Telomeric Trans-Silencing: An Epigenetic Repression Combining RNA Silencing and Heterochromatin Formation

The study of P-element repression in Drosophila melanogaster led to the discovery of the telomeric Trans-Silencing Effect (TSE), a repression mechanism by which a transposon or a transgene inserted in subtelomeric heterochromatin (Telomeric Associated Sequence or TAS) has the capacity to repress in trans in the female germline, a homologous transposon, or transgene located in euchromatin. TSE shows variegation among egg chambers in ovaries when silencing is incomplete. Here, we report that TSE displays an epigenetic transmission through meiosis, which involves an extrachromosomal maternally transmitted factor. We show that this silencing is highly sensitive to mutations affecting both heterochromatin formation (Su(var)205 encoding Heterochromatin Protein 1 and Su(var)3–7) and the repeat-associated small interfering RNA (or rasiRNA) silencing pathway (aubergine, homeless, armitage, and piwi). In contrast, TSE is not sensitive to mutations affecting r2d2, which is involved in the small interfering RNA (or siRNA) silencing pathway, nor is it sensitive to a mutation in loquacious, which is involved in the micro RNA (or miRNA) silencing pathway. These results, taken together with the recent discovery of TAS homologous small RNAs associated to PIWI proteins, support the proposition that TSE involves a repeat-associated small interfering RNA pathway linked to heterochromatin formation, which was co-opted by the P element to establish repression of its own transposition after its recent invasion of the D. melanogaster genome. Therefore, the study of TSE provides insight into the genetic properties of a germline-specific small RNA silencing pathway.


Introduction
Repression of transposable elements (TEs) involves complex mechanisms that can be linked to either small RNA silencing pathways or chromatin structure modifications depending on the species and/or the TE family [1][2][3][4][5][6]. Drosophila species are particularly relevant to the study of these repression mechanisms since some families of TEs are recent invaders, allowing genetic analysis to be carried out on strains with or without these TEs [7,8]. In some cases, crossing these two types of strains induces hybrid dysgenesis, a syndrome of genetic abnormalities resulting from TE mobility [9,10]. In D. virilis, repression of hybrid dysgenesis has been correlated to RNA silencing since small RNAs of the retroelement Penelope, responsible for dysgenesis, were detected in nondysgenic embryos but not in dysgenic embryos [11]. In D. melanogaster, repression of retrotransposons can be established by noncoding fragments of the corresponding element (I factor [12,13], ZAM, and Idefix [14]) and can be in some cases (gypsy [15], mdg1 [16], copia [17], Het-A, TART [18,19], and ZAM, Idefix [C. Vaury and S. Desset, personal communication]) sensitive to mutations in genes from the Argonaute family involved in small RNA silencing pathways. In the same species, strong repression of the DNA P TE, by a cellular state that has been called ''P cytotype'' [10], can be established by one or two telomeric P elements inserted in heterochromatic ''Telomeric Associated Sequences'' (TAS) at the 1A cytological site corresponding to the left end of the X chromosome [20][21][22][23][24]. This includes repression of dysgenic sterility resulting from P transposition. We have previously shown that this P cytotype is sensitive to mutations affecting both Heterochromatin Protein 1 (HP1) [21] and the Argonaute family member AUBERGINE [25]. P repression corresponds to a new picture of TE repression shown, using an assay directly linked to transposition, to be affected by heterochromatin and small RNA silencing mutants.
