Research Article

A Distinct Small RNA Pathway Silences Selfish Genetic Elements in the Germline

See allHide authors and affiliations

Science  21 Jul 2006:
Vol. 313, Issue 5785, pp. 320-324
DOI: 10.1126/science.1129333


In the Drosophila germline, repeat-associated small interfering RNAs (rasiRNAs) ensure genomic stability by silencing endogenous selfish genetic elements such as retrotransposons and repetitive sequences. Whereas small interfering RNAs (siRNAs) derive from both the sense and antisense strands of their double-stranded RNA precursors, rasiRNAs arise mainly from the antisense strand. rasiRNA production appears not to require Dicer-1, which makes microRNAs (miRNAs), or Dicer-2, which makes siRNAs, and rasiRNAs lack the 2′,3′ hydroxy termini characteristic of animal siRNA and miRNA. Unlike siRNAs and miRNAs, rasiRNAs function through the Piwi, rather than the Ago, Argonaute protein subfamily. Our data suggest that rasiRNAs protect the fly germline through a silencing mechanism distinct from both the miRNA and RNA interference pathways.

In plants and animals, RNA silencing pathways defend against viruses (13), regulate endogenous gene expression (4), and protect the genome against selfish genetic elements such as retrotransposons and repetitive sequences (5). Common to all RNA silencing pathways are RNAs 19 to 30 nucleotides (nt) long that specify the target RNAs to be repressed. In RNA interference (RNAi) (6), siRNAs are produced from long exogenous double-stranded RNA (dsRNA). In contrast, ∼22-nt miRNAs are endonucleolytically processed from endogenous RNA polymerase II transcripts. Dicer ribonuclease III (RNase III) enzymes produce both siRNAs and miRNAs. In flies, Dicer-2 (Dcr-2) generates siRNAs, whereas the Dicer-1 (Dcr-1)–Loquacious (Loqs) complex produces miRNAs (711). After their production, small silencing RNAs bind Argonaute proteins to form the functional RNA silencing effector complexes believed to mediate all RNA silencing processes.

Phased sense and antisense siRNAs in vivo. In Drosophila, processive dicing of long dsRNA and the accumulation of sense and antisense siRNAs without reference to the orientation of the target mRNA are hallmarks of RNAi in vitro (12, 13) and in vivo (Fig. 1). We prepared total small RNA from the heads of adult males expressing a dsRNA hairpin (fig. S1A) that silences the white gene via the RNAi pathway (14). white silencing requires Dcr-2 (7), R2D2 (9), and Ago2. siRNAs were detected with a microarray containing TM (melting temperature)–normalized probes, 22 nt long, for all sense and antisense siRNAs that theoretically can be produced by dicing the white exon 3 hairpin (Fig. 1A). Both sense and antisense white siRNAs were detected in wild-type flies but not in dcr-2L811fsX homozygous mutant flies. The Dcr-2–dependent siRNAs were produced with a periodicity of ∼22 nt (Fig. 1B), consistent with the phased processing of the dsRNA hairpin from the end formed by the 6-nt loop predicted to remain after splicing of its intron-containing primary transcript (fig. S1B).

Fig. 1.

Sense and antisense siRNAs accumulate during RNAi in vivo. (A) Microarray analysis of the siRNAs derived from the white exon 3 hairpin RNAi trigger. (B) Fourier transform analysis of the Dcr-2–dependent siRNAs in (A).

Su(Ste) rasiRNAs. Drosophila repeat-associated small interfering RNAs (rasiRNAs) can be distinguished from siRNAs by their longer length, 24 to 29 nt (15, 16). rasiRNAs have been proposed to be diced from long dsRNA triggers (5, 16), such as the ∼50 copies of the bidirectionally transcribed Suppressor of Stellate [Su(Ste)] locus on the Y chromosome (17) (fig. S2B) that in testes silence the ∼200 copies of the protein-coding gene Stellate (Ste) found on the X chromosome.

