We cloned and sequenced small RNAs from mock-inoculated (Control library) and B. japonicum-inoculated (3 h) soybean roots (Bj library). We obtained a total of 159145 sequence reads from the control library and 194855 sequence reads from the Bj library using deep pyrosequencing (454 Lifesciences). For computational analysis and filtering of potentially non miRNA reads, we combined reads from both the libraries. For abundance calculations, reads from Control and Bj libraries were differentiated based on ID tags. Only reads with a recognizable adapter (see Methods) sequence were retained for further analysis ('high quality reads' in Figure 1A). Since 454 sequencing is not directional, we used the 5' adapter sequence to determine the direction of each sequence read. Where necessary, the sequences were converted to their corresponding reverse complements to facilitate computational analysis. Adapter sequences were removed and the sequences were cleaned of low quality reads (see Methods). All sequences 17 nt or longer in length were retained ('17nt or longer' in Figure 1A).

Ideally, the length range of miRNAs in the library would be 20–24 nt while the majority of the miRNAs would be 21 nt in length 23. However, cloning and sequencing artifacts may result in a large number of genuine miRNA sequences that deviate from this expected range. We searched the sequences from both the libraries against miRbase 2526 to identify conserved miRNAs (see Methods for parameters used) and used the length of conserved miRNA sequences in the library as a measure of quality. The assumption is that a higher quality small RNA library will be enriched in 20–24 nt long conserved miRNAs. If a major part of the library was affected by inefficient cloning, sequencing or trimming errors, majority of the conserved miRNAs will be outside the 20–24 nt range.

A total of 3758 sequences in our library matched previously identified miRNAs listed in miRbase ('conserved miRNAs' in Figure 1B). About 68% of these reads were 20–24 nt in length representing possibly full length mature miRNAs (Figure 2). About 28% of the reads were 17–19 nt in length. These are very likely to be genuine miRNAs shortened due to cloning, trimming and/or sequencing artifacts. Majority of the conserved miRNAs in our libraries ranged from 20–21 nt in length (~60%; Figure 2) suggesting that we had high quality small RNA libraries.

The best method to differentiate candidate miRNAs from other RNA sequences in the library is to identify potential hairpin forming pri miRNA precursor sequences in the genome that encompass the mature miRNA sequence 1923. Hence, we searched all reads in our libraries against all public soybean, genomic and expressed sequence tag (EST) sequences available (see Methods) to identify reads with perfect sequence matches to known soybean or other legume sequences. Any read that did not match a known sequence in these databases cannot be used to predict precursors and these were excluded from this stage of processing. After searching for perfect match with known soybean sequences, we retained ~43% of the reads ('Perfect match to WGS' in Figure 1A). It is expected that this would exclude some genuine miRNA sequences; therefore, we determined the number of conserved miRNA sequences excluded to estimate the loss of genuine miRNA sequences. Of the 3758 conserved miRNAs, 2955 (~79%) remained in the analysis set ('Perfect match to WGS' in Figure 1B) indicating that ~21% of genuine miRNAs were lost due to unavailability of comprehensive genome sequence information for soybean.

Small RNA libraries often contain non-regulatory small RNAs including tRNAs, rRNAs, snRNAs and degradation products of protein-coding RNAs. We searched our library against several non-regulatory RNA databases (see Methods) which identified 107640 sequences from both the libraries and all of these were excluded from further analyses (Figure 1A). This excluded just one conserved miRNA (ppt-miR896; 27 reads) from the library (Figure 1B) even though it removed close to 30% of the total sequences.

To further enhance the identification of genuine miRNA sequences, we eliminated all 'orphans' (reads represented only once in both libraries combined) and retained only 'non-orphan' sequences. Differences in read length were taken into account during this process (see Methods). This clean-up process removed only ~2% of genuine miRNA sequences (Figure 1B). We also excluded library sequences that hit more than 20 individual soybean whole genome shotgun (WGS) reads since these are unlikely to be genuine miRNAs 23. Consistently, this step did not exclude a single conserved miRNA (Figure 1B). The above described clean-up processes resulted in 4959 unique sequences (18868 reads; Figure 1A) ranging from 17 to 30 nt in length which were used for computational identification of miRNAs.

