microrna Biogenesis and its Impact on RNA Interference

Transcription

1 microrna Biogenesis and its Impact on RNA Interference Stefanie Grund and Sven Diederichs Contents 1 The microrna Biogenesis Pathway microrna Gene Transcription microrna Editing: Small Changes Affect Many Steps pri-mirna Cleavage by the Microprocessor Complex Nuclear Export of the microrna Precursors by Exportin The RISC Loading Complex Terminal Loop Removal by Dicer Ago2 Jumps the Queue: Generation of the ac-pre-mirna mirna Duplex Unwinding Strand Selection: Who Becomes the Guide? Mediators of RNA Silencing: The Argonaute Proteins Half-Life and Degradation of microrna Implication for RNAi Technology Potentials and Challenges of sirnas as a Tool sirna Versus shrna shrna-mir Library: Transferring microrna Structures to Synthetic shrnas Enhancement of RNAi by microrna Biogenesis Factors microrna Biogenesis in Health and Disease: Basis for RNAi Therapy Concluding Remarks References Abstract micrornas (mirnas) are small, noncoding, single-stranded RNAs that control diverse key cellular pathways at the posttranscriptional level. Their mode of action is translational repression or degradation of target mrnas containing S. Grund and S. Diederichs (*) Helmholtz-University-Group Molecular RNA Biology & Cancer, German Cancer Research Center (DKFZ), Im Neuenheimer Feld 280 (B150), 69120, Heidelberg, Germany Institute of Pathology, University of Heidelberg, Im Neuenheimer Feld 220 (B150), 69120, Heidelberg, Germany s.diederichs@dkfz.de V.A. Erdmann and J. Barciszewski (eds.), RNA Technologies and Their Applications, RNA Technologies, DOI / _15, # Springer-Verlag Berlin Heidelberg

2 326 S. Grund and S. Diederichs complementary sequences. As many mirnas act in crucial cellular pathways, their dysregulation can result in various diseases including cancer. Here, we summarize recent insights into the complex processing pathway generating the mature, functional mirna. Cleavage of the primary mirna transcripts (pri-mirnas) by the microprocessor complex Drosha-DGCR8/Pasha releases nt hairpin structures, the pre-mirnas. After export to the cytoplasm mediated by Exportin-5, the RNase III-like Dicer completes processing in conjunction with its dsrbd partner protein (TRBP in mammals and LOQS in flies). The small RNA duplexes are unwound, and one strand, the guide strand, is incorporated together with Argonaute proteins into the RNA-induced silencing complex (RISC). Multiple studies in recent years have revealed that every step of this processing pathway can be regulated and that certain mirnas do not follow this general processing pathway but use a variety of other processing and regulatory options for their maturation. Importantly, the mirna processing and effector proteins also provide the essential machinery for RNA interference (RNAi). While ectopically delivered RNA (like dsrna, sirna, or shrna) functions as specificity component to knockdown target genes, the processing and effector machinery has to be contributed by the targeted cell. Also, several mirna processing factors can be used to enhance RNAi. Thus, a deeper understanding of mirna processing, regulation, and function is an essential prerequisite to optimize experimental RNAi and enable therapeutic RNAi approaches. Keywords microrna mirna processing mirna biogenesis Argonaute RNAi sirna shrna Abbreviations mirnas pri-mirnas RISC RNAi Pol II Tudor-SN bp nt dsrbd OB-fold shrnas ac-pre-mirna micrornas primary mirna transcripts RNA-induced silencing complex RNA interference RNA Polymerase II Tudor staphylococcal nuclease basepairs nucleotide double-stranded RNA binding domain oligonucleotide/oligosaccharide binding fold short hairpin RNAs Ago2-cleaved pre-mirna

3 microrna Biogenesis and its Impact on RNA Interference The microrna Biogenesis Pathway Since RNAi was first described in the nematode Caenorhabditis elegans (Fire et al. 1998), it has been found to be a widespread phenomenon in eukaryotic organisms. Over the past years, hundreds of mirna genes have been identified in animals, plants, and viruses, making them one of the largest gene families. Several different mirnas can control one mrna target cooperatively and each mirna can bind to many different mrnas. Thus, more than 30% of all human genes are predicted to be subject to mirna regulation (Lewis et al. 2005). Hence, the control of mirna expression is by itself a key step in regulating target mrnas. 1.1 microrna Gene Transcription Analysis of the genomic position of mirnas revealed that the majority resides in intergenic regions (Lagos-Quintana et al. 2003), indicating that they are autonomous transcription units. The mirna genes embedded within known transcripts are primarily found in intronic regions (Lagos-Quintana et al. 2003; Rodriguez et al. 2004; Kim and Kim 2007), although some are found in exonic locations, such as the untranslated regions of mrnas (Rodriguez et al. 2004; Kim and Kim 2007). Animal mirna genes are often localized in close proximity to each other forming clusters (Lau et al. 2001; Lagos-Quintana et al. 2003), which are transcribed polycistronically (Lee et al. 2002). In contrast, this polycistronic arrangement is quite rare in plants with few exceptions (Guddeti et al. 2005; Zhang et al. 2008). Transcription of mirna genes is mediated by RNA Polymerase II (Pol II) as primary mirna transcripts (pri-mirnas) have been shown to contain the hallmarks of Pol II transcripts, a cap structure and a poly(a) tail (Cai et al. 2004; Lee et al. 2004a). Further, expression of mirnas was decreased by the Pol II-specific inhibitor a-amanitin, and Pol II has been shown to associate with mirna promoters (Lee et al. 2004a; Bortolin-Cavaille et al. 2009). Transcription by Pol II enables tissue-specific or developmental control by regulatory transcription factors. In several cases, where the mirna gene is transcribed together with a protein-coding gene as a single transcription unit, the regulated expression pattern appears to be coordinated (Baskerville and Bartel 2005). The largest human mirna gene cluster, C19MC, is embedded in Alu repeat sequences. It was proposed that this cluster is unique in being transcribed by RNA Pol III (Borchert et al. 2006). However, another study claims that C19MC mirnas are encoded within introns of a Pol II transcript (Bortolin-Cavaille et al. 2009). 1.2 microrna Editing: Small Changes Affect Many Steps During transcription, RNAs undergo various maturation processes, such as 5 0 capping, splicing, 3 0 end processing, polyadenylation, and RNA editing. The most

