Genome-wide discovery and validation of Eucalyptus small RNAs reveals variable patterns of conservation and diversity across species of Myrtaceae

Micro RNAs are a class of small non coding RNAs of 20–24 nucleotides transcribed as single stranded precursors from MIR gene loci. Initially described as post-transcriptional regulators involved in development, two decades ago, miRNAs have been proven to regulate a wide range of processes in plants such as germination, morphology and responses to biotic and abiotic stress. Despite wide conservation in plants, a number of miRNAs are lineage specific. We describe the first genome wide survey of Eucalyptus miRNAs based on high throughput sequencing. In addition to discovering small RNA sequences, MIR loci were mapped onto the reference genome and interspecific variability investigated. Sequencing was carried out for the two most world widely planted species, E. grandis and E. globulus. To maximize discovery, E. grandis samples were from BRASUZ1, the same tree whose genome provided the reference sequence. Interspecific analysis reinforces the variability in small RNA repertoire even between closely related species. Characterization of Eucalyptus small RNA sequences showed 95 orthologous to conserved miRNAs and 193 novel miRNAs. In silico target prediction confirmed 163 novel miRNAs and degradome sequencing experimentally confirmed several hundred targets. Experimental evidence based on the exclusive expression of a set of small RNAs across 16 species within Myrtaceae further highlighted variable patterns of conservation and diversity of these regulatory elements. The description of miRNAs in Eucalyptus contributes to scientific knowledge of this vast genre, which is the most widely planted hardwood crop in the tropical and subtropical world, adding another important element to the annotation of Eucalyptus grandis reference genome.


