The proto-oncogene c-Myc is a helix-loop-helix leucine zipper transcription factor which regulates ~10 to 15% of human genes to control cell growth and apoptosis [63]. Deregulation of c-Myc is a hallmark of many cancers and is associated with both activation and repression of transcription of protein coding genes [64]. It is now clear that c-Myc also modulates transcription of miRNA genes [65,66]. c-Myc binds to E-boxes in the promoter and activates transcription of the miR-17-92 cluster [65]. Consistent with the prominence of c-Myc activation in tumors, miR-17-92 is often highly expressed in tumors. The miR-17-92 cluster contains 6 miRNAs, including miR-17-5p and miR-20, which are known to repress the cell cycle regulator E2F1 [65]. Interestingly, c-Myc also promotes E2F1 transcription; suggesting that c-Myc is able to fine tune cell cycle progression via the regulation of both miRNA and mRNA expression (Fig. 2C) [67]. In addition to activating transcription, c-Myc also decreases the expression of several tumor suppressor miRNA genes, including the miR-15a, -29, -34 and let-7 families [66]. Exogenous expression of c-Myc-repressed miRNAs in lymphoma cells reduced cell growth, indicating that down regulation of a subset of miRNAs is an important mechanism of c-Myc-mediated tumorigenesis [66]. ChIP analysis reveals that this repression is mediated, at least in part, through direct binding of c-Myc to miRNA promoters [66]. Interestingly, c-Myc has also been shown to increase transcription of Lin-28B which in turn mediates the post-transcriptional repression of let-7 family members (Fig. 3 and discussion below) [68]. Consistent with the repression of let-7 by c-Myc, increased expression of c-Myc in Burkitt lymphoma cells is associated with decreased let-7 expression [69]. c-myc is also targeted by let-7, suggesting a double negative feed-back loop in which c-Myc represses let-7 transcription and let-7 in turn represses c-Myc (Fig. 3) [69]. Together these studies provide just one example of the dynamically regulated transcriptional control of miRNAs and are illustrative of the biological outcomes which are mediated by altered transcription of miRNA genes. c-Myc is not alone in this arena: the tumor suppressor p53 activates transcription of the miR-34 family, which in turn promotes cell cycle arrest and apoptosis [70,71].

The tissue-specific expression of protein coding genes is controlled by complex transcription factor networks which orchestrate the differentiation and maintenance of cell fate in specialized tissues. Consistently, the transcription of tissue-specific miRNAs is often modulated by the same master regulatory factors which regulate mRNA. For example, skeletal and cardiac muscle differentiation is characterized by the transcriptional activation of muscle specific genes through the cooperation of serum response factor (SRF) and myocyte enhancer factor 2 (MEF2) with muscle-specific basic-helix-loop-helix transcription factors, including Myf-5, MyoD, and Myogenin [72]. SRF and MEF2 have recently been described to also promote the transcription of several myocyte-specific miRNAs, including miR-1, -133, -206, and -208 (Fig. 2C) [54]. Similar to myogenesis, the regulation of haematopoetic cell differentiation is controlled by combinations of lineage-specific transcription factors. miR-223 is highly expressed in the myeloid cell lineage and is activated by the myeloid transcription factors, C/EBP and PU.1 [73]. The coordinated expression of tissue specific miRNAs with mRNAs may serve to enforce cell fate switching during development. Negative regulation of miRNA transcription has also evolved to prevent misexpression of cell type-specific miRNAs. For example, expression of miR-124, which is abundant in brain, is negatively regulated by the binding of the RE1 silencing transcription (REST) factor to the promoter (Fig. 2B) [74]. REST and its cofactor Small C-terminal domain phosphatase 1 (SCP1) are highly expressed in neuronal precursors and non-neuronal genes. The REST complex silences transcription of neuronal genes through the recruitment of histone deacetylases (HDACs) and the methyl CpG binding protein MeCP2 [75]. Therefore, the REST complex prevents the misexpression of neuronal genes, including miR-124, outside of the nervous system. The interplay between the REST complex and miR-124 sets up a feedback loop which stabilizes and enforces the expression of neuronal vs. non-neuronal genes (Fig. 2B) [74,76]. Given that exogenous expression of miR-124 in HeLa cells down-regulates hundreds of non-neuronal targets, the tight control of miR-124 expression is critical for the diversification of neuronal from non-neuronal cells [77].

