Recent systematic analysis of the spatial expression of miRNAs in zebra fish larvae showed that most tissues and organs express a unique miRNA complement 30 . To identify conserved and pig-specific miRNAs, we constructed a small RNA library using pooled RNA from the heart, thymus and liver of the pig. Heart represents a muscle tissue with profound diversity in gene expression 26 , thymus is a vital part of the immune system essential for T-lymphocytes development and maturation, and liver represents a tissue with specialized and restricted function and was found to have low level of transcript diversity 26 . The miRNAs from such a library should reflect a higher abundance of heart-, liver- and thymus-specific miRNAs. Additionally, the library should provide an opportunity to identify miRNAs expressed with low-abundance in these tissues. Small RNAs (15–30 nt) were recovered by size fractionation, and the library was constructed as described previously 31 32 33 .

Of the 20,000 reads obtained from the 454/Roche GS-FLX massively parallel pyrosequencer, ~12,592 quality small RNA sequences ranging from 15 to 30-nt were extracted after removal of the 5' and 3' adapters [see Additional file 1]. The quality of our small RNA library is poor as reflected by the accumulation of degradation products from miRNAs (shorter sequences ranging in size between 15 to 8-nt). While the sequencing depth obtained in this study is certainly not exhaustive, but these many reads should be sufficient to identify many of the novel but conserved miRNA homologs in pig. The low depth of the library could also be due to the technical problems associated with the sequencing. All identical sequences were counted and eliminated from the dataset, and only unique sequences with associated read counts were analyzed further. These unique sequences were subjected to BLAST analysis against the Rfam database to remove the ribosomal RNA, tRNA and snRNA breakdown products. This analysis removed ~304 sequences that represented largely breakdown products of rRNAs and tRNAs (Table 3). A search against RepBase revealed 91 sequences as possible sequences derived from repeats. Thus far, miRBase lists only 55 pig miRNAs that were computationally predicted. A BLAST search with the use of the remaining unique small RNA sequences revealed 24 unique sequences (368 reads) that exactly matched with the pig miRNAs listed in the latest miRBase release 11.0 (April, 2008) (Table 1). Our identification of less than 50% of the pig miRNAs listed in miRBase could be due to the use of highly specialized tissues (heart, thymus and liver) for miRNA library construction. BLAST searches against miRBase were performed with remaining small RNAs to identify pig counterparts of the conserved miRNAs. This analysis revealed 98 new miRNAs (represented by 1158 reads) that are homologs of the conserved miRNAs (Table 2 and Figure 1). Together, our study confirmed the expression of 120 conserved miRNAs (24 predicted miRNAs deposited in the miRBase and 96 new miRNAs) in pig (Tables 1 and 2). A significant number (6728) of small RNAs could not be mapped to the available pig genome as it exists at present. A small portion of these sequences could be potentially pig-specific miRNAs or other important small RNAs but their annotation requires the complete genome sequence.

In the present study, the cloning frequency of newly identified miRNAs varied greatly (1 to > 400) (Tables 1 and 2). Some of the sequences had slight-shifts in their 5' and 3' ends, which could be attributed to processing shifts or enzymatic modifications of miRNAs such as RNA editing 34 , 3' nucleotide additions 35 or sequencing artifacts.