In the course of the study of P cytotype, a new silencing phenomenon has been discovered. Indeed, a P-lacZ transgene or a single defective P element inserted in TAS can repress expression of euchromatic P-lacZ insertions in the female germline in trans, if a certain length of homology exists between telomeric and euchromatic insertions [23,26]. This homology-dependent silencing phenomenon has been termed Trans-Silencing Effect (TSE) [26]. Telomeric transgenes, but not centromeric transgenes, can be silencers and all euchromatic P-lacZ insertions tested can be targets [23,26]. TSE is restricted to the female germline (unpublished data) and has a maternal effect since repression occurs only when the telomeric transgene is maternally inherited [27]. Further, when TSE is not complete, variegating germline lacZ repression is observed from one egg chamber to another, suggesting a chromatin-based mechanism of repression [28]. Recently, an extensive analysis of small RNAs complexed with PIWI family proteins (AUBERGINE, PIWI, and AGO3) was performed in the Drosophila female germline [4]. The latter study showed that most of the RNA sequences associated to these proteins derive from TEs. TSE corresponds likely to such a situation.
Here, we analyze the genetic properties of TSE and show that it has an epigenetic transmission through meiosis, which involves an extrachromosomal maternally transmitted stimulating component. Further, in order to investigate the mechanism behind TSE, we performed a candidate gene analysis to identify genes whose mutations impair TSE. We found that TSE is strongly affected both by mutations in genes involved in heterochromatin formation and in the recently discovered small RNA silencing pathway called ''repeat-associated small interfering RNAs'' (rasiRNA) pathway [3,4,6,29]. In contrast, we show that TSE is not sensitive to genes specific to the classical RNA interference pathway linked to small interfering RNAs (siRNA) or to the micro RNA (miRNA) pathway. This suggests thus that TSE involves a rasiRNA pathway linked to heterochromatin formation and that such a mechanism, working in the germline, may underlie epigenetic transmission of repression through meiosis.

Quantification of TSE
TSE was generated by combining telomeric transgene insertions (from the P-1152 line) as a silencer locus, with various euchromatic P-lacZ transgenes expressed in the germline as targets. Depending on the target, TSE can be almost total ( Figure 1C) or intermediate ( Figure 1E and 1F). When TSE is incomplete, variegation is observed since ''on'' and ''off'' lacZ expression is seen among egg chambers: egg chambers can show strong expression (dark blue) or no expression, but intermediate repression levels are not (or very rarely) found. In addition, a given ovary can present ovarioles showing all possible combinations of on or off egg chambers ( Figure 1E). Simple quantification of TSE is thus possible by determining the percentage of repressed egg chambers ( Figure 1F). We scored the number of repressed chambers among the first five egg chambers of a given ovariole for ten ovarioles chosen at random per ovary. For a given genotype, more than 1,000 egg chambers were classically counted (Table 1). This measure generally produces very reproducible results among replicate experiments allowing accurate quantification of TSE.

Transmission of TSE over Generations
TSE was previously shown to have a maternal effect since strong repression occurs only in the progeny of crosses involving females carrying the telomeric silencer, whereas no or weak repression occurs if the telomeric silencer comes from the father [27,28]. Conversely, the parental origin of the target does not significantly affect TSE. We have thus tested whether TSE exhibits only a maternal effect (during one generation) or maternal inheritance over several generations as well. Two reciprocal G 0 crosses were performed in which the telomeric silencer (P-1152) was maternally introduced (maternal lineage) or paternally introduced (paternal lineage) ( Figure S1). Thus, the G 1 females of the two lineages have the same genotype. These G 1 females were backcrossed with males having the telomeric silencer and the target transgene in order to recover G 2 females having one copy of the telomeric silencer locus and one copy of the target transgene. Such backcrosses were repeated for several generations in order to follow lineages with constant genotypes over generations. At each generation, the percentage of TSE was measured. Figure 2 shows that in the maternal lineage, strong TSE was detected in G 1 (around 70%) and maintained over generations. By contrast, in G 1 of the paternal lineage, the level of TSE was weak (around 10%), a result consistent with the maternal effect previously reported. However, in this paternal lineage, a strong level of repression was not observed in G 2 females either (around 24%), despite the fact that G 1 females were carrying a telomeric silencer. TSE thus shows a maternal inheritance. However, the level of TSE in G 2 is greater than in G 1 in the paternal lineage, and in succeeding generations TSE gradually increases. A total of six generations are necessary, however, to reach a repression level close to that of the maternal lineage. Thus, a memory of the initial maternal effect is observed over generations. In conclusion, TSE is characterized by maternal transmission that gets progressively reinforced over successive generations. It is therefore partially epigenetically transmitted through meiosis.

