Introduction

MicroRNAs are ∼22-nt endogenously initiated non-coding RNA (ncRNA) species that regulate the expression of target mRNAs in post-transcriptional level (Bartel 2004; Kim 2005). Increasing evidence reveals that miRNAs play significant roles in various physiological and pathological processes including development, differentiation, apoptosis, anti-viral defense, and tumorigenesis (Davison et al. 2006; Schickel et al. 2008; Wang et al. 2007). In general, miRNAs are produced from a primary stem-loop structure in nucleus and become functional after step-by-step processing in cytoplasm. The functional miRNAs have to integrate into multiprotein effector complexes, named as RNA-induced silence complex (RISC), to effectively catalyze sequence-dependent cleavage or specific repression of target mRNAs by binding to their 3′UTR regions (Denli et al. 2004; Gregory et al. 2004). Together with other non-coding RNA families, miRNA may be of great importance in articulating gene expression in details (Oulas et al. 2009).

As one of the most important economic insects, silkworm genome has been reported in 2004 (Xia et al. 2004), which provides opportunities for a thorough survey for miRNAs and their functionality. To date, only a few silkworm miRNAs are deposited in miRBase (version 13.0; http://microrna.sanger.ac.uk), and the collection is far from complete as compared with other model insects such as Drosophila. Traditional methods hold their own merits and limitations; for instance, direct cloning is superior for testing highly abundant miRNAs but not those low abundant ones and it is very expensive (Yin et al. 2008). While microarray is a powerful tool for detecting conserved miRNAs in different species and has made important contributions for large dataset surveys and specialized clinical applications, it is sensitive to background noise (Lehmussola et al. 2006). Given the complexity of the regulatory roles played by miRNAs in gene expression, their thorough discovery is of high importance, especially for the low abundance and transiently expressed miRNAs whose expression are often stringently regulated and tissue/developmental stage-specific. The SOLiD sequencing system with a sequence-by-ligation approach provides a powerful small RNA profiling method with high-throughput and accuracy.

Over 20 million raw sequence reads were obtained in the present study and analysis of the genome-matched reads yielded 287 novel candidate silkworm miRNAs, including both miRNA and miRNA* sequences (21%). Several candidates were experimentally validated by stem-loop RT PCR. We also systematically analyzed the potential SNPs and targets of silkworm miRNAs. The identification of silkworm miRNAs and the analysis of their potential targets have demonstrated that miRNAs may play an important role in the development of Bombyx mori, especially in regulating the ecdysone and juvenile hormone signaling pathways.

Materials and methods

Total RNA extraction

We collected tissue samples from silkworms (Dazao) at 14 different developmental stages demonstrated previously (Yu et al. 2008). Total RNA is extracted by using TriPure Isolation Reagent (Roche) according to manufacturer’s protocol.

Library construction and bioinformatics analysis

We isolated small RNA fractions (less than 40-nt) using the FlashPAGETM Fractionator system (Ambion, Austin, TX) and cloned them by SOLiDTM Small RNA Expression Kit (Applied Biosystems, Foster City, CA; Goff et al. 2009) according to the manufacturer's instructions. The sequencing was carried out by SOLiDTM System 2.0.

We used RNA2MAP (RNA_pipeline_0.4.0; http://solidsoftwaretools.com/gf/project/rna2map/) for primary data analysis. We first filtered reads against a dataset containing silkworm Unigene, tRNA, rRNA, Rfam sequences (excluding known miRNAs), and a custom-curated insect ncRNA database. We then attempted to align the remaining raw reads to all 55 known pre-miRNA sequences in miRBase Version 13.0(http://microrna.sanger.ac.uk/; Griffiths-Jones et al. 2008) and the sequences of the miRNA–miRNA* pairs with a 4-nt extension toward both directions.

To predict novel miRNAs, we aligned the yet unmapped reads against the genome sequences (Version 2.0; http://silkworm.genomics.org.cn/; Xia et al. 2008) by setting a 28-nt scanning window and extracted the matches with 100-nt flanking sequences for hairpin structure prediction. We used RNAfold (Denman 1993) for the prediction of candidate novel miRNAs as depicted before (Ambros et al. 2003; Yu et al. 2009).

