Title: Endo-siRNAs depend on a new isoform of loquacious and target artificially introduced, high-copy sequences
Abstract: Article30 July 2009free access Endo-siRNAs depend on a new isoform of loquacious and target artificially introduced, high-copy sequences Julia Verena Hartig Julia Verena Hartig Gene Center, Department of Chemistry and Biochemistry, Ludwig-Maximilians-Universität München, Munich, Germany Search for more papers by this author Stephanie Esslinger Stephanie Esslinger Gene Center, Department of Chemistry and Biochemistry, Ludwig-Maximilians-Universität München, Munich, Germany Search for more papers by this author Romy Böttcher Romy Böttcher Gene Center, Department of Chemistry and Biochemistry, Ludwig-Maximilians-Universität München, Munich, Germany Search for more papers by this author Kuniaki Saito Kuniaki Saito Keio University School of Medicine, Shinjuku-ku, Tokyo, Japan Search for more papers by this author Klaus Förstemann Corresponding Author Klaus Förstemann Gene Center, Department of Chemistry and Biochemistry, Ludwig-Maximilians-Universität München, Munich, Germany Munich Center for Integrated Protein Science (CiPSM), Ludwig-Maximilians-Universität München, Munich, Germany Search for more papers by this author Julia Verena Hartig Julia Verena Hartig Gene Center, Department of Chemistry and Biochemistry, Ludwig-Maximilians-Universität München, Munich, Germany Search for more papers by this author Stephanie Esslinger Stephanie Esslinger Gene Center, Department of Chemistry and Biochemistry, Ludwig-Maximilians-Universität München, Munich, Germany Search for more papers by this author Romy Böttcher Romy Böttcher Gene Center, Department of Chemistry and Biochemistry, Ludwig-Maximilians-Universität München, Munich, Germany Search for more papers by this author Kuniaki Saito Kuniaki Saito Keio University School of Medicine, Shinjuku-ku, Tokyo, Japan Search for more papers by this author Klaus Förstemann Corresponding Author Klaus Förstemann Gene Center, Department of Chemistry and Biochemistry, Ludwig-Maximilians-Universität München, Munich, Germany Munich Center for Integrated Protein Science (CiPSM), Ludwig-Maximilians-Universität München, Munich, Germany Search for more papers by this author Author Information Julia Verena Hartig1,‡, Stephanie Esslinger1,‡, Romy Böttcher1, Kuniaki Saito2 and Klaus Förstemann 1,3 1Gene Center, Department of Chemistry and Biochemistry, Ludwig-Maximilians-Universität München, Munich, Germany 2Keio University School of Medicine, Shinjuku-ku, Tokyo, Japan 3Munich Center for Integrated Protein Science (CiPSM), Ludwig-Maximilians-Universität München, Munich, Germany ‡These authors contributed equally to the work *Corresponding author. Gene Center, Department of Chemistry and Biochemistry, Ludwig-Maximilians-Universität München, Feodor-Lynen-Str. 25, Munich, 81377, Germany. Tel.: +49 89 2180 76912; Fax: +49 89 2180 76945; E-mail: [email protected] The EMBO Journal (2009)28:2932-2944https://doi.org/10.1038/emboj.2009.220 PDFDownload PDF of article text and main figures. ToolsAdd to favoritesDownload CitationsTrack CitationsPermissions ShareFacebookTwitterLinked InMendeleyWechatReddit Figures & Info Colonization of genomes by a new selfish genetic element is detrimental to the host species and must lead to an efficient, repressive response. In vertebrates as well as in Drosophila, piRNAs repress transposons in the germ line, whereas endogenous siRNAs take on this role in somatic cells. We show that their biogenesis depends on a new isoform of the Drosophila TRBP homologue loquacious, which arises by alternative polyadenylation and is distinct from the one that functions during the biogenesis of miRNAs. For endo-siRNAs and piRNAs, it is unclear how an efficient response can be initiated de novo. Our experiments establish that the endo-siRNA pathway will target artificially introduced sequences without the need for a pre-existing template in the genome. This response is also triggered in transiently transfected cells, thus genomic integration is not essential. Deep sequencing showed that corresponding endo-siRNAs are generated throughout the sequence, but preferentially from transcribed regions. One strand of the dsRNA precursor can come