piRNA-guided co-transcriptional silencing coopts nuclear export factors

The PIWI-interacting RNA (piRNA) pathway is a small RNA-based immune system that controls the expression of transposons and maintains genome integrity in animal gonads. In Drosophila, piRNA-guided silencing is achieved, in part, via co-transcriptional repression of transposons by Piwi. This depends on Panoramix (Panx); however, precisely how an RNA binding event silences transcription remains to be determined. Here we show that Nuclear Export Factor 2 (Nxf2) and its co-factor, Nxt1, form a complex with Panx and are required for co-transcriptional silencing of transposons in somatic and germline cells of the ovary. Tethering of Nxf2 or Nxt1 to RNA results in silencing of target loci and the concomitant accumulation of repressive chromatin marks. Nxf2 and Panx proteins are mutually required for proper localization and stability. We mapped the protein domains crucial for the Nxf2/Panx complex formation and show that the amino-terminal portion of Panx is sufficient to induce transcriptional silencing.

Here we show that Panx coopts elements of the nuclear RNA export machinery to trigger transcriptional silencing. Panx is part of a complex that also contains Nuclear Export Factor 2 (Nxf2) and Nxt1/p15. Panx and Nxf2 are interdependent for their protein stability. nxf2 mutants show strong de-repression of Piwi-regulated transposons and severe loss of H3K9me3 at affected loci, similarly to panx mutants. We find that the amino-terminus of Panx delivers the critical silencing signal, as it is necessary and sufficient to trigger the deposition of repressive chromatin marks if tethered to a reporter construct, while its carboxyl-terminal region is involved in the interaction with Nxf2. Nxf2 is closely related to the mRNA export factor Nxf1, which also interacts and functions with Nxt1 (Fribourg et al., 2001;Herold et al., 2001;Herold et al., 2000). Thus, our findings reveal that the evolution of transposon defense mechanisms involved exaptation of the nuclear RNA export machinery.