The most abundant reads from predicted pre-miRNAs were considered as candidate mature miRNAs. After a relatively stringent screening, those with five or more copies in frequency and 25-nt or less in length were considered as candidates of novel miRNAs. The predictions were mapped to the genome for cluster analysis and cross-species comparison against other insect miRNAs and genome sequences involving 12 Drosophila species (http://flybase.org), Anopheles gambiae (http://agambiae.vectorbase.org/index.php), Apis mellifera (http://www.hgsc.bcm.tmc.edu/project-species-i-Apis%20mellifera.hgsc?pageLocation=Apismellifera), and Tribolium castaneum (http://www.hgsc.bcm.tmc.edu/projects/tribolium).

Stem-loop RT PCR

The stem-loop RT PCR experiments are carried out as described before (Chen et al. 2005). All of the designed stem-loop RT primers and gene-specific primers are listed in Supplemental file 1.

Target prediction

We used miRanda (Enright et al. 2003; John et al. 2004) version 3.1 to predict potential targets on 3’-UTR sequences retrieved from silkworm (Dazao) unigenes (ftp://ftp.ncbi.nih.gov/repository/UniGene). The thresholds for candidate target sites were S ≥ 90 and ∆G < −17 kcal/mol (Enright et al. 2003). Those silkworm miRNAs are also scanned against 3′-UTR datasets of Drosophila melanogaster and A. mellifera searching for potential targets. A target of B. mori is considered to be conserved if its orthologous in D. melanogaster or A. mellifera is also predicted to be candidate targets.

Results

Library construction, sequencing and preliminary miRNA identification

We obtained over 20 million raw reads by SOLiD sequencing from a RNA library that is constructed by using small RNA enriched samples from 14 different developmental stages. Our data processing pipeline is shown in Supplemental file 2. Since the decoding of SOLiD sequencing reads is a reference-based methodology at the present stage, we currently rely on database registries for contaminates elimination, known silkworm miRNA identification and genome matching. First, reads corresponding to rRNA, tRNA, sno/snRNA, mRNA, and other non-coding RNAs are removed; 701,913 reads that match this filter with up to one mismatch. Second, the remaining reads are used as an input file for known silkworm miRNA identification, which involves 247,410 hits with up to two mismatches. Third, the reads that have not mapped in the first filtering step or against known silkworm miRNAs are mapped to the silkworm genome in an attempt to identify novel miRNAs. In this final step, 3,219,395 reads were successfully mapped. And these reads matched to the silkworm genome are collected together with 100-nt flanking sequences for candidate miRNA prediction. Finally, in order to guarantee accuracy, we did not attempt to map more reads that do not hit any of the references using looser mapping stringency as they may contain sequencing errors, contaminants or un-sequenced/assembled genomic regions. Taken together, we successfully annotated over 4 million reads.

Known silkworm miRNAs and their abundances

In our library, we detected all of the 55 previously annotated B. mori miRNAs (miRbase Release 13.0) with the exception of bmo-miR-190 (ESM Table 1). This single failure is probably the consequence of several chance events including incompleteness of the sampling (referring to the entire silkworm life span, for instance), the delicate RNA processing procedure, and uneven sequencing depth (Fahlgren et al. 2009). Of the 54 miRNAs identified, 50 are found with both miRNA and miRNA*, and four are accounted for those with either miRNA or miRNA*. Of the 247,410 reads that map to known silkworm miRNAs, 10.4% match the hairpin structure in out of the miRNA and miRNA* regions at a high frequency consistent with the previously reported study (Ruby et al. 2006). It is believed that high-throughput sequencing not only is capable of identifying miRNAs but also provides information on gene expression ('t Hoen et al. 2008) albeit with variable results (Kato et al. 2009). Some miRNAs, such as bmo-bantam, bmo-miR-1 and bmo-miR-263a, are found at a very high frequency (∼10,000×) which alludes to their functional roles in regulating silkworm development (ESM Table 1). However, there are also many miRNAs expressed at relatively low levels, probably representing stringent transient and spatiotemporal expression. Since we constructed a library with pooled RNA samples, it is not possible to come to a conclusion of which stage these miRNAs expressed and accumulated. We also discovered a notable strand bias between miRNAs and their corresponding miRNA*s in known silkworm miRNAs (ESM Table 1) as reported previously in the case of D. melanogaster (Okamura et al. 2008). There may be functional implications in such a bias. In two such examples, bmo-miR-31 and bmo-miR-993a, whose miRNA*s are more abundant than their miRNAs, it is likely that the complementary miRNA*s are more active than their miRNAs.