TSE Requires an Extrachromosomally Transmitted Component
In the previous experiment, two main hypotheses can be proposed concerning the molecular basis of the difference

Author Summary
The genome of the fruitfly was invaded in the last century by a mobile DNA element called the P element. After a transient period of genetic disorders due to P mobility, the P element established a repressive state for its transposition. We have shown that a major component of this repression comes from P copies inserted close to telomeres, the ends of linear chromosomes. One or two P copies inserted in subtelomeric heterochromatin (the DNA region highly compacted by protein complexes) can stabilize around 80 P copies. This finding allowed the discovery of a more general phenomenon called the ''Trans-silencing effect'' in which a transgene inserted in this subtelomeric heterochromatin represses, in the female germline, a homologous transgene, irrespective of the genetic location of the latter. We show that Trans-silencing requires not only the chromosomal copy of the telomeric silencer, but also a maternally transmitted factor whose influence can persist over generations. We have found that this epigenetic silencing is sensitive to mutations in genes involved in heterochromatin formation and in a recently discovered silencing pathway based on small RNAs. Trans-silencing thus provides a tool for mechanistic analysis of gene repression on the basis of chromatin changes combined with small RNA pathways in the germline.
inherited in G 1 between the maternal and paternal lineages of TSE. First, maternally inherited telomeric transgenes may be imprinted while paternally inherited transgenes are not, and imprinting may be necessary for TSE to take place in the zygote. Second, a mother carrying telomeric transgenes may deposit an extrachromosomal factor in the oocyte, which is necessary for TSE to occur. This latter hypothesis was tested by using females hemizygous for the P-1152 telomeric silencer locus and a dominant genetic marker on the homologous chromosome to identify transmission of the chromosome carrying the telomeric silencer or the chromosome devoid of silencer (M5 balancer chromosome). Figure 3 shows that crossing these hemizygous females (''A'' females) with males carrying a target transgene produced control G 1 ''B'' females, which have inherited from their mother both the cytoplasm and a chromosomal copy of the telomeric silencer: in these females TSE is about 65%. However, sisters having inherited the M5 chromosome do not show any repression (''C'' females, TSE ¼ 0%). Thus, the cytoplasm of a P-1152 female without a chromosomal P-1152 copy cannot induce the TSE. Crossing P-1152 ; P-Z-target males with females devoid of telomeric silencer produces a weak repression in the progeny (2.7%) as shown by ''E'' females, a result consistent with the maternal effect of TSE reported previously [27]. Finally, crossing ''A'' females with males carrying a P-1152 telomeric silencer allows recovery of females having maternally inherited only a ''P-1152'' cytoplasm and paternally inherited a P-1152 chromosomal silencer. In that case, strong repression is observed (''D'' females, 75% TSE). Thus the cytoplasmic component (incapable by itself of inducing TSE, as shown with ''C'' females) combined to a paternally inherited telomeric silencer can establish a strong TSE. Moreover, this repression is as strong as if the telomeric silencer was maternally inherited (''B'' females). Consequently, the maternal effect of TSE cannot be attributed to a difference in imprinting between the maternally and paternally transmitted P-1152 telomeric silencers, but rather to an extrachromosomally transmitted factor likely deposited in the cytoplasm of the oocyte, which renders the telomeric (E, F) Incomplete TSE in the G 1 females produced by the cross between P-1152 females and BC69 males. Variegation is observed inside ovaries that show individual ovarioles with both off (lacZ repressed) and on (lacZ not repressed) egg chambers (Ea, Ec, Ed) and ovarioles that are completely off (Ee) or completely on (Eb). (F) For the quantification of TSE, the number of repressed egg chambers among the first five egg chambers of a given ovariole is assayed. A total of ten ovarioles are scored per ovary for more than ten ovaries. doi:10.1371/journal.pgen.0030158.g001  R2A8 3.3% (1,100) Su(var)3-7 R14 /Su(var)3-7 R14 7.5% (1,100) Su(var)3-9 06 /Su(var)3-9 06 58.5% ( Females analyzed carried a maternally inherited P-1152 telomeric silencer locus (from a homozygous P-1152 mother) and a paternally inherited P-LacZ target transgene. They also carried mutant alleles of candidate genes involved in heterochromatin formation or/and small RNA silencing. For heterozygous or heteroallelic mutant genotypes, the maternally inherited allele is written first. The percentage of TSE is given with the total number of egg chambers assayed in parentheses. doi:10.1371/journal.pgen.0030158.t001 silencers (paternally or maternally inherited) capable of establishing TSE.