Nxf2 is a TGS factor that interacts with Panx
To identify proteins associated with Panx, we immunoprecipitated a GFP-Panx fusion protein expressed from its endogenous promoter (Handler et al., 2013; Figure 1-figure supplement 1A) from ovary lysates and identified co-purifying proteins by quantitative mass spectrometry. Three proteins showed the strongest enrichment and significance: Panx, Nxf2 and Nxt1 ( Figure 1A, Figure 1source data 1). Nxf2 is a homolog of the general messenger RNA (mRNA) export factor Nxf1 but was reported previously as being dispensable for canonical mRNA transport in S2 cells eLife digest For an organism to form and grow correctly, it must rely on the genetic information it has received from its parents. DNA, however, is full of elements called transposons that can disrupt this information by moving around the genome and inserting themselves into genes. Changes caused by these invading elements can lead to devastating effects, such as cancer and cell death. To shield their DNA from harm, organisms have evolved regulatory machineries to recognize and correct alterations that may be damaging.
One way cells can protect their DNA is by silencing disruptive transposons using small molecules known as piRNAs. These protective molecules detect transposons as soon as they are active and recruit other proteins to switch them off. However, questions still remain about how specific proteins recruited by piRNAs are involved in this process. In flies, a protein called Panoramix (Panx) is known to trigger transposon silencing, but how it does this is still unclear. Now, Fabry, Ciabrelli, Munafò et al. set out to investigate how Panx silences active transposons in the ovaries of flies and whether other proteins are involved. Like so many other proteins, Panx was found not to work alone but to form a complex with two other proteins, called Nxf2 and Nxt1. The experiments showed that all three components of the complex, named PICTS, are critical for transposon control in flies, but Panx is the engine that drives the machine.
Panx and Nxf2 were found to stabilize each other, and together with co-factor Nxt1 place a mark on the genome at the point where the transposon emerges, effectively switching it off. Notably, although Nxf2 and Nxt1 are part of a family of proteins that export molecules from the nucleus, both these factors appear to have been repurposed to silence transposable elements within the genome.
This work expands our understanding of how cells employ regulatory machineries, like the PICTS complex, to guard against disruptive genetic changes. These mechanisms are often conserved throughout evolution, and the findings presented here may help identify ways to counteract harmful changes caused by transposons in other organisms, including humans. However, more work would be required to deepen our knowledge of how these processes work.  . Green, GFP-Nxf2; magenta, Aubergine; blue, DNA. Scale bar, 10 mm. (D) as (A) but co-purification from GFP-Nxf2 ovary lysates (GFP-Nxf2, n = 4; control, n = 3). (E) Bar graphs showing fold changes in steady-state RNA levels of soma-(mdg1, gypsy) and germline-(HeT-A, burdock) specific transposons in total ovarian RNA from the indicated genotypes (relative to wild-type and normalized to rp49). * denotes P value < 0.05; ** denotes p<0.001 (unpaired t-test). Error bars indicate standard deviation (n = 3). (F) Scatter plots showing expression levels (reads per million sequenced reads) of genes (in grey) and transposons (in purple) from total RNA from ovaries of the indicated genotypes (left, nxf2; right, panx; n = 3; r 2 values represent expression of genes only). Transposons whose abundance change more than four-fold compared to heterozygotes are highlighted in blue.  Herold et al., 2003). Nxf2 contains all domains present in the family defined by Nxf1, namely an amino-terminal region (NTR), an RNA-binding domain (RBD), leucine-rich repeats (LRR), the NTF2-like domain (NTF2) and a Ubiquitin-associated (UBA) domain ( Figure 1B   . While the NTR, LRRs and RBD are typically involved in cargo binding, the NTF2 and UBA domains mediate binding to the Nuclear Pore Complex (NPC) and are required for Nxf1-mediated RNA export Fribourg et al., 2001). Nxt1, also known as p15, is a co-factor of Nxf1 responsible for interaction with the NPC, specifically through the NTF2-fold (Fribourg et al., 2001;Lévesque et al., 2001). Interestingly, Nxt1 was also reported to interact with Nxf2 Herold et al., 2000). Both Nxf2 and Nxt1 were previously identified in screens for piRNA-guided silencing in somatic and germline cells, and their depletion resulted in female sterility (Czech et al., 2013;Handler et al., 2013;Muerdter et al., 2013). Contrary to expectations based upon previous findings (Sienski et al., 2015;Yu et al., 2015), we saw no enrichment for Piwi by mass spectrometry ( Figure 1A), results that are consistent with another recent study (Batki et al., 2019). However, co-immunoprecipitation experiments detected weak but reproducible interactions between Piwi and Panx, Nxf2, and Nxt1, but not with a negative control ( Figure 1-figure supplement 1D), suggesting that low amounts of transposon substrates in unperturbed cells and/or transient associations might push Piwi below the limit of detection by less sensitive approaches. Using CRISPR/Cas9, we generated flies that express a GFP-Nxf2 fusion protein from the endogenous nxf2 locus. GFP-Nxf2 is expressed in follicle and germline cells of the ovary and localizes predominantly to nuclei ( Figure 1C and Figure 1-figure supplement 1C). Mass spectrometry of GFP-Nxf2-associated proteins identified Panx, Nxf2, and Nxt1, as binding partners ( Figure 1D, Figure 1source data 2), implying the existence of a complex containing these three factors, which we named the Panx-induced co-transcriptional silencing (PICTS) complex.
We therefore generated two nxf2 mutant alleles, nxf2 F10* and nxf2 D1* , which harbor premature stop codons that disrupt the nxf2 open reading frame from amino acid 10 onwards ( Figure 1B and To assess the specificity of the impact of nxf2 mutations on the transcriptome, we performed RNA-seq from total RNA of heterozygote and mutant ovaries, using panx mutants for comparison. As reported previously, the expression of protein-coding genes was not generally affected in nxf2 mutants, with only 16 out of 7252 (r 2 = 0.963) being changed more than 4-fold ( Figure 1F) Herold et al., 2003). In contrast, 28 out of 60 transposon families (that were above the expression threshold of 1 rpm) were de-repressed by more than 4-fold in nxf2 mutants. Similar results were obtained for panx mutants: 16 out of 60 transposons were de-repressed and only 6 out of 7252 genes mis-regulated (r 2 = 0.991). ChIP-seq for the H3K9me3 mark from nxf2 mutant ovaries showed reduced methylation levels at the same transposon families that were derepressed according to RNA-seq, such as Het-A, while randomly chosen genomic intervals were not changed ( Figure 1-figure supplement 2A-B).
Ovarian somatic cells (OSCs), cultured in vitro, express a functional piRNA-guided co-transcriptional gene silencing machinery and provide a convenient context for mechanistic studies  (Saito et al., 2009). RNA-seq from OSCs depleted of Piwi, Panx or Nxf2 showed marked de-repression of soma-specific (e.g. mdg1, gypsy, 297) and intermediate transposon families (e.g. blood) when compared to control cells treated with GFP siRNAs ( Figure 1G left; Figure 1-figure supplement 2C). The de-repression of transposon families strongly correlates with the reduction in H3K9me3 levels mapped over their consensus sequences in ChIP-seq samples generated from the same knockdowns ( Figure 1G right). In contrast, no major changes in H3K9me3 were detected over transposons that show no de-repression in these cells upon piwi, panx, or nxf2 knockdown.
We next focused on a set of 233 individual, Piwi-regulated genomic transposon insertions in OSCs (see Materials and methods for details). This enabled analysis of chromatin states on individual loci, including flanking regions, rather than averaging contributions over a consensus sequence ( Ohtani et al., 2013). To test for such dependencies among TGS factors, we depleted Piwi, Panx or Nxf2 in germ cells of flies expressing GFP-Nxf2 ( Figure 2A) or GFP-Panx ( Figure 2B). Germline knockdown of piwi had no effect on the localization of either Nxf2 or Panx. Depletion of Panx, however, led to a pronounced loss of nuclear GFP-Nxf2 in nurse cell nuclei ( Figure 2A). The reciprocal was also true, with GFP-Panx nuclear signal being reduced upon nxf2 knockdown in nurse cells ( Figure 2B). Similarly, the individual depletion of Panx or Nxf2 in follicle cells resulted in a reduction of both proteins (Figure 2-figure supplement 1A-B). To assess whether the observed reduction reflects protein stability rather than mislocalization, we performed western blots on ovarian lysates. Panx protein level was strongly reduced in nxf2 mutant ovaries ( Figure 2C) and GFP-Nxf2 signal was completely lost in homozygous panx mutants ( Figure 2D). Of note, mRNA levels of Nxf2 and Panx were not affected when the other factor was mutated ( Figure 1F), implying regulation at the protein level. Considered together, the localization and stability of Nxf2 and Panx are interdependent in vivo.

