In many eukaryotes, RNA-dependent RNA polymerases (RdRPs) play key roles in the RNAi pathway. They have been implicated in the recognition and processing of aberrant transcripts triggering the process, and in amplification of the silencing response. We have tested the functions of RdRP genes from the ciliate Paramecium tetraurelia in experimentally induced and endogenous mechanisms of gene silencing. In this organism, RNAi can be triggered either by high-copy, truncated transgenes or by directly feeding cells with double-stranded RNA (dsRNA). Surprisingly, dsRNA-induced silencing depends on the putatively functional RDR1 and RDR2 genes, which are required for the accumulation of both primary siRNAs and a distinct class of small RNAs suggestive of secondary siRNAs. In contrast, a third gene with a highly divergent catalytic domain, RDR3 , is required for siRNA accumulation when RNAi is triggered by truncated transgenes. Our data further implicate RDR3 in the accumulation of previously described endogenous siRNAs and in the regulation of the surface antigen gene family. While only one of these genes is normally expressed in any clonal cell line, the knockdown of RDR3 leads to co-expression of multiple antigens. These results provide evidence for a functional specialization of Paramecium RdRP genes in distinct RNAi pathways operating during vegetative growth.
RNAi is a conserved eukaryotic mechanism of gene regulation which can be triggered by different forms of double-stranded RNA (dsRNA) ( 1 ). Since its discovery, the artificial introduction of dsRNA or the expression of inverted-repeat constructs have become powerful tools to inactivate gene expression. The RNAi machinery typically produces small RNAs (sRNAs), 21–28 nt in length, which have been shown to act in a homology-dependent manner. According to their origin and properties, sRNAs can inhibit gene expression at different levels. Post-transcriptional gene silencing (PTGS) can result from mRNA cleavage targeted by siRNAs, which are processed from long dsRNA precursors, or from translation inhibition. The latter is the most frequent mode of action of miRNAs (microRNAs), which are processed from genome-encoded, stem–loop forming transcripts ( 2 ). sRNAs also mediate transcriptional gene silencing (TGS) in organisms as diverse as plants and fungi through mechanisms involving DNA methylation, as seen in Arabidopsis ( 3–5 ), or histone methylation, as first demonstrated in the case of pericentric heterochromatin formation in budding yeast ( 6 ).
Despite the diversity of effector mechanisms, core enzymes involved in RNAi share a high degree of similarity among different organisms ( 1 , 7 ). siRNAs are usually cleaved from dsRNA by RNAse-III-type endonucleases of the Dicer family, yielding duplexes with characteristic 5′-monophosphate ends and 2 nt 3′ overhangs ( 8 ). Such is the case of the primary siRNAs that are cleaved from exogenously introduced dsRNA or artificially expressed hairpins in many systems. Stoichiometric considerations revealed that a few dsRNA molecules per cell are enough to cause RNAi phenotypes ( 9 ), leading to the discovery that the silencing response can be amplified through the action of RNA-dependent RNA polymerases (RdRPs) on the targeted mRNA, which results in the synthesis of secondary siRNAs. cRNA synthesis allows siRNA formation and homology-dependent silencing to spread outside of the initial inducer sequence, a process called transitivity ( 10 ). However, the precise mechanisms involved in RdRP-mediated synthesis of secondary siRNAs appear to vary in different organisms.
In Arabidopsis , both the 5′ and the 3′ fragments of an mRNA that has been targeted in its central region by primary siRNAs can become substrates for RdRP activity, generating dsRNAs that are then cut into secondary siRNAs by Dicer enzymes ( 11 ). While the appearance of secondary siRNAs upstream of the targeted region (3′- to 5′-transitivity) is thought to result from the priming of RdRP activity by the primary siRNAs ( 12 ), 5′- to 3′-transitivity appears to involve unprimed RdRP activity initiated at the 3′-end of the 3′ cleavage fragment ( 11 ). A different mechanism has been described in Caenorhabditis elegans . Targeting an mRNA with primary siRNAs makes it a substrate for an unprimed RdRP activity which directly synthesizes short antisense RNAs, predominantly from the targeted region but also from the 5′ and 3′ regions of the mRNA ( 13 , 14 ). These Dicer-independent secondary siRNAs have a 5′-triphosphate end and are responsible for a potent slicer activity ( 15 ). Although their length may be controlled by other factors, in vitro studies of a purified RdRP from Neurospora revealed two different reactions, one of which synthesizes 9–21-nt RNAs from the entire length of a single-stranded template ( 16 ). In view of these mechanistic differences, the common feature that will be used here as a definition of secondary siRNAs is the fact that they are synthesized by an RdRP from the targeted mRNA, rather than processed from the RNA molecule that initially triggers silencing.
Two entirely distinct RNAi pathways have been described in the ciliate Paramecium tetraurelia ( 17 ). In addition to the meiosis-specific scnRNA pathway, which is involved in epigenetic regulation of genome rearrangements during early development, a constitutively expressed pathway is responsible for homology-dependent gene silencing during the vegetative phase of the life cycle. The latter can be induced experimentally either by feeding cells with an Escherichia coli strain producing dsRNA ( 18 ), as observed in C. elegans ( 19 , 20 ), or by microinjecting 3′-truncated transgenes at high copy numbers into the somatic macronucleus, which leads to the production of aberrantly sized transcripts ( 21 , 22 ). Previous studies showed that both methods result in the accumulation of ∼23-nt siRNAs which depend on the Dicer protein Dcr1 ( 17 , 23 , 24 ). The cloning and sequencing of ∼23-nt siRNAs associated with dsRNA feeding suggested the existence of two different subclasses. One appeared to represent primary Dcr1 cleavage products of the dsRNA trigger since it contained sRNAs matching the entire length of that molecule on both strands, including sequences derived from the plasmid vector used for dsRNA production in E. coli . A more abundant subclass of sRNAs with a 5′-UTR bias, containing a short untemplated polyA stretch at the 3′-end, was strictly antisense to the targeted mRNA and did not include vector-derived sequences, suggesting that it represents RdRP-dependent secondary siRNAs ( 17 ).
