Identification of Mammalian microRNA Host Genes and Transcription Units

RESULTS AND DISCUSSION

The miRNA Registry contains 232 mammalian miRNAs (http://www.sanger.ac.uk/Software/Rfam/mirna/; Griffiths-Jones 2004). Of these we identified 117 miRNAs located in introns of protein-coding genes or long ncRNA transcripts (Table 1, Supplemental Tables A,B). Approximately 40% (90) of all miRNAs are found within introns of protein-coding genes, whereas ∼10% (27) are located within introns of long ncRNA transcripts (Table 1). Interestingly, 30 miRNAs overlap with exons of ncRNAs (Supplemental Table C). In some cases (14), miRNAs are located in either an exon or an intron (`mixed') depending on alternative splicing of the host transcript (Table 1, Supplemental Table D). Where clusters of miRNAs overlap with a single host transcript, the vast majority of miRNAs are located in the same intron or exon (Supplemental Tables A-D). Additionally, 32 miRNAs overlap with two or more TUs transcribed on opposite DNA strands (Supplemental Tables A-D). This observation indicates that miRNAs are commonly associated with complex transcriptional loci. The remaining 70 miRNAs are of uncertain transcriptional origin and were not analyzed further. To sum up, a total of 161 miRNAs are linked to the transcription of mRNAs or ncRNAs. Based on these observations, we propose that miRNAs be classified as exonic (exon-derived) miRNAs or intronic (intron-derived) miRNAs.

A partial listing of protein-coding miRNA host genes is shown in Table 2. These miRNA host genes encode proteins with a broad spectrum of biological roles ranging from embryonic development, to the cell cycle, and physiology (Supplemental Tables A,D). To derive a perspective on the classes of protein-coding genes possibly utilized by miRNAs for their transcription, we surveyed the Gene Ontology (GO) classifications of all miRNA host genes (Gene Ontology Consortium 2001; http://www.geneontology.org/). The two most commonly identified `biological process' classifications for GO annotated miRNA gene hosts are `metabolism' (GO:0008152; 19 of 90; 21%) followed by `cellular physiological process' proteins (GO:00050875; 14 of 90; 16%) and the most common GO `molecular function are `purine nucleotide binding' (GO:0017076; 8 of 90; 9%) and `DNA binding' (GO:0003677; 7 of 90; 8%). Comparison of the most commonly identified GO classifications for miRNA host genes and the entire collection of GO annotated mammalian genes does not reveal disproportionate representation of these classes of genes (data not shown). We also note that several genes involved in human disease are hosts to miRNAs. For instance, nonsense, splicing, frameshift, or deletion mutations in the Chloride Channel Protein 5 (CLCN5) gene cause Dent disease and nephrolithiasis, an X-linked recessive disorder in human patients (OMIM: 300008; Yamamoto et al. 2000). A causative role for the CLCN5-encoded miR-188 has not been explored in this disease.

A significant number of intron-based miRNAs reside within related families of protein-coding genes (Table 3, Supplemental Tables A,D,E). This allowed us to group specific sets of miRNAs into novel families based on host gene identity. In some cases, host gene families aid in the identification of related sets of intronic miRNAs (Table 3). For instance, miR-148b and miR-152 are sequence-related and located in two genes encoding subunits of the coatomer transporter (see Table 3, Supplemental Table E). In other examples, we note sequence-unrelated miRNAs that are found within related families of proteins. For example, miR-105 and miR-224 are located in two genes encoding subunits of the ligand-gated ion channel γ-aminobutyric-acid receptor-α-3 and -ε respectively, although the mature miRNAs are unrelated by sequence (Table 3).