PICTS complex formation is required for TGS
To map the domains of Nxf2 and Panx that mediate their interaction, we expressed various combinations of full-length, truncated, or mutant proteins in S2 cells or in OSCs where native protein expression had been reduced by RNAi ( Figure 3A). Interactions were tested by co-immunoprecipitation and western blot analyses, and the subcellular localization analyzed by immunofluorescence staining. In OSCs, the ability of each construct to rescue transposon de-repression was monitored by qPCR.
Full-length Nxf2 and Panx robustly co-immunoprecipitated ( Figure 3B) and co-localized in S2 cell nuclei ( Figure 3C). Removing the carboxy-terminal half of Panx (Panx-DC) yielded a protein that remained nuclear, while co-expressed Nxf2 remained largely cytoplasmic, and these proteins no longer formed a complex ( Figure 3B-D). When expressed alone in S2 cells, Nxf2 remained predominantly cytoplasmic (ZsGreen-HA in Figure 3C), suggesting that interaction with Panx is necessary for nuclear localization of Nxf2. Panx-DN, in contrast, retained the ability to interact with Nxf2 but failed to localize to the nucleus ( Figure 3B-D). Strikingly, enforced localization of Panx-DN to the nucleus also restored nuclear localization of Nxf2 ( Figure 3D and Deleting only either the coiled coil domain #2 or C-terminal region, which together make up most of the carboxy-terminal half of Panx, reduced co-purification with Nxf2 ( Figure 3D and  Figure 3E). Overall, these results suggest that the N-terminal part of Panx carries its nuclear localization signal that aids proper localization of Nxf2 via interaction with the Panx C-terminal region.
To probe the regions of Nxf2 that are essential for its function, we expressed, in the presence of full-length Panx, Nxf2 proteins that lacked either the regions required for RNA cargo binding or the region essential for its association with the NPC Fribourg et al., 2001;Herold et al., 2001) ( Figure 3A and  The NTF2-like fold was previously shown to mediate the interaction of NXF proteins with Nxt1 (Herold et al., 2000;Kerkow et al., 2012;Suyama et al., 2000). To probe a potential requirement for Nxt1 in silencing, we generated NTF2 domain point mutants in residues involved in the interaction with Nxt1 (Kerkow et al., 2012). All four Nxf2-NTF2 point mutants localized predominantly to the nucleus and co-precipitated quantities of Panx comparable to the full-length control ( Figure 3D and Figure 3-figure supplement 2E-F). Yet, three mutants (#1, #2, and #3) failed to rescue transposon de-repression upon depletion of Nxf2 ( Figure 3E), pointing to an involvement of Nxt1 in silencing. Indeed, reduced amounts of Nxt1 were recovered with the NTF2 mutants #1 and #2, while the NTF2 mutant #4, which rescued transposon expression ( Figure 3E), as well as UBA mutants #1 and #2 showed levels comparable to full-length Nxf2 ( Figure 3G and Tethering of Nxf2 and Nxt1 to RNA triggers silencing Artificial tethering of Panx to nascent RNA or DNA was previously shown to result in co-transcriptional silencing and the concurrent accumulation of repressive chromatin marks (Sienski et al., 2015;Yu et al., 2015). To test whether Nxf2 could induce TGS, we created an integrated sensor comprising the Drosophila simulans ubiquitin promoter driving an HA-tagged ZsGreen transcript with 9 BoxB sites in its 3' UTR in OSCs ( Figure 4A). As expected from previous studies (Sienski et al., 2015;Yu et al., 2015), expression of lN-Piwi did not lead to reduced RNA or protein levels ( Figure 4B and  (Sienski et al., 2015;Yu et al., 2015). lN-Nxf2 caused an even stronger reduction of RNA and protein expression from the reporter ( Figure 4B). FISH experiments supported consistent repression by Panx and Nxf2 (Figure 4-figure supplement 1C). Tethering of lN-Nxt1 also induced reporter repression ( Figure 4B). Strikingly, upon tethering of Nxt1, Nxf2 or Panx, we observed increased levels of H3K9me3 over the reporter ( Figure 4C). These data suggest that Nxf2 and Nxt1, along with Panx, act as key effectors of co-transcriptional silencing and are each sufficient to recruit the downstream silencing machinery.
We also tested whether Nxf2 could silence artificial targets upon tethering to DNA rather than nascent transcripts. Our sensor construct carried 8 LacO sites upstream of the D. sim. ubiquitin promoter, which drives the expression of HA-ZsGreen ( Figure 4D). While tethering of LacI-Piwi did not affect sensor expression, LacI-Panx resulted in robust reductions in both RNA and protein levels ( Figure 4E and Figure 4-figure supplement 1D), as previously reported (Sienski et al., 2015). LacI-Nxf2 also silenced the reporter, reducing both mRNA and protein output, albeit to a lesser extent than LacI-Panx tethering. Surprisingly, LacI-Nxt1 was unable to silence the sensor construct. Repression by tethered Panx or Nxf2 resulted in a striking decrease in H3K4me2 marks over the transcribed parts (i.e. promoter and ZsGreen coding region), while the remainder of the reporter showed low read coverage and little difference in the prevalence of the mark (Figure 4-figure supplement 1E). Conversely, H3K9me3 increased upon repression ( Figure 4F). Of note, the entire reporter sequence (except a small gap around the LacO site and the 3' UTR where the mappability is poor) was prominently decorated with H3K9me3, suggesting spreading of this chromatin mark following initial silencing. The data presented above identified functional elements within Nxf2 and Panx that are required for proper localization, interaction, and Piwi-dependent transcriptional gene silencing. We next examined the ability of Panx and Nxf2 mutants to silence our artificial DNA reporters, thus bypassing Piwi-piRNA dependent target recognition. Neither LacI-Nxf2-DCargo nor LacI-Panx-DN, which were predicted to interact with their full-length partner in OSCs ( Figure 3B-D), were able to repress the sensor ( Figure 4G). Yet, LacI-Panx-DC, which could not interact with Nxf2, substantially reduced RNA and protein expression from the sensor, with its effects as robust as upon DNA tethering of LacI-HP1a ( Figure 4G). This suggests that the amino-terminal part of Panx is necessary and sufficient to enforce silencing of an artificial reporter independent of its interaction with Nxf2.