Although previous work strongly suggests that ∼23-nt siRNAs are responsible for reducing the steady state amount of homologous mRNAs, only limited experimental evidence is available as to whether this occurs through PTGS or TGS in P. tetraurelia . One study used run-on assays to examine transgene-induced silencing of the endogenous T4a gene and concluded that the effect was post-transcriptional ( 22 ). In the case of dsRNA-induced silencing, another study reported northern blot evidence for the cleavage of Upf1 mRNA in the targeted region ( 25 ). Furthermore, the rapid disappearance of silencing phenotypes obtained by this method when dsRNA-producing bacteria are replaced with the normal food bacteria indicates that no lasting modification is set on the gene itself, suggesting that silencing only occurs through PTGS.
The aim of the present study was to investigate the involvement of four Paramecium RdRP genes in dsRNA-induced and in transgene-induced RNAi during vegetative growth. By knocking down the function of each gene in cell lines silenced for reporter genes, we show that two putatively functional RdRP genes are required for dsRNA-induced silencing. These genes are not only necessary for the accumulation of putative secondary siRNAs, but also, unexpectedly, for the accumulation of primary siRNAs processed from the dsRNA trigger. Similar tests show that a third RdRP gene, with a highly divergent catalytic domain, is involved in transgene-induced silencing. The same gene is also necessary for the accumulation of a cluster of endogenous siRNAs mapping in an intergenic region, and for the mutual exclusion that characterizes the expression of surface antigen genes.
MATERIALS AND METHODS
Paramecium strains and cultivation
Experiments were carried out with P. tetraurelia stock 51. Cells were grown at 27°C in wheat grass powder (Pines International Co., Lawrence, KS, USA) infusion medium bacterized with Klebsiella pneumoniae the day before use (unless otherwise stated) and supplemented with 0.8 µg/ml β-sitosterol.
To induce silencing by dsRNA feeding, fragments of the coding region were cloned into the plasmid L4440 and transformed into the RNaseIII-deficient E. coli strain HT115DE3. Positions in the coding sequences (cdss) were as follows: RDR1 : 2423-3162 of accession number GSPATG00024768001 (ParameciumDB, http://paramecium.cgm.cnrs-gif.fr ); RDR2 : 2422-3122 of GSPATG00036857001; RDR3 : 1789-2462 or 548-1315 of GSPATG00006401001; RDR4 : 1750-2459 of GSPATG00018564001; ND169 : 1450-1860 of GSPATG00008337001; ICL7a : 1-580 of GSPATG00021610001 and A 51 : 380-873 of M65163. To induce RNAi by injection of a ND169 transgene, a plasmid (pTI+) containing the entire ND169 cds and 3′-UTR downstream of a bidirectional constitutive promoter was modified by removing the 3′ cds and 3′-UTR downstream from position 1653 of the cds (pTI− construct). GFP on the opposite site of the promoter served as a control for the presence in the macronucleus.
Induction of RNAi and analysis of phenotypes
Transgene-induced silencing was carried out by microinjection of the BglI-linearized pTI− (position 1.593). Paramecium cell preparation and microinjection was carried out as described ( 24 ). Gfp fluorescence level of injected cell lines was observed to estimate transgene copy numbers.
DsRNA-induced silencing was performed by feeding dsRNA-producing E. coli as described ( 26 ). DsRNA-induced co-silencing of two genes was achieved by providing two types of OD-synchronized feeding bacteria in a 1:1 ratio. Specificity of silencing was verified by semi-quantitative RT–PCRs using four different dilutions of cDNA.
Ability of trichocyst discharge was tested by supplementing small amounts of the culture with saturated picric acid (1:2). Expressed surface antigens were determined by adding specific antisera (anti-A 51 , B 51 , C 51 , D 51 , E 51 , G 51 , H 51 , I 51 , J 51 , N 51 or Q 51 , 1:200) to small volumes of the culture (∼50 cells). Immobilization of cells expressing the corresponding antigen was completed within 30 min at room temperature. To determine the percentage of cells expressing one or several surface antigens, immobilization reactions were carried out with one antiserum. Then a second antiserum was added to test for immobilization of remaining cells. Additionally, each antiserum was used separately as a cross-check.
Average division rate was determined from single cells cultured in 200 µl silencing medium ( E. coli producing dsRNA) over 5 days. Cells were counted every 24 h and one cell was transferred to fresh silencing medium.
Total RNA extraction, northern blot analysis and real-time RT–PCR
RNA was extracted using TRIzol® Reagent (Invitrogen, Karlsruhe, Germany), modified by the addition of glass beads. Prior to harvesting, cells were transferred from silencing medium to medium supplemented with K. pneumoniae and 0.8 µg/ml β-sitosterol for 30 min to allow for the complete processing of provided dsRNA. Without this procedure, a smear of (partially) degraded dsRNA makes the identification of single RNA species impossible.
For sRNA northern blots, 20 µg of denatured total RNA were separated on 15% polyacrylamide (acrylamide:bis 19:1)-7 M urea gels, transferred to Hybond N + membranes (GE/Amersham, Braunschweig, Germany) in 20× SSC by vacuum and UV cross-linked. For standard northern blots, RNA was separated on formaldehyde 1.2% agarose gels (20 µg per lane) and blotted in 10 × SSC. Size markers used for valuation of siRNA migration were 17-, 21- and 25-nt unphosphorylated ssRNA oligonucleotides (microRNA Marker, NEB, Frankfurt am Main, Germany) which were detected by specific probes.
Hybridizations of siRNAs were carried out at 42°C in 1× church buffer (7% SDS, 0.25 M sodium phosphate, 1% SDS, 1 mM EDTA, pH 7.2). Membranes were washed in 2× SSC and 0.1% SDS for 5 and 30 min and subsequently with 0.2× SSC and 0.1% SDS for 5 and 30 min at the same temperature. Riboprobes were hybridized at 60°C in 6× SSC, 2× Denhardt’s solution and 0.1% SDS. Membranes were washed at 60°C as described for siRNA northern blots. Labelled membranes were then exposed to phosphor image plates.
Real-time RT–PCR was carried out as described ( 26 ). Briefly, total TRIzol ® -isolated RNA was additionally purified with the RNeasy Micro Kit (Qiagen, Hilden, Germany). RNA (500 ng) was reverse transcribed using an oligo-dT primer. cDNA was amplified with the QuantiTect TM SYBR® Green PCR Kit (Qiagen, Hilden, Germany). Data from surface antigen expression were set in relation to Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) which was found to be constantly expressed during vegetative growth and RDR silencing.
