For the four investigated plant species, we obtained mature miRNA sequences and stem-loop sequences associated with miRNA precursors from miRBase release 15 http://www.mirbase.org 15 yielding 218 miRNAs for Ath, 375 for Mtr, 46 for Bna, and 73 for Cre, respectively. No miRNA-star sequences were considered for analysis. For Ath, cDNA sequence information was obtained from The Arabidopsis Information Resource (TAIR, http://www.arabidopsis.org), genome release 9 16. For Bna, assembled contigs were retrieved from PlantGDB http://www.plantgdb.org/. For Mtr, sequence and annotation information was obtained from The Medicago Genome Sequence Consortium (MGSC, http://www.medicago.org), and here referred to as "Mt3.0". Sequences for Cre were downloaded from the DOE Joint Genome Institute using genome assembly v4.0 and Augustus v5.0 gene models http://genome.jgi-psf.org.

We used smallRNA sequencing data obtained and published for Ath under eight 1117 and for Mtr under two different experimental conditions 12. The specific conditions were for Mtr: a) treatment with the symbiotic fungus mycorrhiza ("Myc") and b) treatment without the fungus ("N-Myc"). The eight conditions for Ath were: full nutrition ("FN"), phosphate starvation ("P"), phosphate starvation after three hours phosphate re-addition ("P+3 h"), nitrogen starvation ("N"), nitrogen starvation after three hours nitrogen re-addition ("N+3 h") (all from 11), and FN from root cells ("root+p"), phosphate starvation from root cells ("root-p"), and phosphate starvation from shoot cells ("shoot-p") (from 17). In total, 15.8 Mill small RNA reads were sequenced for Ath and 13.6 Mill reads for Mtr (2 conditions ("Myc", "N-Myc"). Degradome data to experimentally identify miRNA targets by detecting miRNA induced cleavage products from four conditions in Ath ("FN", "P-12 h", "P-48 h" and "N-48 h") and two conditions in Mtr ("Myc", "N-Myc") were used. For experimental details see 1112.

Normalization of expression values per condition was done to adjust for variable sequencing depth between samples. The sequencing reads mapping on annotated miRNA were normalized to reads per million (RPM) per experimental condition: number of reads per gene/number of total reads * 1E6.

For analyzing the conservation of miRNA families across species, we performed a pairwise global sequence alignment of all single mature miRNA sequences with the program Align0 18. Sequence pairs were considered conserved if the sequence identity was greater than 75, if there was a perfect match of seed sequence (6 nt, positions 2-7), and the two respective identifiers of the pair were classified by miRBase to be in the same MIRNA family.

Verified miRNA-target relationships were extracted from several sources: Supplementary Data of 19 (500 targets), 530 targets in total from the Arabidopsis Small RNA Project, "ASRP" (http://asrp.cgrb.oregonstate.edu, 20), experimental data reported in Supplementary Table two and three of 13, referred to here as "degradomeG" (60 targets), and degradome sequencing data for Ath and Mtr from in-house experiments, called "degradome" (1,154 targets) 1112. To identify miRNA-target relationships from degradome data, the CleaveLand algorithm was used 1421. miRNA-targets were further predicted using the program RNAhybrid 22. The mature miRNA sequence data from miRBase and, on the potential target side, the downloaded cDNAs or assembled ESTs mentioned above were used as input. We used the parameter settings described in 19. We required the minimum free energy of hybridization to be greater than 70% compared to perfect match hybridization; i.e. in concordance with the initial threshold used in 19. Note that for the final set, the authors in 19 used a stricter 75% mfe cutoff. In Arabidopsis and using a 75% mfe threshold level, we obtained 2,967 unique targets transcripts for 218 miRNAs. All RNAhybrid predictions with additional score and mfe information for all four plant species are provided in tabular format as supplementary material (Additional File 1).

Gene Ontology (GO) annotation files were downloaded: for Ath from TAIR 16, for Cre from the DOE Joint Genome Institute http://genome.jgi-psf.org/Chlre4/Chlre4.download.ftp.html, and for Mtr from http://www.medicago.org/genome/downloads/Mt3/. Annotations for Bna were assigned by copying the GO slim term from TAIR for the best hit from a BLAST run against Ath. The calculation of over-representation of GO terms was done by applying the Fisher's Exact Test for count data and the p-values for Molecular Functions and Biological Processes were adjusted for multiple testing applying the Benjamini-Hochberg method 23.