4 328 S. Grund and S. Diederichs frequent form of RNA editing is the conversion of individual adenosines to inosines in double-stranded RNAs by the action of ADARs (adenosine deaminases acting on RNA). In vertebrates, three ADAR family members have been identified (ADAR1 to ADAR3), although ADAR3 has not yet been proven to be catalytically active. In addition, ADAR1 yields two isoforms due to usage of alternative translation initiation codons (Patterson and Samuel 1995). While ADAR2 and the short form of ADAR1 are expressed constitutively and are localized in the nucleoplasm and the nucleolus, the long isoform of ADAR1 is induced by interferon and located primarily in the cytoplasm (Patterson and Samuel 1995; Desterro et al. 2003). Cotranscriptional A! I base conversion (Ryman et al. 2007) results in a sequence different from the one encoded by the DNA template as inosines are interpreted as guanosines by cellular machineries. Since the mirna precursors form stem-loop structures, they were also considered to be potential targets of A! I editing. Indeed, several studies in the past years have revealed that certain mirna precursors are edited by ADAR enzymes. This sequence change has far reaching consequences regarding processing and target site recognition of mirnas as structural and base pairing properties are altered. Nishikura and colleagues have shown that editing of pri-mir-142 interferes with the Drosha cleavage step, which leads to a reduction of mature mir-142 (Yang et al. 2006b). The edited, unprocessed pri-mirna is degraded by the Tudor staphylococcal nuclease (Tudor-SN), which preferentially cleaves dsrna with multiple IU wobble pairs (Scadden 2005; Yang et al. 2006b). In contrast, editing of two other pri-mirnas aids in Drosha processing and increases pre-mirna levels (Kawahara et al. 2008). While an effect of RNA editing on pre-mirna export into the cytoplasm has not been reported yet, it has been shown to abolish a further downstream step, the cleavage by the Dicer-TRBP complex (Kawahara et al. 2007a). Some A! I conversions, however, have no impact on either of the two cleavage steps, leading to the expression of edited mature mirnas if the edited site resides in the mature mirna sequence (Pfeffer et al. 2005). MiRNAs with altered sequences result in targeting of mrnas different from those targeted by the unedited mirnas, especially when the A! I conversion is located in the seed sequence (Kawahara et al. 2007b). Since selection of the guide strand, which is incorporated into the RISC complex, depends highly on the stability of the mirna duplex (Khvorova et al. 2003; Schwarz et al. 2003), and A! I editing is thought to affect stability properties, RNA editing might also affect selection of the functional guide strand. In summary, A! I RNA editing adds another layer of complexity, increasing the pool of cellular mirnas. 1.3 pri-mirna Cleavage by the Microprocessor Complex A typical primary transcript of mirna genes comprises an imperfect stem of 33 basepairs (bp) in length with a terminal loop and adjacent single-stranded

5 microrna Biogenesis and its Impact on RNA Interference 329 sequences (Han et al. 2006) (Fig. 1). The hairpin, containing the future mirna in either the 5 0 or 3 0 half of its stem, is excised by the endonuclease Drosha, which liberates the nt long pre-mirna with a two nucleotide (nt) overhang at its 3 0 -end (Lee et al. 2003; Denli et al. 2004). This process, also referred to as cropping, occurs cotranscriptionally and precedes the splicing reaction (Kim and Kim 2007; Morlando et al. 2008). Notably, the Drosha cleavage reaction is no hindrance for the spliceosome to execute its function because continuity of the intron is not a prerequisite (Dye et al. 2006; Kim and Kim 2007). Drosha is a member of the highly conserved RNase III enzyme family, containing two RNase III domains, a double-stranded RNA binding domain (dsrbd), and a long N-terminal region with unknown function (Filippov et al. 2000; Wu et al. 2000). By intramolecular dimerization, the two RNase III domains form a single processing center with two catalytic sites that each cut one strand of the stem (Han et al. 2004a). Drosha is present in two different complexes. In the smaller complex, referred to as the Microprocessor, Drosha binds to its cofactor DGCR8 (DiGeorge syndrome critical region 8; also known as Pasha in C. elegans and Drosophila melanogaster) (Denli et al. 2004; Gregory et al. 2004; Han et al. 2004a). This interaction is essential for the conversion of pri-mirnas into pre-mirnas (Gregory et al. 2004; Han et al. 2004a). While Drosha is crucial for the catalysis, DGCR8 establishes specificity for pri-mirnas and determines the cleavage site 11 bp distant from the junction of the stem base and the flanking single-stranded RNA (ssrna) (Gregory et al. 2004; Han et al. 2006). Binding of the dsrbds of DGCR8 to the pri-mirna requires the ssrna regions, which are therefore indispensible for pri-mirna processing (Zeng and Cullen 2005; Han et al. 2006). However, several sequence alterations in pri-mirnas of human tumors lead to conformational changes without affecting processing efficiency pointing towards a certain flexibility of the Microprocessor (Diederichs and Haber 2006). Aside from the necessity of the basal segments, the stem and a loop at the end of the stem are important for efficient cleavage (Zeng et al. 2005b; Han et al. 2006). After determination of the cleavage site by DGCR8, Drosha can interact transiently with this preformed RNA protein complex and execute the cut. Drosha-mediated cleavage represents a critical step in mirna biogenesis since this initial processing event defines one end of the mature mirna. Nonetheless, it is not compulsory for the generation of pre-mirnas. Short introns can also form hairpin structures that resemble pre-mirnas. These alternative precursors, termed mirtrons, can be spliced and debranched into pre-mirna-like hairpins. Mirtrons lack the lower stem of pri-mirnas and therefore omit processing by the Microprocessor. The debranched hairpins are then exported and further processed by the canonical mirna biogenesis pathway (Okamura et al. 2007; Ruby et al. 2007) Regulation of the Microprocessor To control Microprocessor activity, Drosha and DGCR8 regulate each other via a regulatory feedback circuit. Protein protein interaction of DGCR8 with Drosha