Candidate miRNAs can be identified from genomic and/or EST sequences by searching for the presence of potential hairpin forming precursors encompassing the mature miRNA sequence present in the library. Using this criterium, candidate miRNAs can also be differentiated from other regulatory RNAs such as siRNAs 192327. For each candidate miRNA sequence that had a perfect match in the soybean genome, we determined the ability of sequences in the surrounding region (at least 30 bp away and not farther than 500 bp on either side) to form a hairpin structure encompassing the miRNA sequence. We identified a total of 227 unique sequences (2626 reads; Figure 1A) that had the ability to form a potential hairpin structure of which 139 were novel.

The 139 novel precursors were subjected to folding (see Methods) and these hairpin structures were manually examined if they fit the criteria (less than 8 unpaired nucleotides and no more than three consecutive unpaired nucleotides of which no more than two were asymmetrically bulged in the 25 bp stem encompassing the mature miRNA sequence) used by 23. We selected 82 sequences that fit the criteria as candidate miRNA genes. The folded secondary structures of three of our candidates are shown in Figure 3 as examples. Finally, we clustered these sequences into 35 different families representing novel candidate miRNAs (Table 1). Of the 35 candidates, 3 had miRNA* sequences in our library (Table 1), and all three paired to their corresponding miRNAs with 2 nt 3' overhangs (Figure 3). This is strong evidence that the miRNA/miRNA* pair originated from a Dicer1-like (DCL1) processing indicating that these are genuine miRNAs 2223. Of the remaining 32 candidates, all of the 5 tested were validated by Northern expression analysis (Table 1, see below) suggesting that most of the novel miRNAs identified in the study could be genuine miRNAs.

For about 55% of the conserved miRNA sequences in the set used to search for hairpin-forming precursors, we identified potential precursor sequences. All of these formed hairpin structures that fit the criteria for pri miRNAs (data not shown). The rest of the conserved miRNAs either lacked full-length cDNAs or corresponding precursors in the EST or WGS databases. Of the conserved miRNAs that we identified, 13 were from novel loci (representing 9 families) previously unidentified in soybean (Table 2). Over all, we identified 12 conserved miRNA families using the above approach. Of these, two families also had miRNA* sequences present in the library. As expected, we were able to validate several of the conserved miRNA families by Northern expression analysis (Table 2, see below). Thus, in addition to novel miRNAs, the study also revealed novel loci for conserved miRNAs from the soybean genome.

As mentioned previously, we searched our library against sequences in miRbase to identify conserved miRNAs. We identified 20 different families of conserved miRNA sequences in our library based on homology (Table 2) comprising a total of 39 loci. Ten of these families (12 loci) had been identified in soybean earlier by screening EST libraries for similarity to conserved miRNAs 28. By constructing a small RNA library, we identified 10 additional families and 27 new loci representing conserved miRNAs in soybean (Table 2).

Using both of these methods above, we identified 55 families of miRNAs in soybean and of these 35 were novel. We also identified novel precursors and loci for conserved miRNAs. The identified miRNA sequences constituted about 3% of the total number of reads (Table 3). This number is low compared to other recent large-scale analyses of small RNA libraries 2223. The lower abundance in our library could be due to the fact that our library was constructed only from root tissue whereas the others were from whole plants or a combination of libraries from different tissues. In addition, our analysis excluded a large number of genuine miRNAs due to the lack of comprehensive genome sequence data.

We examined sequences in both libraries to identify differential abundance if any of the identified candidate miRNAs. Candidate miRNA sequences with at least 10 reads in both libraries combined were analyzed for abundance. As indicated in Table 4, there were several miRNA sequences that were up- or down-regulated in response to B. japonicum treatment in soybean roots. These included novel and conserved miRNAs.