Bioinformatic analyses indicate that miRNAs may have a general propensity to regulate transcription factors [78]. Interestingly, an emerging theme in tissue-specific miRNA regulation is that miRNAs are often involved in complex regulatory networks with the transcription factors that drive their expression. Autoregulation of miRNA expression is found when the transcription factor which regulates miRNA expression is targeted by the miRNA itself. Examples of this type of regulation include Runx1-miR-27a in megakaryopoiesis [79], Pitx3-miR-133b in midbrain dopaminergic neurons [80] and c-Myb-miR-15a in hematopoiesis (Fig. 2C) [81]. This type of regulation allows tight control of miRNA and transcription factor levels. Feed-forward circuits have also been identified. In the heart, HDAC4 represses the MEF2/MyoD regulated transcription of myogenic factors. miR-1 is induced by MEF2/MyoD and silences HDAC4, the resulting decrease in HDAC4 allows increased activity of MEF2/MyoD and thus mediates further expression of miR-1 loci (Fig. 2C) [82]. It is speculated that feed-forward circuits ensure sustained, robust expression of miRNAs and may play a role in promoting tissue-specific miRNA expression.

In addition to the stable changes in miRNA transcription modulated by developmental programs, miRNA can also be rapidly, and in some cases, transiently induced by growth factor signaling. Vascular smooth muscle cells (VSMC) retain the unique ability to undergo phenotypic switching between a differentiated 'contractile' state and a de-differentiated 'synthetic' state. This process is critically important for the repair of vessels following injury, but deregulation of phenotype switching contributes to vascular disorders, including hypertension, balloon-angioplasty-induced restenosis, and atherosclerosis. A critical player in the dedifferentiation of VSMC are the platelet derived growth factors (PDGFs), which promote reduced contractile gene expression, increased migration and increased proliferation of smooth muscle cells [83]. PDGF signaling results in a rapid, transcriptional induction of miR-221 which is required for multiple facets of the synthetic VSM response [58]. Interestingly, miR-222 which is a part of the miR-221 gene cluster, is not co-regulated by PDGF [58]. Induction of miR-221 promotes increased proliferation and migration as well as decreased contractile gene expression through distinct targets, indicating that a single miRNA can extrapolate an extracellular input to mediate multiple biological outcomes [58]. Other examples of growth factor-mediated activation of miRNA genes include miR-132 which is induced in neurons by brain derived neurotrophic factor (BDNF) and miR-192 which is induced in the diabetic kidney by transforming growth factor-β (TGF-β) [84,85]. Rapid induction of miRNAs in response to growth factors may serve to promote a switch in gene expression and are likely to be important for cellular responses upon cytokine stimulation.

The DEAD-box RNA helicases p68 (DDX5) and p72 (DDX17) were identified as components of the large Drosha processing complex by immunoprecipitation-mass spectrometry analysis and subsequently shown to also associate with DGCR8 [41,102]. Knockout of either p68 or p72 in mice is lethal (E11.5 or P2, respectively) and double knockout animals show earlier lethality without obvious specific degeneration of organogenesis [103]. Analysis of mature miRNA levels by microarray indicated reduced steady state levels of ~35% (94 out of 266 surveyed) miRNA in p72 null MEFs vs. control. Most of the miRNAs which were reduced by p72 knockout were similarly reduced by p68 knockout, and combined reduction of p68 and p72 did not further reduce mature miRNA levels; suggesting that the two proteins are largely redundant or function as a heterodimer. Interestingly, while the level of pre-miRNA was also reduced in p72 or p68 knockout cells, the level of pri-miRNA was unchanged. Furthermore, in vitro processing assays showed attenuation of Drosha processing from extracts depleted of p68/p72. In vivo RNA-immunoprecipitation (RNA-IP) studies indicated reduced Drosha binding to pri-miR-199a in the absence of p68/p72. Together, these results strongly support a role for p68/p72 in promoting the Drosha processing of a sub-set of miRNAs [103]. The precise mechanism of p68/p72-mediated processing is unclear, but may involve rearrangement of the RNA hairpin which results in enhanced Drosha recruitment or stability. Alternatively, as p68/p72 are known to interact with a variety of proteins, including RNA processing enzymes and transcription factors, they may serve as a scaffold for the recruitment of multiple protein factors to the Drosha microprocessor [104,105].