Candidate Gene Analysis
Given the variegating phenotype of TSE and the interaction observed above between a maternally transmitted factor and the chromosomal copy of the telomeric silencer, it seemed possible that heterochromatin formation and RNA silencing could be involved in this repression. A candidate gene analysis was thus performed to identify genes whose mutations affect TSE among genes encoding heterochromatin components and actors of the small RNA silencing pathways. For a given assay, a P-1152 telomeric silencer was combined with a P-lacZ target expressed in the female germline, in the absence (TSE positive control) or presence of mutant alleles of the candidate gene. The P-1152 silencer was inherited in each case from a homozygous P-1152 female. When tested in the heterozygous state, the mutant allele was maternally inherited. The first gene tested was Su(var)205 ( Figure 4A-4D), which encodes HP1 [30]. HP1 is a multifunctional protein that binds at centromeres, telomeres, and some scattered sites on the chromosomal arms [31,32]. The TSE positive control produced almost 90% repression ( Figure 4B), whereas one copy of the null allele Su(var)205 5 , corresponding to an almost completely amorphic allele, has a strong negative dominant effect on TSE (15.8% TSE remaining, Figure 4C). The allele Su(var)205 4 encoding a truncated HP1 protein also behaves as a genetically null allele and also strongly impairs TSE (29.5% TSE remaining, Figure 4D). TSE is therefore sensitive to the dose of HP1. The same analysis was performed for SU(VAR)3-7, another nonhistone heterochromatin protein that binds at centromeres and telomeres and is a partner of HP1 [33,34]. SU(VAR)3-7 contains seven zinc finger motifs, which were shown to bind DNA in vitro [35,36]. Again a dose effect was observed since females heterozygous for the Su(var)3-7 R2A8 null allele showed a reduced TSE (less than 40%, Table 1). Since Su(var)3-7 mutants can be homozygous viable, mutations in this gene were also tested at the homozygous state and an almost complete loss of TSE was observed (3.3% TSE remaining, Table 1). Such a result was also obtained with a second Su(var)3-7 mutant allele (Su(var)3-7 R14 , Table 1). Su(var)3-9 encodes a histone methyl transferase (HMT) responsible for the methylation of histone H3 on lysine 9 [37]. This protein is also a partner of HP1. The null allele of this gene tested here at the homozygous state had no significant effect on TSE (58.5% in the mutant versus 62.9% in the wild-type control, Table 1).