6 330 S. Grund and S. Diederichs mirna gene Pol II m 7 G ADAR transcription & editing pre-mrna pri-mirna m 7 G DGCR8 Drosha branched pre-mirtron splicing mrna AAA Drosha cleavage debranching pre-mirna 5 3 Exp-5 RanGTP nucleus cytoplasm Dicer RISC loading complex (RLC) 5 3 Ago2 Dicer TRBP Ago2 TRBP Ago2 cleavage Dicer cleavage 5 3 Ago2 Dicer TRBP ac-pre-mirna Dicer cleavage mirna duplex 5 3 Ago2 Dicer TRBP 5 3 Ago2 Dicer TRBP duplex unwinding & RISC formation mature mirna active RISC 5 Ago2 passenger strand degardation Fig. 1 The microrna biogenesis pathway in vertebrates. MicroRNA (mirna) genes are transcribed by RNA Polymerase II (Pol II) generating primary transcripts (pri-mirnas). Cleavage by the Microprocessor complex consisting of Drosha and DGCR8 results in a 65 nt precursormirna (pre-mirna). Intron-derived mirnas (mirtrons) are generated by splicing and debranching. The intron resembles the hairpin structure of the pre-mirna thereby bypassing the Drosha processing step. After export to the cytoplasm, mediated by Exportin-5 in a Ran-dependent

7 microrna Biogenesis and its Impact on RNA Interference 331 stabilizes Drosha. In turn, Drosha cleaves hairpins within the Dgcr8 mrna resulting in destabilization of Dgcr8 mrna (Han et al. 2009; Triboulet et al. 2009). Apart from this general mechanism, Drosha-mediated cleavage of specific primirnas can be influenced by additional factors. The DEAD-box RNA helicases p68 (DDX5) and p72 (DDX17), which are present in the large Drosha complex, seem to be involved in the processing of a subset of pri-mirnas, as deficiency of these factors decreases their mature mirna expression levels (Fukuda et al. 2007). Some auxiliary processing factors act even on individual mirnas. Transforming growth factor-b (TGF-b) and bone morphogenetic factors (BMPs) activate specific SMAD signal transducers, which then form a complex with the Microprocessor via the RNA helicase p68. This interaction promotes processing of pri-mir-21 into premir-21 by Drosha (Davis et al. 2008). Although the terminal loop seems to be of inferior significance for Microprocessor action per se (Han et al. 2006), it seems to have a fine-tuning role, providing a binding platform for regulatory proteins. The heterogeneous nuclear ribonucleoprotein A1 (hnrnp A1) protein, for example, contributes to the biogenesis of only a single mirna in the mir-17 cluster, mir-18a (Guil and Caceres 2007). By binding to the conserved loop of the hairpin, hnrnp A1 remodels the stem structure and thereby creates a more advantageous cleavage site for Drosha (Michlewski et al. 2008). In contrast, interaction of a stem cell-specific regulator, lin-28, with the loop of pri-let-7 prevents Drosha cleavage (Newman et al. 2008; Viswanathan et al. 2008). The repressive effect of lin-28 seems to be antagonized by the KH-type splicing regulatory protein (KSRP), which promotes maturation of a let-7 and a subset of mirna precursors by binding to the terminal loop (Trabucchi et al. 2009). Thus, it is likely that there are far more factors to be identified, optimizing recruitment and positioning of the Microprocessor thereby controlling processing in a coordinated manner Primary mirna Generation in Plants Plant mirnas are also derived from long, primary transcripts although rather diverse in structure and with longer hairpins than in animals in a stepwise process (Kurihara and Watanabe 2004). Nevertheless, biogenesis in plants holds some substantial differences compared to metazoans. A key characteristic of plant mirna maturation is the lack of a Drosha-like protein, which is highly conserved ä Fig. 1 (Continued) manner, the pre-mirna interacts with the preformed ternary complex of Dicer, TRBP, and Ago2 forming the RISC loading complex (RLC). Dicer removes the terminal loop of the pre-mirna creating a mirna duplex. Pre-miRNAs with a high degree of complementarity are cleaved by Ago2 in their passenger strand, producing a nicked hairpin called ac-premirna, before Dicer cleavage. After unwinding of the mirna duplex, the passenger strand is degraded whereas the functional guide strand is loaded onto Ago2, constituting the RNA-induced silencing complex (RISC), which silences target mrnas by translational repression or mrna cleavage