Discussion
Our data identify Nxf2 and Nxt1 as critical mediators of co-transcriptional gene silencing, acting in concert with Panx to repress loci in response to Piwi-piRNA target engagement ( Figure 4H). The emerging model for piRNA-dependent silencing implies that target recognition by Piwi is necessary to recruit the PICTS complex onto the appropriate nascent RNA targets. Difficulties in detecting stable interactions between Piwi and PICTS components in vivo may arise from a requirement for Piwi target engagement to licence it for recruitment of silencing complexes, as has been suggested previously (Sienski et al., 2015;Yu et al., 2015). The same mechanism may underlie the difficulties experienced in observing Piwi on its target loci by ChIP (Marinov et al., 2015).
We find that Panx and Nxf2 are interdependent for their protein stability and proper subcellular localization, underscoring the fact that correct assembly of the PICTS complex is essential for TGS, while the silencing capacity, per se, resides in Panx. Of note, previous work reported a partial destabilization of Nxf2 in cells depleted of Nxt1 , potentially extending the interdependency to all three proteins. RIP-seq experiments from unperturbed cells found transposon RNAs enriched only with Panx, as reported (Sienski et al., 2015), but not with Nxf2 ( Figure 4-figure supplement 1F), possibly due to low substrate availability combined with an insensitive assay. These results are consistent with another recent report that did not detect transposon enrichment in Nxf2 CLIP-seq from wild-type cells (Batki et al., 2019). However, two other studies identified transposon mRNA association with Nxf2 in CLIP-seq experiments upon depletion of the previously described TGS factor, Mael (Zhao et al., 2019), or by using a stable cell line and depletion of endogenous Nxf2 (Murano et al., 2019). Considered together, these data suggest that Nxf2 might be important for stabilizing the binding of Panx to nascent RNAs. However, precisely how Nxf2 executes this function remains to be fully elucidated. Of note, Murano and colleagues find that Panx interacts with Nxf2, Piwi, Mael and Arx (Murano et al., 2019), which could imply that other TGS factors come into contact with the PICTS complex, although the relationship between these factors and PICTS requires further investigation.
Mutational analyses suggest that Panx and Nxf2 must normally bind Nxt1 to carry out transposon repression. Direct recruitment of any of the PICTS complex components to RNA reporters results in robust chromatin silencing. Upon tethering to DNA, however, Panx induces potent TGS, whereas Nxf2 leads to less prominent effects and Nxt1 shows no silencing capability in our assays. Interestingly, recruitment of the amino-terminal part of Panx alone is necessary and sufficient to induce reporter repression, pinpointing this domain of Panx as harboring the silencing effector function. Future investigations will be crucial to uncover the molecular mechanism by which the Panx amino terminus instructs the downstream chromatin silencing machinery.
Our work, and that of others (Batki et al., 2019;Murano et al., 2019;Zhao et al., 2019) indicates that piRNA-guided co-transcriptional silencing of transposons has coopted several components of the RNA export machinery, namely Nxf2 and Nxt1. Of the four NXF proteins present in flies, only two have thus far been characterized. Interestingly, while Nxf1, acting along with Nxt1, is crucial for canonical mRNA export Fribourg et al., 2001;Herold et al., 2001;Wilkie et al., 2001), Nxf2 has been coopted by the piRNA pathway and functions in co-transcriptional gene silencing. Nxf3, which also is required for transposon repression in germ cells (Czech et al., 2013), is emerging as being critical for the export of piRNA precursors generated from dual-strand clusters in the germline (ElMaghraby et al., 2019; Kneuss et al., 2019. The role of Nxf4, whose expression is testis-specific, is yet to be established. This remarkable functional diversity of NXF family members correlates with tissue-specific expression patterns, and seems conserved in mammals (Yang et al., 2001). However, deciphering how each achieves substrate specificity will be critical to understanding how these homologs can be exclusively dedicated to different targets and confer different fates upon the RNAs that they bind.
Importantly, the fate of the nascent transcript that is detected by Piwi and instructed by PICTS for silencing remains unclear. One hypothesis is that instead of being exported, these targets undergo degradation by the nuclear exosome. Such mechanism would be contrary to yeast, where the TREX complex subunit Mlo3 was shown to oppose siRNA-mediated heterochromatin formation at gene loci (Yu et al., 2018), and suggests that different lineages have evolved different silencing mechanisms. In any case, it is possible that a single transcript from a locus that is marked for silencing might pose a lesser threat than an unsilenced locus and, therefore, not be capable of exerting evolutionary pressure for the detemination of its fate.

Fly stocks and handling
All flies were kept at 25˚C. Flies carrying a BAC transgene expressing GFP-Panx were generated by the Brennecke lab (Handler et al., 2013). Panx frameshift mutants panx M1 and panx M4 were described earlier (Yu et al., 2015). The GFP-Nxf2 fusion knock-in and nxf2 frameshift mutations (nxf2[F10*] and nxf2[D1*]) were generated for this study (see below). Control w 1118 flies were a gift from the University of Cambridge Department of Genetics Fly Facility. For knockdowns we used a stock containing the Dcr2 transgene and a nos-GAL4 driver (described in Czech et al., 2013) and dsRNA lines from the VDRC (panx KK102702 , nxf2 KK101676 , piwi KK101658 ). Fertility of the nxf2 and panx mutant females was scored by crossing ten freshly hatched females to five w 1118 males and counting the number of eggs laid in 12 hr periods and pupae that developed after 7 days.

Generation of mutant and transgenic fly strains
Frameshift mutant alleles of nxf2 were generated by injecting pCFD4 (addgene plasmid # 49411; Port et al., 2014) containing two gRNAs against Nxf2 (generated by Gibson assembly) into embryos expressing vas-Cas9 (Bloomington stock 51323). To generate GFP-Nxf2 fusion knock-in flies, homology arms of approximately 1 kb were cloned into pUC19 by Gibson assembly and co-injected with pCFD3 (addgene plasmid # 49410; Port et al., 2014) containing a single guide RNA into embryos expressing vas-Cas9 (# 51323, Bloomington stock center). Microinjection and fly stock generation was carried out by the University of Cambridge Department of Genetics Fly Facility. Mutant and knock-in flies were identified by genotyping PCRs and confirmed by sanger sequencing.