Background
In the last decade, small non-coding RNAs have emerged as key endogenous regulatory elements in eukaryotic cells. It became clear that part of the so called junk DNA was transcribed into silencing RNAs that take part in an intricate gene regulatory network with highly specific functions [1][2][3]. Among a variety of functional non coding small RNAs (sRNAs), two main classes have been subject of intensive investigation in plants: small interfering RNAs (siRNAs) and micro RNAs (miRNAs). Mature siRNAs are predominantly 24-nt long transcribed by RNA polymerase IV with subsequent double strand synthesis [4,5]. They are fundamental players in cellular defense mechanisms against viruses [6,7] and epigenetic regulation, as the drivers of RNA directed DNA methylation (RdDM), contributing to the silencing of transposable elements and transcriptional gene silencing [8,9]. MiRNAs are 20 to 24-nt long sRNAs involved predominantly in posttranscriptional gene regulation. Single stranded miRNAs primary precursorspri-miRNAsare transcribed by RNA polymerase II from MIR genes [4], and processed by Dicer like 1 enzyme (DCL1) to generate an intermediary precursor (pre-miRNA), typically folded into a stable single strand stem loop structure [1,[10][11][12]. In plants, the mature miRNA is excised from the pre-miRNA, exported from the nucleus and then incorporated into protein RNA induced silencing complex (RISC). RISC is guided to the target messenger RNA (mRNA) by sequence complementarity to miRNA [13,14]. The silencing mechanism is determined by the degree of complementarity between the pair miRNA-target mRNA. Pairing mismatches are frequent in metazoans and miRNAs tend to block target mRNA by hindering its translation whereas high complementarity usually directs mRNA degradation in plants [15]. A wide variety of processes such as plant development [16][17][18][19], flowering [18,20,21], meristem and vascular differentiation [16], disease resistance [22][23][24] and response to abiotic stress [25][26][27] are regulated by miRNAs.
One of the first steps to ascertain how these components engage in complex regulatory networks is the characterization of the small RNA repertoire of a species. In recent years, second generation sequencing provided the technical breakthrough for rapid and comprehensive small RNA discovery including non-conserved and low abundance miRNAs [18,[26][27][28][29][30][31][32][33][34][35][36]. Accordingly, a surge in studies characterizing plant sRNAs was initiated. Despite the vast number of sRNA reads in sequencing data, bona fide plant miRNA genes of a given species are typically numbered in hundreds exemplars. Then extensive postprocessing becomes fundamental to adhere sRNA sequences to a strict set of rules defining miRNA genes, which have foundations in recalling their biogenesis mechanisms [37].
There are many examples of highly conserved miR-NAs families in plants, some present from basal plant species to angiosperms [38]. However it is known that there is great interspecific variability in small RNAs repertoire with several MIR genes lineage or species specific. Studies on miRNA discovery, either by RNA sequencing or in silico prediction, have shown this high diversity in different species [39][40][41][42].
Eucalyptus is a highly diverse genus of the Myrtaceae family. Native to Australia and its northern islands, Eucalyptus species occur predominantly in the southern hemisphere from sea level to alpine tree line and from high rainfall to semi-arid zones [49]. Eucalyptus species are mostly outcrossers [50,51] and their extensive genetic variation has been widely used in breeding programs [52]. Due to their noteworthy high growth rate, wide adaptability, high biomass production and carbon sequestration capabilities, eucalypts became the hardwood crop most widely planted in tropical and subtropical areas, exceeding 20 million hectares around the world [53]. Eucalyptus globulus and Eucalyptus grandis are currently the most extensively planted species among the~700 species described for the genus, widely used as sustainable short fiber source for pulpwood and energy. Despite the great wood quality of E. globulus for pulp production, its temperate origin implies poor adaptation to highly productive tropical environments where E. grandis is the species of choice, making this latter one the most widely planted Eucalyptus species.
The genome of E. grandis was recently sequenced [54] and is available at Phytozome [55]. As part of that landmark development we performed a comprehensive annotation of miRNA genes, now fully described in this report. This additional layer of information will be valuable to promote a deeper genomic understanding of a number of processes such as tree growth, vascular development, phase change and response to environmental stresses, pests and pathogen, where miRNAs are known to be involved, and that currently make up the bulk of investigation both in forest productivity and health. Complementary strategies were used, starting with Illumina-based high throughput small RNA library sequencing (smRNA-Seq) for E. grandis and E. globulus, followed by degradome sequencing analyses for large-scale target mRNA identification. This study represents the first genome wide discovery, mapping and characterization of Eucalyptus miRNAs and should provide useful fundamental information for upcoming studies on gene regulation in what has now been promoted to an additional model species for forest tree genomics.

Small RNA sequencing data
Illumina GAII deep sequencing was carried out for the small RNA fraction of four samples: one leaf (E. grandis BRASUZ1) and three developing xylem samples (E. grandis BRASUZ1 and two E. globulus trees). This experiment resulted in a total of 6,104,498 raw reads ranging from 1,115,404 to 1,766,355 per sample. Pre-processing steps, namely quality screening, adapter and redundancy removal, resulted in a total of 1,857,986 unique sequences (Table 1). Contaminant sequences of ribosomal, chloroplast and tRNA origin were filtered out totaling 1.8 % of the unique sequences.
Read size distribution shows the expected two main peaks at 21 and 24 nucleotides (nt). Twenty-four nucleotide sequences are the most abundant reads in all four samples (Fig. 1). As seen by size ranking of smRNA-Seq data, 24-nt sequences also show extensive sequence diversity with the highest number of unique sequences (clusters) (Additional file 1: Figure S1). Despite the high diversity, each 24-nt cluster exhibits low expression levelnone of the 24-nt read makes up 1 % of the total counts in the size class. Twenty-one nucleotide sequences show an opposite distribution showing less sequence diversity (fewer unique sequences) but the highest counts per cluster observed amongst all sRNA size classes analyzed (from 15 to 28nt). The most abundant sequence within the 21-nt class varies from 10 % up to 40 % of total counts in BRA-SUZ1 leaves sample (Fig. 2).
Conserved miRNAs were often observed among 21-nt reads with highest counts. MiR159a was consistently the most abundant 21-nt sequence in three out of four samples -E. grandis BRASUZ1 leaves (40 %) (Fig. 2a) and developing xylem (10.5 %) (Fig. 2b), and E. globulus C3 developing xylem (18.2 %) (Fig. 2d). In the E. globulus A2 xylem sample, miR159a (6.99 %) was outnumbered by miR172g (9.16 %), but was the second most abundant 21-nt read (Fig. 2c). MiR159a is a highly conserved miRNA as seen by alignment of Eucalyptus miR159a with all plant orthologs present in miRBase (Fig. 3). MiR166 and miR396 are other conserved miRNAs that are featured among the top fifteen 21-nt reads in all four samples. A broad search for conserved miRNAs was carried out and is discussed further.

