Degradative RNAi occurs through a complex series of linked biochemical events involving a large number of protein components, many of which are still being identified. The overall process can be conceptualized as occurring in three stages, including (i) processing of long dsRNAs into short 21–23mer functional siRNAs, (ii) assembly of a mature RISC and (iii) sequence-specific cleavage of target ssRNA (i.e. mRNA) followed by degradation. In the initiator phase, long dsRNAs are processed by Dicer into short 21–23mer siRNAs which have 2-base 3′-overhangs and 5′-terminal phosphate groups (10
). In cell extracts, this process is accelerated by ATP (9
); however, recombinant human Dicer is ATP-independent (42
). In Drosophila
, the potency of dsRNA to trigger an RNAi response decreases as length is shortened from 130 to 29 bp (8
). Recombinant human Dicer efficiently cleaves long dsRNAs but shows some loss of activity as the dsRNA substrate decreases from 100 to 50 bp length (43
). Shorter sequences regain activity and, in fact, RNA duplexes as short as 23mer length are substrates for human Dicer (cleaved to 21–22mer products) (1
). More than one Dicer molecule can bind a 130 bp dsRNA molecule (possibly one on each end). Processing preferentially begins by cleaving duplexes at their ends and eventually results in complete degradation into 21–23 base fragments (42
). It is not clear if sequential cleavage events occur in a processive fashion along a single substrate RNA molecule or if Dicer cleaves, dissociates and re-binds the substrate before cleaving again.
Dicer is a complex 220 kDa protein comprising a dsRNA binding domain (dsRBD), a PAZ domain, a DExH RNA helicase/ATPase domain, and two RNase III class domains (42
). The RNase III domains cooperatively function to cleave substrate dsRNA into smaller 21–23 base fragments and orientation for this cleavage is assisted by flanking dsRNA binding domains, dsRBD and PAZ (33
). Crystal structure of the PAZ domain from human Argonaute eIF2c1 has been determined and suggests that this element specifically functions as a binding site for 2-base 3′-overhangs (44
). Similar conclusions were made for the Drosophila
Ago-2 protein based upon NMR solution structure (45
). The PAZ/PIWI domains seem to serve as anchors that helps spatially oriented bound RNAs in the enzyme active site (46
). Argonaute cleaves the substrate RNA 10 bases away from the PIWI anchor site whereas Dicer cleaves 21–22 bases away from the PAZ anchor site. This model fits well with our experimental observations of dicing patterns for 27mer dsRNA substrates. Blunt 27mer substrates do not provide an optimal structure for binding the Dicer PAZ domain so the ‘anchor’ step takes place with imprecision and results in heterogeneous cleavage products. Asymmetric duplexes with a single 2-base 3′-overhang provide a single favorable PAZ binding site, so these substrates usually have a single unique anchor site and cleavage occurs 21–22 bases away with limited heterogeneity. The presence of a 3′-overhang promotes ‘correct’ dicing of hairpin RNA substrates (21
). Interestingly, the actual base sequence of the 3′-overhang can influence dicing (29
) and not all asymmetric duplexes with a single 3′-overhang show a simple dicing pattern (EGFPS2 R 27/25, ). Adding DNA residues to the blunt end of an asymmetric duplex seems to help direct binding to the 3′-overhang even for ‘unfavorable’ sequences, possibly because the presence of DNA makes the blunt end an even worse structure for PAZ binding (EGFPS2 R 27/25D, ). Thus, asymmetric duplexes with one 2-base 3′-overhang in combination with 3′-DNA residues on the blunt end present Dicer with a substrate that is cleaved into predictable products.