First, we provide an overview of the statistics of miRNAs, their targets, and the genomic context in the respective plant species (Table 1). We based our analyses on the miRNAs deposited in miRBase (v15, see Methods). Interestingly, for the three well annotated species (Ath, Mtr, Cre), a positive correlation was found between the number of miRNAs and the genome size (Pearson correlation coefficient r = 0.91), and even more strongly to the number of transcripts encoded in those genomes (r = 0.99), albeit with only three species, significance cannot yet be established. Thus, the repertoire of miRNAs appears to scale with the size of the transcriptome that is to be regulated. Since for Bna only assembled EST reads were available that are likely overestimating the number of transcripts, Bna was not included in this statistic. However, the average number of targets per miRNA appears to differ quite substantially between species (Table 1). Across all species, every miRNA targets, on average, 46 transcripts indicating a rather large target set per miRNA. In Ath, about one fifth of all transcripts is predicted to be targeted by the known miRNAs, whereas in Cre, about half of all transcripts are under miRNA control, while the percentage is about 10% in Mtr, and even less for Bna (0.7%), although it needs to be noted again that the number of transcripts may be inflated in Bna. Furthermore, miRNAs in Bna were found primarily via sequence similarity searches. Thus, the Bna-specific miRNA set may be missing altogether. Target assignments came primarily from RNAHybrid predictions (94%), and the remaining targets from public domain resources and the used degradome-sequencing data (see Methods). Of the 885 miRNA-target relationships in Ath (269 in Mtr) detected in the degradome dataset by the Cleaveland program (see Methods), 274 were also predicted by RNAHybrid (176 in Mtr) (Note: comparison was done independently of the actual cleavage position). Thus, many real, experimentally determined targets were actually missed by RNAHybrid, especially in Ath.

As reported for plant miRNA target action earlier 24, most miRNA target sites were found to fall within the coding regions (86% in Ath), whereas the 5' and 3'UTR regions are targeted by approximately 7% (in Ath).

Relative to Ath, Bna is evolutionarily the closest species with their supposed common ancestor dating back ~40 Mya, followed by Mtr (~110 Mya), and Cre as the evolutionarily farthest species relative to Ath (~475 Mya) (data taken from 9). According to miRBase (v15), the 712 individual miRNAs belong to 272 miRNA families. For the 47 miRNA families present in Cre, no event of conservation was found with a miRNA family from any of the three other organisms, and there was no conservation reported with any other plant species yet either 59. Only a very small fraction of miRNA families is present in more than one organism, 22 miRNA families in total. A larger fraction of miRNAs appears to be species specific (Figure 1). Bna is an exception, for which we found only one species specific miRNA family, but this may be due to the incomplete genome and miRNA search strategies based on sequence homology to other plant species.

In Ath, most of the 6,788 target transcripts associated with 5,690 genes are targeted by only one miRNA with the frequency of multiple hits decreasing rapidly with increasing number of miRNA targeting a mRNA, observed to follow a power-law (linear in double-logarithmic diagram) (Figure 2A). Transcripts targeted by 3 or more miRNAs were observed to be preferentially associated with developmental processes (GO-slim enrichment p-value (FDR multiple testing corrected) of 1.86E-6), response to abiotic and biotic stimulus (p = 3.46E-4), transcription (p = 4.15.E-3), and response to stress (p = 1.03E-3), compared to those transcriptions targeted by only a single miRNA. Conversely, for the number of targets hit by a single miRNA a relative plateau is observed up until about 20 different targets per miRNA decreasing also in a power-low fashion beyond this number.

Conserved miRNA families were found to target on average more gene transcripts - with the average number of targets summed up across the three species Ath, Bna, and Mtr amounting to 161.4 - than their non-conserved counterparts (106.2), p = 0.073 (Mann-Whitney test). Based on the available quantitative data of miRNA expression via normalized read counts (see Methods), conserved miRNAs were found to be expressed at higher levels than non-conserved miRNAs. In Ath, the average log-2 expression value for conserved miRNAs was 9.25 and significantly higher than the corresponding value for non-conserved miRNAs (3.92, p = 2.2e-5), observed similarly in Mtr with 7.39 for conserved vs. 4.94 average log-2 expression level for non-conserved miRNAs, albeit significance could not be established (p = 0.21).

Regarding biological processes particularly associated with genes targeted by conserved miRNAs, developmental and transcription-related processes are overrepresented (Table 2). Species-specific; i.e. non-conserved miRNAs, are predominantly associated with signal transduction and response to stress processes, as were "unknown biological processes", in fact this category was the most significant one (Table 2).

The available quantitative miRNA expression data derived from normalized read counts allowed us to compare expression levels of conserved miRNAs in the species Ath and Mtr and to probe whether expression levels are also conserved for sequence-conserved miRNAs. Among the various experimental conditions applied in the studies involving the two species, one was nearly equivalent between them: the phosphate-starvation condition in Ath and the non-mycorrhiza conditions in Mtr. Mycorrhiza facilitates phosphate uptake from the soil, thus non-mycorrhiza plants are exposed to phosphate starvation conditions. For both individual miRNAs and averaged per miRNA family, a strong and significant positive correlation was found (Figure 3). Even when comparing average expression levels across all available 8 experimental conditions in Ath and two in Mtr, relative expression levels of sequence-conserved miRNAs appear to be preserved in the two different species as well. It appears plausible that highly expressed miRNAs target more transcripts than miRNAs expressed at low levels. Indeed, a positive correlation between the number of targets and expression level was observed in Ath (r = 0.29, p = 0.02), while no correlation was evident in Mtr (r = - 0.02, p = 0.95). Thus, the true nature of this relationship still remains to be established.