Genome mapping
Mapping of 1,857,986 sRNA unique sequences on the E. grandis reference genome (Phytozome 8.0) was carried out using BWA [56]. Uniquely mapped reads per sample varied from 67.2 % to 87.3 % (Table 2). Mapping data confirmed the high diversity of the 24-nt sRNA sequences previously mentioned in total reads counting. For all samples, 24-nt reads represent the highest number of uniquely mapped sequences. Mapping data was investigated for correlation between sequence size and mapping location in repetitive regions. Mapping data of unique reads from 19 to 26-nt showed sRNA reads located mostly in repeat regions (70.1 % in E. grandis BRASUZ1 leaves, 59.3 % in E. grandis BRA-SUZ1 xylem, 64.3 % in E. globulus A2 xylem and 64.9 % in E. globulus C3 xylem). Size distribution of mapped reads revealed a general tendency to mapping on repetitive regions irrespective of the sequence size (Fig. 4).

Characterization of conserved miRNA sequences
A similarity search of all 20 to 22 nucleotides unique sequences was done using PatMan [57] against miR-Base plant mature miRNA sequences. A total of 303 distinct 21-nt sRNAs showed significant similarity (with at most three mismatches) to an orthologous miRNA sequence in miRBase. Conserved reads with 100 % identity totaled 95 sequences encompassing 25  Table S1). Expression of miRNAs miR156a, miR159a, miR160 and miR172b was assessed by northern blot in leaves and developing xylem of E. dunnii, E. urophylla and E. grandis BRASUZ1 (Additional file 3: Figure S2).
MiRNAs with 22-nt in length constitute a class of sRNAs not as abundant as the 21-nt ones but which make up a subclass with outstanding role in triggering biosynthesis of phased secondary siRNAs known as tasiRNAs (trans-acting siRNAs) or phasiRNAs (phased siRNAs) [58,59]. Sixteen 22-nt conserved miRNA sequences with 100 % identity and 69 with up to three mismatches were identified.

In silico prediction by miRDeep
There are hundreds of MIR genes identified in plant genomes. The massive bulk of smRNA-Seq reads thus requires careful analysis to identify bona fide miRNA genes, as established by a set of strict criteria [37]. The first one is to be excised from a stem loop arm of a single stranded intermediary precursor (pre-miRNA). To test smRNA-Seq data for this criterion, the reference genome of E. grandis BRASUZ1 provided the ideal conditions for a broad computational search. We used miR-Deep2 package for de novo prediction of miRNAs from sequencing data [60]. After a genome-wide search for candidate regions complying with precursor secondary structure constraints, 193 mapped sequences showed to be compatible with a MIR gene locus (Additional file 4: Table S2). Eighty two of these in E. grandis BRASUZ1 leaves sample, 55 in E. grandis BRASUZ1 xylem, 74 and 73 in E. globulus A2 e C3 xylem samples, respectively.
Five of the most abundant 21-nt reads in smRNA-Seq data (euc_sRNA_149582, euc_sRNA_75850, euc_sRNA_ 438131, euc_sRNA_33215 and euc_sRNA_372867) (Fig. 2) had their expression experimentally revalidated by northern blot in three Eucalyptus species -E. dunni, E. urophylla and E. grandis (Additional file 5: Figure S3). Nevertheless, despite the evidences of expression (smRNA-Seq and blot) and mapping on the reference genome of E. grandis, their flanking regions did not meet the requirements for a typical miRNA precursor structure.

