Distinct Populations of Primary and Secondary Effectors During RNAi in C. elegans

See allHide authors and affiliations

Science  12 Jan 2007:
Vol. 315, Issue 5809, pp. 241-244
DOI: 10.1126/science.1132839


RNA interference (RNAi) is a phylogenetically widespread gene-silencing process triggered by double-stranded RNA. In plants and Caenorhabditis elegans, two distinct populations of small RNAs have been proposed to participate in RNAi: “Primary siRNAs” (derived from DICER nuclease-mediated cleavage of the original trigger) and “secondary siRNAs” [additional small RNAs whose synthesis requires an RNA-directed RNA polymerase (RdRP)]. Analyzing small RNAs associated with ongoing RNAi in C. elegans, we found that secondary siRNAs constitute the vast majority. The bulk of secondary siRNAs exhibited structure and sequence indicative of a biosynthetic mode whereby each molecule derives from an independent de novo initiation by RdRP. Analysis of endogenous small RNAs indicated that a fraction derive from a biosynthetic mechanism that is similar to that of secondary siRNAs formed during RNAi, suggesting that small antisense transcripts derived from cellular messenger RNAs by RdRP activity may have key roles in cellular regulation.

Double-stranded RNA (dsRNA)–triggered gene silencing in eukaryotes appears universally to involve 21- to 25-nucleotide (nt) siRNA effectors. In Drosophila and mammals, siRNAs derive primarily from processing of longer duplexes by DICER nuclease, forming 21- to 25-nt duplexes possessing 5′-monophosphates, 3′-hydroxyl groups, and 2-nt 3′ overhangs (1, 2). Along with this “primary” siRNA response, amplification of the RNA trigger population has been proposed to contribute to potency and persistence of gene silencing in several systems (3, 4). Amplification mechanisms are accompanied in some cases by “transitive RNAi” phenomena in which dsRNA matching one mRNA region can silence targets bearing homology to other parts of the mRNA (59). Unlike the situationinplantswhere “spreading” of the effector population occurs bidirectionally relative to the target mRNA (6, 8), transitive RNAi in Caenorhabditis elegans exhibits a strong bias toward sequences upstream of trigger homology (7, 9). Transitive RNAi requires function of a putative RdRP (RRF-1 in C. elegans soma, SDE1/SGS2 in Arabidopsis thaliana), suggesting several conceivable means for secondary siRNA production (6, 7). One possibility is that antisense primary siRNAs could act as primers in the RdRP-mediated synthesis of new dsRNAs on an mRNA template. Alternatively, primary siRNAs may merely guide the RdRP to a target, allowing unprimed synthesis (10, 11) either at the cleaved end of the targeted transcript, at a location close to the trigger-target complex, or at a structure such as a free end that might be revealed as aberrant through consequences of the initial RNA-induced silencing complex (RISC)::target interaction.

To better understand signal amplification in C. elegans, we characterized small RNAs from animals undergoing RNAi against an abundantly expressed endogenous gene, sel-1 (12). After reverse transcription, 245,420 18- to 25-nt RNAs were sequenced by means of single-molecule pyrosequencing (13). Among these sequences, 534 exhibited either a perfect match (428 instances) or single mismatches (106 instances) to sel-1 mRNA (Fig. 1, A and B). A similar analysis of ∼850,000 clones from animals not exposed to dsRNA yielded just one sel-1 small RNA (14). Most sel-1 small RNAs induced during interference (483) had an antisense orientation, consistent with previous hybridization-based analyses (7). Of the 51 sense strand clones, 22 showed complementarity to at least one antisense clone.

Fig. 1.

Distinct RNAi-associated small RNA populations identified by 5′-ligation-dependent and 5′-ligation-independent methods. Small RNAs were isolated from animals in which RNAi had been triggered by dsRNA covering nucleotides 535 to 992 of sel-1 mRNA. (A and B) By the 5′-ligation-dependent method, 51 sense clones (47 within the trigger, 4 downstream) and 483 antisense clones (110 upstream of the trigger, 1 in the trigger/upstream junction, 335 within the trigger, 9 in the trigger/downstream junction, 28 downstream) were obtained. (C) The 127 sel-1–associated small RNAs isolated by the 5′-ligation-independent method. All were antisense: 12 upstream of the trigger, 1 at the upstream/trigger junction, 104 within the trigger, and 3 downstream.

We observed an incomplete bias in siRNA positions relative to the trigger; of 138 antisense siRNAs outside the original trigger, 110 (80%) occurred on the 5′ side. This bias could certainly account for preferential detection of upstream secondary responses in functional and biochemical assays (7, 9). Twenty-eight observed instances of small antisense RNAs completely downstream of the trigger homology were of particular interest, as these would not have been expected if the sole mode of amplification involved extension by RdRP of existing siRNA triggers that hybridize to the target transcript.