Immunoprecipitation from ovary lysates and mass spectrometry
Ovaries from~170 GFP-Panx, GFP-Nxf2 and control flies (3-5 days old) were dissected in ice-cold PBS and lysed in 300 ml of CoIP Lysis Buffer (20 mM Tris-HCl pH 7.5, 150 mM NaCl, 2 mM MgCl2, 10% glycerol, 1 mM DTT, 0.1 mM PMSF, 0.2% NP-40 supplemented with complete protease inhibitors [Roche]) and homogenized using a motorized pestle. Lysates were cleared for 5 min at 16000 g and the residual pellet re-extracted with the same procedure. GFP-tagged proteins were immunoprecipitated by incubation with 30 ml of GFP-Trap magnetic beads (Chromotek) for 3 hr at 4˚C on a tube rotator. The beads were washed 6x with Lysis Buffer and 2x with 100 mM Ammonium Bicarbonate, before TMT-labelling followed by quantitative Mass Spectrometry. TMT chemical isobaric labeling were performed as described (Papachristou et al., 2018).

Analysis of mass spectrometry data
Raw data were processed in Proteome Discoverer 2.1 software (Thermo Fisher Scientific) using the SequestHT search engine. The data were searched against a database derived from FlyBase ('dmelall-translation-r6.24') at a 1% spectrum level FDR criteria using Percolator (University of Washington). For the SequestHT node the following parameters were included: Precursor mass tolerance 20 ppm and fragment mass tolerance 0.5 Da. Dynamic modifications were oxidation of M (+15.995 Da), deamidation of N, Q (+0.984 Da) and static modifications were TMT6plex at any N-Terminus and K (+229.163 Da). The consensus workflow included S/N calculation for TMT intensities and only unique peptides identified with high confidence (FDR < 0.01) were considered for quantification. Downstream data analysis was performed on R using the qPLEXanalyzer package (https://doi.org/10. 5281/zenodo.1237825) as described (Papachristou et al., 2018). Only proteins with more than one unique peptide were considered.

Cell culture
Drosophila Ovarian Somatic Cells (OSCs) were a gift from Mikiko Siomi and were cultured at 26˚C in Shields and Sang M3 Insect Medium (Sigma Aldrich) supplemented with 0.6 mg/ml Glutathione, 10% FBS, 10 mU/ml insulin and 10% fly extract (purchased from DGRC) as described (Niki et al., 2006;Saito, 2014;Saito et al., 2009). Cell identity was authenticated by whole genome DNA sequencing in-house. Gibco Drosophila Schneider 2 (S2) cells were purchased from Thermo Fisher Scientific (catalog number R69007) and were grown at 26˚C in Schneider's Drosophila Media (Gibco) supplemented with 10% heat-inactivated FBS. Cell identity was characterized by Thermo Fisher Scientific through isozyme and karyotype analysis (see product description). OSCs and S2 cells tested negative for mycoplasma contamination in-house. Knockdowns (all siRNA sequences are given in Supplementary file 1) and transfections in OSCs were carried out as previously described (Saito, 2014). In short, for knockdown experiments 10 Â 10 6 cells were nucleofected with 200 pmol annealed siRNAs using the Amaxa Cell Line Nucleofector Kit V (Lonza, program T-029). After 48 hr, 10 Â 10 6 cells were nucleofected again with 200 pmol of the same siRNA and allowed to grow for an additional 48 hr before further processing. For rescue experiments, 5 mg of rescue construct plasmid were added to the second knockdown solution. OSCs were transfected with 10 mg of plasmid using Xfect (Clontech), according to manufacturer's instruction. S2 cells were transfected with 2 mg of plasmid using Effectene (Qiagen), according to manufacturer's instructions.
Co-immunoprecipitation from cell lysates S2 cells or OSCs were transfected with 3xFLAG-and HA-tagged constructs (wild-type and mutants). Cells were harvested 48 hr after transfection and lysed in 250 ml of CoIP Lysis Buffer (Pierce) supplemented with Complete protease inhibitors (Roche). 200 mg of proteins for each sample were diluted to 1 ml with CoIP Lysis Buffer and the 3xFLAG-tagged bait was immunoprecipitated by incubation with 20 ml of anti-FLAG M2 Magnetic Beads (Sigma M8823) for 2 hr at 4˚C on a tube rotator. The beads were washed 3 Â 15 min with TBS supplemented with protease inhibitors. Beads were then resuspended in 2x NuPAGE LDS Sample Buffer (Thermo Fisher Scientific) without reducing agent and boiled for 3 min at 90˚C to elute immunoprecipitated proteins. IPs, unbound fractions and input fractions were diluted to 1x NuPAGE LDS Sample Buffer concentration and reducing agent was added. Samples were boiled at 90˚C for 10 min before separating proteins as described below.

Immunofluorescence in ovaries
Fly ovaries were dissected in ice-cold PBS and fixed in 4% paraformaldehyde (PFA) at RT for 15 min. After two quick rinses in PBS with Triton at 0.3% (PBS-Tr), samples were permeabilized with 3 Â 10 min washes with PBS-Tr. Samples were then blocked in PBS-Tr with 1% BSA for 2 hr at RT and then incubated overnight at 4˚C with primary antibodies in PBS-Tr and 1% BSA. The next day, samples were washed 3 Â 10 min at RT in PBS-Tr and incubated overnight at 4˚C with secondary antibodies in PBS-Tr and 1% BSA. The next day, samples were washed 4 Â 10 min in PBS-Tr at RT and DAPI (Thermo Fisher Scientific #D1306) was added during the third wash. After 2 Â 5 min washes in PBS, samples were mounted on slides with ProLong Diamond Antifade Mountant (Thermo Fisher Scientific #P36961) and imaged on a Leica SP8 confocal microscope (63x and 100x Oil objective). The following antibodies were used: chicken anti-GFP (ab13970), rabbit anti-Piwi (described in Brennecke et al., 2007), mouse anti-Aub , anti-Rabbit-555 (Thermo Fisher), anti-Mouse-647 (Thermo Fisher), anti-Chicken-647 (Abcam).