Probes, riboprobes and oligonucleotide probes
Double-stranded probes for siRNA detection in dsRNA-induced silencing spanned the silencing fragments or the polylinker between the T7 promoters. Transgene-siRNAs were hybridized with a double-stranded probe spanning the entire ND169 cds or the plasmid part adjacent to the shortened ND169 3′-end. Probes were generated by random priming with [α- 32 P] dCTP (3000 Ci/mmol). For strand-specific detection of siRNAs, two adjacent oligonucleotide probes were 5′-end labelled with T4 polynucleotide kinase and [γ- 32 P] ATP (3000 Ci/mmol) (for siRNAs of ND169 : position 1580–1629 and 1631–1679; for siRNAs from the L4440 polylinker: position 80–130 and 165–215; for endogenous siRNAs from scaffold 22, ParameciumDB: position 564980–565029 and 565030–565079). Long transcripts of the ND169 transgene and endogenous mRNA were detected with sense and antisense riboprobes synthesized in vitro from the entire coding region of ND169 cloned into the pGEM-T vector (Promega, Mannheim, Germany). In vitro transcription was carried out using the MAXIscript® T7/SP6 Kit (Applied Biosystems/Ambion, Austin, TX, USA) and [α- 32 P] UTP (800 Ci/mmol).
Biochemical analysis of siRNA
A 22-nt unphosphorylated RNA oligonucleotide and a portion of the same control oligonucleotide, which was 5′-monophosphorylated using T4 polynucleotide kinase were added to total RNA of each individual reaction as a control. Removal of 5′ phosphates was carried out by treating 20 µg of total RNA with 5 U calf intestinal alkaline phosphatase (CIP) (New England Biolabs, Frankfurt am Main, Germany) for 1 h at 37°C. For verification of 5′-monophosphates, 20 µg of total RNA were incubated with 1 U Terminator5′-monophosphate-dependent exonuclease (Epicentre, Madison, WI, USA) for 1 h at 30°C. For characterization of 3′-ends of siRNA, periodate treatment andβ-elimination were carried out with 20 µg of total RNA as described ( 27 ). A 23-nt 5′-triphosphorylated control RNA oligonucleotide [ in vitro transcribed from the SmaI digested vector pGEM-3Zf (Promega, Mannheim, Germany)] was used for control treatment to check Terminator exonuclease specificity. All samples were precipitated and dissolved in DEPC-treated water before loading on a 15% (19:1) polyacrylamide gel.
SDS-PAGE, western blot and immunofluorescence staining
Surface antigens were isolated using a standard salt–alcohol extraction procedure ( 28 ). Gradient SDS polyacrylamide gels (6–15%) with 3% stacking gels were used ( 29 ). Proteins were electro-blotted to nitrocellulose membranes (Schleicher and Schuell, Dassel, Germany) in 15.6 mM Tris–HCl, 120 mM glycine, 1% SDS and 20% methanol. After staining in 0.1% Ponceau S, membranes were blocked in 5% dried milk in TBS (10 mM Tris–HCl, 150 mM NaCl, pH 7.6) and decorated with primary sera and secondary antibodies conjugated with alkaline phosphatase (Promega, Mannheim, Germany). Primary sera were the anti-A 51 , anti-B 51 , anti-D 51 and anti-H 51 polyclonal sera (kind gift of J. Forney). Indirect immunofluorescence staining was carried out as described ( 30 ) using the Y4 monoclonal mouse antibody ( 31 ) (kind gift of Y. Capdeville) and polyclonal anti-D 51 serum.
The Paramecium genome reveals four RdRP candidate genes
Four genes encoding putative RdRPs were identified in the Paramecium macronuclear genome, two of which ( RDR1 and RDR4 ) are paralogs derived from the intermediate whole genome duplication ( 32 ). Figure 1 shows an alignment of the predicted proteins with functional RdRPs from other eukaryotes over a portion of the catalytic domain. Many of the highly conserved amino acids are present in Rdr1 and Rdr2, including the DLDGD motif in which the third aspartatic acid residue was shown to be essential for catalytic activity in other organisms ( 16 , 33 , 34 ). In contrast, many conserved amino acids are missing in Rdr3 and Rdr4. The latter shows a seven-residue gap in the region of the DLDGD motif, whereas Rdr3 has three aspartic acid residues with different spacings. A phylogenetic analysis based on an alignment of the entire catalytic domains indicates that Rdr3 and Rdr4 have greatly diverged from Rdr1 and Rdr2, which are most closely related to the catalytically active Rdr1 protein from the ciliate Tetrahymena thermophila ( 35 ) ( Supplementary Figure S1 ). Expression levels of the RdRP genes during vegetative growth were determined by real-time PCR. All four genes appear to be expressed at low levels ( Table 1 ) and no significant upregulation of any of them was observed when cells were submitted to dsRNA feeding (data not shown). All predicted introns were verified by cDNA sequencing. This further revealed two non-predicted and inefficiently spliced introns in the RDR4 gene, the splicing of which disrupts the integrity of the open reading frame. Thus, Rdr1 and Rdr2 may well be catalytically active, but it is difficult to make any prediction about Rdr3; on the other hand, RDR4 may be a non-functional pseudogene, which would be consistent with its extremely low mRNA level ( Table 1 ) and with the fact that no phenotype was observed upon its silencing, as described below.
|Relative transcript level||2.1 ± 0.8||0.8 ± 0.2||0.6 ± 0.2||0.2 ± 0.2||100 ± 0.0|
|Relative transcript level||2.1 ± 0.8||0.8 ± 0.2||0.6 ± 0.2||0.2 ± 0.2||100 ± 0.0|
Steady state levels were determined in vegetatively growing cells using real-time RT–PCR and were quantified in relation to the transcript level of the housekeeping gene GAPDH .
Rdr1 and Rdr2 are involved in dsRNA-induced silencing
To test whether Paramecium RdRPs are involved in dsRNA-induced silencing, we compared the efficiency of silencing of a reporter gene when each of the RDR genes, or an unrelated gene as a control, was co-silenced by dsRNA feeding. Although RNAi-mediated knockdown of genes involved in RNAi is a self-defeating process with variable outcomes, this approach has been used successfully in several genome-wide RNAi screens ( 36–39 ). The efficiency of the feeding procedure was verified by the accumulation of RDR -specific siRNAs and by the reduction of RDR mRNA levels ( Supplementary Figure S2 ).