8 332 S. Grund and S. Diederichs in animals. One of four Dicer homologs, Dicer-like1 (DCL1; also known as CARPEL FACTORY, CAF), seems to execute not only the Dicer-like maturation step but is also responsible for catalyzing the pri- to pre-mirna conversion (Kurihara and Watanabe 2004). Both cleavage events seem to succeed quite fast, as intermediate products are rarely detected. Due to the predominantly nuclear localization of DCL1 (Papp et al. 2003), mature mirnas are produced in the nucleus and not in the cytoplasm like in animals. The whole processing requires several additional proteins, such as the RNA-binding protein DAWDLE (DDL) aiding in stabilizing the precursor (Yu et al. 2008), the zinc-finger protein SER- RATE (SE) (Lobbes et al. 2006; Yang et al. 2006a), and a double-stranded RNAbinding protein HYPONASTIC LEAVES1 (HYL1) (Han et al. 2004b; Hiraguri et al. 2005). The latter two interact with DCL1 in the so-called D-bodies (Fang and Spector 2007; Song et al. 2007). An interesting particularity in plants is the protection of mature mirna duplexes by the S-adenosyl methionine-dependent methyltransferase Hua Enhancer 1 (HEN1). Modified by methyl groups at the 3 0 end of each strand, mirnas are more stable and can escape degradation by the SMALL DEGRADING NUCLE- ASE (SDN) class of exonucleases (Li et al. 2005; Yang et al. 2006c). 1.4 Nuclear Export of the microrna Precursors by Exportin-5 After the initial cropping by Drosha, the precursor has to be further processed to yield the final mirna. This second cleavage reaction is mediated by the cytoplasmic enzyme Dicer. Owing to the different spatial appearance of the two endonucleases, nucleocytoplasmic transit of the pre-mirna has a pivotal role in mirna biogenesis. Like other noncoding RNAs, pre-mirnas are exported by a member of the karyopherin family of nucleocytoplasmic transport receptors. Several studies demonstrated that Exportin-5 is the main if not only transport factor for nuclear export of pre-mirnas and that the transport process characteristic of karyopherin mediated export depends on the Ran cycle (Yi et al. 2003; Bohnsack et al. 2004; Lund et al. 2004). Binding of Exportin-5 to its cargo requires a stem of at least 16 bp and a short 3 0 -overhang, an end structure characteristic of RNase III cleavage. Improperly processed pre-mirnas with, e.g., 5 0 -overhangs prevent binding of Exportin-5 and thus efficient export, highlighting the importance of precise cleavage by Drosha (Lund et al. 2004; Zeng and Cullen 2004). Notably, Exportin-5 is not only required for nucleocytoplasmic transport of pre-mirnas but also aids in stabilizing the relatively unstable precursor (Yi et al. 2003; Zeng and Cullen 2004). In plants, the role of HASTY, the plant homolog of Exportin-5 is not as well defined as in animals. Since mirnas are completely processed in the nucleus, mature mirnas are found in both the nucleus and the cytoplasm; however, they are more abundant in the latter compartment. Although loss-of-function mutants of HASTY reduce mirna levels, they do not cause accumulation in the nucleus. Hence, till date, there is no direct evidence that HASTY is involved in mirna

9 microrna Biogenesis and its Impact on RNA Interference 333 export. Also, the exact form as which mirnas leave the nucleus either as mirna itself, loaded into the RISC complex, or in complex with their target mrnas remains still unclear (Park et al. 2005). 1.5 The RISC Loading Complex Once in the cytoplasm, the pre-mirna still has to undergo several additional processing steps before the single-stranded mature mirna can guide the effector complex known as RNA-induced silencing complex (RISC) to its target mrna. Maturation of the pre-mirna and RISC loading is performed by the RISC loading complex (RLC). This complex comprises in the human system the RNases and Dicer and the structurally related dsrna binding proteins TRBP and PACT (Chendrimada et al. 2005; Haase et al. 2005; Lee et al. 2006; MacRae et al. 2008). Like Drosha, Dicer is an RNase III enzyme and removes the terminal loop by a staggered cut. Albeit TRBP and PACT are not required for the Dicer-mediated processing step per se, they have a stimulating effect on the cleavage reaction, influence the efficiency of RNA silencing, and are involved in the recruitment of Ago2 (Haase et al. 2005; Lee et al. 2006). Ago2 (also known as eif2c2) acts in the effector phase of RNAi and cleaves the mrna targeted for destruction by the complementary mirna or mediates as well as other human Ago proteins translational inhibition. Ago2 stability and thus also efficiency of RNAi is regulated by hydroxylation (Qi et al. 2008). The general paradigm that RNase III enzymes cooperate with dsrbd proteins holds true for Drosophila Dicer enzymes. Dicer-1 and Dicer-2 interact with Loquacious (Loqs) and R2D2, respectively. While the heterodimer Dicer-1/Loqs processes pre-mirnas, the counterpart Dicer-2/R2D2cleaves long dsrnas into sirnas (Lee et al. 2004b). Loqs is required for efficient pre-mirna processing and confers substrate specificity for pre-mirnas to Dicer-1 (Forstemann et al. 2005; Saito et al. 2005), whereas R2D2 is necessary for RISC loading but not for the cleavage reaction (Liu et al. 2003; Tomari et al. 2004b). The assembly of the RLC and the final RISC differs in flies and humans. In humans, the RLC is formed prior to pre-mirna binding and Dicer is released after mirna incorporation into Ago2 (Gregory et al. 2005; Maniataki and Mourelatos 2005). On the contrary, in Drosophila, Dcr-2/R2D2 binds first the sirna duplex. Thereafter, Ago2 joins the complex and the single-stranded sirna is loaded onto Ago2. This final complex is called holo-risc and still retains Dcr-2/ R2D2 (Pham et al. 2004; Tomari et al. 2004b). For the mirna pathway, however, Dicer-1/Loqs was recently shown to be excluded from the RISC. After processing of the pre-mirna and loading of the duplex onto Ago1, Dicer dissociates from the RLC. Ago1, loaded with the mature mirna, interacts then with GW182, a P body component with a role in mirna-mediated silencing, forming the RISC (Miyoshi et al. 2009).