RNA fluorescent in situ hybridization (RNA FISH)
RNA FISH was performed with Hybridization Chain Reaction (HCR), similar as reported (Ang and Yung, 2016;Choi et al., 2014). OSCs were fixed for 15 min in 4% PFA, washed 2 Â 5 min with PBS and permeabilized for at least 24 hr in 70% ethanol at À20˚C. Ethanol was removed and slides washed twice for 5 min in 2x Saline-Sodium Citrate buffer (SSC). Priming for hybridization was done by incubating for 10 min in 15% formamide in 2x SSC. HCR probes were diluted to 1 nM each in hybridization buffer (15% formamide, 10% dextran sulfate in 2x SSC) and incubated overnight at 37C in a humidified chamber. Excess probes were removed by rinsing twice in 2x SSC and washing once in 30% formamide for 10 min at 37˚C. HCR hairpins conjugated to AlexaFluor-647 (IDT) were heat-denatured and diluted to 120 nM in 5x SSC and 0.1% Tween-20. HCR amplification was carried out for 2 hr at RT in the dark and washed 3 Â 10 min with 5x SSC and 0.1% Tween-20. Nuclei were stained with DAPI for 10 min, followed by 3 Â 10 min washes in 5x SSC. Slides were mounted with ProLong Diamond Antifade Mountant (Thermo Fisher Scientific) and imaged on a Leica SP8 confocal microscope (100x Oil objective). The sequences of all probes are given in Supplementary file 1.

Image analysis
Intensity plot profiles across individual egg chambers were acquired in Fiji (lines displayed). Intensity values for each channel were averaged over 10 pixels and adjusted as a percentage of the highest value along the profile. A threshold of 30% DAPI intensity was set to define nuclei. Individual egg chambers used for analysis are displayed for each channel with inverted LUT.

Tethering experiments
For RNA tethering, OSCs with a stable integration of the sensor plasmid (pDsimUbi-HA-ZsGreen-NLuc-9xBoxB) were generated in the lab. 4 Â 10 6 cells were nucleofected with 5 mg of plasmid expressing lN-3xFLAG-tagged constructs, as described above. After 48 hr, 4 Â 10 6 cells were nucleofected again with 5 mg of the same plasmid and allowed to grow for an additional 48 hr before the relative expression of the sensor was analyzed. For DNA tethering, OSCs were transiently transfected with 8xLacO-pDsimUbi-HA-ZsGreen sensor plasmid and LacI-3xFLAG fusion constructs. Cells were allowed to grow for 72 hr before the relative expression of the sensor was determined.

ChIP-seq from ovaries
Ovaries from 120 to 150 adult flies were dissected in ice-cold PBS, collected in 1.5 ml Bioruptor Microtubes (Diagenode #C30010016), and immediately frozen at À80˚C. Samples were crosslinked in 1 ml A1 buffer (60 mM KCl, 15 mM NaCl, 15 mM HEPES pH 7.6, 4 mM MgCl2, 0.5% Triton X-100, 0.5 mM dithiothreitol (DTT), 10 mM sodium butyrate and complete EDTA-free protease inhibitor cocktail [Roche #04693159001]), in the presence of 1.8% formaldehyde. Samples were homogenized with a micropestle for 2 min and incubated for a total time of 15 min at RT on a rotating wheel. Crosslinking was stopped by adding 225 mM glycine followed by incubation for 5 min on a rotating wheel. The homogenate was centrifuged for 5 min at 4,000 g at 4˚C. The supernatant was discarded, and the nuclear pellet was washed twice in 1 ml A1 buffer and once in 1 ml of A2 buffer (140 mM NaCl, 15 mM HEPES pH 7.6, 1 mM EDTA, 0.5 mM EGTA, 1% Triton X-100, 0.5 mM DTT, 0.1% sodium deoxycholate, 10 mM sodium butyrate and complete mini EDTA-free protease inhibitor cocktail) at 4˚C. Nuclei were then resuspended in 100 ml A2 buffer with 1% SDS and 0.5% N-laurosylsarcosine and incubated for 2 hr at 4˚C with agitation at 1,500 rpm. Chromatin was sonicated using a Bioruptor Pico (Diagenode #B01060010) for 16 cycles of 30 s on/30 s off. Sheared chromatin size peaked at 150 bp. After sonication and 5 min high-speed centrifugation at 4˚C, fragmented chromatin was recovered in the supernatant and the final volume was raised to 1 ml in A2 buffer with 0.1% SDS. 50 ml of the diluted samples were used as DNA input control, in a final volume of 200 ml of A2 buffer with 0.1% SDS. Chromatin for IP was precleared by addition of 15 ml of Protein A/G Magnetic Beads (Thermo Fisher Scientific) suspension followed by overnight incubation at 4˚C. Beads were removed by centrifugation, and anti-H3K9me3 (Active Motif #39161) antibody was added (1:200 dilution) to 5 mg of chromatin and incubated for 4 hr at 4˚C on a rotating wheel. 50 ml of Protein A/G Magnetic Beads were added, and incubation was continued overnight at 4˚C. Antibody-protein complexes were washed 4 times in A3 (A2 +0.05% SDS) buffer and twice in 1 mM EDTA, 10 mM Tris (pH 8) buffer for 5 min at 4˚C on a rotating wheel. Chromatin was eluted from the beads in 200 ml of 10 mM EDTA, 1% SDS, 50 mM Tris (pH 8) for 30 min with agitation at 1,500 rpm and then reversecrosslinked overnight at 65˚C, together with the input DNA. IP and input samples were treated with 2 ml of Proteinase K (Thermo Fisher Scientific #EO0491) for 3 hr at 56˚C. DNA was purified using the MinElute PCR purification Kit (Thermo Fisher Scientific), according to manufacturer's instructions, and resuspended in 30 ml water. Recovered DNA was quantified with Qubit 4 Fluorometer (Thermo Fisher Scientific) and analysed with Agilent Bioanalyzer 2100 High Sensitivity DNA Chip (Agilent). DNA libraries were prepared with NEBNext Ultra II DNA Library Prep Kit for Illumina (NEB), according to manufacturer's instructions. DNA libraries were quantified with KAPA Library Quantification Kit for Illumina (Kapa Biosystems) and deep-sequenced with Illumina HiSeq 4000 (Illumina).