from spliced mRNA, whereas the opposite strand derives from independent transcripts in antisense orientation. Introduction The appearance of a new selfish genetic element in an organism's genome is often accompanied by amplification and mobilization, leading to a highly mutagenic situation until the cells regain control. Most eukaryotic organisms carry a large proportion of transposons in their genome, indicating that the taming of selfish genetic elements was often not immediately successful (Orgel and Crick, 1980; Pace and Feschotte, 2007). In Drosophila, zebrafish, rat and mouse germ line cells, the piRNA system can efficiently repress transposons (for review, see Malone and Hannon, 2009), but only if there is a pre-existing, maternally contributed pool of piRNAs with corresponding sequence (Blumenstiel and Hartl, 2005; Brennecke et al, 2008). If not, the resulting progeny has severely impaired fertility due to frequent transposon mobilization. The necessary maternal pool of piRNAs to prevent this phenomenon (called hybrid dysgenesis) can be built up slowly over the course of several generations (for review, see Chambeyron and Bucheton, 2005). Although an amplification loop has been proposed that can explain the maintenance of corresponding sense and antisense piRNA populations, we currently do not understand how a piRNA response is initiated de novo (for review, see Aravin et al, 2007; Hartig et al, 2007; Malone and Hannon, 2009; O'Donnell and Boeke, 2007). Even though germ cells are the ultimate target for selfish genetic elements, they are not the only cells that can be faced with a newly emerging transposon threat. Certain viruses, for example, can carry transposable elements in their genome, which will become integrated into the host cell genome on establishment of latency (van Oers and Vlak, 2007). Silencing of transposable elements is, therefore, also important in somatic cells to reduce the possibility of transition into the germ cell lineage (Chalvet et al, 1999; Pelisson et al, 2002). The recent discovery of endo-siRNAs points to one mechanism by which transposons can be silenced in the somatic cells of Drosophila (Chung et al, 2008; Czech et al, 2008; Ghildiyal et al, 2008; Kawamura et al, 2008), and a similar class of RNAs has been detected in mouse ES cells (Watanabe et al, 2006, 2008; Babiarz et al, 2008; Tam et al, 2008). The transposon-matching endo-siRNAs are comparable with piRNAs in the sense that they induce homology-dependent repression, but their biogenesis is clearly different: a roughly equal distribution of sense-matching and antisense-matching endo-siRNAs implies a double-stranded precursor. This is substantiated by the dependence of endo-siRNAs on the enzyme Dicer-2 (Dcr-2) that also processes dsRNA to induce ‘classical’ RNA interference (for review, see Golden et al, 2008). In the case of mouse oocytes, a generation of dsRNA in trans from two distinct loci (e.g. a gene and a corresponding processed pseudogene) has been proposed (Tam et al, 2008; Watanabe et al, 2008). Despite these insights into endo-siRNA biogenesis, one fundamental question remains unsolved: can a newly appearing selfish genetic element be recognized without the need for a pre-existing pool of small RNAs and an inactive genetic remnant or pseudogene? And if yes, how are the double-stranded precursor molecules for endo-siRNAs generated? Loqs normally functions together with Dcr-1 in miRNA biogenesis (Forstemann et al, 2005; Jiang et al, 2005; Saito et al, 2005; Park et al, 2007), whereas the usual partner for Dcr-2 is R2D2 (Liu et al, 2003). The dependence of endo-siRNAs on dcr-2 in combination with loqs, but not r2d2, was unexpected, as it clearly distinguishes the dsRNA-mediated response against transposons from the r2d2-dependent response that occurs after infection with an RNA virus (Wang et al, 2006). In addition, it suggested significant overlap between the miRNA and siRNA biogenesis pathways at the level of the processing step, whereas so far an exchange had only been described at the Argonaute loading step (Forstemann et al, 2007; Tomari et al, 2007). The complex could be verified biochemically (Czech et al, 2008), but it is unclear which of the Loqs protein isoforms is