Regulation of gene expression is a fundamental process in both prokaryotic and eukaryotic organisms. Multiple regulatory mechanisms are in place to control gene expression at the level of transcription, post-transcription and post-translation to maintain optimal RNA and protein expressions in cells. miRNAs (microRNAs) are abundant short 21–23 nt non-coding RNAs that are key regulators of virtually all eukaryotic biological processes. The levels of miRNAs in an organism are crucial for proper development and sustaining optimal cell functions. Therefore the processing and regulation of the processing of these miRNAs are critical. In the present chapter we highlight the most important steps of miRNA processing, describe the functions of key proteins involved in the maturation of miRNAs, and discuss how the generation and the stability of miRNAs are regulated.
Historically we had come to understand that gene expression was controlled by transcription factors and other protein co-factors. However, in the last decade it has become more apparent that gene regulation is highly complex due to the discovery of ncRNAs (non-coding RNAs) as key factors in regulating gene expression (Chapter 1). Approximately 15 years ago a new pathway was starting to be elucidated which underpins gene expression in animals and plants. This regulatory pathway is mediated by small ncRNAs, known as miRNAs (microRNAs; also abbreviated to miRs). The first descriptions of these miRNAs were termed stRNAs (small temporal RNAs) and were implicated in Caenorhabditis elegans development . In subsequent studies these small RNAs were shown to be common and abundant in plants and animals [2–4], but do not exist in the bacterial kingdom. They are now proven to be the key regulators of gene expression in virtually all biological processes.
The current model for the biogenesis of miRNAs involves several distinct steps. In the nucleus, there is the initial transcription of the miRNA loci and then processing by the RNase III enzyme Drosha. This cleaved product is then transported into the cytoplasm for a second processing event mediated by another RNase III enzyme known as Dicer. The end result of this enzymatic processing is the production of a 21–22 nt dsRNA (double-stranded RNA) species which represents the mature miRNA. The mature miRNA is then loaded on to the Argonaute family of proteins to form the RISC (RNA-induced silencing complex) which is the effector for mediating miRNA activity.
In the present chapter we describe the key steps of miRNA processing and formation of the regulatory competent complex, and also summarize the regulation of many of these steps. We discuss the studies carried out in animal systems and describe the crucial differences between the plant and the animal miRNA pathways.
Transcription and processing of pri-miRNAs (primary miRNAs) in the nucleus
The majority of miRNAs are transcribed by RNA Pol II (RNA polymerase II) which generates a typical RNA transcript containing a 5′ cap and a polyA tail . However, in special cases, RNA Pol III has also been shown to transcribe miRNAs from viral genomes .These RNA polymerases make longer transcripts known as the pri-miRNAs. miRNAs are embedded in the hairpin-structured motifs of these pri-miRNAs. miRNAs are encoded by ncRNAs or introns of protein-coding transcripts. In the initial nuclear processing step for animals, these stem-loops are recognized by the microprocessor complex  consisting of Drosha (an RNase III enzyme) and DGCR8 (Di George Syndrome critical region gene 8) . DGCR8 binds to the stem-loop base and then positions Drosha which cleaves the double-stranded stem-loop approximately 11 bp from a less-structured single-stranded base to generate a 100–70 nt hairpin structure known as a pre-miRNA (precursor miRNA) . Plants, however, do not have a Drosha homologue but they have multiple Dicer genes (see below) and one of them generates pre-miRNAs . One of the unique features of Drosha processing is that it generates RNA ends containing a 5′ phosphate group and a 2 nt 3′ overhang, which are characteristic of RNase III activity and are essential for the recognition of the pre-miRNA by further processing enzymes. Apart from its major role in miRNA biogenesis Drosha also recognizes pri-miRNA-like hairpins in protein-coding genes such as Brd2, Mbnl1, Wipi2 and DCGRG, cleaving them at these structures to regulate their expression [11,12]. Other factors known to associate with the microprocessor are the DEAD box RNA helicases p72 and p68, which are needed to process specific subsets of pri-miRNAs .