10 334 S. Grund and S. Diederichs 1.6 Terminal Loop Removal by Dicer Once the pre-mirna joins the RLC, Dicer cleaves the precursor removing the terminal loop. The cleavage defines the other end of the mature mirna and results in a 22 nt long mirna duplex with 2 nt 3 0 -overhangs at each end (Bernstein et al. 2001; Provost et al. 2002). While in invertebrates this step necessitates ATP (Zamore et al. 2000; Nykanen et al. 2001), dicing is ATP-independent in mammalian cells (Provost et al. 2002; Zhang et al. 2002). Studies with dicer mutants in different organisms have shown the significance of the Dicer reaction for functional RNAi and its requirement for embryonic development (Knight and Bass 2001; Bernstein et al. 2003). Dicer proteins are evolutionary conserved throughout the eukaryotic kingdom except budding yeast (Bernstein et al. 2001). In some organisms, even multiple Dicer homologs exist, dedicated to mainly one aspect of RNA silencing, like generation of sirnas or mirnas. In contrast to Arabidopsis and Drosophila, where two Dicer isoforms can share the duties (Lee et al. 2004b; Xie et al. 2004), the single Dicer enzymes in nematodes and humans process both dsrna and mirna precursors. As members of the RNase III family, Dicer enzymes comprise two neighboring RNase III domains responsible for the catalytic reaction and a dsrbd that interacts with dsrna in vitro (Provost et al. 2002). The N-terminus harbors in addition a DExH/DEAH box RNA helicase domain, a PAZ domain and a domain of unknown function (DUF283). The RNA-binding PAZ domain, named after the Piwi, Argonaute, and Zwille proteins, is also found in the Argonaute protein family and adopts a topology related to the oligonucleotide/oligosaccharide-binding fold (OB-fold). The binding pocket for dsrnas with a two nucleotide 3 0 -overhang within the PAZ domains enables the anchoring and recognition of pre-mirnas processed by Drosha (Song et al. 2003; Lingel et al. 2004; Ma et al. 2004). An additional loop in the PAZ domain of the Dicer family alters the electrostatic potential of the surface surrounding the binding pocket compared to the Argonaute PAZ. Due to this difference, substrate recognition and transfer of the substrate to other complexes might differ between the two PAZ-domain families (Macrae et al. 2006). Notably, some Dicer proteins, such as Drosophila Dicer-2, Arabidopsis DCL-4, or Schizosaccharomyces pombe Dicer, do not contain a PAZ domain. In these species, adaptor molecules might compensate the lack of a PAZ domain providing an alternative to recognize pre-mirnas or the organism like in the case of S. pombe does not encode mirnas. The crystal structure of the parasite Giardia intestinalis Dicer shed light onto the question how the Dicer cleavage site is determined. G. intestinalis Dicer is a minimalist among the Dicer enzymes as it consists only of a PAZ domain and two consecutive RNase III domains. However, it generates RNA duplexes of nt from dsrna in a similar way to human Dicer (Zhang et al. 2002; Macrae et al. 2006). Unlike the nuclear RNase III Drosha, which is dependent on the ruler activity of its partner DGCR8/Pasha, Dicer is capable of measuring the distance of about 25 nt from the 3 0 -end on its own. This length is defined by the distance

11 microrna Biogenesis and its Impact on RNA Interference 335 between the PAZ domain that binds the and the 5 0 -end and the RNase IIIa domain, forming the catalytic center (Zhang et al. 2004; Macrae et al. 2006). The connecting helix between both domains is predicted to form in all Dicer homologs, and thus also the mechanism of cleavage site determination might be conserved (Macrae et al. 2006) Control Mechanisms of the Dicer Cleavage Dicer activity is also subject to diverse regulatory mechanisms. One striking example is the tissue-specific Dicer cleavage of pre-mir While the precursor is expressed ubiquitously, the mature mir-138 is found only in certain cell types. As the precursor is exported to the cytoplasm, the lack of mature mir-138 seems due to inhibition of Dicer although the mechanism and involved factors are unclear (Obernosterer et al. 2006). Another regulatory layer is mediated by the Dicer interaction partner TRBP. In the absence of TRBP, the N-terminal DExD/H-box helicase domain of human Dicer displays a low rate of substrate cleavage. Binding of TRBP stimulates Dicer-mediated catalysis possibly by inducing a conformational change. This mechanism could prevent unintentional activity of free Dicer before incorporation into the RLC (Ma et al. 2008). Additionally, its product, the mature let-7 mirna, controls Dicer activity creating a negative feedback loop. Let-7 downregulates Dicer by binding to complementary sequences in the dicer mrna (Forman et al. 2008; Tokumaru et al. 2008). In vitro, maturation of pre-let-7 itself is inhibited by the presence of Lin-28 (Rybak et al. 2008), a factor that has also regulatory function in the Microprocessor step (see above). Furthermore, Lin-28 mediates 3 0 -terminal uridylation of pre-let-7 in the cytoplasm. The elongated tail of pre-let-7 impedes Dicer cleavage and targets the precursor for degradation constituting another regulatory mechanism for Dicer activity for a specific mirna (Heo et al. 2008). Lastly, KSRP, also known as transcription factor FBP2 with regulatory roles in the nucleus, acts as an antagonist of Lin-28 and promotes Dicer processing of a subset of pre-mirnas (Trabucchi et al. 2009). 1.7 Ago2 Jumps the Queue: Generation of the ac-pre-mirna After joining the preformed RLC, certain pre-mirnas undergo an additional endonucleolytic processing step prior to Dicer cleavage (Diederichs and Haber 2007). Ago2 cleaves in highly complementary stems the prospective passenger strand, which is not designated to become the mature mirna, 12 nt from its 3 0 -end. The generated processing intermediate, termed Ago2-cleaved pre-mirna (ac-pre-mirna), joins then the canonical pathway and is processed by Dicer. The biological function of the ac-pre-mirna is still unsolved but is speculated to influence strand selection or to alleviate removal of the passenger strand by analogy to sirna processing (Matranga et al. 2005; Rand et al. 2005).