binding to Dcr-2. As only one isoform, Loqs-PB, is essential for miRNA biogenesis together with Dcr-1, the overlap between the miRNA and the endo-siRNA pathways may, in fact, be smaller than what was initially assumed. In an increasingly complex repertoire of small RNAs (Core et al, 2008; He et al, 2008; Preker et al, 2008; Seila et al, 2008; Fejes-Toth et al 2009; Lee et al, 2009; Taft et al, 2009), the importance of distinguishing each class during its biogenesis and of ensuring functional specificity is obvious. This is indeed an almost daunting task, as the various classes of small RNAs all share a common chemical structure and populate similar size ranges. In this study, we show that biogenesis of endo-siRNAs depends on a new isoform of the dsRNA-binding domain protein Loquacious (also known as R3D1), which is distinct from the isoform acting in the miRNA pathway. We describe a cell-culture model in which an artificial plasmid sequence, integrated at high-copy number, has become subject to endo-siRNA-mediated repression. The artificial, as well as an endogenous transposon target for endo-siRNAs are silenced through a post-transcriptional mechanism, indicating that repression is direct, and deep sequencing analysis showed corresponding endo-siRNAs. The repressive response also occurred in transiently transfected cells. Thus, Drosophila endo-siRNAs can mount a de novo response without a need to integrate into the host cell genome or the need for any pre-existing sequence that can serve as a template for the production of small RNAs. Results Production of Drosophila endo-siRNAs depends on a new isoform of loqs The observation that endo-siRNA-mediated silencing depends on loqs, in combination with dcr-2, was a surprise because it suggested that a hybrid complex with components of the canonical miRNA and siRNA biogenesis pathways exists. However, this interpretation may have been an oversimplification because the loqs gene can give rise to, at least, three different mRNAs, each coding for a protein with distinct properties (Forstemann et al, 2005; Jiang et al, 2005; Saito et al, 2005). Only one isoform, called Loqs-PB, is an essential partner of Dcr-1 in the biogenesis of miRNAs (Jiang et al, 2005; Park et al, 2007). We carried out 3′ RACE experiments and detected a new mRNA variant of loqs (loqs-RD), in which an alternative polyadenylation in the third intron leads to a new protein isoform that lacks the 3rd dsRBD of Loqs-PB/PA and contains 22 amino acids of new protein sequence (Figure 1A and Supplementary Figure 1, Loqs-PD). Figure 1.The Loqs-PD isoform is primarily responsible for endo-siRNA biogenesis. (A) Schematic diagram of the four loqs mRNA and protein variants currently known. Start codons are indicated by vertical green bars, stop codons by vertical red bars. The regions from which dsRNA was derived to trigger isoform-specific RNAi are indicated below the mRNA variants. The dsRBD motifs are indicated in the cartoon drawings of the protein isoforms (diagram analogous to Forstemann et al, 2005). (B) Western Blot to confirm depletion of specific Loqs protein isoforms. (C) Northern blot to detect a hairpin-derived endo-siRNA (called CG4068 B in Okamura et al, 2008b) and the bantam miRNA; 2S rRNA serves as a loading control. Download figure Download PowerPoint Isoform-specific knockdowns are possible if the corresponding mRNA contains a stretch of unique sequence that is amenable to RNAi. In the case of the loqs gene, it is possible to target the loqs-RC and loqs-RD RNA individually, the loqs-RB RNA together with the loqs-RC RNA, loqs-RA, loqs-RB and loqs-RC together through the common sequence towards the 3′ end and finally all four loqs isoforms simultaneously with dsRNA directed against the amino-terminus of Loqs. Detection of Loqs protein isoforms on immunoblots has shown three bands of distinct sizes, which had been assumed to correspond to Loqs-PB, Loqs-PA and Loqs-PC (Forstemann et al, 2005). Using our isoform-specific knockdown, we could show that the smallest of these bands predominantly contains the new Loqs-PD isoform (Figure 1B). To determine which of the four Loqs protein isoforms is required for endo-siRNA biogenesis, we depleted individual Loqs protein isoforms