Regulation of pri-miRNA processing
Since miRNAs are key regulators of developmental processes, the cell cycle and immune responses, their processing is also tightly regulated at the different steps of biogenesis. There are an increasing number of mechanisms and proteins that are implicated in the regulation of processing for the pri-miRNAs. First, Drosha and DGCR8 can also regulate each other post-transcriptionally via cleaving a hairpin sequence found on the DGCR8 transcript, thereby controlling the activity of the microprocessor . It has been shown that transcription factors such as the Smads can bind to the microprocessor via the p68 subunit and enhance Drosha cleavage . In a similar manner the p53 tumour suppressor protein binds to p68 and increases processing for specific pri-miRNAs such as miR-145 . A similar mechanism has been described with BRCA1, another tumour suppressor protein, which facilitates the processing of a subset of pri-miRNAs by directly binding to the pri-mRNAs Drosha and p68. It also interacts with p53 and Smad3, two proteins that regulate signalling and miRNA processing . Hormone-mediated regulation can also occur via the oestrogen receptor which associates with the p68 and p72 helicases . It may be that these RNA helicases act as the scaffold for recruiting factors to enhance the processing of specific pri-miRNAs.
RNA-binding proteins such as KHSRP (KH-type splicing regulatory protein) can also regulate pri-miRNA processing of a subset of miRNAs . It binds with a high affinity to single-stranded GGG-triplet motifs within the terminal loop and promotes microprocessor cleavage of, for instance, the let-7a primary transcript. The stimulatory activity of KHSRP can be abrogated by another RNA-binding protein, hnRNPA1 (heterogeneous nuclear ribonucleoprotein A1), which competes for the same binding site at the terminal loop . hnRNPA1 has a dual function since it can also promote pri-miR-18 processing in the miR-17-92 cluster . Lastly, ADARs (adenosine deaminases that act on RNA) are RNA-editing enzymes that deaminate adenosines to create inosines in dsRNA structures. These ADARs can edit a pri-miRNA transcript sequence which then can prevent Drosha processing . These examples clearly suggest a tightly controlled mechanism and also indicate that certain subsets of miRNA families are only processed by binding of specific co-factors to the microprocessor unit. After nuclear processing the next phase occurs in the cytoplasm, with the pre-miRNA being exported in a Ran-GTP-dependent manner through the nuclear export receptor exportin-5 . This transport is specific as exportin-5 recognizes the 3′ overhang and the duplex nature of the pre-miRNA .
Pre-miRNA processing in the cytoplasm
In the cytosol, the pre-miRNA is recognized by the second RNase III enzyme Dicer  and its partner, the RNA-binding protein [TRBP (trans-activation response RNA-binding protein) and PACT in humans/R2D2 and Loquacious in flies, for instance] [24–27]. Dicers in general contain an N-terminal helicase domain, a PAZ domain, two RNase III domains and a C-terminal dsRNA-binding domain. Dicer binds the 3′ overhang and also the 5′ end to correctly position the pre-miRNA for cleavage by its PAZ domain and cleave each of the strands with 2 nt offset with the two RNase III catalytic centres. This process liberates an intermediate product which is a 21–23 nt dsRNA which contains 3′ overhangs at both ends . Recent biochemical and structural studies also suggest that in higher eukaryotes the N-terminal helicase domain is responsible for binding the pre–miRNA loop .
Regulation of pre-miRNA processing
Dicer processing of certain pre-miRNAs has been shown to be influenced by the developmentally regulated protein lin-28 in C. elegans and mammals [30,31]. This protein has a high specificity for the terminal loop of the let-7 miRNA family and, when bound to the precursor, recruits TUTases (terminal uridyl transferases). These enzymes mediate uridylation of precursor let-7 at its 3′-end which then prevents Dicer processing and promotes degradation of the pre-miRNA . Lin-28 can also be translocated to the nucleus and binds the terminal loop of let-7 primary transcripts to block processing by either physically impeding Drosha binding or sequestering the primary structure into nucleoli . In a twist, Dicer is also regulated by let-7 and therefore there is a negative-feedback mechanism which regulates the stoichiometry of Dicer and its product . Furthermore, MCPIP1 (monocyte chemoattractant protein-1-induced protein 1) is an endo-RNase that regulates miRNA levels by cleaving the loop sequences of a group of miRNAs. This regulation is likely to be important in cancer and immune responses .