Functional interpretation of the ESI-MS dicing data assumes that the cleavage properties of endogenous human Dicer will parallel the patterns observed using purified recombinant human Dicer in vitro
. In Drosophila
, Dicer functions as a heterodimer with the RNA binding protein R2D2, which forms a complex with Dicer and the siRNA duplex (48
). A possible human ortholog for R2D2 has recently been identified as TRBP (49
). SiRNA-mediated depletion of TRBP resulted in loss of recruitment of Ago-2 and destabilization of Dicer. Behavior of human Dicer in vitro
in the absence of other RNAi pathway proteins such as TRBP may be different from its properties in vivo
. Since the actual functional potency of RNA duplexes when transfected into cells follows predictions made based upon dicing patterns obtained for the same duplexes in vitro
, it seems unlikely that in vivo
dicing patterns will be significantly different from those observed in vitro
Dicer plays more than one role in the RNAi pathway. In addition to endonuclease cleavage of long dsRNA into siRNAs, Dicer is involved with entry of the siRNA into RISC and participates in RISC assembly (25
has two Dicer proteins, a Dicer-1 nuclease that is involved in miRNA processing and a Dicer-2 nuclease that is involved in siRNA processing (22
). Mutants lacking Dicer-2 activity have defective RISC assembly, even when provided with 21mer siRNAs that do not require cleavage (23
). We theorize that the RISC assembly function of Dicer is involved with the increased potency seen for Dicer–substrate RNA duplexes compared with short 21mer siRNAs; providing Dicer with a ‘substrate’, rather than a ‘product’, may improve efficiency of the RISC entry step.
While association with Dicer is required for entry of siRNAs into RISC in Drosophila
, the pathways available for RISC loading may be slightly different in mammals. In one study, single-stranded RNA was shown to be capable of directing sequence-specific target cleavage via RNAi pathways in HeLa cell extracts while only duplex siRNAs could direct target cleavage in Drosophila
extracts. Further, immunodepletion of Dicer from the HeLa cell extracts did not block target cleavage triggered by siRNAs (50
). In another study, a mouse ES cell line was established that was homozygous for disruption of the dcr-1
gene and had no functional Dicer activity. As expected, these cells were deficient for miRNA production and could not initiate an RNAi response from shRNA compounds, both of which require Dicer processing. Unlike the Drosophila
Dicer-2 mutants, however, the Dicer-deficient mouse cells could support RNAi if provided exogenous 21mer duplex siRNAs (51
). Thus two independent studies suggest that Dicer is not required for siRNA entry into RISC in mammalian cells. However, Dicer may still play some role in siRNA loading into RISC. In studies performed in both human and mouse cells, Doi and colleagues demonstrated that knockdown of Dicer using RNAi triggered by 21mer siRNAs significantly reduced the efficiency of siRNA-mediated silencing of a luciferase reporter target (52
). In a study that reconciles some of the apparent differences between the human and Drosophila
systems, Chendrimada et al
) reported that RNAi-mediated knockdown of either Dicer or TRBP, the proposed human R2D2 ortholog, reduced efficiency of siRNA-mediated silencing of a luciferase reporter target in a human cell line. Taken together, these studies suggest that RISC loading and functional triggering of an RNAi response is more efficient when Dicer and TRBP are present, even though Dicer is clearly not required for siRNA loading into RISC in mammals.