12 336 S. Grund and S. Diederichs 1.8 mirna Duplex Unwinding As soon as the mirna duplex is generated by Dicer, Dicer and its binding partners TRBP and PACT dissociate from the mirna. Since only one strand is finally retained in the active RISC complex, the duplex has to be separated into their component strands: the guide strand, complementary to the target and mediating RNAi, and the nonfunctional passenger strand, which is subsequently degraded. Separation of double-stranded RNA molecules is usually mediated by helicases using energy derived from ATP hydrolysis. Albeit several helicases have been shown to associate with proteins that act in RISC formation or activity and influence RNA silencing, a direct involvement in strand unwinding could till date not be proven. For example, the DEAD box helicases Gemin3/4, RCK/p54, and the putative DExD-box helicase MOV10 interact with Argonaute proteins and are required for posttranscriptional silencing (Mourelatos et al. 2002; Meister et al. 2005; Chu and Rana 2006). MOV10 homologs in flies (Armitage) or plants (SDE-3) play also a role in RNAi (Cook et al. 2004; Tomari et al. 2004a). RHA/ DHX9, a member of the DEAH-containing family of RNA helicases unwinds dsrna (Lee and Hurwitz 1992) and aids in active RISC loading by promoting the association of sirna with Ago2 (Robb and Rana 2007). Although this observation points towards a role of RHA in sirna duplex unwinding, it is unclear whether this possible function also applies to mirnas. For mirna let-7 unwinding, the ATP-dependent helicase p68/ddx5 is sufficient in vitro. In accordance with this, the lack of p68/ddx5 inhibits let-7 mirna function (Salzman et al. 2007). These findings and the multitude of potential factors in the unwinding process suggest that specific helicases may regulate particular subclasses of mirnas. Whether they participate directly in unwinding the duplexes or whether they rather remodel the RLC to facilitate mirna loading remains to be investigated. As RISC assembly and RISC loading can be accomplished in an ATP-independent manner (Gregory et al. 2005; Maniataki and Mourelatos 2005; MacRae et al. 2008), the general necessity of helicases in this process is challenged. Another factor implicated in strand separation is the endonuclease Ago2. In the sirna pathway, the effector protein Ago2, which is loaded with the sirna duplex, cleaves the passenger strand to reduce the internal stability (Matranga et al. 2005; Rand et al. 2005). This destabilization is necessary for strand dissociation and facilitates removal of the passenger strand. Mismatches and unpaired bulges in some mirna hairpin stems could render the cleavage-assisted strand separation mechanism not only feasible but also unnecessary due to their inherent thermodynamic instability. However, Ago2 is able to cleave the passenger strand of si-like mirna precursor stems with a highly complementary sequence like the let-7 mirna family (Diederichs and Haber 2007) (see above). Thus, Ago2-mediated passenger strand cleavage and generation of the ac-pre-mirna could facilitate strand dissociation and RISC activation also in the mirna pathway at least for mirna precursors that resemble sirnas with a high degree of complementarity in the middle of the hairpin stem. In Drosophila, a novel complex of Translin and

13 microrna Biogenesis and its Impact on RNA Interference 337 Trax, called C3PO (component 3 promoter of RISC), aids in unwinding of sirna duplexes and removes the remnants of the passenger strand caused by the Ago2- mediated nick. Thereby, C3PO supports active RISC formation and enhances RNAi (Liu et al. 2009). For mirna duplexes loaded onto the weak endonuclease Drosophila Ago1, a slicer-dependent strand separation is not possible. In this case, structural features of mirnas mismatches in the seed and the 3 0 mid region are pivotal for strand unwinding (Kawamata et al. 2009). If this mechanism holds true also for the mammalian nonslicer Ago proteins, Ago1, Ago3, and Ago4, has to be investigated. 1.9 Strand Selection: Who Becomes the Guide? In theory, unwinding of the mirna duplex yields two different mature mirnas with the potential to become the effector strand. However, the two strands are not equally competent in entering the RISC. Therefore, only one strand from each duplex, the so-called guide strand, is predominantly incorporated into the RISC. The remaining strand, named passenger strand, is primarily targeted for degradation, although both strands can be functionally active (Ro et al. 2007; Okamura et al. 2008). Which strand is chosen to participate in RNAi and which is condemned to be destroyed is predestined by the thermodynamic properties of the base pairs at the ends of the duplex. The strand whose 5 0 -end is less stably paired to its counterpart is loaded into RISC (Khvorova et al. 2003; Schwarz et al. 2003). In Drosophila, the binding partner of Dicer-2, R2D2, senses the functional asymmetry of the two sirna strands and binds the strand with the greater doublestranded character. R2D2 orients the Dcr-2/R2D2 heterodimer on the sirna in such a way that the correct strand is handed to Ago2, thereby facilitating RISC loading (Liu et al. 2003; Tomari et al. 2004b) Mediators of RNA Silencing: The Argonaute Proteins RISC is the effector complex that silences target transcripts posttranscriptionally by degradation or translational inhibition. Aside from the mature mirna and the target mrna, this ribonucleoprotein complex contains a member of the Argonaute family as core protein, which binds to the single-stranded mirna, and also includes accessory factors, such as GW182, aiding in the effector step. The Argonaute family is the largest protein family involved in RNAi and can be divided into the Ago and the Piwi subfamily. While members of the Ago subfamily are ubiquitously expressed, expression of Piwi proteins is restricted to the germ line (Sasaki et al. 2003). The whole Argonaute family is defined by three characteristic domains, the PAZ, the MID, and the PIWI domain. While the MID domain anchors the guide strand by binding to its 5 0 -phosphate (Ma et al. 2005; Parker et al. 2005;