We further tested genes involved in RNA silencing. Three primary RNA silencing pathways have been discovered so far: miRNA, siRNA (classically termed ''RNA interference''), and rasiRNA pathways. These pathways may partially overlap, since they have certain actors in common, but they differ in terms of the biogenesis of short RNAs. miRNAs derive from hairpin RNAs and target numerous essential genes; siRNAs derive from double-strand RNAs and can serve as a defence mechanism against parasites such as viruses; and rasiRNAs derive from interactions between two complementary types of PIWI-interacting RNAs and have numerous targets but are more specific of TEs and heterochromatic sequences [3,4,6,29]. We tested three genes (aubergine [aub], homeless [hls], and armitage [armi]) involved in both the siRNA pathway [38,39] and the rasiRNA pathway in Drosophila ovaries [3,4,29]. Mutations of these genes can induce defaults in siRNA-guided cleavage or in the production of rasiRNAs depending on the target of the silencing [3,39,40]. These mutations are also responsible for disruption of embryonic axis specification linked to disturbance of microtubule polarisation, but this effect is mediated by a mechanism different from rasiRNA silencing itself [41]. The first gene tested is aub, an RNA binding protein and member of the Argonaute family ( Figure  4E-4H) [42]. No significant dose effect of aub was detected ( Figure 4G and unpublished data), but the two heteroallelic aub mutant genotypes tested completely abolish TSE ( Figure  4H; Table 1). The second one is hls, which encodes an RNA helicase [43]. Again no dose effect was observed (unpublished data), but a heteroallelic hls mutant genotype completely abolishes TSE (Table 1). The third one is armi, which encodes a putative RNA helicase [44,45]. No dose effect was observed (unpublished data), but a complete loss of TSE was observed in a heteroallelic mutant context (Table 1). Finally piwi, Figure 2. The Level of TSE Is Epigenetically Transmitted through Meiosis Two kinds of G 1 females were generated that carry the telomeric silencer P-1152 and the P-otu-lacZ transgene (P-Co1) used as a TSE target (mating schemes are shown with the complete genotypes on Figure S1). In one lineage, P-1152 was maternally introduced, whereas it was paternally introduced in the other lineage. Because of the maternal effect of TSE, the first lineage (maternal lineage) corresponds to a TSE-positive context in G 1 , whereas the second (paternal lineage) corresponds to a TSEnegative context in G 1 . G 1 females from the two lineages were crossed with P-1152; P-Z-target males to generate in both cases P-1152/ þ ; P-Ztarget/þ G 2 females. Thus G 2 females from the maternal and paternal lineages have the same genotype but differ by the cytoplasm inherited in G 1 . This (P-1152/ þ ; P-Z-target/þ) genotype was maintained over generations and TSE was measured at each generation. The evolution of TSE is shown in relation to succeeding generations. The blue and pink lines correspond to the lineages in which the P-1152 telomeric silencer locus was maternally and paternally inherited in G 1 , respectively. Confidence intervals of the percentages (a ¼ 5%) are given. doi:10.1371/journal.pgen.0030158.g002 another member of the Argonaute family [46] shown to be involved in the rasiRNA pathway [3,4,6,29,47] was tested. This gene is also necessary for germline stem-cell renewal [48] and was shown to be required for post-transcriptional transgene silencing (a phenomenon termed cosuppression) [49]. No dose effect on TSE was observed (unpublished data). Because piwi has deleterious effects on ovary structure at the homozygous state, we tested the effects of piwi mutant alleles at the heterozygous state in combination with a Su(var)205 mutant. In these combinations, piwi mutant alleles aggravate the negative effect of reduction of the dose of HP1: the single Su(var)205 mutant shows 29.5% TSE, whereas the double heterozygotes show 10.0% and 5.6% for piwi 1 and piwi 2 respectively (Table 1). These results indicate that the siRNA and/or the rasiRNA pathways are involved in TSE.