Novel miRNAs and their cross-species conservation

In a search for novel candidates, we evaluated reads with perfect match to the silkworm genome that fell within miRNA-like hairpin structures, and a total of 2,036 qualified hairpins represented by 1,306,704 short reads were identified (see details in the “Materials and Methods” section). Comparison with a recent silkworm miRNA identification study based on the sequencing-by-synthesis method reveals an overlap of 16 miRNAs between the two datasets (Zhang et al. 2009). Besides, these candidate novel miRNAs also included six overlaps with our former discoveries published recently (Supplemental file 3) (Yu et al. 2009). We obtained 287 novel miRNA candidates mapped with 116,494 short reads in their stem regions(provisional named with prefix bmo-miR-p), in which about 20% of candidates contain both miRNA and miRNA* sequences (Table 1) and the remaining reads were presented either in miRNAs or miRNA*s (Supplemental files 4 and 5). We believe that the detection of miRNA*s is a strong clue, albeit not absolute, for the formation of precursor hairpin structures and adds weight to the authenticity of the predicted candidates (Fahlgren et al. 2007; Sunkar et al. 2008). These novel candidates displayed a concentrated length distribution between 21 and 25 nt with a peak at ∼23 nt as well as a robust U bias at 5′ ends, which is a characteristic of miRNAs produced from dsRNA precursors by Dicer. To confirm the expression of the identified miRNAs, we performed stem-loop RT PCR and validated 36, out of 43 attempted, of the candidate novel B. mori miRNA sequences (Fig. 1). The remaining seven candidates were excluded due to non-specific amplification.

Table 1 Summary of novel miRNA candidates with both miRNAs and miRNAs
Full size table
Fig. 1
figure 1

Stem-loop RT PCR. Experimental validation of a subset of the predicted miRNAs

Full size image

To determine whether the novel candidates are conserved among other insect species, we searched their sequences against other insect miRNAs and genome sequences, including T. castaneum, A. gambiae, A. mellifera, and twelve Drosophila species. As a result, 81 silkworm candidates are identified as cross-species conserved with high confidence; there are eight homologs that are already reported in other insects (bmo-miR-316, bmo-miR-1000, bmo-miR-998, bmo-miR-996, bmo-miR-11, bmo-miR-308, bmo-miR-306, and bmo-miR-1175; directly named according to their homologs). The other 73 candidates are mapped to at least one or more insect genomes and foldable into eligible hairpin structures (Supplemental file 6). The remaining candidates do not have cross-species homologs, which probably represent silkworm-specific miRNAs.

SNPs in the silkworm miRNAs

We also found single-nucleotide polymorphism phenomena in our miRNA library. We identified SNP sites in both “seed” region and outside of the “seed” region on the mature miRNAs (Fig. 2a). Most of the SNP sites are located in the outside of the “seed” region suggest that the “seed” region of miRNA is more conservative than the other region of miRNA. We failed to recognize an obvious SNP bias in the known and the novel candidate miRNAs because of the arbitrary distribution of the observed SNPs. Although some of the SNPs in mature miRNAs may be false positive resulting from assembling or sequencing errors, we believe that the highly abundant ones may come from individual differences at DNA level, which need to be further validated and distinguished from post-transcriptional modification events, such as miRNA editing (Ohman 2007).

Fig. 2
figure 2

Possible SNPs and clusters in silkworm miRNAs. a The most representative single-nucleotide polymorphisms in the known miRNAs and the new candidate miRNAs in silkworm are illustrated. The mature miRNAs, hairpin locations, and putative SNPs are in green, yellow, and red, respectively. The folding energy (dG), mature miRNA length, and frequencies are also shown. b Candidate novel miRNAs located in cluster structures are shown in appropriate direction and color-coded (not drawn to scale)

Full size image

Cluster analysis

One of the typical characteristics of miRNA is that closely juxtaposed miRNA genes are inclined to transcribe from a common promoter to generate polycistronic primary structures (Baskerville and Bartel 2005; Lee et al. 2004). In our study, 31 of the novel miRNA candidates are also grouped into 13 clusters; most of them are composed of two miRNAs (Supplemental file 7). The top cluster is made up of six miRNA molecules, and each has a high detection frequency (Fig. 2b). Surprisingly, some of the previously annotated miRNAs located in cluster structures that are proposed to co-express showed a strong inconsistency in our results. For example, the empirical expression level of bmo-miR-1 is nearly forty times higher than that of bmo-miR-133 and the two are 16 kb apart in B. mori. The reason for this disparity remains unknown.

