From our libraries, we detected the expression of 133 of the 154 previously annotated C. elegans miRNAs (miRbase release 11.0; Additional data file 2). While we did not detect 21 of the previously reported miRNAs (we suspect that most of these undetected miRNAs may not actually encode miRNAs at all 2329 or may be annotated incorrectly; detailed results are shown in Additional data file 3), we did obtain 125 clones of a very rare miRNA, lsy-6, expressed in only one pair of neurons in the C. elegans head 30. These findings demonstrate the significant sequencing depth of our survey. Conversely, the maximum number of clones we obtained for a single miRNA was 12,295,951 (miR-58; Additional data file 2), which highlights the high dynamic range of miRNA expression that can be surveyed using deep-sequencing technology such as that from Solexa.

Two miRNAs, miR-58 and miR-1, which showed the highest expression in our total libraries, were abundantly expressed in animals of all developmental stages we examined, from embryo to young adult of hermaphrodites, and in young adult males (Figure 2). Although the function of mir-58 in C. elegans remains unknown, we speculate that it has a general housekeeping role. Similarly, C. elegans miR-1 has a broad and generalized role, as it is involved in the function of neuromuscular junctions 31, and a mir-1 homologue in Drosophila has an important role in muscle development 32.

The number of sequence reads for a particular miRNA is known to be proportional to the molecular abundance of that species 33. Thus, the number of sequence reads of each unique miRNA in each sample is a reasonable measure of stage-specific expression during development (Figure 3). We controlled for library differences by normalizing these values to the total number of reads that matched to the C. elegans genome in each sample (Additional data file 4). The raw data for the number of reads is available in Additional data file 2. Finally, we confirmed by RT-PCR the relative stage-specific expression levels of the ten known miRNAs with highly dynamic expression patterns (Additional data file 5).

About 16% of known miRNAs showed major changes in expression at some point during development (for example, between embryo and the mid-L1 stage; Figure 3a, b). We define here 'a major change' as more than a tenfold difference in the number of reads. For example, the let-7 miRNA exhibited a major increase in expression around the mid-L4 stage, as did one of the let-7 family members, miR-48, from the mid-L3 stage (Figures 2 and 3a). Additionally, another well-characterized miRNA, lin-4, showed a large increase in expression from the mid-L2 stage (Figures 2 and 3a). These observations correspond to previously published results 3435 and support the validity and reliability for our small RNA libraries and our analysis.

It is interesting that we were able to clone multiple members of the let-7 and lin-4 families from stages where they were not previously known to be expressed (Additional data file 4). For example, we detected small numbers of clones to both let-7 and lin-4 in embryonic stages, many hours earlier than they had been observed previously. It is unclear if these miRNAs function during these earlier stages, since no embryonic phenotypes are known for let-7 or lin-4 null mutants 636. Conceivably, this could also represent maternal inheritance or a small bleed-through from the adults to the embryos during preparation.

Of the 24 miRNAs with major changes in expression, some had particularly dynamic expression patterns. For example, miR-71 is dramatically up-regulated from the embryo to the mid-L1 stage and then quickly down-regulated at the mid-L2 stage, and again gradually but significantly up-regulated after the mid-L4 stage (Figure 3a; Additional data file 5). Given its temporal regulation, this miRNA might be involved in control of developmental timing, like lin-4 and let-7. Another interesting case is the expression of miR-77, miR-85, miR-240 and miR-246, which is very low or completely absent in earlier developmental stages but increases after the mid-L4 and young adult stages (Figure 3b; Additional data files 4 and 5), implying a potential role in adult functions like reproduction, metabolism or aging. A recent report by Martinez et al. 37 also mentioned that some of these miRNAs, including miR-85 and miR-240, are temporally regulated during development, mirroring our results. We highlight additional developmentally regulated miRNAs in Additional data file 4.

The different sexes of animals result from different developmental pathways, which specify and maintain cell differentiation of the animal as male rather than female or hermaphrodite. Males in C. elegans have several distinct features and tissues, including mating organs in the tail and a male-specific germline, generating only sperm. In addition, males exhibit a smaller overall body size and different behavior compared to hermaphrodites. To assess those miRNAs preferentially expressed in males or in hermaphrodites, we generated and sequenced a cDNA library from small RNAs of young adult males (him-8 (e1489) mutants crossed with dpy-28 (y1); see Materials and methods). We found that about 12% of known miRNAs exhibited major differences in expression in hermaphrodites and in males (Figure 4; Additional data file 4). The correlation between miRNA expression levels in males and hermaphrodites is shown in Additional data file 6. Interestingly, most of the differentially expressed miRNAs are more abundant in males than hermaphrodites, which may reflect their expression in male-specific organs, for example, the rays used in copulation.