Target prediction and degradome sequencing
Prediction of transcripts targeted by miRNAs adds another level of in silico confirmation, providing clues about potential biological processes being regulated. Target prediction was performed for all mapped sequence candidates with a compatible precursor secondary structure using psRNATarget [61]. The Eucalyptus transcript database from Phytozome was used for reverse complementary matching between smRNA-Seq data and potential target transcripts. Functional annotation of targets was retrieved from the BioMart tool available in Phytozome. The number of predicted targets for each miRNA candidate varied from one to twenty transcripts. The enrichment for specific protein domains in predicted target mRNAs is shown in Fig. 5 and indicates the prevalence of signature domains for NB-LRR (NB-ARC, leucine rich repeat) disease resistance genes, ion transporters, SBP (squamosa binding proteins) transcription factors and PPR (pentatricopeptide repeat) proteins. Predicted targets also include transcripts related to wood formation such as cellulose synthases and cytochrome P450 which is involved both in biosynthetic and detoxification pathways.
To get a glimpse of miRNA binding site conservation on MYB transcription factors, the orthologs were searched in Phytozome database, the transcripts scanned for the presence of miR159a binding site, using psRNATarget, and the ones with at least 20 bases flanking the binding the site were used to create a sequence logo (using WebLogo 3 [62]) highlighting the miRNA binding site at the target transcripts (Fig. 6).
To experimentally confirm in silico predicted targets, a degradome sequencing experiment was performed. Libraries of 3' cleavage products were prepared from mRNA samples of leaf and developing xylem from an adult E. grandis BRASUZ1 tree. Illumina HiSeq sequencing of both samples resulted in 27,244,395 raw reads. After filtering for low quality reads, including no adapter, no insert and sequences smaller than 18 nucleotides, the total number was reduced to 26,387,851 (97.53 %) valid sequence reads. Annotation of filtered reads resulted in 18,257,616 total and 2,887,536 unique sequences. Non coding RNAssuch as ribosomal, transporter, small nuclear and small nucleolaraccounted for around 2.67 % of the total read count. Sharing of unique sequences between the samples represented 9.64 % of the reads.
Analysis of degradome sequencing data with Cleave-Land pipeline identified 189 cleaved transcripts targeted by 21-nt sRNA sequences from leaves and 324 from xylem smRNA-Seq data (Additional file 7: Table S4; Additional file 8: Table S5). Considering the 22-nt sRNAs, the number of pairs was 149 and 248, respectively. Targets included a set of transcription factors such as MYB, GRAS and SBP families, cation transporter genes and ARFs (auxin response factors) as reported before for other plant species [63,64] and matching the in silico results predicted by psRNATarget.
A wide variety of other transcripts involved in diverse physiological processes such as cytochrome P450, TIR-NB-ARC disease resistance genes and nuclear transport factor 2 (NTF2) were also identified in the degradome data. Some results corroborate known targets of conserved miRNAs such as a SBP transcription factor cleaved by miR156 [65][66][67], an ARF transcript, by miR160 [68,69] and the transcription factor AP2 (apetala 2), by miR172 [70,71]. The degradome sequencing data further supported two newly identified Eucalyptus miRNAs, which displayed two distinct transcripts of cellulose synthase as targets, detected only in the xylem sample (Eucgr.D00476.1 and Eucgr.H00939.1 in Phytozome).