We performed Northern analysis of selected miRNA candidates with two goals. 1. To experimentally validate these miRNAs and 2. To study the abundance of these miRNA during the early phases of nodulation. All of the 5 novel miRNAs tested were detectable by Northern expression analysis (Figure 4). Of the 10 conserved miRNAs tested, 9 were detectable by Northern expression analysis except miR167 (Figure 5). Thus, we were able to validate a majority of the identified miRNAs.

We performed a time course analysis of miRNA expression from 0 to 12 h after inoculation with B. japonicum cells. Interesting patterns of expression were observed for various miRNAs (Figures 4 and 5). A set of miRNAs tested were transiently up-regulated at 1 or 3 h after B. japonicum inoculation and were gradually down-regulated back to basal levels by 12 h (e.g. miR168, miR172). Significant up-regulation at 3 h post inoculation and sustained induction was observed for another set of miRNAs (e.g. miR159, miR393). Down-regulation of a set of miRNAs was also observed in response to B. japonicum (e.g. miR160, miR169). As expected, we also observed a number of miRNAs that remained unaffected by B. japonicum inoculation (e.g. miR1510, miR1509). Interestingly, we observed coordinated expression of certain miRNAs implicated in signal transduction of the plant hormone auxin (see Discussion).

We sought to identify potential targets silenced by the identified miRNAs in soybean. Criteria and methods developed by Bartel, Carrington and Weigel labs 293031 have enabled well-defined identification of potential targets for plant miRNAs with very few false-positives. We used the miRU online search utility 32 to identify predicted targets of candidate miRNAs and then examined the results manually to narrow down potential targets that fit the recently developed criteria 3031. The potential targets of both known and novel miRNAs were identified with varying levels of confidence. The results of the analysis are presented as supplementary data [see Additional file 1]. Some miRNA families had multiple targets, often functionally divergent with homologous sequences. In these targets, the putative miRNA binding-sites were often located in highly conserved regions.

For a subset of miRNAs, we experimentally verified the cleavage of selected targets by 5'-RACE analysis. We isolated polyA RNA from mock-inoculated and B. japonicum-inoculated (3 h) soybean roots and subjected the RNA to a 5'-RACE reaction (see Methods). We examined the presence of miRNA-directed cleavage products by PCR using primers specific to the target of interest. We were able to verify the cleavage of target miRNA for four of the identified miRNA families (Figure 6). The targets of miR166 and miR393 are well-conserved in other species. Consistently, we also observed cleavage of transcripts encoding a homeodomain protein [TGI:TC221756; 4 clones] and a TIR1-like protein [TGI:TC225843; 3 clones] by these miRNAs respectively. The identified targets of miR168 [TGI:BG882680; 1 clone] and miR396 [TGI:TC206710; 2 clones] in soybean seem to be non-conserved. These targets encode a putative protein kinase and a cysteine protease respectively. In all these cases, the site of cleavage corresponded to the position between the 10^th and 11^th nt of the miRNA.

Construction and high throughput sequencing of small RNA libraries seems to be the most efficient method for miRNA identification especially in species for which complete genome sequence information is not available. However, effective authentication of miRNAs is still best achieved by analyzing potential hairpin forming precursors identified from genomic or EST sequences. This approach identified the most number of novel miRNA sequences in our analysis (Table 1) and all 5 novel miRNAs tested were validated by Northern analysis (Figure 4). Based on our estimates using conserved miRNAs as internal control, this approach (analysis of hairpin forming precursors) identified about 49% of the conserved miRNAs in our library (Figure 1B). Failure to identify the rest of the conserved miRNAs was primarily due to non availability of either genome sequence data (~21%) or assembled genome shotgun sequences (~27%; Figure 1B). Analysis of our library using additional sequence data that might be available in the future 33 will lead to the identification of additional novel miRNAs. Another approach to the identification of miRNAs (without cloning) is based on similarity to previously known miRNAs in other species 2834. Obviously, this approach has the limitation of not being able to identify new or species-specific miRNA families. This approach has successfully identified 22 miRNAs from soybean ESTs. Another bioinformatics approach to identify novel miRNAs from species with little or no genome sequence information is to search for EST sequences with potential ability to form hairpin structures which has not been attempted in soybean to our knowledge.