Exon-exon junctions offer a unique opportunity to unequivocally distinguish de novo synthesis of antisense nucleic acids from an mRNA template (15, 16). We found 50 sel-1 small antisense RNA sequences that span exon/exon junctions. Of these, 43 fall within the trigger [458 base pairs (bp) of sel-1 cDNA sequence] and thus could have derived directly from triggering dsRNA. Six antisense exon-exon junction sequences upstream of the trigger were recovered (four matching perfectly and two with single mismatches). These imply de novo copying of the mature mRNA template (16).

The apparent scarcity of sel-1 siRNAs suggested that the procedure for cloning small RNAs (including ligation of linkers to 3′ and 5′ ends) might underrepresent the siRNA population (12). To analyze small RNA termini in detail, we used a number of structure-specific treatments. Treatment of RNA with periodate followed by β elimination results in a shift on a denaturing acrylamide gel, indicating at least one unmodified (cis-diol) 3′ terminus (17). Ribonuclease T (RNaseT) requires a 3′-hydroxyl to degrade single-stranded RNA. Finally, Terminator exonuclease preferentially degrades substrates with a single 5′-phosphate. Although sel-1 siRNAs are susceptible to both β elimination and RNaseT reactions, they are resistant to Terminator (Fig. 2A). Control synthetic 25-nt sel-1 RNA oligonucleotides s with 5′-monophosphate and 3′-OH were sensitive to all three treatments. We surmised that sel-1 siRNAs are blocked at their 5′ ends.

Fig. 2.

Small RNAs associated with RNAi are modified at their 5′ termini. (A) sel-1 small RNAs are protected from 5′-exonucleolytic attack but not from 3′ attack. Small RNAs were extracted from wild-type animals undergoing RNAi for sel-1; treated with the 5′-exonuclease Terminator, the 3′-exonuclease RNaseT, or sodium periodate plus β elimination; and subjected to Northern analysis (20% 19:1 acrylamide:bis-acrylamide gel) with probe directed against the sel-1 trigger sequence (nt 535 to 992) (blue arrowheads). A control 25-nt synthetic sel-1 RNA (5′-monophosphate) was added to each reaction (red arrowhead). (B) Size distributions of various classes of cloned small RNAs with perfect matches to sel-1 sequence.

We next asked if we could design a cloning protocol that would not be biased by the structure at the 5′ end on an siRNA. The resulting protocol (fig. S3) avoids both (i) the requirement for ligation of the 5′ end of the RNA and (ii) the possibility that modified 5′ ends on small RNAs could affect enzymatic treatments of the paired cDNA strand. We detected 127 sel-1 antisense sequences and zero sense sequences from 1612 total clones using this protocol (Fig. 1C). For sel-1 antisense sequences, this represents a 40-fold enrichment compared to the 5′-ligation-dependent cloning method, providing further evidence for a prominent population of 5′-blocked siRNAs.

Secondary siRNAs are still recovered in 5′-ligation-dependent cloning, albeit inefficiently, as indicated by the representation of sequences outside the trigger (presumably most siRNAs within the trigger are also secondary). Notably, we found that small antisense segments cloned with a 5′-ligation-independent procedure were on average 1 nt longer than those cloned with a 5′-ligation-dependent procedure (Fig. 2B; figs. S4 and S5). The substantial increase in incidence of sel-1 clones that followed 5′-ligation-independent cloning indicates that the vast majority of small sel-1 RNAs are modified on their 5′ ends, while at most 2 to 3% (SOM Text) have simple 5′-phosphate termini that are exposed in vivo or produced by 5′ cleavage during the cloning procedure. An assumption that sense and antisense are roughly equal in the primary siRNA pool leads to primary siRNA estimates of <0.6% of the total sel-1 siRNA population and <0.05% of the total 21- to 25-nt RNAs in the animal.

The two methods of cloning were selective for different classes of endogenous small RNAs (table S1) (18, 19). microRNAs (miRNAs) appeared much less frequently with the 5′-ligation-independent cloning method, seemingly replaced by endogenous small RNAs corresponding to antisense sequence from coding regions. This analysis suggests that miRNAs and small antisense RNAs could be comparably abundant in C. elegans, with 5′ modification of the small antisense RNAs accounting for the predominance of miRNA clones in libraries derived using ligation-dependent schemes. We observed 612 out of 245,420 clones from the 5′-ligation-dependent method and 9 out of 1612 clones from the 5′-ligation-independent method that were perfect antisense copies of exon/exon junctions, suggesting synthesis by RdRP acting on an mRNA template (fig. S6).

To further characterize the modification of siRNAs in C. elegans, we used a ligation assay and Terminator 5′-exonuclease treatment (both requiring a 5′-phosphate). Sensitivity of the predominant fraction of the siRNAs could be restored by sequential treatment with alkaline phosphatase (which removes any number of 5′-phosphates) and T4 polynucleotide kinase (which adds a single 5′-phosphate), suggesting that the 5′ modification was likely to involve additional 5′-phosphate groups on the siRNA (Fig. 3, A and B). How many phosphates do these molecules have on their 5′ ends? Examining relative gel mobilities of the native and dephosphorylated siRNAs, using a variety of gel porosities (and using a series of synthetic RNA markers with different numbers of phosphates), indicated that the predominant fraction of the untreated siRNAs have triphosphate 5′ termini (Fig. 3, C and D).