Conservation within Myrtaceae
Interspecific variability in sRNA loci was observed from mapping data of E. grandis and E. globulus sRNA reads on the E. grandis reference genome. In our smRNA-Seq data, the number of mapped sequences differed by 20 % between the two species (Table 2). In order to extend the inquiry of sRNA conservation within Myrtaceae, we expanded the analysis to other species and genera. A comparative analysis of Eucalyptus and Eugenia uniflora smRNA-Seq data (available at the NCBI Gene Expression Omnibus, GEO, accession number GSE38212) was carried out. Mapping of sRNA reads of E. uniflora to the E. grandis genome totaled 1,392,334 (34.7 %) of the total unique sequences. Size sorting of mapped sequences showed a higher conservation of 21-nt sRNAs and lower of 24-nt, 61.6 % and 21.1 % respectively (Table 3). Same tendency was observed when overall unique sequences were compared. Considering only perfect matches (100 % identity), 21-nt common sequences were 10 fold higher than 24-nt and allowing 1 mismatch, 7 fold higher. Higher conservation was also observed for 22 nucleotides sRNAs when compared to 24 nucleotides.
Northern blot analysis of some of the most expressed sRNA reads in the smRNA-Seq data was carried out in order to investigate conservation within the Myrtaceae family. Five sRNAs probes were hybridized against RNA from completely developed leaves of seventeen Myrtaceae species, including six Eucalyptus species (E. grandis, E. botryoides, E. brassiana, E. globulus, E. pellita and E. resinifera), a hybrid E. urograndis (E. urophylla x E. grandis), in addition to ten other species of different genera of Myrtaceae -Corymbia citriodora (previously classified as Eucalyptus citriodora), Eugenia uniflora, Psidium cattleyanum., Psidium guajava, Syzygium cumini, Melaleuca lateritia, Eugenia calycina, Eugenia dysenterica, Campomanesia pubescens and Syzygium malaccense. Three diverse selected outgroup species -Glycine max, Lycopersicum esculentum and the gymnosperm Pinus taedawere also included in the experiment. Expression of two sRNAs (euc_sRNA_149582, euc_sRNA_ 75850) were not detected in any of the outgroup species (Fig. 7a) but were consistently detected in all Myrtaceae samples (Fig. 7b). The other three sRNAs (euc_sRNA_ 438131, euc_sRNA_33215 and euc_sRNA_372867) are potentially Eucalyptus specific as their expression were confirmed in three Eucalyptus species -E. grandis, E. dunnii and E. urophylla -(Additional file 5: Figure S3) but not in the outgroups G. max, O. sativa, L. esculentum and P. taeda nor in the other Myrtaceae species tested -C. citriodora and E. uniflora (Fig. 7a). Secondary structure prediction for genome flanking regions of these sRNAs failed to confirm them as miRNAs.