and then measured the levels of an endo-siRNA derived from the long hairpin-forming gene, CG4068, by northern Blotting (Czech et al, 2008; Kawamura et al, 2008; Okamura et al, 2008b). Only depletion of the loqs-RD transcript correlated with a reduced production of the CG4068 ‘B’ endo-siRNA (Figure 1C), whereas the biogenesis of the bantam miRNA was not impaired when loqs-RD was depleted. Depletion of loqs-RB+RC, as expected, led to an increased abundance of pre-bantam. Co-IP of Loqs with Dcr-2 has been described previously (Czech et al, 2008) but the antibody employed recognized all Loqs protein isoforms. Using cDNA constructs coding for only one of the splice variants, we could distinguish Loqs-PA, PB and PD in transfected S2 cells and determine the potential of each isoform to interact with Dcr-2 by co-immunoprecipitation. Epitope-tagged Loqs-PD, similar to tagged R2D2, was able to associate with Dcr-2 but not with Dcr-1, though the extent of Dcr-2 association varied between experiments (Supplementary Figure 2). This is consistent with the previous observation that on the level of the endogenous protein, the smallest Loqs isoform (identified as PD in this paper) does not co-immunoprecipitate with Dcr-1 (Forstemann et al, 2005). In contrast, tagged Loqs-PB and Loqs-PA can associate with Dcr-1, and also to some extent with Dcr-2. As only Loqs-PD is required for endo-siRNA generation, it seems that the Dcr-2–Loqs-PD complex is exclusively responsible for endo-siRNA generation. Stably integrated transgenes can be subject to repression that depends on endo-siRNA biogenesis factors To further validate the importance of the Loqs-PD isoform for endo-siRNA-dependent silencing, we attempted to generate a reporter system that closely resembles natural endo-siRNA targets and, at the same time, provides a convenient read-out. Our reasoning was that some of the canonical features of transposable elements, such as multicopy insertion and the formation of repetitive regions, are shared by transgenes that have integrated into the host cell genome after transfection and selection of stable cell culture lines. Indeed, Ago2-dependent repression of a stably integrated GFP expression plasmid in Drosophila cells has been described (Siomi et al, 2005). We examined a clonal cell line (called 63N1) that expresses GFP (Figure 2A) and tested the changes in GFP levels on impairment of known miRNA/siRNA pathway components. From the three RNaseIII enzymes Drosha, Dcr-1 and Dcr-2, only the latter seemed to be involved in repression of GFP. Depletion of the cytoplasmic dsRBD protein, Loqs, also resulted in a de-repression of GFP, whereas depletion of its homologue, R2D2, seemed to increase repression. Finally, Ago2 is the main effector protein mediating this response, although depletion of Ago1 also resulted in a slight de-repression (Figure 2B). These are precisely the genetic requirements of the endo-siRNA pathway in Drosophila (Czech et al, 2008; Kawamura et al, 2008; Okamura et al, 2008b). If the GFP transgene has caused an endo-siRNA response, then the new Loqs-PD isoform should be required for its repression. Using isoform-specific knockdowns, as described in Figure 1A, we observed that GFP fluorescence did not change upon a knock down of loqs-RA+RB+RC or loqs-RC alone. In contrast, specific targeting of loqs-RD as well as targeting loqs-RD and loqs-RC together led to an even stronger de-repression than knock down of all loqs variants (Figure 2C). Thus, Loqs-PD seems to be required for both hairpin-derived and repetitive element-derived endo-siRNAs. Figure 2.Stably integrated transgenes are subject to endo-siRNA-mediated repression. (A) Our experimental system is based on transfection of a GFP expression construct together with an antibiotic resistance plasmid to allow for selection of stable transformants, followed by selection of stable, single-cell-derived clones. (B) Individual genes were depleted through RNA interference to define the genetic requirements for GFP repression. The control dsRNA was directed against DsRed; values are represented as mean±s.d. (n=3). (C) Effects of isoform-specific RNAi experiments on the GFP-expression in our stable cell line. The applied RNAi trigger is indicated below