ChIP-seq from OSCs
For ChIP from OSCs we adapted a protocol by Schmidt and colleagues (Schmidt et al., 2009). In short, 10 Â 10 6 OSCs were crosslinked in 1% formaldehyde for 10 min. Crosslinking was quenched by addition of glycine solution, followed by three washes in ice-cold PBS. Crosslinked cells were either snap-frozen in liquid nitrogen and stored at À80˚C or processed immediately. Cells were resuspended in 1 ml buffer LB1 (50 mM HEPES-KOH pH 7.5, 140 mM NaCl, 1 mM EDTA, 10% glycerol, 0.5% Igepal CA-630, 0.25% Triton-X 100, EDTA-free protease inhibitor cocktail [Roche]) and incubated on ice for 10 min while inverting several times. Cells were centrifuged for 5 min at 2,000 g at 4˚C. Supernatant was discarded and pellet resuspended in 1 ml LB2 10 mM Tris-HCL pH 8.0, 200 mM NaCl, 1 mM EDTA, 0.5 M EGTA, EDTA-free protease inhibitor cocktail). Cells were incubated on ice for 5 min and centrifuged again. Isolated nuclei were resuspended in 300 ml sonication buffer LB3 (10 mM Tris-HCL pH 8, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.1% Na-Deoxycholate, 0.5% N-lauroylsarcosine, EDTA-free protease inhibitor cocktail) and transferred in 1.5 ml Bioruptor Microtubes (Diagenode). Chromatin was sonicated using a Bioruptor Pico (Diagenode) for 16 cycles of 30 s on/30 s off. Sheared chromatin size peaked at 150 bp. The lysate was cleared by high-speed centrifugation at 4˚C. 100 ml Protein A Dynabeads (Thermo Fisher Scientific) were incubated with 5 ml H3K9me3 (Active Motif #39161) or H3K4me2 antibody (Millipore # 07-030) over night at 4˚C while rotating. The cleared lysate was split in two equal fractions and a 5 ml input fraction was saved for further processing. Lysate volumes were adjusted to 300 ml with LB3 and Triton-X 100 was added to a final concentration of 1%. Lysates were incubated with either H3K9me3 or H3K4me2 coated beads over night at 4˚C while rotating. Washing, reverse-crosslinking, DNA purification and library preparation was done as described above (ChIP-seq from ovaries).

RNA isolation
Cell pellets or fly ovaries were lysed in 1 ml Trizol and RNA was extracted using RNeasy mini prep column (Qiagen), according to manufacturer's instructions.
qPCR analysis 1 mg of total RNA was treated with DNAseI (Thermo Fisher Scientific), according to manufacturer's instructions. Reverse transcription was performed with Superscript III First Strand Synthesis Kit (Thermo Fisher Scientific), using oligo(dT) 20 primers, according to the manufacturer's instructions. Real-time PCR (qPCR) experiments were performed with a QuantStudio Real-Time PCR Light Cycler (Thermo Fisher Scientific). Transposon levels were quantified using the DDCT method (Livak and Schmittgen, 2001), normalized to rp49 and fold changes were calculated relative to the indicated controls. All oligonucleotide sequences are given in Supplementary file 1.

RIP-seq from ovaries
Ovaries from~100 GFP-Panx or GFP-Nxf2 flies (3-5 days old) were dissected in ice-cold PBS and fixed with 0.1% PFA for 20 min, followed by quenching with equal volumes of 125 mM Glycine. Fixed ovaries were lysed in 200 ml of RIPA Buffer (supplemented with complete protease inhibitors (Roche) and RNasin Plus 40 U/ml) and homogenized using a motorized pestle. Lysates were incubated 3 min at 37˚C with 4 ml of Turbo DNase, incubated 20 min at 4˚C on a tube rotator and sonicated with a Bioruptor Pico (3 cycles of 30 s on/30 s off). Lysates were pre-cleared using 40 ml of Pierce Protein A/G beads for 1 hr at 4˚C and GFP-tagged proteins were immunoprecipitated by incubation with 50 ml of GFP-Trap magnetic agarose beads (Chromotek) overnight at 4˚C. An aliquot of pre-cleared input lysate was saved for RNA isolation and library preparation. Following 3 washes in 150 mM KCl, 25 mM Tris (pH 7.5), 5 mM EDTA, 0.5% NP40, 0.5 mM DTT (supplemented with protease inhibitors and RNasin Plus 1:1000), IP and input samples were reverse crosslinked in 1x Reverse Crosslinking buffer (PBS, 2% Nlauroyl sarcosine, 10 mM EDTA, 5 mM DTT) and Proteinase K. RNA isolation was performed using Trizol and 100 ng of input or IP RNA were used for library preparation using the SMARTer stranded RNA-seq Kit (Clontech). DNA libraries were quantified with KAPA Library Quantification Kit for Illumina (Kapa Biosystems) and deep-sequenced with Illumina HiSeq 4000 (Illumina).