Because, we pooled the total RNA from the heart, liver and thymus for library construction, matching miRNAs to tissue sources is not possible. Nevertheless, by comparing the tissue-specific expression patterns represented by the sequence counts, we can suggest the relative tissue contribution of at least a few miRNA families. For instance, the miR-1 family has the highest frequency (411 times) in our sequences (Table 2) and the highest level of expression in the heart, but was barely detected in thymus and liver (Figure 2). Therefore the total miR-1 count in our sequences could be derived largely from heart tissue. As another example, miR-122 is represented by 126 reads (Table 1). It is a liver-specific miRNA, as determined by a small RNA blot analysis (Figure 2B; 36 , and therefore, its total count could be attributed to the liver source. Some other miRNAs are ubiquitously expressed in the heart, liver and thymus, so their counts in our sequences could be attributed to the 3 tissue sources. For instance, let-7 is represented by 445 reads and miR-26 by 177 reads (Tables 1 and 2), and these two miRNAs are ubiquitously expressed in the heart, liver and thymus (Figure 3A and 3B). Thus, miRNA families (e.g., miR-1 and miR-122) that are specifically or highly expressed in any one of the 3 tissues, or miRNAs that are expressed ubiquitously (e.g., let-7 and miR-26) in all 3 tissues, show a far greater frequency than other miRNAs. A few notable exceptions are miR-499, an miRNA abundantly expressed in the heart (Figure 2A), which is represented by only one read (Table 2), and the miR-133 family, which is preferentially and abundantly expressed in the heart (Figure 2), and represented by only 7 reads (Table 1). These observations suggest that cloning frequency does not always reflect the true abundance. This could be attributed to the biased cloning, although what causes the biased cloning of small RNAs is unknown.

In animals and plants, miRNAs exist as multigene families (two or more closely related sequences and vary by 1 or 2 nt only). With use of small RNA blot analysis or probe-based miRNA micro array, discriminating individual members within a miRNA family is difficult. The expression information of a member of a miRNA family is useful for understanding its functions and for practical use for gene manipulations 37 . The cloning frequency-based determination of miRNA abundance can be used to assess the relative expression of each member within a miRNA family. The miR-1 family is represented by three members (miR-1a, miR-1b and miR-1c) in diverse animals (miRBase). We cannot ascertain whether the miR-1 family is also represented by three members in pig because of the lack of complete genome information, but is possible because we found miR-1a, miR-1b and miR-1c homologs in our library (Table 2). Of these three members, miR-1a and miR-1c are represented by 312 and 72 reads, respectively, whereas miR-1b is represented by a lower number (22 reads). The let-7 family has 10 members in diverse animal species (miRBase). Here, we found evidence for the expression of all 10 let-7 members in pig. Interestingly, the expression abundance varies among the let-7 family members (Tables 1 and 2); let-7a and let-7j, each have 80 reads; similarly, let-7b, let-7c and let-7e have almost the same number of reads (63–64); let-7d, let-7f and let-7j have 18 to 32 reads; and let-7h, let-7i and let-7k have a lower number of reads (5–9) (Tables 1 and 2). The miR-98 sequence differs from that of the let-7 family by one nt at position 11 from the 5' end, thus miR-98 is also a member of the let-7 family. miR-98 is represented by 5 reads in our sequences (Table 2). Hence the let-7 miRNA family is represented by 11 members, and this study provides the evidence for the expression of all 11 let-7 family members in pig. Similarly, we found all members of the miR-15, miR-16, miR-18 and miR-133 families in our sequences, suggesting that all members belonging to these miRNA families are expressed in these three (heart, liver and thymus) tissues.

Several recent studies found a number of lineage-specific miRNAs 18 19 or species-specific miRNAs 18 20 21 . Small RNAs in our library lacking annotations in miRBase but having perfect matches to the available pig genome are novel miRNA candidates. Some of the conserved miRNA homologs had appeared only once in our sequences (Table 1 and 2). Several candidate miRNAs are identified and some of them could be even detected using small RNA blot analysis (data not shown). These sequences have several loci in the pig genome. Fold-back structures could be predicted for some loci, but not for all. However, these evidences are not adequate to annotate them as miRNAs.