The chosen reporter genes were ND169 , whose product is involved in membrane fusion during trichocyst ejection, and the A 51 gene, encoding a frequently expressed surface antigen. Figure 2 shows the silencing efficiency of the reporter genes, as determined by phenotypic analysis. Feeding cells exclusively with bacteria producing ND169 or A 51 dsRNA resulted in close to 100% of cells in the cultures showing the expected phenotypes. When cells were fed a mix of these two bacterial strains in equal amounts, or a mix of each of them with bacteria producing dsRNA homologous to ICL7a (encoding a non-essential component of the cytoskeleton), phenotypes were only observed in 60–85% of cells, which is likely due to the lower amount of each ingested dsRNA. Similar results were obtained when cells were fed a mix of each reporter dsRNA with either RDR3 or RDR4 dsRNA. In contrast, feeding cells a mix of each reporter dsRNA and either RDR1 or RDR2 dsRNA resulted in a significantly lower fraction of cells showing the expected phenotypes (10–15%), which surprisingly indicates that RDR1 and RDR2 are involved in dsRNA-induced silencing. To determine whether the reduced silencing efficiency correlates with a lower amount of silencing-associated siRNAs, total RNA was extracted from these cultures and analysed on northern blots.
Two classes of dsRNA-induced siRNAs depend on Rdr1 and Rdr2
Hybridization with gene-specific probes readily revealed silencing-associated siRNAs for both reporters. In these northern blots, some lanes showed a ∼1-nt ladder background signal. This likely resulted from ongoing degradation of ingested bacterial dsRNA because it was even more apparent when the cultures were not subjected to a brief starvation period to allow complete digestion of food vacuoles before RNA isolation (see ‘Materials and Methods’ section). Figure 2 C shows that short RNAs associated with A 51 silencing migrated with an apparent size between 22 and 23 nt. Considering that the RNA oligonucleotides used as markers were 5′-OH (and migrated ∼0.5-nt faster than their 5′-monophosphate counterparts in the type of gel used, see below), these A 51 siRNAs appear to have the same size as previously described P. tetraurelia siRNAs ( 17 ) and will be referred to as ∼23-nt siRNAs. These were also present when the A 51 surface antigen gene was co-silenced with the other reporter ( ND169 ) or with RDR3 or RDR4 , but, consistent with the phenotypic analysis, A 51 siRNAs were clearly less abundant in cultures that were also fed RDR1 or RDR2 dsRNA.
SiRNAs associated with ND169 silencing were also easily detected in double-silencing tests involving RDR3 , RDR4 or ICL7a , but were almost completely absent in those involving RDR1 or RDR2 . However, ND169 siRNAs migrated as a major band with an apparent size between 21 and 22 nt, i.e. ∼1-nt faster than A 51 siRNAs; taking into account the unphosphorylated state of markers, this band will be referred to as ∼22-nt siRNAs. A weaker signal was also observed ∼1-nt above the main band (this is more clearly seen with strand-specific probes in Figure 3 ). To determine whether the ∼23-nt signal corresponded to a distinct subpopulation of siRNAs, the same blot was stripped and rehybridized with a probe specific for the plasmid vector sequences present in the dsRNA molecule (polylinker sequences on both sides of the ND169 insert in the recombinant plasmid, in between the convergent T7 promoters; see Figure 2 F). Strikingly, the vector-specific probe only revealed ∼23-nt siRNAs ( Figure 2 E). Since these sequences are absent from the macronuclear genome, the ∼23-nt siRNAs must be processed from the ingested dsRNA. Nevertheless, Figure 2 E clearly shows that these siRNAs, like the ∼22-nt ND169 siRNAs, were almost completely missing when either RDR1 or RDR2 was co-silenced with ND169 .
The siRNAs associated with ND169 silencing were further characterized by hybridization with strand-specific oligonucleotide probes. Interestingly, the ∼22-nt siRNAs could only be detected with a sense probe of the ND169 sequence used for dsRNA production, indicating that they were strictly antisense ( Figure 3 A). In contrast, the ∼23-nt siRNAs, although much less abundant, could be detected with ND169 probes from both strands. This is more clearly seen in RNA samples from cultures that were only fed ND169 dsRNA, where both the ∼22-nt and the ∼23-nt siRNAs are more abundant than in double-silencing cultures ( Figure 3 A, ND lane on the right). Similarly, the ∼23-nt siRNAs homologous to vector sequences were derived from both strands of the dsRNA ( Figure 3 B).
The absolute strand bias of the ∼22-nt siRNAs, together with the fact that they did not contain any vector sequence, strongly suggests that they are not processed from the symmetrical dsRNA ingested, but represent secondary siRNAs synthesized by an RdRP activity from the endogenous ND169 mRNA. Conversely, the much less abundant ∼23-nt siRNAs, which can be found on both strands of the entire dsRNA including vector sequences, likely represent primary Dcr1 cleavage products. This further suggests that dsRNA-induced silencing of the A 51 gene only generates primary siRNAs, a difference with ND169 silencing which may be explained by the special regulation of surface antigen genes (see ‘Discussion’ section). The results of our functional tests of RDR genes thus unexpectedly indicate that both primary and secondary siRNAs depend on the RDR1 and RDR2 genes.
Different biochemical tests were then carried out to characterize the 5′- and 3′-ends of both classes of siRNAs ( Figure 3 C). As controls for the efficiency of reactions, small amounts of 5′-monophosphate and 5′-OH versions of a synthetic RNA oligonucleotide, both with a 3′-OH end, were added to total RNA from a culture that was fed ND169 dsRNA. The sample was first treated with phosphatase (CIP), and the same northern blot was then successively hybridized with probes specific for the polylinker part of the dsRNA, for the antisense strand of the ND169 sequence and for the synthetic oligonucleotide. CIP treatment resulted in an upward shift of the 5′-monophosphate control to its 5′-OH position (∼0.5-nt slower). The same shift was observed for the ∼23-nt siRNAs with both the polylinker and the ND169 probes, and for the ∼22-nt siRNAs with the ND169 probe. Both classes of siRNAs were also completely degraded after treatment with the 5′-monophosphate-specific exonuclease Terminator, as was the 5′-monophosphate version, but not the 5′-OH version, of the control. Thus, both types of siRNAs appear to have 5′-monophosphate ends.