Microarray analysis of total small RNA isolated from fly testes revealed that Su(Ste) rasiRNAs detectably accumulate only from the antisense strand (Fig. 2A), with little or no phasing (fig. S2A). As expected, Su(Ste) rasiRNAs were not detected in testes from males lacking the Su(Ste) loci (cry1Y) (Fig. 2A). Su(Ste) rasiRNAs were also absent from armitage (armi) mutant testes (Fig. 2A), which fail to silence Ste and do not support RNAi in vitro (18). armi encodes a non–DEAD-box helicase (19) homologous to the Arabidopsis thaliana protein SDE3, which is required for RNA silencing triggered by transgenes and some viruses (20), and depletion by RNAi of the mammalian Armi homolog Mov10 blocks siRNA-directed RNAi in cultured human cells (21). Normal accumulation of Su(Ste) rasiRNA and robust Ste silencing also require the putative helicase Spindle-E (Spn-E), a member of the DExH family of adenosine triphosphatases (ATPases) (16, 22) (Fig. 2B and fig. S2B).

Fig. 2.

rasiRNAs derived from the endogenous silencing trigger Su(Ste). (A) Microarray analysis of Su(Ste) small RNAs in wild-type, cry1Y, and armi mutant testes. (B) Northern analysis of the most abundant Su(Ste) rasiRNA in testes mutant for RNA silencing genes.

The accumulation in vivo of only antisense rasiRNAs from Su(Ste) implies that sense Su(Ste) rasiRNAs either are not produced or are selectively destroyed. Either process would make Ste silencing mechanistically different from RNAi. In support of this view, mutations in the central components of the Drosophila RNAi pathway—dcr-2, r2d2, and ago2—did not diminish Su(Ste) rasiRNA accumulation (Fig. 2B). Deletion of the Su(Ste) silencing trigger (cry1Y) caused a factor of ∼65 increase in Ste mRNA (Fig. 3A), but null or strong hypomorphic mutations in the three key RNAi proteins did not (Fig. 3B).

Fig. 3.

(A to D) RNA expression from selfish genetic elements was measured in homozygous mutants relative to heterozygotes for Ste silencing in testes [(A) and (B)] and for the repeated locus mst40, the LTR retrotransposons roo, mdg1, and gypsy, and the non–LTR retrotransposons I-element and HeT-A in ovaries [(C) and (D)]. (E) RNA expression from selfish genetic elements in dcr-1Q1147X null mutant clones generated by mitotic recombination in the ovary.

Fly Argonaute proteins can be subdivided into the Ago (Ago1 and Ago2) and Piwi [Aubergine (Aub), Piwi, and Ago3] subfamilies. Unlike ago1 and ago2, the aub, piwi, and ago3 mRNAs are enriched in the germline (2325). Aub is required for Ste silencing (16) and Su(Ste) rasiRNA accumulation (26). In aubHN2/aubQC42 trans-heterozygous mutants, Su(Ste) rasiRNAs were not detected by microarray (fig. S2B) or Northern analysis (Fig. 2B), and Su(Ste)-triggered silencing of Ste mRNA was lost completely (Fig. 3B). Even aubHN2/+ heterozygotes accumulated less of the most abundant Su(Ste) rasiRNA than did the wild type (Fig. 2B). That the Ago subfamily protein Ago2 is not required for Ste silencing, whereas the Piwi subfamily protein Aub is essential for it, supports the view that Ste is silenced by a pathway distinct from RNAi. Intriguingly, Su(Ste) rasiRNAs hyperaccumulated in piwi mutant testes, where Ste is silenced normally (Figs. 2B and 3B and fig. S2B).

Mutations in aub also cause an increase in sense, but not antisense, Su(Ste) RNA (16); our results suggest that antisense Su(Ste) rasiRNAs can silence both Ste mRNA and sense Su(Ste) RNA, but that no Su(Ste) rasiRNAs exist that can target the antisense Su(Ste) transcript. Our finding that Su(Ste) rasiRNAs are predominantly or exclusively antisense is essentially in agreement with the results of small RNA cloning experiments, in which four of five Su(Ste) rasiRNAs sequenced were in the antisense orientation (15), but is at odds with earlier reports detecting both sense and antisense Su(Ste) rasiRNAs by non-quantitative Northern hybridization (16).