In order to identify novel miRNAs, we first filtered out sequence reads corresponding to all annotated RNA molecules, including miRNAs, mRNAs and other small non-coding RNAs. We then used the miRDeep program 38 to predict which of the remaining sequence reads might be miRNAs. This analysis revealed 66 novel miRNA candidates (Additional data file 7). In addition, we found the 'star sequence' for 24 of these candidates in our sequence reads (highlighted in red in Additional data file 7). Mature miRNAs are processed from the stem of a hairpin precursor, and the star sequence corresponds to the section of this hairpin that remains hybridized to the mature form (with approximately 2-nucleotide 3' overhangs) throughout much of miRNA biogenesis 33. The presence of these star sequence reads thus strongly suggests that at least these 24 novel candidates are bona fide miRNAs. We further examined the expression of five of these candidates using RT-PCR in both wild-type N2 and alg-1(gk214) mutant backgrounds. It is known that the two Argonaute family members alg-1 and alg-2 are essential for miRNA processing, but have no role in the RNA interference (RNAi)-mediated silencing pathway including siRNA (small interfering RNA) production 3940. Indeed, mature let-7 miRNA transcripts were less abundant in the alg-1 mutant background, as were those of all five novel miRNA candidates tested (Figure 5a). This was also confirmed in the alg-1 RNAi background (data not shown). These observations indicate that these five candidates are indeed true miRNAs. Computationally predicted secondary structures of the primary miRNA transcripts (pri-miRNAs) of these novel miRNAs are shown in Figure 5b.

Furthermore, of the 66 novel miRNA candidates, 20 may fall into known miRNA families since they had the same core target-binding ('seed') sequence found in other miRNAs in other species (Figure 6a; Additional data file 7). One of the novel miRNAs verified by RT-PCR, miR-2209a, has the seed sequence common to the bantam miRNA family, which is known to function in apoptosis 41. Further, we found that this novel miRNA is clustered on chromosome IV together with another four novel miRNA members, including miR-2208b-5p, miR-2208b-3p and miR-2209c (Additional data file 7). Also, these clustered novel miRNAs had similar expression patterns, falling into the male-enriched group (see below; Figures 6b and 7; Additional data file 8). Moreover, another validated novel miRNA, miR-2212, was genomically clustered on chromosome X with a known miRNA, miR-1819, and both showed male-enriched expression (Figures 6b and 7; Additional data files 4 and 8).

To visualize broad trends in the temporal expression of both previously identified and our newly identified miRNAs, we performed a simple hierarchical clustering. (Figure 7). We found that the 199 miRNAs detectable in our analysis assort into roughly five groups: those expressed primarily at the embryonic stage, those enriched in males, and those primarily expressed in early, middle, and late larval development.

Interestingly, we found that genomically clustered miRNAs are not necessarily co-expressed at the same levels. Some sets of miRNA map to specific chromosomal clusters, as in the case of miR-35 to miR-41, which have redundant functions in embryonic development 42 and are abundantly expressed in the embryonic stage (Figures 3b and 7). Genomically clustered miRNAs are thought to be transcribed as a single transcript and then individual pre-miRNA are subsequently processed out. We found that although these miRNAs have generally similar expression patterns during development (Figure 7), the absolute expression levels are strikingly different (Additional data file 4). Perhaps, then, clustered miRNAs may be differentially controlled at the transcriptional level and/or during subsequent processing.

Our analysis of the changes of miRNA expression during development may provide helpful information in identifying the target genes for these miRNAs. Coupling this data set with several of the studies describing mRNA expression profiles during development and aging of C. elegans 4344 could provide correlations pointing to potential miRNA-target pairs, since changes in expression of miRNAs may cause reciprocal expression patterns of their target genes during development of C. elegans. (Although miRNAs that form imperfect duplexes with their targets inhibit protein production in animals, miRNA binding can also result in degradation of the target mRNA in C. elegans 45; indeed, microarray analysis has proven to be an effective way to find genes modulated by miRNAs 46.)

Another class of C. elegans non-coding small RNAs, 21U-RNAs, have important functions in transposon silencing in the germline and maturation of gametes 242526. More than 15,000 unique 21U-RNA sequences have been reported in C. elegans, the vast majority of which map to either intergenic or intronic regions on chromosome IV 2325. As expected from their function in germline development, our results confirmed recent studies that show prominent accumulation of 21U-RNAs in the young adult stage (Figure 1; Additional data files 1 and 9) 242526.