In addition to the miRNAs in protein coding genes, we also noted 66 miRNAs within TUs that lack a significant protein-coding potential (Table 1). More detailed analysis of these miRNA host TUs revealed a number of previously classified long ncRNAs (see Tables 4,5 and Supplemental Tables B-D). These include: BIC (Tam 2001), Deleted in Leukemia 2 (DLEU2; Migliazza et al. 2001), Noncoding RNA in Rhabdomyosarcoma (NCRMS; Chan et al. 2002), Synapse-Specific-7H4 (7H4; Velleca et al. 1994), FLJ34037 (Ota et al. 2004), and numerous FANTOM2 long ncRNAs (Okazaki et al. 2002; Numata et al. 2003). Only the BIC RNA has previously been noted as a carrier for a miRNA (Lagos-Quintana et al. 2002). These types of ncRNA transcripts are sometimes referred to as mRNA-like ncRNAs (mlncRNAs; Erdmann et al. 2000) because they share properties with mRNAs such as splicing (e.g., BIC: Tam 2001), polyadenylation (e.g., Air: Sleutels et al. 2002), long transcription units (e.g., Xist: Hong et al. 1999), and possibly also spatio/temporally restricted expression (e.g., Ntab: French et al. 2001). The molecular/genetic function of most noncoding mRNA-like RNAs is unclear but some, such as Xist and Air, have been studied in greater detail.

As well as the previously defined ncRNA TUs noted above, we identified dozens of novel TU hosts for miRs with little protein coding potential (see Methods, Tables 4,5, and Supplemental Tables B-D). For example, EST and cDNA data in ENSEMBL allowed the identification of an mlncRNA TU host containing miR-100, let-7a-2, and also miR-125b-1 in an intron (Table 4, Fig. 1A). Sequence alignment of the human/mouse host transcript 1 (ENSESTT00000036713/ENSMUSESTT00000030830; data not shown); and transcript 2 (ENSESTT00000036715/2610203C20Rik) does not reveal significant homology (Fig. 1B). Nevertheless, the miR-100, let-7a-2, and miR-125b-1 transcription locus is remarkably similar in human and mouse in both arrangement and structure (Fig. 1A). The possibility exists that some miRNA hosts of this type may be so-called `inside out' genes analogous to some small nucleolar RNA (snoRNA) hosts (Tycowski et al. 1996), in which transcription of the host gene serves purely to yield the intron-encoded RNAs. It is also plausible that some mlncRNA hosts containing miRNAs in their introns serve as functional RNAs.

A significant number of miRNAs overlap exons of novel noncoding transcripts (Tables 1,5; Supplemental Tables C,D). For example, the miR-22-predicted hairpin precursor is contained entirely within exon 2 of a noncoding TU, and the splicing pattern is generally conserved in human and mouse, despite the lack of protein-coding potential (Fig. 2A,B). Significant sequence conservation between the mouse and human miR-22 TU is observed within the 5′ end and within the miR-22-predicted hairpin precursor (Fig. 2B). Such high overall conservation in the mammalian miR-22/TU may indicate an important role in biogenesis or localization. Although high sequence conservation within the predicted hairpin precursors of mammalian miRNAs is a common feature (Lim et al. 2003), it is unclear whether the extended sequence similarity observed in miR-22/TU is a more common phenomenon for the exonic class. An alignment of the mouse and human miR-155/BIC does not show such extended conservation beyond the predicted hairpin precursor (data not shown). In addition to miR-22, a further 29 other miRNAs are contained within exons of transcripts (Table 5). In the majority of cases, exonic miRNAs were identified within mlncRNA transcripts, but in two rare cases we noted that the miRNA overlapped with the 3′ UTR of a protein-coding transcript (Supplemental Table D).

The prevalence of miRNAs within introns or exons of TUs prompted us to ask whether it might be possible to derive reliable miRNA expression patterns from host gene expression data. By analogy to mammalian snoRNAs, intronic miRNAs may be cotranscribed with the host gene by inclusion in introns of their pre-mRNAs and excision from debranched introns by an exonuclease (Tycowski et al. 1993). Similarly to yeast snoRNAs, mammalian miRNAs of exonic transcription origin could be derived from ncRNA transcripts by endonucleolytic cleavage within duplex regions (Chanfreau et al. 1998). If parallel transcription of host genes and miRNAs takes place, similar expression patterns for host transcript and miRNAs would be observed. To test this possibility, we analyzed the expression of five predicted miRNA host transcripts by RT-PCR from several organs in adult mice and compared their expression pattern to that of their corresponding exonic or intronic miRNA (see Methods and Table 6). In four out of five cases, we note an exact correlation between the host gene/transcript expression (Table 6) and the previously reported miRNA expression in mouse adult organs (Sempere et al. 2004). This result suggests that miRNA expression may be reliably derived from host transcript expression information, and strongly suggests that overlapping transcripts in the genome are likely hosts for miRNAs.

The above result impelled us to derive likely miRNA expression pattern information from all host genes identified here. We curated host gene expression data from a number of publicly available sources including primary literature and from public gene expression pattern databases (see Methods). This analysis allowed the identification of physiologically regulated, signal transduction, and tissue/organ specifically transcribed miRNA hosts (see Supplemental Tables A-D for a full description of expression patterns of miRNA hosts). For example, miR-33 is in an intron of SREBP2 (Table 2), an important gene in cholesterol homeostasis that is tightly regulated by sterols at the transcriptional level (Sato et al. 1996). In addition, we identified a cluster of three intronic miRNAs (miR-25, -93, and -106b) that are likely linked to the transcription of DNA Replication Licensing Factor MCM7 (Table 2), a gene that is quiescent in nondividing cells and is upregulated just prior to S phase of the cell cycle (Fujita et al. 1996). Interestingly, human miR-7-3 is contained within intron 2 of Pituitary Gland Specific Factor 1 (Table 2), a gene that is expressed in the pituitary gland (Tanaka et al. 2002). This mature miRNA has two other identical copies in the genome, one of which (miR-7-1) is possibly linked to the expression of a ubiquitously expressed gene, HNRPK (Supplemental Table E). This illustrates that expression pattern information for identical miRNAs may be confounded by distinct but overlapping patterns of transcription. In such cases where multiple paralogous miRNAs exist in the genome, it may prove useful to distinguish the expression of individual miR copies by analyzing the expression of their respective host genes (see Supplemental Table E for a list of similar/paralogous microRNAs contained in different host genes).

The current dogma in the miRNA field is that compartmentalized processing of all miRNAs to mature miRs occurs with Drosha in the nucleus and Dicer in the cytoplasm (for review, see Bartel 2004; He and Hannon 2004). In two seminal studies, it was demonstrated that miRNAs are dependent on Drosha for processing in HeLa cells (Lee et al. 2003) and that enrichment of a particular pri-/pre-miRNA occurs in the nucleus while the mature miR is found in the cytoplasm (Lee et al. 2002). However, we note that the miRNAs tested in these studies are intronic class miRNAs (i.e., miRs-15a, -16-1, -20, -30a; Supplemental Tables A,B), a `mixed' miRNA (miR-21) (Supplemental Table D), and several undefined transcription class miRNAs (data not shown). In lieu of the findings in this study of two types of transcription miRNA classes in mammals, it will be very interesting to revisit these experiments in the context of experimentally verified exonic transcription-class miRNAs. It seems likely that commonalities in processing and maturation by Drosha and Dicer occur for all transcription classes of miRNAs; however, the existence of exonic miRNAs in mammals implies that differences of biogenesis may also take place. In this regard, it is noteworthy that we found miR-206 within an exon of the synapse-associated mlncRNA, 7H4 (Table 4, Supplemental Table C; Velleca et al. 1994; Numata et al. 2003). 7H4 was originally identified by Velleca et al. (1994) as an RNA selectively found in the motor endplate of the rat skeletal neuromuscular junction. Detection of 7H4 at the neuromuscular junction raises the intriguing possibility that the unprocessed form of miR-206 may be transported outside of the nucleus as a pri-mir-206/7H4 RNA. Future studies will be required to determine whether some miRNAs can be actively transported to subcellular compartments outside of the nucleus embedded within their host transcripts.

METHODS