A third RNA silencing pathway in flies. Is germline RNA silencing of selfish genetic elements generally distinct from the RNAi and miRNA pathways? We examined the expression of a panel of germline-expressed selfish genetic elements—three long terminal repeat (LTR)–containing retrotransposons (roo, mdg1, and gypsy), two non-LTR retrotransposons (I-element and HeT-A, a component of the Drosophila telomere), and a repetitive locus (mst40)—in mutants defective for eight RNA silencing proteins. All selfish genetic elements tested behaved like Ste: Loss of the RNAi proteins Dcr-2, R2D2, or Ago2 had little or no effect on retrotransposon or repetitive element silencing (Fig. 3, C and D). Instead, silencing required the putative helicases Spn-E and Armi plus one or both of the Piwi subfamily Argonaute proteins, Aub and Piwi. Silencing did not require Loqs, the dsRNA-binding protein required to produce miRNAs (Fig. 3, C and D).

The null allele dcr-1Q1147X is homozygous lethal, making it impossible to procure dcr-1 mutant ovaries from dcr-1Q1147X/dcr-1Q1147X adult females (7). Therefore, we generated clones of dcr-1Q1147X/dcr-1Q1147X cells in the ovary by mitotic recombination in flies heterozygous for the dominant female-sterile mutation ovoD1 (27). We measured RNA levels, relative to rp49 mRNA, for three retrotransposons (roo, HeT-A, and mdg1) and one repetitive sequence (mst40) in dcr-1/dcr-1 recombinant ovary clones and in ovoD1/TM3 and dcr-1/ovoD1 nonrecombinant ovaries. The ovoD1 mutation blocks oogenesis at stage 4, after the onset of HeT-A (28, 29) and roo rasiRNA production. Retrotransposon or repetitive sequence transcript abundance was unaltered or decreased in dcr-1/dcr-1 relative to ovoD1/TM3 and dcr-1/ovoD1 controls (Fig. 3E). We conclude that Dcr-1 is dispensable for silencing these selfish genetic elements in the Drosophila female germline.

roo is the most abundant LTR retrotransposon in flies. We analyzed roo silencing in the female germline with the use of microarrays containing 30-nt probes, tiled at 5-nt resolution, for all ∼18,000 possible roo rasiRNAs (Fig. 4, A and B); we corroborated the data at 1-nt resolution for those rasiRNAs derived from LTR sequences (fig. S3A). As observed for Su(Ste) but not for white RNAi, roo rasiRNAs were nonhomogeneously distributed along the roo sequence and accumulated primarily from the antisense strand (Fig. 4A and fig. S3A). In fact, the most abundant sense rasiRNA peak (asterisk in Fig. 4, A and B) corresponded to a set of probes containing 16 contiguous uracil residues, which suggests that these probes nonspecifically detected fragments of the mRNA polyadenylate [poly(A)] tail. Most of the remaining sense peaks were unaltered in armi mutant ovaries, in which roo expression is increased; this result implies that they do not contribute to roo silencing (Figs. 3C and 4A). We detected no phasing in the distribution of roo rasiRNAs (fig. S3B).

Fig. 4.

roo rasiRNAs in ovaries arise mainly from the antisense strand and require the putative helicase Armi for their accumulation. (A and B) Microarray analysis of roo rasiRNAs in (A) wild-type and (B) armi mutant ovaries. The asterisk marks a peak arising from nonspecific hybridization to poly(A)-containing RNAs. (C) Northern analysis for roo rasiRNA [peak 1 in (A)] in ovaries heterozygous (+/–) or homozygous mutant (–/–) for components of the RNA silencing machinery. The open red arrowhead highlights pre–miR-311 accumulation in loqsf00791; the solid red arrowhead highlights loss of mature miR-311. miR-311 is expressed predominantly in the germline.

As for Su(Ste), wild-type accumulation of antisense roo rasiRNA required the putative helicases Armi and Spn-E and the Piwi subfamily Argonaute proteins Piwi and Aub, but not the RNAi proteins Dcr-2, R2D2, and Ago2 (Fig. 4, B and C). Moreover, accumulation of roo rasiRNA was not measurably altered in loqs f00791, an allele that strongly disrupts miRNA production in the female germline (Fig. 4C).