To test if there are functional differences with regard to 21U-RNAs in the sperm, we further examined the expression of 21U-RNA in wild-type hermaphrodites together with males (dpy-28(y1);him-8(e1489)) at the young adult stage. Although the overall mapping pattern of 21U-RNAs on chromosome IV seemed unchanged in each strain, their abundance was significantly decreased in males (dpy-28;him-8) compared to wild-type hermaphrodites (Figure 8 - note that the scale in wild-type (top) is tenfold greater than that in male (bottom); Additional data file 9). This reveals that sperm and/or their progenitors produce a number of the piRNAs/21U-RNAs, but the level may be lower than that in the oocyte germline in C. elegans.

Approximately 44% of known 21U-RNAs on chromosome IV are genomically clustered within 10 bp with other 21U-RNAs (see below), implying that expression of 21U-RNAs in each cluster is controlled in a similar manner, and one would expect that these clustered 21U-RNAs might show similar changes in expression in both male and hermaphrodite germlines compared to 21U-RNAs mapping outside the clusters. Interestingly, though, we did not detect common patterns in expression of 21U-RNAs in the clusters; that is, 21U-RNA abundance was routinely different for 21U-RNAs in the same cluster, although 21U-RNAs in a genomic cluster appears to be transcribed from the same strand (data not shown).

In the course of our analysis, we identified approximately 10,000 21-nucleotide sequence reads starting with a uracil that have not been previously annotated (Additional data file 10). These reads are referred to here as 21nt-U-RNA for descriptive purposes to differentiate them from previously identified 21U-RNAs. Of these 21nt-U-RNA sequence reads, about 40% mapped to chromosome IV while the remaining approximately 6,100 reads mapped to other chromosomes, ranging from 7% of reads in chromosome X to nearly 16% in chromosome I (Figure 9; Additional data file 10). While many of the 21nt-U-RNA reads on chromosome IV mapped to the two distinct regions observed for known piRNAs/21U-RNAs, similar clustering was not apparent on other chromosomes (Figure 9). To determine whether these sequence reads represent new members of the piRNA/21U-RNA family, we searched for characteristic features of previously described 21U-RNAs. Although 21U-RNAs generally share little sequence identity other than the uracil at their 5' termini and specific localization on chromosome IV, it has been shown that the sequences upstream of 21U-RNAs contain an 8-nucleotide core consensus motif, CTGTTTCA, centered within a larger motif 23. About 14% (562), of our 21nt-U-RNAs on chromosome IV had a complete consensus motif in their upstream larger motif (the 43-nucleotide regions, -20 to -63 bp upstream from 5' termini of each 21nt-U-RNA, were analyzed.), whereas only a few 21nt-U-RNAs on other chromosomes had this 8-nucleotide motif (Additional data file 10). This result is consistent with the chromosome IV-biased localization of known piRNAs/21U-RNAs. We therefore believe that the 21nt-U-RNAs reads that map to chromosome IV and contain the core motif are indeed new piRNA/21U-RNAs (Additional data file 11; note that 10 of the 562 novel 21U-RNAs (21nt-U-RNAs) map to multiple loci on chromosome IV).

While we have not shown that these RNAs associate with Piwi proteins like PRG-1, we suspect that these are very likely to be novel piRNAs/21U-RNAs for several reasons: first, these RNAs are abundantly expressed in the L4 and young adult stages (Additional data file 12; consistent with known 21U-RNAs); second, they are transcribed from the same two distinct regions of chromosome IV as known 21U-RNAs (Additional data file 12); third, they contain the core motif associated with bone fide 21U-RNAs; and fourth, most of them partially overlap with known or other novel 21U-RNAs (see below). Also, approximately 8% of these novel 21U-RNAs were detectable in other libraries obtained by 454 sequencing from different biological sources (ADL and FS, unpublished result).