To test for a possible 2′- O -methylation on the3′-terminal ribose, which is mediated by the Hen1 methyltransferase in diverse classes of short RNAs ( 40 ), we used periodate treatment and subsequent β-elimination, which will cleave the last nucleotide only if it is both 2′-OH and 3′-OH. As can be seen with the control, this results in a downward shift of ∼1.5 nt because of the retention of the 3′ phosphate. The same shift was observed for both classes of siRNAs and no resistant fraction could be detected, indicating that the 3′-terminal ribose was unmethylated.
Transgene-induced RNAi depends on Rdr3
To test the involvement of RDR genes in transgene-induced silencing, we injected an ND169 truncated construct (pTI−) lacking the last 229 bp of the cds and the 3′-UTR into the macronucleus ( Figure 4 A). This resulted in a trichocyst non-discharge phenotype in ∼100% of the cells, indicating that the endogenous ND169 gene was silenced. Wild-type cultures and cells injected with the pTI+ transgene, containing the untruncated ND169 gene, showed wild-type trichocyst discharge. The silencing of each RDR gene was then induced by dsRNA feeding as in the above experiments. As shown in Figure 4 B, the pTI− phenotype was neither altered in the ICL7a and A 51 dsRNA feeding controls nor by the feeding of RDR1 , RDR2 and RDR4 dsRNAs. In contrast, the feeding of RDR3 dsRNA had severe effects, suppressing the silenced phenotype in ∼80% of cells. Northern blots of total RNA samples revealed ND169 transgene-induced siRNAs migrating at ∼23-nt in cultures without dsRNA feeding and with A 51 dsRNA feeding as a control ( Figure 4 C). These siRNAs were not detectable in wild-type cells or pTI+ injected cells and were clearly reduced in pTI− cultures when RDR3 was silenced, suggesting that their accumulation is responsible for silencing of the endogenous ND169 gene. Hybridization of the same blot with a probe specific for the plasmid sequences located between the truncated ND169 gene and the end of the injected linear molecule revealed siRNAs migrating slightly slower (∼0.5 nt) than those detected with the ND169 -specific probe; these were also clearly reduced in RDR3 silencing samples ( Figure 4 D). All attempts to test the strand specificity of transgene-induced ND169 siRNAs with oligonucleotide probes failed. We were only able to detect them by hybridization with probes of the entire ND169 gene, which may be due to their low density or uneven distribution over the length of the construct.
Biochemical analysis of the 5′- and 3′-ends was carried out as described above. The bulk of transgene-induced siRNAs showed a ∼0.5-nt slower migration after CIP treatment and disappeared after Terminator treatment, indicating that the vast majority (if not all) of them have a 5′-monophosphate end ( Figure 4 E). The residual signal seen at ∼23 nt on the blot may simply be part of the 1-nt ladder background. In contrast to dsRNA-induced siRNAs, the migration of transgene-induced siRNA was unaffected after periodate treatment and β-elimination. Successful β-elimination was shown by the shift of the control oligonucleotide in the same sample. Although the treatment resulted in a reduced signal for both transgene-induced siRNAs and the control oligonucleotide, no shortened version of the siRNAs migrating ∼1.5-nt faster was detectable even after a longer exposure. We therefore conclude that transgene-induced siRNAs carry a modification on either the 2′ or the 3′ of the 3′-terminal ribose.
Although the same reporter gene ( ND169 ) was used here to study dsRNA- and transgene-induced silencing, ∼22-nt siRNAs were clearly not present among transgene-induced siRNAs. Our data thus indicate differences between the two methods used to induce RNAi in Paramecium , both in terms of RdRP requirements and in the associated siRNAs, suggesting that the mechanisms involved are at least partially distinct. This conclusion is further supported by an experiment in which efficient silencing of the A 51 gene was induced by dsRNA feeding in a pTI− injected culture in which transgene-induced silencing of the endogenous ND169 gene was inhibited by RDR3 silencing ( Figure 4 F). Simultaneous RDR3 and A 51 dsRNA feeding resulted in a loss of the ND169 silencing phenotype comparable with single RDR3 dsRNA feeding and, in contrast to RDR1 or RDR2 silencing, still led to efficient silencing of A 51 .
Transgene-induced siRNAs are likely derived from aberrant transcripts of the truncated transgene. To test whether Rdr3 is involved in the synthesis of long antisense cRNA, a northern blot of total RNA samples from a pTI− injected cell line, with or without dsRNA-induced silencing of RDR3 , was hybridized with sense and antisense ND169 probes ( Figure 4 G). While the antisense probe only detected the ∼1.8-kb ND169 mRNA in wild-type cells and, in much greater amounts, in pTI+ injected cells, pTI− samples contained a ∼3.2-kb sense transcript, which corresponds to the length between the promoter and the end of the linear transgene. The sense probe revealed a similar-sized antisense transcript specifically in pTI− samples. However, neither the sense nor the antisense aberrant transcripts were decreased in abundance upon dsRNA-induced silencing of RDR3 or A 51 . The antisense transcript of the transgene may arise from promiscuous transcription of the transgene as such full-length RNA was previously also observed for promoter and terminatorless sequences injected at high copy numbers into the Paramecium macronucleus ( 22 ). We conclude that the role of Rdr3 in the accumulation of transgene-induced siRNAs is likely downstream of the synthesis of sense and antisense aberrant transcripts.
Rdr3-dependent endogenous siRNAs
During RdRP dsRNA feeding experiments, we noticed reduced growth rates of RDR3 -silenced cultures compared with others. Precise measurements after long-term dsRNA feeding revealed a very significantly reduced division rate of 1.4 ± 0.3 division/day, compared with 3.6 ± 0.6 division/day for cells that were fed RDR1 , RDR2 , RDR4 , ND169 , A 51 , ICL7a or GFP dsRNAs ( Figure 5 A). However, this effect became obvious only after 4 days of continuous silencing, raising the possibility that RDR3 silencing results in the progressive deregulation of endogenous genes important for vegetative growth. Although RNAi-mediated regulation of endogenous genes has not been reported in Paramecium so far, a small-scale sRNA sequencing study identified a cluster of endogenous siRNAs in a region between two convergent genes on scaffold 22 ( 17 ). We therefore tested whether their accumulation depends on Rdr3. Figure 5 B and C show that silencing of RDR3 resulted in a decreased abundance of these siRNAs, in comparison with an ICL7a silencing control, after 5 days of continuous silencing, i.e. at about the same time as the growth rate defect became clearly apparent. The endogenous siRNAs migrated at ∼23 nt, showing two distinct bands separated by <0.5 nt. Although sequencing had only revealed siRNAs from the top strand ( Figure 5 B), a weaker signal was also detected on the complementary strand, which likewise decreased after RDR3 silencing ( Figure 5 C).