the bars. Values represent the mean±s.d. (n=3). (D) Comparison of the extent to which the transgene is repressed between cell lines 63-6 and 63N1 (two independently derived clones from the same parental cells using the same expression plasmid). (E) Quantitative PCR to determine the copy number of the inserted expression plasmid. All values are expressed relative to the value for GAPDH (CG12055), numerical values are indicated above each bar. The red lines in (B–D) mark a value of 1 (=no change relative to the control). Download figure Download PowerPoint The 63N1 cell line has mounted a bona fide endo-siRNA response against the integrated transgene, but the extent to which this occurs varies between clones. In a previous publication, a cell line (called 63–6) derived independently from the same stock of parental S2 cells and transfected with the same expression plasmid showed only marginal response towards depletion of Dcr-2, Loqs and Ago-2 (Forstemann et al, 2007). We re-examined this cell line and could corroborate that there is a minor, but significant increase in GFP levels on depletion of Dcr-2 (1.2-fold, P<0,01) and loqs-RD (1.1-fold, P<0.02). One potential difference between the two cell lines is the number of plasmid copies that have integrated in the genome. Therefore, we compared the copy number with isolated genomic DNA by quantitative PCR. We amplified a region from the ubiquitin promotor that drives expression of GFP and compared the results with GAPDH and two other single-copy genes. Relative to GAPDH, the ubiquitin promotor was 8-fold more abundant in the cell line with the blunted response, whereas it was 42-fold more abundant in the 63N1 cell line (Figure 2E). This suggests that a high-copy number may favour the establishment of robust repression. GFP mRNA is a direct target of the endo-siRNA response Endogenous RNA targets of the endo-siRNA pathway become more abundant on depletion of Dcr-2 or Ago2. The simplest interpretation of these results is that endo-siRNAs—like siRNAs—induce the post-transcriptional degradation of mRNAs, and this capacity has indeed been shown (Chung et al, 2008; Czech et al, 2008; Ghildiyal et al, 2008; Kawamura et al, 2008; Okamura et al, 2008b). To prove that the GFP transgene is a direct target of the small RNAs, and to rule out that upregulation of GFP is an indirect consequence of endo-siRNA loss leading to increased transcription of the GFP-coding gene, we examined transcription and degradation rates of GFP mRNA in vivo by pulse labelling newly synthesized RNA (Johnson et al, 1991; Kenzelmann et al, 2007; Dolken et al, 2008). To verify the procedure (depicted in Figure 3A), we determined the ratio of the hsp70 mRNA, which has a half-life of 15–30 min in S2 cells (Petersen and Lindquist, 1988), to the stable rp49 mRNA in each RNA fraction. As expected, we found that relative to rp49, hsp70 mRNA was significantly more abundant in the newly synthesized RNA fraction than in unfractionated RNA (<1 h versus total RNA, Figure 3B). In addition, we examined the effect of our dcr-2 RNAi in comparison with our control RNAi against DsRed. As shown in Figure 3C, knock down reduced the dcr-2 mRNA levels by about twofold in total RNA (cDNA synthesis was primed with random hexamers, thus we potentially also detected mRNA degradation fragments), but it was much more pronounced (20-fold) in the flow-through fraction (>1 h old) of our RNA separation procedure. We did not observe any significant change of dcr-2 mRNA in the newly transcribed fraction. The 1 h time window for labelling seems, therefore, appropriate to separate transcriptional from post-transcriptional events. Figure 3.GFP mRNA is a direct target of the endo-siRNA pathway. (A) Overview of the pulse labelling and purification method adapted from Dolken et al (2008). (B) To verify the RNA fractionation procedure, the ratio between the short-lived hsp70 and the long-lived rp49 mRNAs was determined through qRT–PCR. The separation procedure can extract newly synthesized RNA because the <1 h fraction is enriched for the short-lived message. (C) Further validation was obtained by determining the ratio of dcr-2 mRNA (normalized to rp49) between control and dcr-2 RNAi treated cells in each fraction. The