Potential targets for silkworm miRNAs

The expression of miRNAs is generally considered to be negatively correlated with their mRNA targets. To gain an overview on the functions of the silkworm miRNAs, we performed miRanda (Enright et al. 2003) to predict potential targets. The thresholds for candidate target sites were S ≥ 90 and ∆G < −17 kcal/mol, and those unqualified targets are not considered for further analysis. Silkworm miRNAs are scanned against 3′-UTR datasets of B. mori, D. melanogaster, and A. mellifera, respectively, for candidate targets prediction. As a result, we obtained 172 conserved targets that involved in a broad range of biological functions, such as transcription factors, signal transduction, metabolism enzymes, and cancer-related genes (Supplemental file 8). Many genes involved in several levels of the ecdysone cascades and the downstream transcription factors are predicted, including EcR, Broad-Complex, pre-prothoracicotropic hormone, nuclear receptor HR3 and nuclear orphan receptors(data not shown). We also obtained several target genes that are involved in the biosynthesis and metabolism of ecdysone such as cytochrome P450, ecdysteroid-phosphate phosphatase and ecdysone-20-hydroxylase.

Discussion

The miRNAs have been generally considered as negative regulators of gene expression in cellular. Identifying the total number of miRNA genes in a species, especially those low-abundance and species-specific ones, is helpful for appreciating the breadth of miRNA functions. MiRNAs have been identified in diverse animal species, and the large-scale miRNA identification in insects with experimental approaches has mainly been performed in fly. Herein, we constructed, sequenced, and analyzed a small RNA library made from a pooled RNA samples from multi-developmental stages of the domesticated silkworm (Dazao) based on the SOLiD high-throughput sequencing system. In total, we identified 287 novel miRNA candidates, in which 59 of the candidates contain both miRNA and miRNA* sequences. Eight of the 287 novel candidates are already reported in other insects. We show that the method is effective in identifying low-abundance and species-specific small RNAs. Additionally, we experimentally verified 36 of the candidate novel miRNAs. Compared to the previously identified silkworm miRNAs, the abundance of predicted novel miRNAs are much lower as indicated by their frequencies, which may account for tissue- or species-specific expression. And, as expected, most of the candidate novel miRNAs are not cross-species conserved. However, it is also likely that several novel non-coding RNA molecules may exist, generating analogous forms to miRNAs by forming hairpin structures, such as endogenous siRNAs, which cannot be excluded due to the lack of database information. All together, deep sequencing is a rapid and effective approach to uncover rare and species-specific miRNAs and the identification of a near complete set of miRNAs for a species is of fundamental importance to establish a basis for unraveling the complex miRNA-mediated regulatory networks.

We noticed a strand bias between miRNAs and their corresponding miRNA*s in known silkworm miRNAs. In the cases of bmo-miR-31 and bmo-miR-993a, the miRNA* strands are more abundant than their miRNA strands, it is likely that the complementary miRNA*s are more active than their miRNAs and thus maintaining a high concentration. Recently, reports sprang up and suggested that transcripts from the passenger strand of pre-miRNAs can also profoundly affect many biological pathways. Kim et al. demonstrated that miR-199a* is able to target MET (mesenchymal–epithelial transition factor) proto-oncogene and its downstream effector ERK2 (extracellular signal-regulated kinase 2) and inhibited the proliferation, motility and invasive capabilities of tumor cells (Kim et al. 2008). In the case of pre-miR-146a, Jazdzewski et al. proposed that miR-146a* from the passenger strand may regulate many genetic processes in thyroid cancer (Jazdzewski et al. 2009).