We also identified conserved miRNAs in our library based on homology to known miRNAs listed in miRbase and this method is completely independent of the availability of genome sequence. This approach identified 20 families of miRNAs (Table 3) largely due to previous identification and characterization of miRNAs from other plant species that have their complete genome sequenced 1517181922. As expected, a number of conserved miRNAs were independently identified by the presence of potential hairpin forming precursor as well (Table 2).

Another approach to identify miRNAs in the absence of a genome sequence is to identify miRNA:miRNA* pairs in the small RNA sequence libraries themselves. We identified several conserved miRNAs using this approach and at least one novel miRNA which was validated by Northern expression analysis (miRX001; Figure 4). This approach may be promising for identifying regulatory RNAs in species with little or no genome sequence information. The most efficient method in identifying novel miRNAs seems to be the identification of potential hairpin forming precursors encompassing the mature miRNA sequence of interest (obtained by cloning) in either genomic or EST sequences. Combined use of these approaches might be the key to unravel novel miRNA loci and families in species with little or no genome sequence data.

In addition to the identification of miRNA sequences, we also examined the regulation of several known and novel miRNAs in response to B. japonicum inoculation in soybean roots. Thus far, only one miRNA has been shown to affect nodule development. The target of miR169, a CCAAT-binding transcription factor MtHAP2-1 has been shown to be critical for nodule development in Medicago truncatula 35. Since miRNAs largely act as 'early' regulators of signal transduction by modulating the levels of transcription factors, we chose to construct the library from soybean roots 3 h post rhizobial inoculation. A number of gene expression changes have been observed at this time point in soybean roots 36. We observed differential regulation of miRNAs that target a wide variety of target genes encoding transcription factors, receptor kinases, proteases, water channels and metabolic enzymes among others in response to B. japonicum inoculation [see Additional file 1]. The differential regulation of a subset of these miRNAs were also verified independently and in more detail by Northern analysis (Figures 4 and 5). In general, for about half of the miRNAs tested, the expression patterns observed by Northern analysis were consistent with the abundance of miRNA reads in the libraries. Discrepancies could be due to differences in cloning efficiency of certain miRNAs 21. In addition, our analysis excluded a portion of the library due to unavailability of genome sequence data. This would also affect the abundance calculations due to missing family members/loci. In general, large-scale sequencing of small RNA libraries seems to be a valid method to study miRNA abundance.

Our study has provided several candidate miRNAs to test for their role in nodulation. For example, transient up regulation of miR172 (Figure 5) which regulates a putative Apetala2-like transcription factor, down-regulation of miR166 and miR396 (Figure 5) targeting an HD-ZIPIII-like transcription factor and a cysteine protease respectively are potentially novel regulatory elements during nodulation. We also identified down-regulation of miR169 (Figure 5) shown to play a role in nodulation earlier 35. At least two putative miR169 targets predicted in soybean are highly identical (not shown) to the M. truncatula HAP2-1 functionally shown to play a role in nodulation. Functional analyses of these novel target genes in nodulation will yield interesting and useful information on their role in nodule signaling and development.