Discussion
Our results highlight the pivotal importance of a careful analysis and raw data filtration of the massive amounts of sequence data produced by next generation sequencing. The large initial number of reads is progressively reduced after consecutive analyses. Ultimately, 1,405,134 (75.63 % of total valid reads) 21-nt unique sequences mapped unambiguously to the E. grandis BRASUZ1 reference genome. Ninety five 21-nt unique sequences are conserved micro RNAs with 100 % identity to sequences available in miRBase. Three hundred and three have an orthologous sequence in miRBase with up to 3 mismatches and thus potentially correspond to new isoforms of conserved miRNA families. While the identification of conserved miRNAs can be easily accomplished by similarity search with previously described orthologous sequences, the annotation of new miRNAs requires the adherence to a set of strict criteria such as the presence of a compatible secondary structure of precursor sequence and target prediction. This process was highly benefited by the availability of a high quality E. grandis reference genome, possibly the last one generated exclusively by Sanger sequencing. Among all 21-nt mapped sequences, 193 have a compatible miRNA precursor secondary structure as predicted by miRDeep2. From this subset, 163 21nt sequences had at least one predicted target by psRNA-Target, complying with the criteria to annotate a miRNA. As seen by the read size distribution, 24-nt reads were the most abundant size class in the smRNA-Seq data with up to 3.75 times the number of 21-nt reads. The predominance of 24-nt sRNAs, mainly represented by siRNAs, is well known in angiosperms and had been previously reported for several species [4,[72][73][74]. Noteworthy is that, in addition to being the most abundant class overall, 24-nt sRNA Eucalyptus sequences constitutes by far the most diverse group, with the greatest number of clusters. The 24-nt overall high sequencing read counts observed is therefore explained by its diversity of unique sequences even with few reads per cluster when compared to 21-nt reads. Altogether this is in agreement with the premise that small RNA repertoire in plants is dominated by a vast number of 24-nt small interfering RNAs (siRNAs) [15]. Contrary to that, 21-nt readswhich include both siRNAs and miRNAsare less diverse, with fewer unique sequences, but exhibit the highest read counts per cluster in sequencing data.
As heterochromatic 24-nt sRNAs predominantly silence transposons and repeat regions by directing DNA methylation at complementary sites in the genome [15], one would expect to see a higher proportion of these sRNAs mapping to repetitive regions. In fact this seemed to be an overall tendency for all Eucalyptus sRNAs from 19 to 26-nt as observed for all samples in the smRNA-Seq.
Interspecific variability in small RNA content of the genome was evidenced by the mapping data. E. grandis BRASUZ1 displayed 20 % higher proportion of uniquely mapped sequences to its own genome when compared to E. globulus. Pairwise comparison of smRNA-Seq data for all samples corroborated this assertion: the most similar samples were from developing xylem of different trees of E. globulus. Pairwise comparison also highlighted the tissue specificity of sRNA expression, as the most diverse repertoire of sRNA reads were from distinct tissues-developing xylem and leavescollected simultaneously from the same E. grandis BRASUZ1 tree (data not shown).
It is suggested that conserved miRNAs usually have higher expression and that lineage specific ones are expressed at lower levels or in specific tissues, developmental stages or conditions [29,75]. Counting of our 21-nt sequences reinforced this concept, as conserved miRNAs were consistently seen among the most abundant reads. Though a large number of 21-nt sRNAs that are not conserved miRNAs were also highly expressed, in silico analysis of these sequences showed that they do not fit canonical miRNA criteria being representatives of another class of sRNAs.
Recently, a large scale identification of miRNAs was performed in Eugenia uniflora [34]. A comparative analysis between smRNA-Seq data from Eucalyptus and E. uniflora indicated a high conservation within Myrtaceae family. This high conservation probably arises mostly from conserved miRNAs sequences as evidenced by higher conservation of 21-nt sequences. MiRNAs commonly have non-related targets and are frequently involved in housekeeping conserved pathways. As miRNA silencing relies on sequence complementarity to heterologous targets, this class tends to suffer more selective evolutionary pressure. On the other hand, siRNAs usually silence related targets or even their own origin loci, acting in cis. It is suggested that siRNAs suffer little or no selective pressure to maintain sequence conservation resulting in high evolutionary rates [76]. Our experimental results of northern blot for 21-nt sRNAs (other than miR-NAs) assessed in various Myrtaceae species and outgroups suggested a tendency of sequence conservation of highly expressed sRNAs within the family.
Target prediction showed predominance of NB-LRR proteins, the most common disease resistance genes in plants, known to be highly regulated by sRNAs [77]. Transcription factor families, as SBP and MYB, were also abundantly present in target prediction. MYB proteins are known as transcription factors related to wood formation [78]. The R2R3-MYB gene family, as an example, is supposed to control lignification during xylogenesis (wood growth) [79,80]. Targets involved in biosynthetic pathways were also predicted such as cytochrome P450. These proteins play a key role in the synthesis of structural polymers as lignins [81] which together with cellulose are the two basic wood components.
A recent study on Eucalyptus grandis miRNAs investigated the relationship between alterations in miR156 and miR172 expression, associated with vegetative phase change [19,82,83], and adventitious root induction during development [84]. Inquiries like this highlight the potential of miRNA investigation in diverse biological pathways with a vast impact on plant development and productivity among other aspects. That study also conducted a profile on miRNA expression in cuttings describing 40 known and 8 novel miRNAs, including one (Cluster_41475) also present in our data (Scaffold11 -40625339-40625414).