The roles played by various protein and their interactions with the siRNA during RISC assembly are currently better defined in Drosophila
than in mammals. In Drosophila
, cooperative interaction of the R2D2 protein and Dicer-2 is required for entry of a siRNA into RISC (48
). R2D2 preferentially binds the end of the siRNA duplex having greater thermodynamic stability and specifically associates with the 5′ end of whichever strand is present at this end. Conversely, Dicer binds the 5′ end of the strand on the opposing end (24
). This complex is joined by a variety of other protein components as RISC assembly proceeds. Eventually, the siRNA duplex is unwound and R2D2 exits RISC with its associated single-strand (the ‘passenger strand’), leaving Dicer and its associated strand in RISC. The retained strand later serves to direct sequence-specific targeting (the ‘guide strand’). Ago-2 replaces R2D2 at the 3′ end of the retained guide strand (24
) and is the Argonaute protein family member that functions as the actual ‘slicer’ endonuclease activity in RISC (53
Thermodynamic bias leads to preferential association of R2D2 on one end of an siRNA duplex and Dicer on the other end, and thereby directs as to which strand remains in RISC (‘guide strand’) and which strand is ejected (‘passenger strand’). The terminal base sequence therefore plays a significant role in determining sense versus antisense strand targeting. Other factors may also influence which strand enters RISC. Elbashir observed that the orientation of Dicer processing of a model 52mer RNA duplex defined which strand directed target cleavage. Long single-stranded overhangs block dicing from starting at that end. A blunt 52mer duplex directed cleavage of both sense and antisense strand targets. A 52mer duplex with 20 base overhangs on both ends was inactive and did not direct cleavage of either strand targets. Asymmetric 52mer dsRNA duplexes showed strand bias. If the 20-base 3′-overhang extended from the sense strand, Dicer processing started from the blunt end and antisense strand sequence preferentially directed cleavage of a sense target. Conversely, if the 20-base 3′-overhang extended from the antisense strand, Dicer processing started from the blunt end and sense-strand sequence preferentially directed cleavage of an antisense target (8
The difference of functional potency that we see between ‘R’ versus ‘L’ version asymmetric Dicer–substrate duplexes probably relates to interactions between the same protein components within the RNAi pathway. In the case of asymmetric 27mers, the blunt end is unfavorable for Dicer binding and the dicing process starts at the end with a 2-base 3′-overhang. For ‘L’ duplexes, the 3′-overhang resides on the sense (top) strand. For ‘R’ duplexes, the 3′-overhang resides on the antisense (bottom) strand. ‘R’ duplexes generally show increased potency in directing silencing of sense-strand targets, a pattern consistent with the observations of Tuschl and co-workers (8
). In the model of RISC formation proposed by Zamore and co-workers (24
), Dicer associates with the 5′ end of the strand that is retained in RISC and R2D2 associates with the 5′ end of the discarded passenger strand. The recent discovery of the required role for TRBP in mammalian RNAi opens the possibility that this protein serves a function similar to the Drosophila
). For 21mer siRNAs, where no dicing occurs, thermodynamic end stability might be the dominant factor directing which protein binds which end. For longer dsRNAs, where dicing occurs, polarity of the dicing reaction may also affect final protein binding patterns. Binding of a 3′-overhang may spatially orient the siRNA cleavage product in the correct position for association of the 5′ end of that strand with Dicer (and therefore remain in RISC) and allow access of the other end to associate with a human R2D2 ortholog. This model assumes that some fraction of Dicer/siRNA complexes remain intact following cleavage and directly enter a developing RISC complex. If Dicer freely dissociated from the siRNA product after cleavage and later rebound on the basis of thermodynamic end stability, then polarity of the dicing reaction should not influence potency or confer strand bias. The nascent siRNA produced by Dicer cleavage has symmetric 2-base 3′-overhangs. Any effects introduced by structure of the asymmetric substrate should be lost if product siRNA and Dicer separate. This model is also consistent with the pattern of increased potency observed for asymmetric short siRNAs by Khvorova and co-workers (29
). Here, binding of an antisense-strand 3′-overhang by the PAZ domain may similarly serve to favorably orient the 5′ end of the antisense strand within Dicer for subsequent retention in RISC, in this case without the need for a cleavage event.
Strand bias introduced by Dicer processing of asymmetric duplexes confers a relative, not absolute, advantage to retention of the 3′-overhang strand in RISC. For example, the strand targeting experiments shown in demonstrate that use of the ‘R’ versus ‘L’ form duplexes significantly alter the ratio of ‘S’ versus ‘AS’ strand targeting; however, in all cases tested both ‘S’ and ‘AS’ strand targeting still occurs. In a more biologically relevant example, the active strand in miRNAs can be derived from either the top or bottom strand of the precursor miRNA and strand selection seems to be primarily determined by thermodynamic asymmetry rules without regard for the direction of Dicer processing (54
). The relative contribution of various factors that contribute to preferential strand loading into RISC is complex.
Use of thermodynamic end stability rules (38
) and empiric design parameters (28
) has helped speed widespread use of RNAi in mammalian biology by making more potent reagents easier to obtain. The asymmetric duplexes described here similarly improve design of Dicer–substrate RNAs by exploiting the functional polarity introduced by Dicer processing.