The role of p68/p72 in miRNA processing is further supported by the positive regulation of Drosha processing mediated by the p68-interacting Smad proteins. The Smads are the signal transducers of the Transforming Growth Factor-β (TGF-β) family signaling cascade. In the canonical pathway, ligand binding to the type I and type II TGF-β receptors promotes the nuclear accumulation of receptor-Smads (R-Smads) in association with the common-Smad (co-Smad), Smad4 (Fig. 4). The complex of R-Smad and co-Smad bind to sequence elements within the promoter of target genes to positively or negatively regulate gene transcription. TGF-β and its family member, Bone Morphogenetic Protein 4 (BMP4) are particularly important for the differentiation of VSMCs. Treatment with either BMP4 or TGF-β increases expression of contractile smooth muscle genes; this process is due, at least in part, to the miR-21-mediated repression of programmed cell death protein-4 (PDCD4). miR-21 is rapidly induced by BMP4 and TGF-β in VSMC which results in a subsequent decrease in PDCD4 and increased smooth muscle gene expression [106]. Interestingly, although knockdown of the R-Smads prevented upregulation of mature and pre-miR-21 in response to BMP4, no alteration in pri-miR-21 transcription could be detected. Furthermore, BMP4 could increase the expression of pre- and mature miR-21 driven by a heterologous promoter, suggesting that miR-21 is post-transcriptionally regulated at the Drosha processing step by the Smads. The identification of R-Smads as binding partners of p68 by yeast-two-hybrid suggested that R-Smads could be associated with the Drosha complex [107]. Consistently, co-immunoprecipitation and RNA-ChIP studies indicate Smad is present in a complex with Drosha and p68 on the pri-miR-21 hairpin following BMP4 or TGF-β stimulation (Fig. 4) [106]. An increase in Drosha binding to pri-miR-21 was also observed following ligand treatment, suggesting that Smads may promote the association of Drosha with miRNA hairpins. These results indicate that in addition to transcriptional regulation mediated by the Smads, TGF-β can regulate gene expression through miRNA processing (Fig. 4). A distinguishing feature of the two pathways is the requirement for the co-Smad4. While Smad4 is required for transcriptional responses to TGF-β, the association of R-Smads with the Drosha processing machinery does not require Smad4. Knockdown of Smad4 in VSMC did not affect induction of miR-21; furthermore, miR-21 is strongly induced by TGF-β in the Smad4-null MDA-MB-468 breast cancer cell line [106]. Several other miRNAs are post-transcriptionally induced by BMP and TGF-β, suggesting that rapid modulation of miRNA levels may play an important role in cellular response to cytokine signaling (B. Davis and A. Hata, unpublished observation) [106]. Intriguingly, knockdown of Smad nuclear interacting protein 1 (SNIP1) in HeLa cells reduces the expression of several mature miRNAs, including miR-21 [108]. Furthermore, SNIP1 is found in complex with RNA processing enzymes [109] as well as Drosha [108]. The Arabidopsis homologue of SNIP1, DAWDLE, is required for efficient pri- to pre-miRNA processing and is thought to promote the access or recognition of pri-miRNA by DCL1 [108]. These results suggest that Smads may be present in large, multi-protein complexes which regulate mature miRNA levels through alteration of miRNA processing in a context dependent manner.

The tumor suppressor protein p53 was recently identified to modulate miRNA processing through association with p68 and Drosha, similar to the Smads. Under conditions of DNA damage, several miRNAs, including miR-143 and miR-16, are post-transcriptionally up-regulated [110]. p53 is required for this process as p53-null HCT116 cells do not increase miRNA in response to DNA damage. Co-immunoprecipitation studies indicate that p53 is present in a complex with both Drosha and p68, and addition of p53 to in vitro processing assays could enhance Drosha processing [110]. Interestingly, several p53 mutants that have been previously linked to oncogenic progression suppressed miRNA expression [110]. These results indicate that the association of p68/Drosha with accessory factors, such as p53 or Smads, may be particularly important for the rapid induction of miRNAs in response to extracellular stimuli. It intriguing to speculate that other p68-interacting proteins, such as estrogen receptors, MyoD, and Runx2 [111-113], might also have a role in the regulation of miRNA processing.