The plant hormone auxin regulates a number of developmental and physiological processes including nodulation. Several miRNA loci have been suggested to play a role in auxin homeostasis/signaling. For example, miR167 regulates the transcript levels of auxin response activator ARF8 2937 and miR160 regulates the transcript levels of auxin response repressors ARF10, 16 and 17 38. ARF8 and ARF17 regulate the transcription of GH3-like genes which encode auxin-amino acid conjugating enzymes that might regulate free auxin levels in the plant. ARF8 negatively regulates free auxin levels by upregulating these conjugating enzymes 39 whereas ARF17 downregulates them presumably increasing free auxin levels 40. We observed down-regulation of an miR160 family member in response to B. japonicum suggesting that ARF17 levels increase and perhaps free auxin levels as well. ARF17 levels have also been shown to be regulated by ARGONAUTE1 (AGO1), a core component of the RISC complex 41. Increased levels of ARF17 transcripts were observed in ago1 mutants of Arabidopsis 40. We also observed increasing levels of an miR168 family member (presumably down regulating AGO1 levels) in response to B. japonicum. The opposite regulation of AGO1 and ARF17 levels in soybean roots is consistent with previous observations that AGO1 might negatively regulate ARF17 levels.

Other miRNAs suggested to play a role in auxin signaling include miR393 that regulates the auxin receptor TIR1 and miR164 that regulates the levels of a NAC1 transcription factor 29. These miRNAs were up- and down-regulated respectively in response to B. japonicum inoculation in soybean roots. These observations suggest that endogenous regulatory RNAs might modulate auxin signaling and/or homeostasis during nodulation. While it still needs to be confirmed if the corresponding targets are silenced and what role they play in nodule signaling and development, these data provide an entry point for the study of miRNA regulation of early auxin signaling during nodulation.

Soybean (Glycine max cv. Williams82) seeds were surface sterilized by treating with 8% Clorox for 4 min followed by 70% ethanol for 4 min. The seeds were then rinsed three times with sterile deionized water. Surface sterilized seeds were planted in a 4" pot filled with 1:3 (v/v) mixture of sterilized vermiculite and perlite (Hummer International, St Louis, MO) and watered with nitrogen free plant nutrient solution (N^- PNS). B. japonicum cells (USDA110; a kind gift from Dr. Gary Stacey, University of Missouri, Columbia, MO) were grown in Vincent's rich medium 42 and resuspended in N^-PNS to a final concentration of 10⁸ cells ml^-1 (OD₆₀₀ = 0.08; 43). One-week-old seedlings were flood-inoculated with the above B. japonicum cell suspension (15 ml/pot). For mock-inoculation, seedlings were watered with N^-PNS. After 3 h of treatment, seedlings were uprooted, rinsed briefly in sterile deionized water to remove vermiculite/perlite particles, immediately frozen in liquid N₂ and stored at -80°C.

Small RNA isolation and library construction was performed as described by 19 except that the PCR products were directly sequenced instead of cloning in to a plasmid vector. RNA molecules corresponding to 15 – 30 nt in size were isolated from mock-inoculated and B. japonicum-inoculated roots and adapters ligated to the 3' and 5' ends sequentially as previously described 1944. These molecules were subjected to a cDNA synthesis reaction followed by PCR amplification 19. One sequencing run was performed on 3 μg of nucleic acid from each library by 454 Life sciences 45 resulting in a total of 354,000 small RNA reads (Figure 1A).

The adapter sequences (5'-AAACCATGGTACTAATACGACTCACTAAA-3' and 5'-TTTTCTGCAGAAGGATGCGGTTAAA-3') and low quality regions were subsequently removed using Lucy 46 with default settings except two parameters (-minimum 16 and -threshod 90). In addition, contaminants were removed using Seqclean 47 based on GenBank's Univec database 48. These sequences were searched against soybean WGS traces (~700,000 reads) downloaded from the National Center for Biotechnology Information (NCBI) trace archive (49;9/21/2006), draft genomes of Medicago truncatula (50;9/25/06) and Lotus japonicus (51;9/21/06) and TIGR soybean Gene Index (52;Ver. 12) using MegaBLAST to identify sequences with perfect matches to at least one sequence in these databases. The resulting ~153000 sequences were searched for perfect sequence matches to non-regulatory RNA sequences in the following databases using MegaBLAST, to exclude non-regulatory RNA sequences: soybean rRNA sequences in GenBank, plant large subunit-rRNAs 53, Arabidopsis rRNA, tRNA, snoRNA sequences 54, TIGR Fabaceae_repeats (Ver. 2), and soybean chloroplast genome (NC_007942). For elimination of 'orphan' sequences, we searched for reads that were represented only once in both libraries combined. Reads that had a perfect match to another read except for overhangs were considered to represent the same sequence. A total of 26722 "non-orphan" sequences (5872 unique reads) remained for further analysis. After excluding reads that had a match to more than 20 WGS reads, we retained 18868 reads (4959 unique reads).