Biochemical analysis of endogenous siRNAs was carried out as described above. After treatment with CIP, the blot indicated a ∼0.5-nt slower migration of the top strand siRNAs ( Figure 5 D); the bottom strand siRNAs were too difficult to detect in these experiments. After treatment with Terminator, the signal was greatly reduced; however, some residual amount of sRNAs could be seen. Since the control oligo was completely digested, this suggests that a minor fraction of cluster 22 siRNAs may not have 5′-monophosphorylated ends. Moreover, the signal remaining after Terminator treatment corresponded to the upper band of the doublet seen with native siRNAs ( Figure 5 B), suggesting that the two bands may represent 5′ modification isoforms. These endogenous siRNAs were sensitive to periodate treatment and β-elimination, as indicated by the appearance of a ∼1.5-nt faster migrating band ( Figure 5 D). Because the treatment resulted in a decrease of the total signal, we cannot rule out the existence of a minor fraction of insensitive molecules. Nevertheless, these endogenous siRNAs clearly differ from Rdr3-dependent transgene-induced siRNAs, for which no evidence for sensitive molecules was obtained.
Exclusive expression of variable surface antigens is controlled by Rdr3
In addition to the loss of endogenous siRNAs, we found that the expression of surface antigen genes was altered during silencing of RDR3 . This multigene family encodes alternative surface proteins whose expression is mutually exclusive: only one gene is normally expressed at a time. The expressed gene defines the serotype of the cell, which is stably maintained during vegetative divisions; this can easily be identified by the immobilization reaction observed upon addition of specific antiserum (see ‘Materials and Methods’ section). RDR3 dsRNA feeding resulted in a destabilization of the serotype in vegetative clones. Figure 6 A shows the percentage of cells reacting to specific sera during RDR3 silencing, starting with a pure A 51 -expressing culture. After 48 h, >80% of cells did not react to anti-A 51 serum any more, but an increasing percentage of cells were immobilized by anti-D 51 serum. In the following days, a growing fraction of cells did not react to any sera tested. This fraction reached 100% after 8 days of RDR3 silencing, which would seem to indicate that no antigen was expressed.
We therefore directly examined surface antigen mRNAs and proteins. Besides A 51 , the most frequently expressed antigens are B 51 and D 51 . Real-time RT–PCR revealed a strong upregulation of steady state mRNA levels for the B 51 and D 51 genes after 24 h of RDR3 silencing ( Figure 6 B). Surprisingly, the A 51 mRNA was still present at relatively high levels during the entire experiment, even when cells were no longer immobilized by anti-A 51 serum. Thus, different genes seemed to be co-transcribed. To test whether the corresponding proteins were produced, surface antigens were isolated from RDR3 -silenced and control cultures and run on SDS–polyacrylamide gels. Silver staining revealed different proteins in the size range of surface antigens, ∼200 kDa ( Figure 6 C). In contrast to RDR3 -silenced cultures, the control culture fed with ICL7a dsRNA showed a single band. Western blots confirmed that this was A 51 , and further showed that A 51 , B 51 , D 51 and H 51 were all produced in RDR3 -silenced cultures ( Figure 6 D). The presence of multiple antigens at the surface of individual cells was revealed by indirect immunofluorescence staining, which showed an even distribution of the A 51 and D 51 antigens on the entire cell surface ( Figure 6 E). This differs strikingly from reports of transient co-occurrence of two surface antigens at the surface during serotype switching, which showed distinct distributions of the old and new antigens on the cilia and cell cortex ( 41 , 42 ). The simultaneous presence of many antigens at the surface could possibly explain the lack of immobilization by corresponding antisera. Indeed, this reaction requires cross-linking of antigens by antibody molecules ( 43 ), which could become inefficient if the concentration of each antigen on the ciliary membrane is too low. This is the first report of a disruption of the mutual exclusion that characterizes the regulation of surface antigen gene expression in Paramecium .
RdRPs are not as widely distributed among eukaryotes as other RNAi components. They are found neither in mammals nor in the unicellular organisms Trypanosoma and Chlamydomonas ( 7 ). Other organisms, such as Arabidopsis thaliana and C. elegans , have several RdRP homologs with specialized functions in different silencing mechanisms ( 11 , 44 ). In contrast, the genomes of Schizosaccharomyces pombe and T. thermophila only contain a single RdRP gene ( 35 , 45 ). Surprisingly, we identified four putative RdRPs in the genome of the ciliate P. tetraurelia . The closest homolog of Rdr1 and Rdr2 in other organisms is T. thermophila’s Rdr1 ( 35 ). Together with the close relationship of Paramecium ’s Dcr1 and Tetrahymena ’s Dcr2 ( 17 ), this suggests that RNAi mechanisms are at least partially conserved in these two ciliates. The highly divergent catalytic domain of the Rdr3 protein raises the question of its possible function. In Paramecium , homology-dependent gene silencing can be triggered by dsRNA feeding or by injection of 3′-truncated transgenes ( 18 , 21 , 22 ). In this study, we used northern blots to characterize sRNAs associated with both pathways and also analysed their RdRP requirements. We used dsRNA feeding to knock down each RdRP gene, and the results appear to indicate that different RdRPs are required for these two pathways. Formally, we cannot exclude the involvement of more RdRPs in each pathway because of the unpredictable efficiency of recursive RNAi. However, other results do indicate that dsRNA- and transgene-induced RNAi are different mechanisms. This is based on the appearance of different classes of sRNAs with different biochemical properties ( Table 2 ). Similarly, it has recently been shown that different sets of Piwi proteins are involved in these two mechanisms (K. Bouhouche et al. , submitted for publication). However, the genetic requirements of the two mechanisms are not entirely different since Dcr1 was found to be involved in both pathways (17, S.M. and M.S., unpublished results).
|dsRNA primary||dsRNA secondary||Transgene induced||endogeneous|
|Standard specifity||s/as||as||n.d. a||s/as|
|dsRNA primary||dsRNA secondary||Transgene induced||endogeneous|
|Standard specifity||s/as||as||n.d. a||s/as|
a Experimental attempts to show strand specificity of transgene-induced siRNAs failed, see text for details. s, sense; as, antisense.