Loading and activation of miRNAs
Regulatory competent miRNAs are incorporated into members of the Argonaute multiprotein family and form the RISC or miRISC (miRNA-induced silencing complex). Argonaute proteins are structural homologues of RNase H enzymes  and characteristically contain an N-terminal domain that is involved in the unwinding of double-stranded small RNAs , a PAZ domain (similar to the Dicer PAZ domain with similar function), an MID domain that orientates the miRNA by accommodating the 5′ phosphate of the miRNA and a PIWI motif which is the catalytic domain . All plant Argonautes are cleavage-competent ; however, many animal Argonautes have lost this activity and are now involved in cleavage-independent gene regulatory mechanism(s) (Chapter 3). All regulatory small RNAs, including miRNAs, are single-stranded when they regulate gene expression, but are generally loaded on to Argonaute proteins as double-stranded molecules; therefore a mechanism must exist to unwind the double-stranded Dicer product and eliminate one of the two strands. The strand selection of regulatory miRNAs is not random. One strand of the miRNA (the guide or trigger strand) is preferentially loaded on to an Argonaute protein compared with the other strand [the (*) star sequence or passenger strand] of the same miRNA duplex. The strand whose 5′-end is located at the thermodynamically weaker end of the duplex will be the guide strand and it will recognize and bind to cognate RNA targets [40,41].
The RLC (RISC loading complex) and miRLC (miRNA loading complex), in general, contain Dicer, RNA-binding proteins (R2D2, Loquacious in flies and TRBP in mammals) and an Argonaute protein . In the case of siRNAs (small interfering RNAs; similar regulatory RNAs to miRNAs, however they are perfectly double-stranded and processed from mainly long dsRNAs), Dicer and the RNA-binding protein recognize the asymmetry of the siRNA and load them on to a specific Argonaute . However, this sensing has not been verified in the case of miRNAs. For instance, the existence of miRLC has been demonstrated in mammals , but it was also shown that Dicer is not required for the recognition of the asymmetry of the miRNA .
Recombinant Argonaute can only bind single-stranded RNAs efficiently . The Hsp (heat-shock protein) 70/90 chaperone machinery facilitates the transfer of the Dicer-cleaved double-stranded product to the Argonaute protein by helping to form a conformation that is capable of accommodating double-stranded small RNAs . The unwinding of double-stranded miRNAs is an energy-dependent step whereby the two strands of the duplex are wedged open by the N-terminal domain of the Argonautes. This duplex is then unwound to remove the passenger strand . The single-stranded guide-RNA-activated Argonautes are now primed and ready to engage in diverse gene regulatory functions. A summary of these steps is shown in Figure 1.
Canonical processing of miRNAs and the regulation of miRNA processing in animal cells
Regulation of mature miRNA stability
In general, miRNAs and siRNAs are very stable in the cellular environment with a half-life of several days. However, several miRNAs in specific cells and/or in a defined stage of the cell cycle can show accelerated turnover. Sequence requirements for rapid degradation have been mapped in some miRNAs. Also, RNA-modifying enzymes that tag miRNAs for degradation and nucleases that turn over mature miRNAs have been identified and characterized in plants and animals. The turnover rate of miRNAs can also be regulated by the target mRNA. This regulation depends on the level of the complementarity of an miRNA and its target; more extensive base pairing leads to the degradation of the miRNA through a mechanism called tailing and trimming (mechanisms involved in miRNA turnover are reviewed in detail by Ruegger and Grosshans ).
Alternative pathways for miRNA biogenesis
In both C. elegans and Drosophila melanogaster, there are miRNAs embedded in introns of mRNA-encoding genes which are generated via the action of RNA splicing and lariat-debranching enzymes to produce pre-miRNA-like hairpins known as mirtrons (Figure 2) . These RNA hairpin structures are not processed by Drosha, but are structural mimics resembling pre-miRNAs. These mirtrons then enter the canonical miRNA pathway during nuclear export and are then cleaved by Dicer to generate a mature miRNA. In mammalian cells, mirtrons were found to be well-conserved and expressed in diverse mammals . The discovery of a Drosha-independent processing pathway suggests that mirtrons were the first set of miRNA genes to have evolved for mammalian gene regulation. Remarkably, there are even some mirtrons such as miR-1225 and miR-1228 which are produced in the absence of splicing. These are termed simtrons and they do not require DGCR8, Dicer, exportin-5 or Ago2 for maturation .
Alternative miRNA processing pathways in animal cells
Another miRNA which does not require Dicer processing is miR-451. The primary transcript of miR-451 is cleaved via the microprocessor to liberate a precursor which is too small to act as a Dicer substrate, but instead is loaded into Ago2 and is cleaved by the Ago2 catalytic centre . In this instance it is the pre–miRNA secondary structure which determines the processing pathway. Recent studies have indicated that it is a combination of the terminal loop, but more importantly, the nucleotide identity at the 5′-end which sorts the miR-451 precursor on to Ago2 for miRNA maturation .