Single-nucleotide polymorphisms (SNPs) are a novel class of functional variations that may affect the generation and function of a miRNA and result in a change of target mRNAs. Importantly, an increasing number of studies have supported an association between miRNA SNPs and tumorigenesis (Chin et al. 2008; Horikawa et al. 2008; Jazdzewski et al. 2009). We focused on the identification of SNPs in mature miRNA regions and obtained a set of SNP sites with high frequencies which increased their authenticity. A few of the SNP sites is located in the 5′ seed region that implies their conservation is relatively higher than the other parts of a mature miRNA. Recently, a growing body of work has illustrated that SNPs in the sequences flanking the pre-miRNAs may affect processing by Drosha/Dicer and result in the impaired/enhanced expression of mature miRNAs (Duan et al. 2007; Hu et al. 2008; Sun et al. 2009). In animals, the “seed” region is considered as the key binding site for target identification, and a single-nucleotide polymorphism in this region may lead to loss or gain of function and greatly change the mediated genes depending on whether they interfere, degenerate, and generate a binding site (Jazdzewski et al. 2009). In short, miRNA SNPs have brought in a new field of research in miRNA functional studies and should be taken into account in the future functional assays.

One of the typical characteristics of miRNA is that closely juxtaposed miRNA genes are inclined to transcribe from a common promoter to generate polycistronic primary structures. It was demonstrated that the secondary structure of a clustered pre-miRNA is more similar to its neighboring pre-miRNAs in the same cluster, when compared to other sequences outside clusters (Leung et al. 2008). Lately, reliable evidence has been published (Kim et al. 2009) to support the hypothesis that miRNA genes of the same cluster may play related biological functions. Young-Kook Kim and colleagues reported that miRNAs in two clusters (miR-106b ∼93∼25 and miR-222∼221) suppressed the expression of the Cip/Kip family members of Cdk inhibitors in gastric cancer tissues. The increasing amount of evidence has led us to speculate that the organization as well as the functional association of clustered miRNAs should be an effective way to coordinate genomic resources and has survived long-term natural selection.

Our work shows that most of the potential targets of silkworm miRNAs are not conserved between species and are highly enriched in ecdysone and juvenile hormone signaling pathway regulators as well as hormone metabolism-related proteins. Ecdysone signaling regulation participates in controlling various biological pathways and plays a critical role in the developmental transitions of insects (Cruz et al. 2007; Horike and Sonobe 1999). Generally, hormone signals bind to the ecdysone receptor (EcR) and trigger the regulation of the expression of different transcription factors, such as the zinc finger proteins of the Broad-Complex and other kinds of nuclear hormone receptors, which in turn modulate central regulators of different physiological processes. Conversely, juvenile hormones perform an important antagonistic function to the biological processes controlled by ecdysone and results in the maintaining of larval modality, preventing the metamorphosis and development of the adult imaginal disk (Enright et al. 2003).

Interestingly, considerable silkworm miRNAs are predicted to aim at more than one hormone signaling pathway genes and/or hormone biosynthesis-related proteins, such as EcR, Broad-Complex, nuclear receptor HR3 isoform A, cytochrome P450, ecdysteroid-phosphate phosphatase, and nuclear orphan receptors. Therefore, we are prompt to speculate that miRNAs may directly or indirectly fine-tune a series of ecdysone-biosynthesis-related genes in order to regulate various developmental states in the life cycle of the silkworms. These predictions as well as speculations only come true when supported by experimental evidences. The accumulating knowledge about miRNA function will contribute to our better understanding of the complex gene regulatory networks at both transcriptional and post-transcriptional levels.

It has been reported that the expression of some miRNAs, such as let-7 and miR-34, are affected by ecdysone in Drosophila during metamorphosis (Sempere et al. 2002; Sempere et al. 2003). Besides, let-7 has been detected to express temporally and spatially in cultured cell lines of silkworm that showed a clear association with ecdysone pulse and a variety of biological processes (Liu et al. 2007). However, the mechanism of how bmo-let-7 responds to ecdysone and other hormone pathways in silkworm is still unknown. In our study, let-7 was highly expressed (represent by 4,683 short reads) and was predicted to target juvenile hormone acid methyltransferase (JHAMT) and hemolymph juvenile hormone binding protein (hJHBP), both of which were associated with juvenile hormone activity (Niwa et al. 2008; Wieczorek et al. 1996). Besides, miR-34 were predicted to aim at several nuclear receptor superfamily members including EcR, HR3 isoform A and orphan receptor, which played key roles in ecdysone regulation mechanisms (Eystathioy et al. 2001; Sutherland et al. 1995). Further experiments of gain-of-function and loss-of-function are still needed and will determine how many of these predicted targets are genuinely and directly targeted by miRNAs in silkworm.