DsRNA feeding triggers Rdr1- and Rdr2-dependent accumulation of primary and secondary siRNAs
We found that two distinct classes of siRNAs with apparent sizes of ∼23 and ∼22 nt are associated with dsRNA-induced silencing and that both classes depend on Rdr1 and Rdr2. The ∼22-nt class appears to represent secondary siRNAs, because these molecules are strictly antisense to the targeted mRNA and are not detectable in the vector part of the ingested dsRNA, suggesting that they are synthesized using the mRNA as a template. The absence of this class when dsRNA feeding targeted the A 51 surface antigen gene is consistent with this hypothesis and may be due to discontinued production of the mRNA: silencing of a member of the antigen family triggers a serotype shift in Paramecium ( 26 ), which is normally accompanied by transcriptional or co-transcriptional silencing of the previously expressed gene (see below). This would perfectly explain why only the ∼23-nt siRNAs can be detected after dsRNA-induced silencing of A 51 . Our results further indicate that secondary siRNAs are more abundant than primary ones, consistent with the results of a small-scale sequencing study in Paramecium ( 17 ) and with studies of secondary siRNAs in C. elegans ( 46 ). However, Paramecium secondary siRNAs are monophosphorylated, which stands in contrast to the triphosphorylated 5′-end of sRNAs resulting from unprimed RdRP activity in C. elegans ( 13 , 14 ). It is still possible that Paramecium secondary siRNAs are initially synthesized with 5′-triphosphate ends, followed by the removal of the γ and β phosphates by a Pir-1-like RNA phosphatase ( 47 , 48 ). Alternatively, the 5′-monophosphate of Paramecium secondary siRNAs may indicate that they are cut by Dicer from long dsRNA as demonstrated for plant secondary siRNAs ( 11 ). In Paramecium , a polarized Dicer activity, which could be due to its interaction with the RdRP as suggested in Tetrahymena ( 35 ), would allow selective stabilization of the antisense strand after cleavage of the dsRNA.
It remains unclear why the ∼22-nt siRNAs were not observed in previous analyses of dsRNA-induced silencing of the ND7 gene. A small-scale sequencing study of associated siRNAs only examined gel-purified fractions between 23 and 26 nt ( 17 ), which may be part of the reason. This study also provided evidence for a distinct subclass of siRNAs with features suggestive of secondary siRNAs (i.e. strictly antisense to mRNA and not observed for the vector part of ingested dsRNA; one was found to match the mRNA upstream of the region targeted by the dsRNA). However, these ND7 secondary siRNAs migrated at 23–24 nt and were found to carry a short untemplated polyA tail. Whether polyadenylation of secondary siRNAs is only observed at the specific stage of the life cycle when these were analysed (early conjugation) will need to be addressed by further sequencing studies.
It is surprising that any RdRP should be necessary for dsRNA-induced RNAi. One possible explanation may be that amplification of the trigger dsRNA by RdRPs may be necessary to generate dsRNA ends that are efficiently recognized by the Dicer enzyme. Indeed, studies in the related Tetrahymena showed that Dcr2 requires a 5′-triphosphate to generate discrete siRNA products ( 35 ). In our experiments, the dsRNA resulting from T7 Pol transcription in E. coli is 5′-triphosphorylated, which would allow only one siRNA duplex to be cleaved by Dcr1 if this had the same specificity as Tetrahymena ’s Dcr2. Because the dsRNA produced in E. coli contains vector-derived sequences on both sides of the insert, cleaving of a single siRNA duplex from each end would not elicit silencing of the target gene. After cleavage, the remaining portion of the dsRNA would be 5′-monophosphorylated, and an RdRP activity might be required to provide new 5′-triphosphate ends by copying each strand. In this hypothesis, the generation of primary siRNAs would be a cyclic process of RdRP activity and Dicer cleavage, each cycle moving the cleavage sites closer to the centre of the initial dsRNA.
Alternatively, it is conceivable that the dsRNA present in the food vacuoles during digestion of E. coli is not imported as such into the cytoplasm, but in the form of ssRNA. RdRP activity would then be required to provide dsRNA substrates for Dcr1. This hypothesis is not exclusive of the first one, since the remaining portion of the dsRNA after the first round of Dcr1 cleavage would still be 5′-monophosphorylated. So far, the mechanism of RNA import from food vacuoles to cytoplasm is unknown and genes similar to C. elegans SID1 , encoding a transmembrane protein necessary for dsRNA uptake ( 49 , 50 ), cannot be found in the Paramecium genome.
Whatever the reason may be, our results indicate that both Rdr1 and Rdr2 are required for the accumulation of primary siRNAs and for dsRNA-induced silencing. Because the effects of each knockdown are the same, there is no evidence that the two proteins have distinct functions. The requirement of these proteins for primary siRNAs also precludes any conclusion as to their possible involvement in the synthesis of secondary siRNAs, since this also depends on the targeting of the mRNA by primary siRNAs.
Transgene-induced silencing differs from dsRNA-silencing
Several lines of evidence indicate that the mechanism underlying transgene-induced silencing are at least partially distinct from those of dsRNA-induced silencing. First, Rdr3 is required for transgene-induced silencing and no evidence was obtained for the involvement of Rdr1 or Rdr2, even in experiments where both RDR1 and RDR2 were co-silenced (data not shown). A high degree of functional specialization of RdRPs in the recognition of different types of RNAs triggering RNAi may help to avoid widespread and unspecific silencing. Secondly, although transgene-induced siRNAs have the same apparent size as dsRNA-induced primary siRNAs (∼23 nt), they differ from the latter in that they do not trigger the synthesis of ∼22-nt secondary siRNAs homologous to the ND169 gene which was used to study both silencing methods. Thirdly, of all the sRNAs analysed in this study, only transgene-induced siRNAs carry a modification of the 3′ ribose. In the related Tetrahymena , the meiosis-specific scnRNAs are 2′- O -methylated on the 3′-ribose by the RNA methyltransferase Hen1 ( 51 ), which is also present in the Paramecium genome. This modification was shown in other systems to protect sRNAs from addition of untemplated nucleotides and degradation ( 40 , 52 ). Finally, distinct mechanisms are also supported by the finding that different (though overlapping) sets of Piwi proteins are involved in the two silencing methods (K. Bouhouche et al. , submitted for publication).