sno-miRNAs [snoRNA (small nucleolar RNA)-derived miRNAs]
snoRNAs are an abundant class of ncRNAs mostly found in introns with diverse cellular functions. These small RNAs are transcribed by RNA Pol II and processed by exonucleolytic trimming after splicing. Typically, they guide RNA modifications and are found in nucleoli and Cajal bodies in most eukaryotic cells. The use of deep sequencing discovered that several miRNAs, such as small RNAs, were aligned to loci regions encoding the C/D and H/ACA box snoRNAs. However, so far, only one snoRNA-derived small RNA has been shown to regulate endogenous targets via a similar mechanism to miRNAs . These sno-miRNAs, like the mirtrons, appear to be independent of Drosha, but may require Dicer for final maturation. It may also be possible that some sno-miRNAs are processed in the same manner as miR-451 and only require Ago2. One of the intriguing questions is why are miRNAs embedded in snoRNAs? There is speculation that sno-miRNAs may represent an ancient version of the modern day miRNA and are continuously evolving. It has also been shown that C/D and H/ACA box snoRNAs are a new family of mobile genetic elements . The insertion of new snoRNA copies may be a safeguard to protect the biological activity of sno-miRNAs if and when the parental miRNA is either mutated or damaged. It is also equally plausible that many sno-miRNAs are the result of accidental processing of hairpin–like structures and have not gained function yet.
miRNAs can also be derived from pathogens such as viruses. These viral miRNAs are encoded in the viral genome and were first discovered in the herpes virus families . To-date 26 mammalian viruses have been described to harbour viral miRNAs including the human EBV (Epstein–Barr virus), herpes simplex virus 1/2 , MDV (Marek's disease virus)  and KSHV (Kaposi's sarcoma-associated herpes virus) . The majority of viral miRNAs from DNA viruses are processed via the canonical pathway, but the route taken by adenoviruses is somewhat different. The initial transcription is done by RNase III to generate a highly structured stem-loop precursor of 160 nt known as the VA1. This RNA is not processed by Drosha, but instead is transported by exportin-5 into the cytoplasm for Dicer processing. This processing is very inefficient with less than 1% of VA1 RNAs being processed into functional miRNAs . The biological role of VA1-derived miRNAs remains to be determined, but VA1 can inhibit Dicer function and prevent the processing of cellular precursor miRNAs . Another pathway used for viral miRNA biogenesis was demonstrated for the murine γ-herpes virus MHV68 . This pri-miRNA structure is a combination of a tRNA-like structure linked to one or two ∼60 nt pre-miRNA hairpins. This fusion RNA is transcribed by RNase III and cleaved by the enzyme RNase Z, which normally functions in the maturation of tRNA 3′-ends. This processing liberates the pre-miRNA which is then exported and further processed by Dicer in the cytoplasm. To-date the majority of studies have focused on dsDNA (double-stranded DNA) viruses and very few studies have yielded any evidence for viral miRNAs from RNA viruses. A recent study has shown that the BLV (bovine leukaemia virus), within its RNA genome, contains five miRNAs which are also transcribed by RNase III . The initial transcription yields pri-miRNAs of only 55 nt, structurally similar to pre-miRNAs, and therefore bypasses Drosha processing.
We have accumulated a substantial amount of information regarding the processing of small RNAs, particularly miRNAs. This is due to the combination of biochemical characterizations, structural studies and high-throughput approaches such as deep sequencing protein-bound small RNAs and their intermediates. We expect to gain a greater insight into pri-miRNA processing by the crystallization of Drosha, the only key protein in the biogenesis pathway which lacks detailed structural studies. There is also increasing evidence that the regulation of the small RNA biogenesis pathway is crucial to cell-cycle regulation and responds to pathogen attack. In the future, we anticipate that more and more miRNA regulatory proteins will be discovered in specialized cell types and full animal systems. Given that there are also alternative non-canonical pathways, it is entirely plausible that different cellular or stress stimuli may govern the pathway by which specific miRNAs are processed.
miRNA maturation is a stepwise process that starts from long structured RNAs and results in a species of 21–23 nt long single-stranded RNA which encodes their biological functions.
To gain function the miRNAs have to be associated with an Argonaute protein forming the minimal RISC.
The maturation of miRNAs is a highly regulated process. RNA binding and signalling proteins with function(s) in the regulation of cell cycle and immune response facilitate and/or suppress miRNA synthesis at different steps of miRNA production.
Alternative maturation pathways exist, but these are specific to certain miRNA families.
Work of the authors is supported by the Australian Research Council and Congressionally Directed Medical Research Programs, Department of Defence. N.T. is an UTS chancellor post-doctoral fellow and G.H. is an ARC Future Fellow.