An additional example of miRNA specific processing regulation is mediated by hnRNP A1 which is required for the processing of a member of the miR-17-92 cluster, miR-18a. Addition of hnRNP A1 to in vitro processing assays dramatically increases the conversion of pri-miR-18a to pre-miR-18a. Conversely, depletion of hnRNP A1 lowers pre-miR-18a levels, but does not alter the expression of the other 5 clustered miRNAs [60]. Further studies indicate that direct binding of hnRNP A1 to the loop region of miRNA hairpins causes the structural rearrangement of the hairpin to generate a more favorable Drosha cleavage site. The addition of 2'O-methyl oligonucleotides ('looptomiRs') complementary to the loop region inhibits hnRNP A1 binding and prevents hnRNP A1-mediated Drosha processing. The loop region of miR-18a is well conserved throughout evolution, emphasizing the importance of the hnRNP A1-loop interaction for the processing of miRNA. Interestingly, approximately 14% of human miRNAs contain highly conserved loop regions, suggesting that processing regulation by hnRNPs may extend well beyond miR-18a. Consistently, the processing of two other miRNAs bearing conserved loops, miR-101 and let-7a-1 is also prevented by complementary looptomiRs (Fig. 3) [114]. The human genome encodes at least 19 hnRNPs, many of which have distinct RNA binding properties and tissue distribution [115,116]. Indeed, miR-101 and let-7a-1 were found to associate with hnRNP L and hnRNP I, respectively [114]. Although the functional requirement for the association of other hnRNPs is unknown, it suggests that the large family of hnRNPs could serve as accessory factors in the regulated processing of a variety of miRNAs. The differential expression of hnRNP proteins as well as their sub-cellular localization may serve to regulate miRNAs in a tissue or context-dependent manner. For example, phosphorylation of hnRNP A1 by p38 MAPK has been reported to promote the cytoplasmic localization of hnRNP A1 [117,118]. These results suggest that, similarly to smads, hnRNP A1 may control miRNA processing in response to extracellular stimuli.

The importance of the loop region in the regulation of miRNA processing is further supported by recent studies which indicate KSRP directly interacts with G-rich regions present within the loop of a sub-set of miRNAs to promote both Drosha and Dicer mediated miRNA processing [119]. Transient knockdown of KSRP in HeLa cells strongly reduced the expression of a group of mature miRNAs, including let-7a, and miR-206 (Fig. 3). Knockdown of KSRP altered cellular responses mediated by these miRNAs, including proliferation and skeletal muscle differentiation. KSRP was found to be present in both the Drosha and Dicer complexes, and knockdown of KSRP reduced the processing activity of Drosha and Dicer [119]. The interaction of KSRP with sequential miRNA processing steps may serve to promote the synergistic activation of processing for a subset of miRNA. Alternatively, KSRP association with either the Drosha or Dicer complexes may be sufficient to alter the miRNA processing in response to cellular stimuli. For example, lipopolysaccharide (LPS) treatment of macrophages dramatically increases mature miR-155, without significantly altering expression of the primary transcript. Mobility shift assays indicate KSRP interacts with pri-miR-155, and knockdown of KSRP prevents LPS-mediated increase of miR-155 [120]. Together these results suggest a role for KSRP in promoting the post-transcriptional regulation of miR-155 in response to LPS. Although the mechanism of this regulation is unknown, it is intriguing to speculate that the sub-cellular localization or modification of KSRP could be regulated by LPS. Indeed, phosphorylation of KSRP Serine 193 by Akt promotes nuclear accumulation of KSRP through interaction with 14-3-3ζ [121]. An important future challenge is the determination of the signaling pathways which modulate the activity of miRNA processing accessory factors such as KSRP and hnRNPs.

In addition to the positive regulation of miRNA processing discussed above, negative regulation of the Drosha processing of some miRNAs has also been reported. The let-7 family of miRNAs is often present as multiple copies in the genome; the human genome contains 9 mature let-7 sequences derived from 12 precursors [122]. While most let-7 primary transcripts are highly expressed throughout development, the mature let-7 is detectable only in highly differentiated cells, suggesting a mechanism of post-transcriptional regulation [122]. Analysis of the primary miRNA sequences of the human let-7 family as well as let-7 from divergent species indicated the presence of conserved nucleotide sequences present in the hairpin loop, suggesting a possible regulatory role for this region [123,124]. While pri-let-7 is processed efficiently in vitro using cell extracts derived from differentiated cells, processing is reduced in extracts derived from undifferentiated cells. The addition of short oligonucleotides corresponding to the loop region to in vitro processing reactions prevents inhibition of Drosha processing in cell extracts, presumably by competing for inhibitor binding [123]. Affinity purification and mass-spec identified Lin-28 as the let-7 interacting protein responsible for processing inhibition [123,125]. Furthermore, crosslinking, followed by electrophoretic mobility shift assay (EMSA) confirmed that the loop region interacts with Lin-28, and disruption of the Lin-28 binding site within the loop prevented Lin-28-mediated repression of let-7 (Fig. 3) [123,124]. Further support for the physiological relevance of let-7 regulation by Lin-28 comes from the observation that Lin-28 and mature let-7 show reciprocal expression patterns during development.