Often, sequencing-based miRNA expression profiling has been used as a tool to measure the relative abundance of miRNAs 35 60 . Our sequence analysis in this study indicated that miR-1 family (miR-1a, miR-1b and miR-1c) has the highest abundance (411 sequence reads). In agreement with this observation, miR-1 is the most abundantly expressed miRNA in the heart but not in the liver or thymus (Figure 3), two other tissues used for miRNA library generation. A similar picture has emerged for miR-122, a liver-specific miRNA and one of the highly represented miRNAs in our sequences. However, this correlation may not always be the case. For instance, miR-133 is represented only by 4 clones (two reads each for 133a and 133b) in our sequences, which indicates a 100-fold lower expression level compared with that of miR-1 family, if cloning frequency taken as a measure of expression. However, our small RNA blot analysis indicated a different picture as miR-133 was detected as abundantly as miR-1 in the heart (Figure 2). These two miRNA genes – miR-1 and miR-133 – exist as a cluster and thus are always expressed together in mouse 42 . The discrepancies between the cloning frequency and small RNA blot results for miRNA-1 and miR-133 could not be attributed to the RNA source because the same RNA samples were used for both experiments (cloning and small RNA blot analysis). We also used approximately a similar amount (activity) of ³²P-labelled probe for detection of miR-1 and miR-133. In our view, small RNA blot analysis provides a more reliable measure of abundance. Therefore, the above discrepancy could be largely due to biased cloning efficiencies of these two miRNAs. In Arabidopsis, a similar discrepancy between the cloning frequency and small RNA blot results was observed for miR398 previously 31 . What causes the biased cloning is not clear but could be attributed to the differences in the 5'- and 3'-end nucleotides, which are involved in ligating with the 5' and 3' adapters, respectively, or formation of secondary structures or adoption of a structure that prevents the exposure of 5' or 3' ends of the miRNA.

The expression of miRNAs is tightly regulated both in time and space. Tissue-specific miRNAs or a high level of expression of miRNAs only in certain tissues implies that these miRNAs play critical roles in the tissues where they are expressed. The observation that miR-22, miR-26b, miR-126, miR-29c and miR-30c are ubiquitously expressed in 14 different tissues of pig is interesting. Similarly, let-7, miR-98, miR-16 and miR-130a are abundantly expressed in 13 of the 14 tissues (except in pancreas) (Figure 3A). These results suggest a very important role for these miRNAs in the regulation of constitutive processes in diverse tissues.

The miR-17-92 cluster (polycistronic miRNA gene) encodes six miRNAs (miR-17, miR-18a, miR-19a, miR-20a, miR-19b-1, and miR-92-1) located in the third intron of a ~7-kb primary transcript known as C13orf25 61 . Although miR-18a and miR-20a are likely derived from the same primary-transcript, the expression levels of these mature miRNAs are not similar (Figure 2D). This finding suggests that their processing varies in different tissues. Expression profiling studies have also revealed widespread overexpression of the miR-17-92 cluster in diverse tumor subtypes, including both hematopoietic malignancies and solid tumors such as those derived from breast, colon, lung, pancreas, prostate, and stomach 58 59 .

Several miRNAs (miR-1, miR-133, miR-499, miR-208, miR-122, miR-194, miR-18, miR-142-3p, miR-101 and miR-143) have distinct tissue-specific expression patterns. Most animal miRNAs were thought to regulate several dozens of genes involved in diverse pathways and processes 62 . Additionally, many targets of these miRNAs are regulatory molecules such as transcription factors, which in turn may affect the expression of a large number of genes. Therefore, the impact of a miRNA expressed or not expressed in a specific tissue has a profound effect on the overall gene expression profiles in a particular tissue.

The use of three specialized tissues for library generation is a major limitation for the discovery of miRNAs in pig. Generating more small RNA libraries from diverse tissues should identify additional conserved and porcine-specific miRNAs. While our manuscript was in preparation, Kim et al., 29 reported 19 new pig miRNAs; 17 of these (let-7a, let-7b, miR-15a, miR-15b, miR-16, miR-17, miR-21, miR-23a, miR-23b, miR-24, miR-29a, miR-30b, miR-34a, miR-106b, miR-107, miR-130a, miR-140, miR-145, miR-152, miR-185, miR-199a, miR-210 and miR-221) are conserved miRNA homologs, and the remaining 2 are miRNA* sequences (miR-140* and miR-199b*). We have identified 11 of these miRNAs but did not find the remaining six miRNAs, possibly because these were not expressed or were expressed only at very low levels in the heart, liver and thymus tissues we used for the library.