Interestingly, an analysis of Tetrahymena endogenous ∼23–24-nt siRNAs bound by five different Piwi proteins showed that only those bound by the Twi8 protein were modified on the 3′ ribose ( 53 ). The same study found that a Twi8–GFP fusion localized to the macronucleus, while two other Twi proteins binding ∼23–24-nt siRNAs localized in the cytoplasm. Furthermore, the sequencing of Twi8-bound siRNAs showed them to be derived from both strands of mRNA-producing loci, leading the authors to propose that Twi8 may be mediating co-transcriptional, rather than post-transcriptional, regulation ( 53 ). The modification of the 3′ ribose of transgene-induced siRNAs in Paramecium thus raises the possibility that this process affects transcription of the endogenous gene in the macronucleus. The absence of secondary siRNAs could then be due to impairment of mRNA production. Conflicting with this hypothesis, one study concluded that transgene-induced silencing of the T4a gene was post-transcriptional on the basis of run-on assays indicating that its 3′-UTR was continuously transcribed ( 22 ). However, it is now known that another gene (GSPATG00004256001) is present very close downstream of T4a in a convergent orientation, with overlapping 3′-UTRs and only 26 bp between the two stop codons. Since the probe used for the run-on assay of the T4a 3′-UTR was double stranded, it may well have been measuring the ongoing transcription of the downstream gene.
Rdr3 is also involved in the accumulation of endogenous siRNAs
In our study, northern blots revealed endogenous siRNAs previously identified by cloning and sequencing ( 17 ), which map to an intergenic region between convergent genes on scaffold 22. We found that these endogenous siRNAs, whose apparent size is similar to that of transgene-derived siRNAs, also depend on Rdr3. Strand-specific probes indicated that they derive from both strands, although the bottom strand was more difficult to detect by hybridization. A partial strand bias may explain why small-scale sequencing only revealed siRNAs from the top strand ( 17 ). In contrast to transgene-derived siRNAs, however, these endogenous siRNAs were not modified on the 3′ ribose, at least for the majority of top-strand siRNAs. In view of the apparent diversity of endogenous siRNAs in Tetrahymena , further sequencing, biochemical and functional characterization of Paramecium endogenous siRNAs will be required to ascertain their relationship with the transgene-induced silencing pathway.
RNAi controls antigenic variation in Paramecium
As mentioned above, the surface antigen gene family is controlled by unknown mechanisms ensuring that only one gene is expressed at a time. Several studies concluded that surface antigen genes are regulated at the level of transcription ( 54–57 ), although cases of post-transcriptional regulation have also been described ( 26 , 58 ). It is not known whether transcriptional and post-transcriptional control mechanisms interact. Importantly, studies of chimeric constructs indicated that the promoters of these genes are not sufficient to mediate mutual exclusion of expression, and that this also involves at least the beginning of the cdss ( 54 , 55 ).
Our experiments showed that the silencing of RDR3 results in the simultaneous expression of multiple surface antigens at the cell surface. Rdr3 is therefore somehow involved in maintaining most surface antigen genes silent. Similarly, antigenic variation in Giardia lamblia was shown to be controlled by RNAi ( 59 ). In wild-type cells, several surface antigen genes are transcribed simultaneously, and all but one are silenced post-transcriptionally. Silencing of Giardia’s RdRP and Dicer leads to a disruption of this system and to expression of multiple surface antigens ( 59 ). Although this is superficially similar to our results, it is clear from previous run-on assays that the regulation of Paramecium surface antigen genes is not entirely post-transcriptional ( 54–57 ), suggesting that Rdr3-mediated silencing may affect transcription itself.
One possibility is that surface antigen genes are co-transcriptionally silenced. The probes used in previous run-on assays were usually not directly located in 5′ part of the long (7–8 kb) cdss, so that they do not formally demonstrate the absence of transcription of the 5′ parts of the genes which may also be below the detection limit ( 54 , 56 , 57 ). In this hypothesis, Rdr3-dependent siRNAs could affect the elongation of transcription, rather than its initiation. There is increasing evidence from plants and fungi that sRNA-mediated heterochromatin formation actually requires transcription of the targeted loci, and that nascent transcripts are cleaved co-transcriptionally ( 6 , 60 ). In many organisms, the process results in the targeting of H3K9 methylation. In Tetrahymena , however, this histone modification only occurs in developing macronuclei, and results in the elimination of modified chromatin ( 61–63 ). In contrast, H3K27me3 is present in the macronucleus of vegetative cells, and this modification is also thought to be required in the developing macronucleus as an intermediate in the process of scnRNA-directed H3K9 methylation ( 62 ).
Whether or not Rdr3-mediated silencing of Paramecium surface antigen genes involves histone modifications, an RNAi-based regulation mechanism may explain the long-known maternal inheritance of serotypes during sexual reproduction, which implies that the particular expression pattern observed in the maternal macronucleus is transferred through the cytoplasm to the developing zygotic macronuclei ( 64–66 ). The identification of surface antigen-related siRNAs in different serotypes would certainly provide new insight into this question. Biochemical studies will also be required to determine whether Rdr3, despite its highly divergent catalytic domain, has RdRP activity or whether it is only required as part of a complex for the dicing of dsRNA formed by the pairing of sense and antisense RNAs.
Supplementary Data are available at NAR Online.
Deutsche Forschungs Gemeinschaft (grant SI 1397/1-1 to M.S.); work in E.M.’s laboratory was supported by grants from the Agence Nationale de la Recherche (ANR-08-BLAN-0233) and Equipe FRM (Fondation pour la Recherche Médicale). Funding for open access charge: Deutsche Forschungs Gemeinschaft (grant SI 1397/1-1).
Conflict of interest statement . None declared.
We wish to thank all members of the laboratory for their help and Helmut Schmidt for support of the work. We are grateful to John Cullum, Micah Dunthorn and Amanda Huckins for comments on the manuscript.