To determine which of these two pathways is responsible for TSE, several other mutations were tested. r2d2 is involved in the siRNA, but not the rasiRNA pathway [3,50,51]. R2D2 is a double-stranded RNA (dsRNA) binding protein involved in siRNA loading onto RISC [50]. In ovaries of females homozygous for the r2d2 1 null allele, the siRNA pathway is severely affected, as shown by a dsRNA-initiated RISC (or RNA-induced silencing complex) assay [52]. These ovaries show somewhat abnormal ovarioles ( Figure S2), but quantification of TSE was nonetheless possible. The loss of function for r2d2 does not affect TSE (Table 1). A second protein DICER-2 (DCR-2), which combined with R2D2 is responsible for dsRNA cleavage, is also involved specifically in the siRNA pathway [53]. The loss of function of Dcr-2 is fertile. Unfortunately, we were not able to test the effect of a Dcr-2 mutant allele on TSE. Indeed, the control staining (in absence of a telomeric silencer) of the two different P-lacZ target transgenes that we tested in homozygous Dcr-2 L811fsX mutants intriguingly showed almost no staining in ovaries, thus making it impossible to assay any target repression by a telomeric silencer in this context. Finally, a mutant allele of loquacious (loqs), a dsRNA binding protein involved in the miRNA pathway was tested. The loqs f00791 allele corresponds to a hypomorphic allele that alters miRNA-induced silencing in all tissues where it has been tested, including ovaries [54]. This allele is viable, allowing us to test homozygous mutant females. No effect on TSE was detected (Table 1). In conclusion, TSE strongly depends on the function of two genes involved in heterochromatin formation and four genes involved in the rasiRNA silencing pathway but is not affected by mutations of genes involved in the siRNA or miRNA pathways.

Discussion
Our study of the regulatory mechanism of a TE that recently invaded the D. melanogaster genome allowed the discovery of a master site for establishing strong transposon repression in the germline; indeed, one or two P elements inserted in TASs are sufficient to repress the mobility of the whole P family [21]. TSE was further discovered [26], a mechanism by which a transgene inserted in subtelomeric heterochromatin has the capacity to repress a homologous transgene located in euchromatin. This phenomenon appears general since it can be induced by transgenes inserted at the telomere of the X chromosome as well as autosomal telomeres, and the targets can be located on all major chromosomes [26,28]. TSE corresponds thus to a useful tool to investigate the properties of the telomeric master site of P repression. Here, we show that TSE requires genes encoding proteins involved in heterochromatin formation (including HP1, a major component of heterochromatin) and proteins involved in the rasiRNA silencing pathway. Another TE master regulatory locus has been described in Drosophila close to the centromere of the X chromosome. It was first shown to regulate the gypsy retrotransposon and referred to as the flamenco locus [55]. This locus was further characterized as regulating together two other retrotransposons, ZAM and Idefix, and proposed as a repression center for multiple TEs, known as ''COM'' (Centre Organisateur de Mobilisation) [14]. Repression by this locus is based on RNA silencing, but differs from TSE in that repression of gypsy, ZAM, and Idefix occurs in the somatic follicle cells of the ovary. This point is important since in somatic follicle cells, AUBERGINE, a major actor of the rasiRNA pathway, is not expressed [4]. Conversely, in the germline, where TSE is active, all known components of the rasiRNA machinery (AUBERGINE, PIWI, and AGO3) are present. Consequently, TSE is particularly appropriate to investigate the genetic properties of a complete rasiRNAbased repression machinery.

What Is the Molecular Support of Epigenetic Transmission of TSE?
Here, we show that TSE exhibits a maternal memory that can be detected for six generations and has thus a partial epigenetic transmission through meiosis (Figure 2). The genetic analysis indicates that potentialization of the telomeric silencer (''potentialization step'') appears to be a prerequisite for repression of the target transgene by the telomeric silencer (''target repression step''). For the potentialization of the telomeric silencer, both an extrachromosomal maternally transmitted factor, produced by females carrying a telomeric silencer, and a chromosomal copy of the telomeric silencer are necessary (Figure 3). The molecular nature of this extrachromosomal maternally transmitted factor, responsible for the maternal effect, is not known at the moment but it may correspond to rasiRNAs since TAS homologous rasiRNAs have been reported (see Table S1 of [4]). We have attempted to identify short RNAs made by telomeric transgenes by northern blot on ovaries but did not detect them; this may be due to a low concentration of such short RNAs. We propose two nonmutually exclusive mechanisms. In the first, which derives from the ''rasiRNA positive loop model'' proposed by Brennecke et al. (''ping-pong model'') [4] and Gunawardane et al. [29], females deposit PIWI-interacting RNAs (piRNAs) in the oocyte produced by telomeric transgenes inserted in tandemly repeated TAS heterochromatic sequences. As proposed in their model, these RNAs would be primarily small antisense RNAs associated with PIWI or AUBERGINE. These piRNAs would interact in the embryo with sense RNAs produced by the telomeric transgenes and result in the production of small sense RNAs associated with AGO3, which in turn would promote the production of a new antisense piRNA, resulting in a positive feedback loop. Such a loop can explain the twocomponent interaction shown in Figure 3 (see below).