The precise mechanism of Lin-28 inhibition of Drosha processing is unclear. Lin-28 has a strong affinity for let-7, perhaps suggesting Lin-28 could prevent interaction with Drosha/DGCR8 by simple competition. Alternatively, analogous to hnRNP A1 discussed above, Lin-28 interaction with the loop region could rearrange the secondary structure of the hairpin and, in this case, discourage Drosha processing. It is interesting to note that Lin-28, KSRP, and hnRNPs all interact with conserved nucleotides present within the loop of pri-miRNAs. The balanced interaction of proteins with the miRNA loop may allow for the fine-tuning of miRNA levels during development or in response to extracellular cues. For example, although KSRP interacts strongly with pri-let-7g in differentiated cells, high levels of Lin28 block KSRP association in undifferentiated p19 cells [119].

Repression of Drosha processing of let-7 family members is also mediated by NF90 and NF45 proteins (Fig. 3) [126]. In vitro and in vivo processing assays indicate that the NF90/NF45 complex reduces pre-miRNA while increasing pri-miRNA levels. The NF90/NF45 family of proteins interacts with small double-stranded RNAs, including the adenovirus VA RNA [127] consistently, the NF90/NF45 complex binds pri-miRNAs with high affinity. EMSA analysis suggests that the high affinity association of NF90/NF45 complex with pri-miRNA precludes interaction with DGCR8, thus reducing production of pre- and mature miRNA [126]. Unlike Lin-28, which specifically regulates let-7, the NF90/NF45 complex may be more promiscuous, as the processing of 3 other analyzed miRNAs is also modulated.

Altered Drosha processing has been described for the tumor suppressor miR-7 in glioblast4oma [128]. While the primary miR-7 transcript is expressed at equal levels in normal and tumor samples, the precursor and mature miRNA is dramatically reduced in glioblastoma tumor samples and cell lines. Restoration of miR-7 levels to primary cell lines reduced the viability and invasiveness by targeting the EGF Receptor and inhibiting the Akt signaling cascade [128]. As the levels of several other miRNA were unaffected, the deregulation of processing seems specific to miR-7. Although the mechanism is unknown, reduced pre-miR-7 in the tumor could arise from either (i) increased expression of a processing inhibitor or (ii) decreased expression of a processing activator. It will be interesting to determine if miR-7 is regulated by one of the mechanisms described above, or if yet-unidentified mechanisms are responsible for the altered regulation of miR-7 in cancer cells.

Finally, it has been reported that the total levels of Drosha and DGCR8 in the cell are tightly controlled and may play a role in the regulation of miRNA processing. Increased Drosha expression was identified in late stage cervical cancer samples and was associated with poor prognosis of esophageal cancer patients [129,130]. Interestingly, despite a 2–7 fold increase in Drosha expression in cervical cancer samples, only miR-31 was increased, while the other differentially expressed miRNAs were decreased [130]. DGCR8 was originally identified as a gene located within a common chromosomal deletion at ch22q11, which gives rise to DiGeorge syndrome, including learning disabilities and heart defects [131]. Mouse models of this chromosomal deletion exhibit a moderate decrease of 59 mature miRNAs [132]. Given the critical role of Drosha and DGCR8 in mediating miRNA processing, it is surprising that relatively minor changes have been found in miRNA levels upon dramatic alteration of DGCR8/Drosha levels. These findings may be explained, at least in part, by the recent finding that the total levels of Drosha and DGCR8 are coupled in an intricate feed-back circuit. In addition to cleaving pre-miRNA hairpins, the microprocessor complex can also promote cleavage of hairpin structures within annotated protein coding genes which are not further processed by Dicer [133,134]. This type of cleavage allows Drosha to modulate protein coding gene expression independent of miRNA production. Indeed, tiling microarray analysis in drosophila S2 cells following Drosha depletion indicates that many annotated protein coding genes, including DGCR8, are targeted for degradation through this pathway [133]. Therefore, Drosha maintains a highly regulated level of DGCR8 through the microprocessor-mediated cleavage of DGCR8 mRNA. Additionally, DGCR8 stabilizes Drosha protein levels and ensures the tight coupling of the core microprocessor proteins. This regulatory loop seems to be highly effective in vivo as in the DGCR8 heterozygous knockout mouse the protein levels of DGCR8 are only modestly reduced despite a 50% decrease in genomic DGCR8 [134]. The presence of this mechanism suggests that the relative levels of DGCR8 and Drosha must be closely coupled for accurate miRNA processing. Interestingly, in vitro processing experiments using recombinant proteins indicate that as little as a 3-fold excess of DGCR8 over Drosha dramatically inhibits processing activity of Drosha [41]. In addition to alteration of protein levels, the overall activity of the Drosha/DGCR8 complex may be altered under some circumstances. Cells grown to high density exhibit increased pre- and mature miRNA expression without alteration of Drosha or DGCR8 protein level. Furthermore, extracts from high-density cells promoted the more efficient conversion of pri- to pre-miRNA in in vitro processing assays [135]. Although the mechanism of this increased processing is unknown, it is possible that post-transcriptional modifications or association with accessory factors could alter the activity of Drosha or DGCR8. For example, the binding of heme to DGCR8 promotes dimerization of DGCR8 and facilitates pri-miRNA processing [136].