Small RNA-seq library preparation
Small RNA libraries were generated as described previously (Jayaprakash et al., 2011). Briefly, 18to 29-nt-long small RNAs were purified by PAGE from 10 mg of total ovarian RNA. Next, the 3' adapter (containing four random nucleotides at the 5' end) was ligated overnight using T4 RNA ligase 2, truncated KQ (NEB). Following recovery of the products by PAGE purification, the 5' adapter (containing four random nucleotides at the 3' end) was ligated to the small RNAs using T4 RNA ligase (Abcam) for 1 hr. Small RNAs containing both adapters were recovered by PAGE purification, reverse transcribed and PCR amplified prior quantification using the Library Quantification Kit for Illumina (Kapa Biosystems) and sequenced on an Illumina HiSeq 4000 (Illumina). All adapter sequences are given in Supplementary file 1.

RNA-seq library preparation
1 mg of total RNA was used as input material for library preparation. The NEBNext Poly(A) mRNA magnetic Isolation Module (NEB) was used to isolate poly(A) RNAs. Libraries were generated with the NEBNext Ultra Directional RNA Library Prep kit for Illumina (NEB) according to manufacturer's instructions. The pooled libraries were quantified with KAPA Library Quantification Kit for Illumina (Kapa Biosystems) and sequenced on an Illumina HiSeq 4000 (Illumina).

RNA-seq, small RNA-seq, RIP-seq and ChIP-seq analysis
Raw fastq files generated by Illumina sequencing were analysed by a pipeline developed in-house. In short, the first and last base of each 50 bp read were removed using fastx trimmer (http://hannonlab.cshl.edu/fastx_toolkit/). RIP-seq reads were first aligned against rRNAs and mapped reads discarded. High-quality reads were aligned to the Drosophila melanogaster genome release 6 (dm6) downloaded from Flybase using STAR (Dobin et al., 2013). For transposon-wide analysis, genome multi-mapping reads were randomly assigned to one location using option '-outFilterMultimapNmax 1000 -outMultimapperOrder Random' and non-mapping reads were removed. Alignment files were then converted back to fastq format with samtools (Li et al., 2009) and re-aligned to the transposon consensus sequences allowing multi-mappers that were assigned to a random position. Generated bam alignment files were indexed using samtools index. For genome-wide analyses, multi-mapping reads were removed to ensure unique locations of reads. Normalization was achieved by calculating rpm (reads per million) using the deepTools2 bamCoverage function (Ramírez et al., 2016) with 10 bp bin sizes. The scaling factor for transposon mapping reads was calculated from reads that aligned to transposon consensus sequences relative to genome aligned reads. Reads mapping to genes were counted with htseq (Anders et al., 2015) and transposon derived reads were calculated using a custom script (available with this article as Source code 1). Metaplots flanking euchromatic insertion sites and transposon coverage plots were calculated by deepTools2 with bin sizes of 10 bp and 50 bp, respectively. Stranded RNA libraries were trimmed, aligned and indexed as described above. Alignment files were split in sense and antisense reads using samtools view. Normalization of the split alignment files as well as feature counting was performed as described above. For transposon expression analysis only sense reads were considered. Differential expression analysis was performed using a custom build R script (available with this article as Source code 2). Adapters from raw small RNA fastq files were clipped with fastx_clipper (adapter sequence AGATCGGAAGAGCACACGTC TGAACTCCAGTCA) keeping only reads with at least 23 bp length. Then the first and last four bases were trimmed using seqtk (https://github.com/lh3/seqtk). Alignment and normalization were performed as described above. Only high-quality small RNA reads with a length between 23 and 29 bp were used for further analysis of piRNA profiles. piRNA distribution was calculated and plotted in R. For piRNA coverage plots over TEs, only the 5' position of reads was plotted.

Generation of annotation files for RNA-seq and ChIP-seq analysis
The locations of euchromatic transposon insertions in OSCs were derived from Sienski et al. (2012) and updated to dm6 genome release coordinates using the UCSC liftOver tool. Transposon consensus sequences were downloaded from Flybase. Mappability tracks for dm6 with 50 bp resolution were calculated as described (Derrien et al., 2012). Piwi-dependent OSC insertions were defined by comparing H3K9me3 signal intensities of siRNA-mediated knockdowns for gfp and piwi. Signal was counted by htseq using a customized GTF file including the locations of all euchromatic TE insertions in OSCs and reads were normalized to rpm. TE insertions were annotated as Piwi-dependent if the ratio of normalized signal intensity of GFP knockdown versus Piwi knockdown was higher than 2.

Plotting and data visualization
Random genomic windows for box plots of H3K9me3 ChIP-seq data were calculated using BEDtools' random function (Quinlan and Hall, 2010) with bin size 5,000 bp, bin number 1000 and random seed number 800. 100 random windows were chosen (number 200-300) and analysed along with ChIP-seq data for de-repressed TEs and those not affected. Welch two sample t-test was applied for statistics. Metaplots of euchromatic TE insertions as well as TE coverage plots for RNA-seq and ChIP-seq data were generated with deepTools2 and Adobe Illustrator. Scatterplots for differentially expressed transposons and genes were generated with R package ggplot2. Heatmaps were calculated with deepTools2 and data plotted in R. For scatterplots, only TEs and genes with a scaled read count larger than 1 (rpm >1) were used in the analysis and included in plots.

Quantification and statistical analysis
Statistical analysis applied to qPCR data sets was calculated by unpaired (two sample) t Test. The number of biological replicates is indicated in the figure legends. Statistical analysis applied to data sets displayed as box plots (Figure 1-figure supplement 2A) was calculated by Welch two sample t-test.

Data availability
Sequencing data reported in this paper has been deposited in Gene Expression Omnibus under ID code GSE121661. Mass Spectrometry data has been deposited to PRIDE Archive under ID code PXD011415.

Data availability
Sequencing data reported in this paper has been deposited in GEO under accession number GSE121661. Mass Spectrometry data has been deposited to the PRIDE Archive (accession number PXD011415).
The following datasets were generated: