress often occurs concurrently with drought stress. In the present study, we found that miR398 was down-regulated under drought stress. This would lead to increases in activities of SODs. The drought-induced down-regulation of miR398 in M. truncatula is consistent with the results in maize 16 , but it is contrast to the results reported by Trindale et al. (2010) 15 and Kantar et al. (2011) 34 . The differences in the expression of miR398 among different studies may results from differences in species, extent and duration of drought stress in different studies.

Plants suffering from water-deficit often display reduced uptake of mineral nutrients. In this context, miR399 negatively regulates the concentration of inorganic phosphate (Pi) by targeting PHO2, a type of E2 conjugase, and overexpression of miR399 in Pi-replete conditions represses expression of E2 conjugase, leading to an increase in Pi concentration in leaves in Arabidopsis 23 . Expression of miR399 is reduced under Pi-deprived conditions to facilitate accumulation of Pi in plants. It has been verified that miR2111 is up-regulated by Pi starvation 65 . In our study, the expression miR399 and miR2111 was similar under drought stress. These results indicate that the function of miR2111 may be as important as miR399 in regulation of nutrient acquisition.

Exposure of plants to a moderate stress induces resistance to other stresses, a phenomenon known as cross adaptation, which has been found in different combinations of stresses 70 . For example, osa-miR821 isolated from virus-infected rice tissues is also expressed in roots of salt-stressed plants, while it is not expressed in healthy, non-stressed plants 71 . Under drought stress, we found that there was up-regulation of miR2089 and miR2118, whose targets may be proteins associated with disease resistance. It is envisaged that these miRNAs may enhance the ability of drought tolerance through unknown mechanisms associated with cross adaptation in plants. Future work aiming at functional elucidation of these miRNAs is warranted by over-expressing these miRNAs in M. truncatula.

Four Medicago truncatula (cv Jemalong A17) seedlings were grown in a pot (diameter 10 cm) filled with vermiculite: peat soil (2:1) under controlled conditions (26°C day/20°C night, 14-h photoperiod, and 50% relative humidity). Drought stress was initiated by withholding water supply to 3-week-old seedlings for varying periods after seedlings were fully watered. The severely wilted leaves appeared on the 14th day after the water withholding. The soil water content was reduced from approx. 60% to 8.1% during the drought stress. Shoots under drought stress were harvested after withholding of water for 6, 8, 10 and 12 d, and mixed averagely as drought treatment materials (DS). The relative water content for leaves collected after 6, 8, 10 and 12 d drought stress was 87.4%, 78.8%, 75.2% and 68.3%, respectively. At the same time, shoots of M. truncatula seedlings grown under normal watering conditions were also harvested as control (CK). The relative water content for control leaves was 91.4%. The measure method of relative water content was described by Catsky (1960) 72 .

To construct small RNA libraries, total RNA was extracted from shoots of control (CK) and drought stress (DS) using the Trizol (Invitrogen) according to the manufacturer's protocols. For each sample, the 18-30 nt small RNAs were ligated with 5' and 3' RNA adapter by T4 RNA ligase (TaKaRa) after they were purified by electrophoretic separation on a 15% TBE-urea denaturing PAGE gel, and at each step purified by urea PAGE gel electrophoretic separation. The RNA was subsequently transcribed to single-stranded cDNA using Superscript II reverse transcriptase (Invitrogen). Thereafter the cDNA was used as templates for double-stranded cDNA synthesis by PCR amplification using primers that anneal to adapters. The purified DNA was sequenced on a Solexa sequencer (Illumina). The raw data have been submitted to Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/projects/geo/) and the accession number is GSE29154.

After the sequence tags from Solexa sequencing went through the data cleaning by removing the low quality tags (i.e., tags less than 18 nt and tags whose adaptors were null) and contaminants (adaptors and polyA), the length distribution and common/specific sequences between two samples were analyzed. After removing adaptor sequences of the left high-quality small RNA reads with exact matches to the adaptor sequences, reads were mapped to the M. truncatula genome sequence (Mt3.0) downloaded from the website http://www.medicagohapmap.org/downloads_genome/Mt3 using SOAP 54 . rRNAs, tRNAs, snRNAs and snoRNAs were removed from the matched sequences through BLASTn search 73 using NCBI Genebank database http://www.ncbi.nlm.nih.gov/blast/Blast.cgi/ and Rfam database http://www.sanger.ac.uk/Software/Rfam/. Mismatches were not allowed in the above two approaches. The unique sequences left were aligned with known miRNAs from miRBase 17 http://www.mirbase.org/ 55 . The potential candidate miRNAs were identified by folding the flanking genome sequence of unique small RNAs using the mfold web server 57 . Parameters were set based on the criteria for annotation plant miRNAs 58 . Target predictions were performed using the psRNATarget http://bioinfo3.noble.org/psRNATarget/ and the srna-tools http://srna-tools.cmp.uea.ac.uk/plant/ 64 , through aligned with genome of M. truncatula and A. thaliana.

The frequency of miRNAs of two libraries was normalized to one million by total number of miRNAs in each sample (Normalized expression = Actual miRNA count/Total count of clean reads*1,000,000). miRNAs whose normalized expression of two libraries is smaller than one were removed, because their expression levels are too low. The fold change between treatment and control was calculated as: Fold-change = log₂ (DS/CK). The statistical analysis was performed according to Poisson distribution. The p-value was calculated by the following formula.

RT-qPCR was used to validate the results obtained from the high-throughput sequencing of miRNAs. RNA that was isolated using the Trizol (Invitrogen) as described above was reversely transcribed using One Step PrimeScript miRNA cDNA Synthesis Kit (TaKaRa). This kit adds ploy (A) to the 3' end of miRNAs and start to reversely transcribe. The reverse transcription was led by a kind of special oligo-dT ligated with known sequence at its 5' end. RT-qPCR was performed using SYBR Premix Ex Tag II (TaKaRa) and all the primers used were listed in Additional file 5. Small nuclear RNA U6 was used as internal control. Subsequently, RT-qPCR was performed using Stratagene M × 3000P instrument.