Are roo rasiRNAs not made by dicing? Loss of Dcr-2 or Dcr-1 did not increase retrotransposon or repetitive element expression, which suggests that neither enzyme acts in rasiRNA-directed silencing. Moreover, loss of Dcr-2 had no detectable effect on Su(Ste) rasiRNA in testes or roo rasiRNA in ovaries (Figs. 2B and 4C). We measured the amount of roo rasiRNA and miR-311 in dcr-1/dcr-1 ovary clones generated by mitotic recombination. Comparison of recombinant (dcr-1/dcr-1) and nonrecombinant (ovoD1/TM3 and dcr-1/ovoD1) ovaries by Northern analysis revealed that roo rasiRNA accumulation was unperturbed by the null dcr-1Q1147X mutation (Fig. 5A and fig. S4). Pre–miR-311 increased and miR-311 declined by a factor of ∼3 in the dcr-1/dcr-1 clones (Fig. 5B and fig. S4), consistent with about two-thirds of the tissue corresponding to mitotic dcr-1/dcr-1 recombinant cells. Yet, although most of the tissue lacked dcr-1 function, we observed improved, rather than diminished, silencing for the four selfish genetic elements examined (Fig. 3E). Moreover, the dsRNA-binding protein Loqs, which acts with Dcr-1 to produce miRNAs, was also dispensable for roo rasiRNA production and selfish genetic element silencing (Fig. 3, C and D, and Fig. 4C). Although we cannot exclude the possibility that dcr-1 and dcr-2 can fully substitute for each other in the production of rasiRNA in the ovary, previous biochemical evidence suggests that none of the three RNase III enzymes in flies—Dcr-1, Dcr-2, and Drosha—can cleave long dsRNA into small RNAs 24 to 30 nt long (10, 30, 31).

Fig. 5.

(A) Northern analysis of mitotic recombinant dcr-1Q1147X homozygous mutant ovaries and nonrecombinant controls. roo rasiRNAs were detected with a mixture of five hybridization probes. Arrowheads are as in Fig. 4C. (B) Quantification of the data in (A), normalized to the 2S rRNA loading control.

Animal siRNA and miRNA contain 5′ phosphate and 2′,3′ hydroxy termini (32, 33). We used enzymatic and chemical probing to infer the terminal structure of roo and Su(Ste) rasiRNAs. RNA from ovaries or testes was treated with calf intestinal phosphatase (CIP) or CIP followed by polynucleotide kinase plus ATP. CIP treatment caused roo (Fig. 6A) and Su(Ste) (fig. S5) rasiRNA to migrate more slowly in polyacrylamide gel electrophoresis, consistent with the loss of one or more terminal phosphate groups. Subsequent incubation with polynucleotide kinase and ATP restored the original gel mobility of the rasiRNAs, indicating that they contained a single 5′ or 3′ phosphate before CIP treatment. The roo rasiRNA served as a substrate for ligation of a 23-nt 5′ RNA adapter by T4 RNA ligase, a process that requires a 5′ phosphate; pretreatment with CIP blocked ligation (Fig. 6B), thus establishing that the monophosphate lies at the 5′ end. The rasiRNA must also contain at least one terminal hydroxyl group, because it could be joined by T4 RNA ligase to a preadenylated 17-nt 3′ RNA adapter (Fig. 6B). Notably, the 3′ ligation reaction was less efficient for the roo rasiRNA than for a miRNA in the same reaction (22% versus 50% conversion to ligated product).

Fig. 6.

roo rasiRNAs are modified at their 3′ terminus and associate with Piwi subfamily Argonaute proteins. (A) Chemical and enzymatic probing of roo rasiRNA structure. roo rasiRNAs (peak 1 in Fig. 4A) were detected by Northern hybridization. The membrane was then stripped and reprobed for miR-8. (B) roo rasiRNAs can serve as a 3′ or a 5′ substrate for T4 RNA ligase. Solid arrowheads, 5′ ligation products; open arrowheads, 3′ ligation products. (C and D) roo rasiRNAs associate with myc-tagged Piwi (C) and GFP-tagged Aub protein (D), but not with Ago1. I, input; S, supernatant; B, bound.