Tissues from 10 month-old crossbreed were collected and stored at -80°C by the DeSilva laboratory as part of the ongoing porcine gene expression project. Three tissues (heart, liver and thymus) used for the generation of small RNA library were obtained from one animal. The experiments were approved by the Institutional Animal Care and Use committees. Total RNA from different tissues was isolated with use of Trizol reagent (Invitrogen, Carlsbad, CA) following the manufacturer's instructions. The RNA pellet was dissolved in de-ionized formamide for effective denaturation of the RNA. Small RNAs of the desired size range (15–30 nt) were gel-purified by resolving in denaturing 15% polyacrylamide gel. These small RNAs were sequentially ligated with the 3' and 5' adapters as described in our previous reports 31 32 33 . Reverse transcription reaction was performed using the RT primer (AAGGATGCGGTTAAA), subsequently performing PCR using the reverse and forward (TACTAATACGACTCACTAA) primers. The purified PCR product using phenol/chloroform extraction and ethanol precipitation were shipped to OU's genomics facility for sequencing. Manufacturer instructions were followed for preparation of DNA samples for sequencing on the 454/Roche GS-FLX 63 . PCR products containing the 454 adaptors were quantified and diluted prior to amplification by emPCR 63 . After emPCR enrichment, the DNA was loaded onto a 454/Roche GE-FLX for massively parallel pyrosequencing.

Our computational methods for analyzing 454 small RNA library was reported previously 64 . In brief, all small RNA reads without perfect matches to the most proximal 11 nt of both adaptor sequences were first removed. Reads corresponding to repeats were removed using the Einverted and Etandem programs in the EMBOSS package, respectively. The unique small RNAs were aligned to RepBase (version 13.04, obtained from http://www.girinst.org) and known non-coding RNAs (rRNAs, tRNAs, snRNAs, snoRNAs, etc., obtained from http://www.sanger.ac.uk/Software/Rfam/ftp.shtml) with NCBI BLASTN. In total, 304 small RNAs mapped to ncRNAs and 83 small RNAs mapped to RepBase were removed from further analysis. Then the small RNAs were mapped to the reported miRNAs in the miRBase (version 11), obtained from miRBase 65 . Small RNAs that matched known miRNAs of pig or other animal species resulted in identification of conserved miRNA homologs in pig. This analysis indicated that 120 unique small RNAs represented by 1562 reads matched with the conserved miRNAs listed in the miRBase.

Small RNA blot analysis was performed to determine the expression patterns of 22 conserved miRNAs. Total RNA (15–30 μg) was resolved on denaturing 15% polyacrylamide gel, along with labeled RNA markers. RNA was electrophoretically transferred to Hybond-N+ (Amersham) membranes, and the membranes were UV cross-linked and baked for 1 h at 80°C. DNA oligonucleotides complementary to small RNA sequences were end-labeled with γ-32P-ATP with T4 polynucleotide kinase (Invitrogen) used as a probe. Blots were pre-hybridized for at least 1 h and hybridized overnight with PerfectHYB Plus buffer (Sigma) at 38°C. Blots were then washed three times at 50°C and autoradiographed.

Note: While our manuscript is under review, two other papers have reported the identification of new miRNAs in pig using bioinformatics approach 66 and experimental approach 67 . Huang et al., 66 have bioinformatically predicted pig miRNAs and some of them were validated using micro array 66 . In another study, Sharbati-Tehrani et al., 67 cloned the small RNAs from pig but sequenced only small number of concatenated clones 30 . Sequence analysis has identified 11 miRNAs despite the fact that the sequencing depth is too low. Sharbati-Tehrani et al., 67 have reported 4 new miRNAs (miR-326, miR-423-3p, miR-484 and miR-451,) that could not be identified in our study, which could be attributed to the use of very specialized tissues (Jejunium, spleen, ileum and kidney) for their study.