14 338 S. Grund and S. Diederichs Yuan et al. 2005; Wang et al. 2008b), the PAZ domain (as described above) provides a binding pocket for dsrnas with 3 0 overhangs (Song et al. 2003; Lingel et al. 2004; Ma et al. 2004; Wang et al. 2008b) suggesting that the mirna after generation by Dicer could be directly handed over from the Dicer-PAZ to the Argonaute-PAZ domain. The PIWI domain structurally resembles an RNase H fold and provides the endonuclease activity for mrna target cleavage (Song et al. 2004; Rivas et al. 2005; Yuan et al. 2005). Structural studies with a ternary complex consisting of a Thermus thermophilus argonaute protein, a guide DNA and a target RNA revealed the processes within this complex that happen upon binding to the RNA target (Wang et al. 2008a, 2009). The binding of the target RNA starts with basepairing to the 5 0 end of the guide DNA, which is anchored in the MID domain. Progressing duplex formation liberates the 3 0 end of the guide DNA from its anchoring site in the PAZ domain. This release leads to a conformational change, which brings the cleavage site of the target RNA in close proximity to the catalytic residues in the PIWI domain. Although most organisms encode several Argonaute proteins (ranging from one in S. pombe to 27 in C. elegans), which are functionally not redundant, many of them are endonucleolytically inactive. In humans, only Ago2, also known as eif2c2, is equipped with the so-called slicer activity (Meister et al. 2004). However, Ago2 is not the only Argonaute protein associated with mirnas. The remaining three members of the Ago subfamily, Ago1, Ago3, and Ago4, interact with mirnas as well (Meister et al. 2004). How mirnas are partitioned between effector complexes with different Ago proteins and what are the functional consequences of this sorting in the mammalian system is currently unknown. More about sorting of small RNAs in different RISC variants is known in flies and plants. In Drosophila, sirnas and mirnas are generated by two different Dicer enzymes (Lee et al. 2004b). It was assumed that due to their different biogenesis, the two classes of small RNAs are predetermined to be loaded into a particular RISC variant, mirnas into Ago1-RISCs and sirnas into Ago2-RISCs (Okamura et al. 2004; Saito et al. 2005). In contrast, Zamore and colleagues could show that Argonaute loading is not coupled to the biogenesis pathway of mirnas or sirnas but that sorting in one or the other complex depends rather on the intrinsic structure of the small RNA (Forstemann et al. 2007; Tomari et al. 2007). The heterodimer Dcr-2/R2D2 enforces incorporation of perfectly matched short RNAs (e.g., sirnas) into Ago2-RISC, while bulged or mismatched mirnas are excluded. In analogy, a yet unidentified mechanism sorts nonperfectly matched mirnas into Ago1-RISCs. For small RNAs with a structure in between that of a completely complementary sirna duplex and a typical mirna duplex with bulges and mismatches, both RISCs compete (Tomari et al. 2007). Importantly, sorting into one or the other RISCs determines the fate of the target mrna as the two Argonaute proteins are functionally specialized: Ago1-RISCs only silence imperfectly matched targets whereas Ago2-RISCs, which have the stronger catalytic capacity, silence targets that are fully complementary to the guide strand (Forstemann et al. 2007). Notably, only those guide strands irrespective from which arm of the pre-mirna they originate are capable of associating with

15 microrna Biogenesis and its Impact on RNA Interference 339 Drosophila Ago2, which have a 5 0 -end derived from an accurate cleavage by Drosha or Dicer (Seitz et al. 2008). In Arabidopsis, sorting of small RNAs is directed by the 5 0 -terminal nucleotide; e.g., joining AGO1, the Argonaute protein prevailing in the mirna pathway, requires a uridine at the 5 0 -end (Seitz et al. 2008) Regulation of Ago Activity and Ago-Mediated Silencing Mechanisms Once loaded onto an Argonaute protein, the mirna guides the complex to a fully or partially complementary mrna target. Target recognition is considered to require primarily full complementarity of nucleotides 2 7, termed the mirna seed sequence, but other nucleotides have been shown to aid in this process (Doench and Sharp 2004; Brennecke et al. 2005; Grimson et al. 2007). The shortness of the seed region enables each mirna to regulate a large number of target genes. The extent of complementarity between the mirna and the target is likely the key determinant for the mechanism of regulation. Highly complementary target sites which particularly occur in plants but rarely in animals cause slicing and subsequent decay of the target by and 3 0 -exonucleases (Zamore et al. 2000). For sirnas, the 5 0 -end of the guide strand defines the position of the target cleavage between the nucleotides paired to bases 10 and 11 of the sirna (Elbashir et al. 2001b,c). Most animal mirnas form imperfect hybrids with their target mrna, which precludes endonucleolytic cleavage, due to central mismatches (bases 9 12) (Elbashir et al. 2001c). Instead, they promote translational attenuation or exonucleolytic mrna decay. The underlying molecular mechanism of this silencing pathway is still under debate. Accounting the multitude of steps that have been proposed to be targeted by RISC action, it is likely that more than one mechanism is involved. For example, evidences exist that translation inhibition can be mediated at the level of translation initiation as well as elongation. Additionally, models for premature termination or cotranslational protein degradation have been suggested (Eulalio et al. 2008; Wu and Belasco 2008). By triggering the removal of the target poly(a) tail and subsequent decapping, mirnas can cause exonucleolytic mrna decay (Behm-Ansmant et al. 2006; Wu et al. 2006; Eulalio et al. 2007b). The enzymes for deadenylation and decapping localize to processing bodies (P bodies), cytoplasmic sites of mrna turnover (Eulalio et al. 2007a). Since Argonaute proteins, mirnas, and their targets colocalize with these foci, mirna targets could be directed to P bodies to be separated from the translation machinery and be accessible for the decay components (Liu et al. 2005; Pillai et al. 2005; Sen and Blau 2005). Targeting of Ago2 to P bodies is regulated by its phosphorylation status (Zeng et al. 2008). The environment of the P bodies per se is not essential for target silencing as mirna function is not or only marginally affected by the loss of P-bodies. Hence, enrichment of the RISC components in P bodies seems to be the consequence and not the cause of silencing (Eulalio et al. 2007a). Considering their core function in translation inhibition by various means, it is not surprising that Argonaute proteins are mainly localized in the cytoplasm and enriched in processing bodies. Nevertheless, nuclear functions for Ago proteins