RNA from ovaries or testes was reacted with NaIO4, then subjected to β-elimination, to determine whether the rasiRNA had either a single 2′ or 3′ terminal hydroxy group or had terminal hydroxy groups at both the 2′ and 3′ positions, as do animal siRNA and miRNA. Only RNAs containing both 2′ and 3′ hydroxy groups react with NaIO4; β-elimination shortens NaIO4-reacted RNA by one nucleotide, leaving a 3′ monophosphate terminus, which adds one negative charge. Consequently, NaIO4-reacted, β-eliminated RNAs migrate faster in polyacrylamide gel electrophoresis than does the original unreacted RNA. Both roo (Fig. 6A) and Su(Ste) (fig. S5) rasiRNA lack either a 2′ or a 3′ hydroxyl group, because they failed to react with NaIO4; miRNAs in the same samples reacted with NaIO4. Together, our results show that rasiRNAs contain one modified and one unmodified hydroxyl. Because T4 RNA ligase can make both 3′-5′ and 2′-5′ bonds (34), we cannot currently determine the blocked position. Some plant small silencing RNAs contain a 2′-O-methyl modification at their 3′ terminus (34).

rasiRNAs bind Piwi and Aub. Drosophila and mammalian siRNA and miRNA function through members of the Ago subfamily of Argonaute proteins, but Su(Ste) and roo rasiRNAs require at least one member of the Piwi subfamily for their function and accumulation. To determine whether roo rasiRNAs physically associate with Piwi and Aub, we prepared ovary lysate from wildtype flies or transgenic flies expressing either myc-tagged Piwi or green fluorescent protein (GFP)–tagged Aub protein (25, 35); immunoprecipitated them with monoclonal antibodies (mAbs) to myc, GFP, or Ago1; and then analyzed the supernatant and antibody-bound small RNAs by Northern blotting (Fig. 6, C and D, and fig. S6). We analyzed six different roo rasiRNAs. All were associated with Piwi but not with Ago1, the Drosophila Argonaute protein typically associated with miRNAs (36); miR-8 (Fig. 6D and fig. S6C), miR-311, and bantam immunoprecipitated with Ago1 mAb. No rasiRNAs immunoprecipitated with the myc mAb when we used lysate from flies lacking the myc-Piwi transgene (fig. S6B).

Although aub mutant ovaries silenced roo mRNA normally, they showed reduced accumulation of roo rasiRNA relative to aub/+ heterozygotes (Fig. 4C), which suggests that roo rasiRNAs associate with both Piwi and Aub. We analyzed the supernatant and antibody-bound small RNAs after GFP mAb immunoprecipitation of ovary lysate from GFP-Aub transgenic flies and flies lacking the transgene. roo rasiRNA was recovered only when the immunoprecipitation was performed with the GFP mAb in ovary lysate from GFP-Aub transgenic flies (Fig. 6D and fig. S6D). The simplest interpretation of our data is that roo rasiRNAs physically associate with both Piwi and Aub, although it remains possible that the roo rasiRNAs are loaded only into Piwi and that Aub associates with Piwi in a stable complex. The association of roo rasiRNA with both Piwi and Aub suggests that piwi and aub are partially redundant, as does the modest reduction in roo silencing—by a factor of 3.3 ± 0.6 (average ± SD, n = 3)—in piwi but not in aub mutants (Fig. 3C). Alternatively, roo silencing might proceed through Piwi alone, but the two proteins could function in the same pathway to silence selfish genetic elements.

Our data suggest that in flies, rasiRNAs are produced by a mechanism that requires neither Dcr-1 nor Dcr-2, yet the patterns of rasiRNAs that direct roo and Ste silencing are as stereotyped as the distinctive siRNA population generated from the white hairpin by Dcr-2 (Fig. 1A) or the unique miRNA species made from each pre-miRNA by Dcr-1. A key challenge for the future will be to determine what enzyme makes rasiRNAs and what sequence or structural features of the unknown rasiRNA precursor lead to the accumulation of a stereotyped pattern of predominantly antisense rasiRNAs.

Supporting Online Material

Materials and Methods

Figs. S1 to S7

Table S1


References and Notes

View Abstract

Navigate This Article