Fig. 3.

Small RNAs associated with RNAi exhibit 5′-triphosphate termini. (A) Ligation assay shows that sel-1 small RNAs possess multiple 5′-phosphates. Samples were sequentially treated with calf intestinal alkaline phosphatase (CIP), T4 polynucleotide kinase (T4 PNK), and/or ligase and analyzed by Northern blotting. Efficient ligation [leading to disappearance of the major siRNA bands (27)] requires 3′-OH/5′-monophosphate termini and was only observed after sequential treatment of siRNA populations with CIP (removing any number of 5′-phosphates) and T4 PNK (restoring a single 5′-phosphate). (B) Sensitivity to Terminator exonuclease, which preferentially degrades RNA with 5′-monophosphate termini. A 25-nt synthetic sel-1 control RNA with 5′-OH terminus was included as an internal control (red arrowhead). Gels (A) and (B) used a 19:1 acrylamide:bis-acrylamide composition under which phosphatase-treated siRNAs comigrate with untreated siRNAs. These species are resolved on 75:1 acrylamide:bis gels (C), where untreated species migrate more quickly than 5′-OH RNAs. (D) sel-1 small RNAs possess 5′-triphosphate termini. Control in vitro–synthesized RNAs with 5′-OH (lanes 1 and 6), 5′-monophosphate (lane 2), 5′-diphosphate (lane 3), 5′-triphosphate (lane 4), and 5′-tetraphosphate (lane 5) termini serve as mobility standards (blue arrowheads). With a high bis-acrylamide ratio (9:1 acrylamide:bis 12%), the size contribution of each 5′-terminal phosphate confers a prominent change in mobility such that a “ladder” of differentially 5′-phosphorylated species is observed. In vivo sel-1 siRNAs were treated with CIP (lane 7) or mock-treated (lane 8) with a darker exposure of these lanes in the right panel (red arrowheads). The distance migrated through the gel relative to the 5′-OH–terminated species (lanes 2 to 6 were compared to lane 1 and lane 8 was compared to lane 7), given by the numbers in arbitrary units below each lane, demonstrated that in vivo siRNAs migrate as 5′-triphosphate–terminated RNAs.

The results presented here define an RNA population produced de novo during RNAi in C. elegans as a pool of 5′-triphosphate–terminated small antisense molecules templated by the mature mRNA target and covering sequences both upstream and downstream of the original dsRNA trigger. On the basis of this work and previous literature, our current working model for amplified gene silencing in C. elegans is that rare primary siRNAs, formed from a long dsRNA trigger, act as guides (presumably in an Argonaute-dependent manner) to recruit RdRP to targeted transcripts. This recruitment leads to de novo synthesis of short antisense RNAs that must be stripped off the template mRNA and incorporated into complexes that are capable of finding additional silencing targets. This model differs from other examples of RdRP action, such as in the generation of ta-siRNAs in A. thaliana where repeated DICER activity on long RdRP-generated dsRNAs produces a phased distribution of small RNAs (20).

Although ongoing RNAi is certainly important for de novo synthesis of antisense siRNAs, this process appears to contribute by providing guidance to the RdRP rather than priming activity. Primary silencing targets may or may not be degraded; whatever their fate, however, they remain intact for a sufficient period to be substrates for RdRP activity upstream and (somewhat less efficiently) downstream of the targeting site. This model is consistent with the biochemical properties of characterized cellular RdRPs in that these enzymes are capable of unprimed (as well as primed) synthesis (10, 11). Initiation at 3′ ends of potential templates has been reported for fungal and plant RdRPs; this might allow initial Argonaute-mediated cleavage of mRNA targets to yield ready RdRP substrates.

Previous investigations of siRNA structure revealing double-stranded character, 3′ overhangs, and 5′-monophosphate termini were performed in organisms whose genomes do not encode canonical RdRPs (2, 21, 22). In addition, crystal structures of Argonaute proteins [key executioners in the RNAi pathway (2325)] indicate considerable specificity in recognizing specific 5′ structures in RNA. It is possible that 5′-triphosphate antisense RNAs are themselves inactive in gene silencing, requiring either removal of two terminal phosphates or of the entire first base for activity. Alternatively, triphosphate-terminated small RNAs may be active directly as silencing triggers, potentially through distinct members of the Argonaute family that might recognize guide RNAs with a 5′-triphosphate.

One feature of the proposed mechanism is the involvement of dsRNA trigger only at the earliest stage of the process (production of primary siRNAs). Following this stage, the double-stranded character of the original trigger plays no role in the reaction. Given diverse structural features that could target an aberrant mRNA for RdRP activity, such a system would permit analogous machineries (each involving RdRP, helicase, and Argonaute activities) to serve in amplified surveillance processes triggered by both aberrant mRNA structure (3) and dsRNA (26).

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S6

Tables S1 and S2


References and Notes

View Abstract

Navigate This Article