Following translocation into the cytoplasm, the pre-miRNA is cleaved near the terminal loop by the RNase type III protein Dicer to generate a ~22-nt double-stranded miRNA [146,147] (Fig. 1). Like Drosha, Dicer generates 2-nt 3' overhangs upon cleavage. Dicer is highly conserved throughout evolution and present in nearly all eukaryotic organisms; the mammalian Dicer is composed of multiple domains including the DExD/H-box helicase, DUF283, dsRBD, PAZ and ribonuclease domains [148]. Structural analysis of the Giardia intestinalis Dicer, composed of the PAZ and ribonuclease domains, indicates that the PAZ domain binds the 3' overhangs at the end of double-strand RNA to position the RNA for cleavage by the two catalytic RNase III domains [149]. Interestingly, replacement of the PAZ domain with a U1A RNA binding domain resulted in altered end-recognition but maintained the length of generated miRNAs [149]. These findings indicate that Dicer serves as a molecular ruler to cleave miRNA at a specific distance from the ends generated by Drosha [150]. The Giardia Dicer likely represents a minimal functional Dicer protein and the additional domains present in mammalian Dicer may serve to allow more complex regulation of Dicer activity. Indeed, analysis of the processing ability of Dicer mutants indicates that the helicase domain may be especially critical for the processing of hairpins containing thermodynamically unstable regions [151].

Dicer is associated with the double-stranded RNA binding proteins TAR RNA binding protein (TRBP) and protein kinase R-activating protein (PACT) [152,153]. Depletion of either of these cofactors decreases the steady-state levels of Dicer protein [153,154]. Furthermore, truncation mutations of TRBP are found in carcinoma and associated with both decreased miRNA levels, as well as dramatic destabilization of Dicer [155]. Dicer protein levels can be rescued by wild type TRBP, suggesting that the interaction between Dicer and TRBP is a critical determinant of Dicer stability [155]. Additionally, the RNA binding protein fragile × mental retardation protein (FMRP) is associated with multiple components of the miRNA pathway, including Dicer [156]. Interestingly, the phosphorylation of FMRP prevents association with Dicer, but promotes interaction of FMRP with pre-miRNA [157]. Although the consequences of FMRP phosphorylation on microRNA processing are unclear, these results suggest a mechanism for the regulation of Dicer activity through the phosphorylation of associated proteins.

Although recombinant Dicer can catalyze the efficient and specific cleavage of RNA in vitro, these miRNAs are inefficiently incorporated into the RISC complex [45]. Furthermore, an exogenous pre-miRNA hairpin is more effective at promoting the silencing of targets than a duplex RNA, suggesting that the coupling of Dicer processing and RISC loading is critical for the proper function of miRNAs [45]. The primary component of the RISC complex, and the effectors of miRNA-mediated repression are the Argonaute (Ago) proteins. The human genome contains eight Ago-family proteins (Ago1-4 and Piwi1-4) [158]. Although structurally similar to Agos, the Piwi family members interact with piRNAs and function in germline maintenance; reviewed in [159]. While all of the Ago proteins have the ability to interact with small RNAs, Ago2 is the only one with RNA cleavage activity and is thought to play a prominent role in miRNA-mediated silencing.