Conclusions
This work provides the first comprehensive genome-wide discovery and characterization effort of miRNA in species of Eucalyptus. High throughput smRNA-Seq with in silico and experimental evidences allowed the characterization of conserved and novel miRNAs. Due to the lack of biological replicates in our smRNA-Sequencing data, an addition that would have allowed further quantitative analyses, we limited our study to a qualitative survey of miRNAs. Nevertheless, the data presented lays the foundation for forthcoming differential miRNA expression analyses.
The availability of a high quality genome sequence for E. grandis was a key asset to carry out a robust precursor structure prediction and this in turn provided improved experimental evidence to support the discovery of bona fide novel miRNAs. When a reference genome is not available, precursor secondary structure prediction is dependent on the availability of expressed sequence tags (ESTs) a method that has limitations. Furthermore, by using smRNA-Seq obtained from the same exact tree whose genome is the reference genome sequence, the analysis considerably improved the number of mapped sequences. The smRNA-Seq data from E. globulus on the other hand, provided solid evidence confirming the interspecific variability in the small RNA repertoire even between related species belonging to the same subgenus Symphyomyrtus. Availability of a large transcript database of the target species also highly favored the identification and characterization of targets. The identification, mapping and characterization of miRNAs loci described in this work directly contributed to the annotation of the Eucalyptus grandis genome [54] adding another layer of information to the current reference genome.

Plant material, RNA extraction and sequencing
Plant material for smRNA-Seq includes the same tree genotype used by JGI-DOE (Joint Genome Institute -USA Department of Energy) for whole genome sequencing: E. grandis (BRASUZ1 tree), a selfed tree (S 1 ) from Suzano Group (Brazil). RNA from four biological samples was prepared for deep sequencing experiments: developing xylem and leaves of E. grandis BRASUZ1 and developing xylem of two unrelated E. globulus trees (named A2 and C3). RNA extraction was performed with an adapted CTAB protocol [85]. Library preparation for Illumina GAII sequencing used single end kit and barcoded adapters to multiplex samples in one lane run, all performed by Fasteris SA (Switzerland). For experimental validation via northern blot, developing xylem and leaves from two trees of E. dunnii and two of E. urophylla and leaves from two clones of E. grandis BRASUZ1 were employed. All RNA samples were obtained from fully developed leaves from adult trees.

Analytical pipeline
A custom-made computational pipeline was developed to process sRNA sequencing data. A pre-processing step cleans the sequences by trimming sequencing adaptors, quality screening and contaminant removal (including ribosomal, chloroplast and tRNA). Processed reads were size sorted, quantified (tag counting) and used to create a non-redundant read set using UCLUST [86]. The non-redundant reads were mapped to the E. grandis reference genome available in Phytozome [55] using BWA [56], with parameters -n 1. In order to identify conserved miRNA sequences, a similarity search using PatMan [57] was carried out against the subset of plant-specific mature sequences from miR-Base release 19 [43,87], allowing at most three mismatches (paramenters -e 3). Novel miRNAs were predicted using the miRDeep2 pipeline [60] (mapper parameters -e -j -l 19 -o 16), which performs a genome wide search for potential miRNA precursors based on the extension of genome regions around mapped reads followed by secondary structure prediction and stability evaluation. Messenger RNA target prediction was performed using psRNATarget [61] applying default settings against the E. grandis transcript database from Phytozome server (version 8.0) [55].

Degradome sequencing
Total RNA from E. grandis BRASUZ1 developing xylem and leaves was used for parallel analysis of RNA ends (PARE) library preparation followed by Illumina HiSeq sequencing (BGI -Hong Kong). Analysis of degradome sequencing was performed by CleaveLand 3 pipeline that outputs potentially cleaved sRNA targets from both degradome and smRNA-Seq data [88].