16 340 S. Grund and S. Diederichs have been reported in worms, flies, plants, and S. pombe (Lippman and Martienssen 2004; Matzke and Birchler 2005). Evidence is accumulating that human Ago proteins also function in the nucleus. They were detected in nuclear fractions (Robb et al. 2005), and Ago1 and Ago2 have also been shown to associate with the promoter region of the progesterone receptor (Janowski et al. 2006). Additionally, sirnas directed against promoter regions resulted in Ago1- or Ago2-dependent transcriptional repression (Morris et al. 2004; Janowski et al. 2006; Kim et al. 2006). Importin- 8 is one factor involved in the transport of Ago2 into the nucleus (Weinmann et al. 2009). In addition to the effector proteins, mature mirnas have also been detected in the nucleus. The human mir-29b is even predominantly detected in the nucleus (Meister et al. 2004; Hwang et al. 2007). Nuclear import of mir-29b is mediated by a 3 0 -terminal hexanucleotide motif, which acts as a nuclear localization element (Hwang et al. 2007). These examples of reimport of a mirna into the nucleus and the nuclear localization and action of effector proteins raise the possibility that RISC-dependent gene silencing could occur also in humans at the transcriptional level Half-Life and Degradation of microrna For steady-state mirna expression levels, their stability, turnover, and degradation could be as important as the regulation of mirna maturation. So far, rather little is known about this aspect of the mirna life cycle. Generally, mature mirnas are rather stable, as they persist long in the cell after depletion of mirna processing factors (Lee et al. 2003; Gregory et al. 2004). However, mirna levels can drop rapidly under certain conditions, albeit the mechanism is yet unidentified (Pedersen et al. 2007). One factor known to play a role in mirna homeostasis is Ago2, as mirna levels depend on the amount of Ago2 in the cell (Diederichs and Haber 2007). Exoribonucleases responsible for mirna degradation have till date only been discovered in two organisms: SDN1 in Arabidopsis and XRN-2 in C. elegans (Ramachandran and Chen 2008; Chatterjee and Grosshans 2009). XRN-2 aids in mirna release from the Argonaute complex and mediates mirna turnover. Both separation and degradation are antagonized by the presence of target RNA implicating a coordination of mirna and target RNA levels (Chatterjee and Grosshans 2009). An alternative way to impede mirna activity could be the shielding of mirna-binding sites by interaction with RNAbinding proteins (Bhattacharyya et al. 2006; Kedde et al. 2007). 2 Implication for RNAi Technology The discovery of gene silencing by small RNA molecules, termed RNAi, was a milestone in biology by providing a novel level of gene expression regulation. Beyond the scientific importance, RNAi opened up the possibility for experimental

17 microrna Biogenesis and its Impact on RNA Interference 341 and therapeutic applications. However, since the small RNA molecules, mirnas or sirnas, provide only the specificity and depend on the endogenous mirna pathway to mediate RNAi, a detailed knowledge of mirna biogenesis is mandatory to use RNAi successfully as tool (Fig. 2). mirna gene shrna-mir m 7 G shrna construct pri-mirna AAA shrna pre-mirna 5 3 nucleus Exp-5 Ran GTP cytoplasm Dicer Dicer cleavage 5 3 Ago2 Ago2 cleavage sirna 5 3 ac-pre-mirna mirna duplex 5 3 Dicer 5 3 Dicer cleavage duplex unwinding & RISC formation mature mirna active RISC 5 Ago2 passenger strand degardation Fig. 2 RNAi takes advantage of the endogenous mirna biogenesis machinery and effector pathway. To use RNAi as a tool, artificial RNAs can be used, which enter the endogenous pathway at different steps. While shrnas or shrna-mirs, which are transcribed from a plasmid by Polymerase III or II, undergo (almost) the entire processing pathway, synthetic sirna duplexes join the pathway only in the cytoplasm. For simplicity, only those factors improving RNAi are shown. Exp-5; Exportin-5

18 342 S. Grund and S. Diederichs 2.1 Potentials and Challenges of sirnas as a Tool In basic research, a fundamental approach to investigate the biological role of a particular gene is the use of loss-of-function studies. In vertebrates, the lack of methodologies for a gene knockout was a major drawback to investigate deletion phenotypes. With the revolutionary discovery of RNAi by Mello and Fire in 1998, a powerful tool for silencing of specific gene products became available (Fire et al. 1998). An initial obstacle for successful application in mammalian cells was the existence of an antiviral response triggered by foreign dsrnas. Accumulating dsrnas that are longer than 30 bp activate two enzymes: protein kinase R (PKR) and 2 0,5 0 -oligoadenylate synthetase (2 0,5 0 -AS). Active PKR phosphorylates the translation initiation factor eif2a, which leads ultimately to a global inhibition of translation, whereas 2 0,5 0 -AS causes activation of RNase L, a ribonuclease that degrades mrnas nonspecifically. The nonspecific effects of the immune response would disguise any sequence-specific regulation generated by RNAi. The solution turned out to be quite simple: Long dsrnas are processed to nt short sirnas, which mediate RNAi (Hammond et al. 2000; Zamore et al. 2000), and the dsrna-induced nonspecific responses are dependent on at least 30 bp length of the dsrna (Minks et al. 1979; Manche et al. 1992). Hence, attempts to use short sirnas in analogy to the successful application of synthetic sirnas in Drosophila (Elbashir et al. 2001b) have also been made in the mammalian system (Caplen et al. 2001; Elbashir et al. 2001a). Indeed, the small RNAs were sufficient to mediate RNAi and due to their shortness without causing unspecific inhibitory effects caused by PKR activation. Since this breakthrough, RNAi has been established as a standard methodology for gene silencing in mammalian cells and opened up new possibilities for gene-targeted therapy approaches. However, it turned out that RNAi triggered by artificial sirnas is not as specific as originally assumed. Off-target effects have been observed in a concentrationdependent manner likely caused by induction of certain signaling pathways or knockdown of genes other than the desired target (Sledz et al. 2003; Persengiev et al. 2004). Another reason for off-target effects might be the possibility that both strands can be loaded into RISC (Ro et al. 2007; Okamura et al. 2008) resulting not only in silencing of the target by the guide strand but also in undesired repression of mrnas complementary to the passenger strand. Studies on requirements for correct strand selection revealed that the strand with the less stably paired 5 0 -end is chosen for incorporation into RISC (Khvorova et al. 2003; Schwarz et al. 2003). Hence, off-target effects caused by the complementary sirna strand can be minimized by designing an asymmetric sirna with high internal instability at the 5 0 -end of the desired guide strand or by chemically modifying the 5 0 -end of the passenger strand to prevent phosphorylation and thus RISC loading (Khvorova et al. 2003; Schwarz et al. 2003). Further, targeting requires primarily complementarity between the target and the seed sequence comprising nucleotides 2 7 of the sirna (Doench and Sharp 2004; Brennecke et al. 2005; Grimson et al. 2007). Therefore, sirnas although absolutely complementary to their desired target region might