In vivo, Ago2 associates with Dicer and TRBP/PACT to form the RISC Loading Complex (RLC) which allows the tight coupling of Dicer cleavage to the incorporation of miRNA into the RISC complex [153]. In vitro reconstituted RLC composed of recombinant Dicer, TRBP, and Ago2 efficiently catalyzes pre-miRNA cleavage in an ATP-independent manner [160,161]. Cleavage by Dicer results in an unstable double-stranded RNA composed of the active guide strand (miRNA) and the passenger (miRNA*) strand (Fig. 1). An important role of the RLC is the unwinding of the double-stranded miRNA, which is followed by incorporation of the guide strand into the miRNA-containing ribonucleoprotein (miRNP) complex and degradation of the passenger strand. The mechanism of strand selection in humans is unclear, however, the guide strand generally exhibits lower thermodynamic stability of the 5' end [162]. Deep sequencing of small RNAs indicates that in most cases, the ratio of miRNA:miRNA* strand is ~100:1, suggesting a possible mechanism of active degradation of miRNA* but not miRNA [163]. Although driven by thermodynamic properties, the assembly of RNA into the RISC complex may be subject to additional regulation as the ratio of miRNA:miRNA* can vary dramatically depending on the specific properties of the miRNA duplex. Furthermore, the ratios of miRNA:miRNA* are altered in different tissues and developmental stages [164,165]. These findings suggest that differential strand selection could represent a yet unappreciated mechanism of miRNA regulation. Recombinant RLC can mediate strand selection of pre-let-7a, indicating that this process is intrinsic to the Dicer-TRBP-Ago2 complex, but may be modified by additional factors in a context dependent manner [161].

The presence of a pre-formed Ago2-Dicer complex may be particularly important for miRNAs containing a high degree of complementarity along the hairpin. These miRNAs are cleaved by Ago2 in the middle of the presumptive passenger strand prior to Dicer processing [166]. This nicking may facilitate strand selection and dsRNA unwinding of some miRNAs. However, as Ago2 is the only Ago family member with detectable endonuclease activity, strand-selection must also occur without cleavage of the passenger strand [167,168]. Following the loading of Ago with the single-stranded guide miRNA, Dicer dissociates and the effector complex (miRNP) capable of selecting miRNA targets is formed through the association of additional proteins [160,169].

The relatively low levels of pre-miRNA relative to pri-miRNA or mature miRNA suggest that Dicer is generally very efficient at processing pre-miRNA [26,170]. In comparison to Drosha, which often requires the addition of accessory factors to promote processing, regulation of miRNAs at the Dicer step often involves inhibition of Dicer activity. Early evidence for post-transcriptional control of miRNA at the level of Dicer processing comes from studies of miRNA expression in colorectal neoplasia. Mature miR-143 and miR-145 are expressed at much lower levels in tumor samples than normal tissue. However, northern blots indicate that the level of pre-miR-143 or -145 is not significantly different in the tumor samples, suggesting altered dicer processing in the colorectal tumor samples [171]. Mature miR-143 and miR-145 are also decreased in several cancer cell lines as well as in B-cell malignancies [171,172]. These miRNAs act as tumor suppressors through the repression of the mitogen activated protein kinase ERK5 [173]. Although the mechanism of processing control is unknown, these results suggest that restoration of normal miRNA processing may represent a novel therapeutic option for the treatment of some tumors.

miRNA processing by Dicer is also differentially regulated during normal development and tissue specification. For example, while pre-miR-138 is ubiquitously expressed in all tissues and HeLa cells, the mature miR-138 is only detected in adult mouse brain and fetal liver [174]. An alteration in nuclear export can be ruled out because the precursor is effectively exported into the cytoplasm, suggesting regulation of Dicer processing. Furthermore, while pre-miR-138 is processed efficiently by Dicer in vitro, the addition of HeLa cell extract inhibits the processing of pre-miR-138. HeLa extract does not alter the processing of other miRNAs normally expressed in HeLa cells [174]. These results suggest that HeLa cells, and most tissues, express a yet-unknown factor which specifically inhibits Dicer activity on miR-138.