For the identification of conserved miRNAs, library sequences were searched against plant miRBase (55;Version 10), which contained 1284 mature miRNA/miRNA* and 1220 corresponding precursor sequences with the BLASTN parameters: -W 7 -S 1 -F F, bit scores > 30. These sequences were then clustered to generate miRNA families. miRbase had eleven soybean miRNA families listed and all of them were identified via computational approaches 3456.

The genomic and cDNA sequences that had a region with perfect sequence match to at least one of the 4959 unique sequences from above were used for novel miRNA discovery. Those redundant soybean WGS sequences hit by small RNA sequences were assembled using CAP3 56. The consensus sequences with other cDNA sequences together were aligned with each given small RNA to search for its complementary sites (i.e. potential corresponding miRNA/miRNA*) within upstream or downstream 500-bp window using WU-BLAST (gapped WU-BLASTN parameters: M = 1 N = -1 S = 10 W = 6 Q = 1 R = 1). The candidate pri miRNA sequences were folded using MFOLD 57 and the resulting structures were manually examined for the criteria proposed by 23 i.e. to be designated as candidate miRNA, a 25 bp region of the stem centered around the mature miRNA should have less than 8 unpaired nucleotides (together on both arms) and no more than three consecutive unpaired nucleotides, of which no more than two were asymmetrically bulged.

RNA isolation and Northern analysis were performed essentially as described by 31. Total RNA was isolated from soybean roots inoculated with B. japonicum cells at different time points (0, 1, 3, 6 and 12 h) using Trizol (Invitrogen, Carlsbad, CA) reagent. Ten μg of total RNA was separated on a Criterion 15% TBE-Urea gel (Bio-Rad Labs, Hercules, CA), transferred to positively charged nylon membranes (Roche, Indianapolis, IN) using a Criterion wet transfer blot apparatus (Bio-Rad Labs, Hercules, CA) and fixed on the membranes using an UV cross-linker (Stratagene, La Jolla, CA). Oligonucleotide probes (IDT Inc., Coralville, IA) complementary to miRNAs were end-labeled with γ-³²P ATP (Perkin Elmer, Shelton, CT) using Optikinase (USB Corporation, Cleveland, OH) and purified using a nucleotide removal column (Qiagen, Valencia, CA). RNA blots were pre-hybridized for at least 1 h with Perfect Hybplus (Sigma, St Louis, MO) hybridization buffer and hybridized with the labeled probes for 16–24 h at 38°C. Blots were then washed twice (10 min ea) with 2× SSC 0.2% SDS at 42°C, exposed to phosphorimager screens (Amersham, Piscataway, NJ) for 1 h and imaged using a Typhoon phosphorimage analyzer. Blots with a higher background were additionally washed once at 50°C and re-exposed.

For experimental validation of target cleavage, total RNA was extracted from soybean roots at 0 and 3h after B. japonicum inoculation. PolyA RNA was isolated from these preparations using polyAtract mRNA isolation system III (Promega, Madison, WI). PolyA RNA was subjected to a 5'-RACE reaction using Gene Racer core kit (Invitrogen, Carlsbad, CA) omitting calf intestinal phosphatase and tobacco acid pyrophosphatase treatments. Gene-specific reverse primers were designed from targets to be verified and used in combination with the generacer 5' primer to amplify cleaved transcripts. PCR products were cloned in to pCR4 TOPO TA vector (Invitrogen, Carlsbad, CA) and sequenced (Cogenics, Houston, TX).