post-transcriptional effect of ‘classical’ RNAi affects only the flow-through fraction (i.e. >1 h old). (D) Quantification of GFP, mdg1, 297 and gapdh transcripts by qRT–PCR. The values represent the fold change of each RNA (normalized to rp49) between control treated cells (=RNAi against DsRed) and dcr-2 RNAi-treated cells. The horizontal line indicates a value of 1 (=no change relative to control). Values represent the mean±s.d. Download figure Download PowerPoint If de-repression of our GFP transgene occurred by a transcriptional mechanism, then inhibition of endo-siRNA biogenesis should cause an increased abundance of target mRNAs in the <1 h RNA fraction. However, if a post-transcriptional mechanism is responsible, the increase should be stronger in the >1 h RNA fraction. We observed a trend towards increased mRNA levels for our GFP transgene and the mdg1 transposon (an endogenous target of the endo-siRNA pathway) in the total RNA fraction on knock down of dcr-2. After fractionation, we could detect significant changes in the >1 h fraction for GFP and mdg1 (Figure 3D, P<0,05, two-tailed z-test, n=3). We did not detect any significant changes in the newly transcribed RNA fraction, indicating that the increased GFP levels caused by impaired endo-siRNA biogenesis are due to an increased stability of the GFP mRNA. As a positive control for changes in the rates of transcription, we used a UAS–GFP vector with and without co-expression of the Gal4 transcription factor. This resulted in an increase of GFP mRNA in all three RNA fractions (Supplementary Figure 3). A second natural endo-siRNA target, the transposon 297, showed no significant upregulation in any of the three fractions. One possible reason for this may be that endo-siRNAs directed against 297 are exceptionally abundant (as determined by deep sequencing; reads antis. to GFP: 8405; reads antis. to mdg1: 4146; reads antis. to 297: 35083) and that our dcr-2 depletion may not have been efficient or long enough for these endo-siRNAs to decay. Detection and analysis of transgene-derived endo-siRNAs If GFP repression is due to mRNA targeting by endo-siRNAs, then small RNAs with complementarity to GFP should be present in the reporter cells. We isolated 18–30-nt long RNAs from the parental S2 cells as well as the stable GFP-expressing clone 63N1, then analysed them by deep sequencing. After removal of adapter sequences and length selection (20–24 nt size window), we obtained 1 098 002 reads for the parental cell library and 2 678 671 reads for the 63N1 cell library. Surprisingly, 359 845 reads from the parental cell library and 1 112 196 reads in the 63N1 library corresponded to miR-184. This exceeds by far the frequency with which miR-184 has been detected in other studies and we, therefore, disregarded these reads during further analysis, leaving 738 157 reads in the parental cell library and 1 566 475 reads in the 63N1 cell library. After mapping to the sequence of our GFP expression construct, we found 16 424 corresponding reads in the 63N1 cell library and 167 reads in the parental cell library (see Table I). The size distribution of these reads showed a sharp peak at 21 nt (Supplementary Figure 4), consistent with the length preference of Ago2. When we analysed from which positions in our construct the corresponding reads were derived, we found that not only the GFP-coding region but also many other regions of the construct gave rise to small RNAs (Figure 4A). Overall, sense- and antisense-matching reads were equally represented (8019 sense reads versus 8405 antisense reads). This is an indication that the small RNAs directed against the construct derive from a double-stranded precursor (Czech et al, 2008; Ghildiyal et al, 2008; Kawamura et al, 2008; Okamura et al, 2008a). We also re-analysed the published deep sequencing data from the 63–6 cell line (Seitz et al, 2008) for corresponding small RNA reads. Similar to 63N1 cells, we found reads derived from many regions of the construct (Figure 4B), though the white gene produced many more endo-siRNAs than it did in the 63N1 cells. This could reflect differences between the cell lines (e.g. due to integration sites) or differences in the deep sequencing techniques (pyrosequencing versus sequencing by synthesis). Figure 4.Deep sequencing of transgene-derived endo-siRNAs. S2 cell-derived small RNA reads were mapped to the sequence of the transfected GFP expression plasmid (horizontal axis; values above axis: quantification of sense reads, below: quantification of antisense reads). (A) Small RNAs sequenced from 63N1 cells. (B) Small RNAs sequenced from 63–6 cells (Seitz et al, 2008). In this experiment, the cells had been treated with dsRNA corresponding to a short stretch of the GFP mRNA; this region is indicated. Download figure Download PowerPoint Table 1. Deep sequencing reads Parental 63N1 pKF63 IP Ago2 (GSM280089) Total reads (processed) 738 157 1 566 475 199 418 1 552 793 CG4068 397 707 159 3303 pCASPER-2 78a 6930 151 50564 pKF63 167a 16424 759 NA norm. CG4068 1 1 1 1 norm. pCASPER-2 0.2a 9.8 0.9 15.3 norm. pKF63 0.4a 23.2 4.8 NA NeoR 1 180 — NA a Size selection of the small RNA from parental, 63N1 and pKF63 transfected cells before adapter ligation was done on the same gel; a small amount of cross contamination may have occurred. The selection plasmid conferring resistance to G418 (NeoR) was co-transfected only in 63N1 cells. Thus, analysis of the reads matching the neomycin resistance gene can give an independent indication of the extent of cross contamination (last row). The loading order was: parental ∣ 63N1 ∣ pKF63. Using the 63-6 cell dataset, we could show that the transgene-derived endo-siRNAs are indeed loaded into Ago2 complexes. If a small RNA resides in Drosophila Ago2, it will be 2′-O-methyl modified and become resistant to β-elimination at its 3′ end (Horwich et al, 2007). By comparing the read counts from a mock-treated RNA sample with a β-eliminated RNA sample, we could assess the extent to which the endo-siRNAs derived from our construct were loaded into Ago2 complexes (Supplementary Figure 5). We found 510 corresponding reads in the mock-treated small RNA population and 4167 reads in the β-eliminated reads (excluding the reads that derive from the dsRNA directed against GFP with which the cells had been treated in this study). This enrichment (8.17-fold) is identical to the 8.2-fold enrichment calculated for endo-siRNAs corresponding to transposable elements in this data set (Chung et al, 2008). Transient transfection also leads to an endo-siRNA response What is triggering the production of double-stranded RNA from the expression construct in the stable cell lines? One possibility is that a head-to-head arrangement of plasmid concatemers is produced upon integration and that read-through transcription yields double-stranded RNA, as has been proposed for natural transposon clusters (Siomi, 2008). Alternatively, it may be the high-copy number per se that is somehow triggering dsRNA production. To assess whether integration is a prerequisite for endo-siRNA generation, we examined a small RNA library generated from transiently transfected cells. The deep sequencing data were processed, as described before, and a total of 199 418 reads remained after the removal of adapter sequences and length selection. We found 759 reads corresponding to the plasmid sequence. After normalization to small RNAs derived from the CG4068 hairpin, this corresponds to a roughly fivefold lower abundance in the transiently transfected cells relative to the 63N1 cells (Table I). However, this comparison should be regarded as a rough estimate because normalization of deep sequencing libraries is not straightforward; furthermore, the total read number was considerably lower for the transfected cells than for the 63N1 cells. Nonetheless, the abundance of small RNAs in transiently transfected cells seems too high to be caused by rare cells within the population that have already integrated the plasmid in their genome. To corroborate the fact that transient transfection can lead to endo-siRNA production, we analysed a published small RNA library derived from S2 cells that had been transfected with an expression plasmid for epitope-tagged Ago2. The library contains small RNAs isolated after immunoprecipitation of the tagged Ago2, among them siRNA-sized reads mapping to the transfected plasmid (G Hannon, personal communication). As both our GFP expression vector