In addition to inhibition of the Drosha processing step discussed above, Lin-28 also inhibits the Dicer processing of let-7 family members (Fig. 3) [125,175]. Lin-28 shuttles between the nucleus and cytoplasm, but is commonly enriched within the cytoplasm, suggesting that the cytoplasm may be the primary location of the Lin-28:let-7 interaction. Addition of purified Lin-28 decreased the association of let-7 with Dicer, and reduced Dicer-mediated in vitro processing, suggesting that Lin-28 can compete with Dicer for access to let-7 [125]. An additional cytoplasmic blockade of let-7 processing by Lin-28 has recently been described by Heo and colleagues (Fig. 3) [176]. In hepatocellular carcinoma cell lines, Lin-28 promotes the 3'-uridylation of pre-let-7 which inhibits Dicer processing and promotes degradation of pre-let-7. This activity requires the zinc-finger domains of Lin-28, as mutations in this region maintain interaction with let-7 but do not induce uridylation [176]. The identity of the uridylating enzyme recruited by Lin-28 awaits further identification. Although Lin-28 promotes the uridylation of only let-7 family members, it will be interesting to determine if other pre-miRNAs are similarly regulated by the addition of oligonucleotides.

The total levels of Dicer may serve as an important control point in miRNA biogenesis. A careful analysis of the Dicer 5'UTR indentified multiple alternatively spliced leader exons that are expressed in a tissue specific manner [177]. Although differential splicing alters the translation efficiency of dicer in vitro, the biological relevance of alternatively spliced dicer isoforms is unknown. Altered expression of Dicer is observed in several types of human cancer, for example, Dicer is increased in prostate tumors, as well as in a Burkitt's lymphoma derived cell line [178-180]. Conversely, Dicer expression is decreased in non-small-cell-lung cancer, and reduced Dicer correlates with poorer prognosis [181]. These conflicting trends may be indicative of cell type differences or tumor stage. Analysis of precursor lesions of lung adenocarcinoma showed increased Dicer expression, while advanced invasive adenocarcinoma showed decreased protein levels [182]. The importance of controlled Dicer levels in normal physiology is further emphasized by the identification of let-7 target sites within dicer [8,183]. As the maturation of let-7 is dependent on Dicer, this interaction represents a negative feedback loop which may serve to regulate the total rate of miRNA biogenesis (Fig 3).

Dicer may also be regulated post-transcriptionally. The catalytic activity of Dicer processing of some miRNAs is increased by removal of the amino-terminal helicase domain, indicating that this domain may serve an autoinhibitory function [184]. In adult mouse neurons, the protease Calpain has been suggested to enhance Dicer activity, perhaps indicating that the proteolytic removal of the Dicer helicase domain could serve as a regulatory mechanism in vivo [185]. Finally, association with TRBP may activate Dicer by promoting the structural rearrangement of the Dicer helicase domain [184,186]. A more complete understanding of the regulation of Dicer activity is an important challenge for future studies.

The total levels of the Ago proteins within the cell also contribute to global miRNA regulation and biogenesis. Ectopic expression of Ago1-4 resulted in a dramatic increase of mature miRNAs [166]. In contrast, mouse embryonic fibroblasts and hematopoietic cells from Ago2 knockouts showed reduced levels of all mature miRNAs tested [166,187]. Some accumulation of pre-let-7 was observed in Ago2 knockout cells, suggesting that the presence of Ago2 in the RLC may enhance Dicer cleavage [166]. The dramatic increase in mature miRNA mediated by increased Ago expression could also indicate that Ago proteins are limiting and serve to stabilize miRNA. If this is the case, one might expect that Ago protein levels may be linked to the level of small RNA present within the cell. Indeed, a retrospective study of publically available microarray data from siRNA treated cells indicated an increase in Ago2 mRNA independent of the siRNA sequence [188]. Ago2 mRNA and protein expression was also found to be increased in a subset of breast cancer cells. Additionally, treatment with Epidermal Growth Factor (EGF) enhanced Ago2 stability, suggesting that alteration of Ago2 levels can occur at both the transcriptional and post-transcriptional level [189]. Recently, Ago proteins have been identified as substrates of the α-(P4H-α(I)) and β-(P4H-β) subunits of the type I collagen prolyl-4-hydroxylase (C-P4H) [190]. The hydroxylation of Ago2 at proline 700 enhanced Ago2 stability, while knockdown of the hydroxylating enzymes P4H-α(I) or P4H-β reduced Ago2 protein level [190]. The P4H-α(I) subunit is rate-limiting for the formation of active C-P4H and has been reported to be up-regulated by a variety of factors, including hypoxia and transforming growth factor-β. Given that these treatments each induce many miRNAs, one could hypothesize that concomitant up-regulation of Ago-2 by C-P4H hydroxylation may serve to stabilize the newly synthesized miRNAs.