Of the 562 novel piRNAs/21U-RNAs we identified, 438 partially overlap other 21U-RNAs; either of their termini is located within 10 bp of another 21U-RNA terminus (although not separated by 10 nucleotides as in the case of Drosophila piRNAs; Figure 10a; Additional data file 11). Note also that approximately 43% of the 21U-RNAs on chromosome IV recently reported in Batista et al. 25 partially overlap (Figure 10a; Additional data file 9 - reads that overlap other 21U-RNAs are marked with a dagger). Interestingly, we noticed longer sequence reads in our libraries that encompassed mature 21U-RNAs (Figure 10a; a list of all longer transcripts detected is available in Additional data file 13). In total, 910 21U-RNAs were found in such longer reads, which corresponds to about 6% of the previously annotated and novel 21U-RNAs. These 21U-RNAs are marked with an asterisk in Additional data files 9 and 11. Similar longer reads were also detected in other small RNA libraries from 454 sequencing (ADL and FS, unpublished result), suggesting that they are biological products but not artifacts of Solexa sequencing. One possible explanation for the presence of longer 21U-RNA transcripts could be that they are by-products due to errors in 21U-RNA biogenesis - for example, read-through transcription and/or aberrant processing. For example, in the case of miRNAs, we also detected various larger sequence variants in our libraries (Additional data file 3). Alternatively, they may represent intermediates in 21U-RNA biogenesis. For example, original 21U-RNA transcripts may be longer in length and are processed to 21 nucleotides by an unknown mechanism. Indeed, in all cases we examined, the most abundant sequences were 21 nucleotides in length (Figure 10a; Additional data file 13), and a significant portion of these longer transcripts had an extension to their 3' side rather than the 5' side (Figure 10b). Additionally, the production of these longer 21U-RNA reads also appeared to be temporally regulated during development; they were abundant at the later stages of development, as in the case of 21-nucleotide mature 21U-RNAs (Figure 10c). Although the mechanism controlling 21U-RNA expression is still not clear, these observations lead us to speculate that precursor 21U-RNA transcripts are longer in length.

Wild-type N2 strains were cultured under standard conditions 49 at 20°C and used to prepare RNAs from each developmental stage (time after stage L1: mid-L1 (4 h), mid-L2 (14 h), mid-L3 (25 h), mid-L4 (36 h); and young adult (48 h). RNAs enriched for small RNA species (less than 200 nucleotides) were prepared using the mirVana miRNA Isolation kit (Ambion/Applied Biosystems, Austin, TX, USA) with the small RNA enrichment procedure. For library preparation from young adult males, dpy-28 (y1);him-8 (e1489) double mutants cultured at 23°C were used to obtain male populations after backcrossing six times to wild-type N2, and RNAs were purified at 40 h after stage L1. him-8 (e1489) mutants produce XO males and XXX hermaphrodites at 37% and 6% frequency, respectively, in addition to XX hermaphrodites 50. However, XX and XXX hermaphrodites can not survive at 23°C in the dpy-28 (y1) background 51, and the resulting surviving population of dpy-28;him-8 double mutants is almost all XO males at this temperature. For validating novel miRNA expression, total RNAs were isolated from N2 wild-type worms, alg-1 (gk214) mutants and N2 wild-type worms on both L4440 (empty vector) and alg-1 RNAi at the young adult stage.

cDNA libraries for small RNAs were made from 10 μg of RNA from an enriched small RNA fraction using the DGE-Small RNA Sample Prep Kit (Illumina, San Diego, CA, USA) according to the manufacturer's instructions. The same amount of cDNA was sequenced on a Genetic Analyzer from Illumina. The data from the miRNA reads we mentioned above were uploaded to the Genome Expression Omnibus database together with the raw Solexa sequence results [GEO:GSE13339]. The 66 novel miRNA candidates and the 552 unique piRNAs/21U-RNAs have GenBank accession numbers (shown in Additional data files 7 and 11).

The expression of some of the known miRNAs were confirmed by quantitative RT-PCR using a TaqMan Small RNA Assay (Applied Biosystems, Foster City, CA, USA) with the RNAs at concentrations of 0.4 ng/μl (enriched small RNAs) and 2 ng/μl (total RNAs), according to the manufacture's instruction. For validating the expression of novel miRNA candidates, 10 ng/μl of total RNAs was used, and the results were normalized to the expression level of U18. The results were further confirmed using independently prepared RNA samples.

The number of sequence reads for miRNAs and 21U-RNAs was assessed from the raw sequence data from Solexa sequencing using perfect sequence matching to known miRNAs (miRBase release 11.0) and 21U-RNAs 25 (Additional data files 2, 4 and 9). For examining the proportion of each non-coding RNA species, including rRNAs, tRNAs, snRNAs, and snoRNAs, sequence reads that matched to the C. elegans genome (WS190) were extracted by the SOAP program (a maximum of 2 bp mismatches were allowed in the alignment) 52, and the number of sequence reads perfectly corresponding to each RNA species was determined using BLASTN against a database of non-coding RNAs from WormBase 53. To compare the differential expression of small RNAs across development, the number of reads in each sample was normalized to the total number of reads that matched to the C. elegans genome in each sample. The Cluster 3.0 program was used to cluster the miRNAs (after normalizing each gene's expression vector to have a 2-norm of 1). The Java TreeView program 54 was then used to visualize these clusters. The miRDeep program 38 was used for finding novel miRNA candidates, and the RNA fold program was used for predicting secondary structure of primary miRNA transcripts of novel miRNAs.