A second mechanism can be proposed for the potentialization step. Indeed, RNA silencing and heterochromatin formation have been shown to be connected since ''short RNA-dependent heterochromatin formation'' pathways have been described in several species (fission yeast [56], ciliates [57], plants [58], and Drosophila [59], for a review see [60][61][62]). According to this model, maternally transmitted small RNA molecules may modify the chromatin structure of the chromosomal copy of the (paternally or maternally transmitted) telomeric transgenes in the embryo and thereby confer to the telomeric transgene the capacity to be a silencer. Recently, RNA-dependent heterochromatin formation was observed at Drosophila telomeres since mutations of homeless were shown to cause an opening of chromatin at the level of telomeric retroelements in Drosophila ovaries (M. Klenov and V. Gvozdev, personal communication). Such a model is consistent with the results of our candidate gene analysis since mutations in two genes involved in heterochromatin formation (Su(var)205 and Su(var)3-7) strongly impair TSE. HP1 could play a role in heterochromatin formation, not only as a component of heterochromatin but also indirectly: indeed, in Schizosaccharomyces pombe CHP1 (a chromodomain protein) has been shown to be part of a complex that drives small RNAs to chromatin in order to modify histone H3 methylation [63]. Such a role for the HP1 chromodomain protein in a similar complex could also exist in Drosophila.
How can maternal memory of TSE over six generations be explained? Because of the modality of TSE transmission over several generations, it can be proposed that establishment of the hypothetical extrachromosomal component discussed above requires a number of generations to reach a sufficient concentration to establish strong repression. Under the ''rasiRNA positive loop model'' proposed by Brennecke et al. [4] and Gunawardane et al. [29], when males carrying a telomeric silencer are crossed with females bearing no silencer, the cytoplasm of the oocytes would lack piRNAs homologous to the transgene. However, synthesis of these RNAs would begin in these G 1 females stimulating this pathway and increasing the piRNA concentration. At each generation, females would transmit to their daughters a higher concentration of rasiRNA molecules than at the previous generation. At generation six, a sufficient level would be reached to allow strong TSE. Under the ''RNAdependent heterochromatin formation'' model, when males carrying a telomeric silencer are crossed with females bearing no silencer, again the cytoplasm of the oocytes would lack small RNAs homologous to the transgene. However, synthesis of these RNAs would begin in these G 1 females stimulating heterochromatinization of the telomeric silencer. This change in the chromatin state would stimulate the production of rasiRNAs at the locus also establishing a positive loop, i.e., reciprocal stimulation between heterochromatin formation and production of small RNAs. Such a reciprocal positive loop between heterochromatin formation and small RNA silencing has been proposed for gene silencing in S. pombe [64,65]. Again, at each generation, females would receive from their mother and transmit to their daughters a higher concentration of RNA molecules than at the previous generation until a sufficient level is reached at generation six.

How Does the ''Target Repression Step'' Occur?
Repression of the target transgene by the telomeric silencer was shown to be homology dependent [23,26,28]. TSE however does not require homology at the level of the promoters of the telomeric and euchromatic insertions since a defective telomeric P element lacking the P promoter can repress a P-lacZ transgene in which lacZ is driven by a heatshock promoter [23]. The TSE decision appears to be established at the cystoblast stage and stably maintained since the 15 deriving nurse cells are roughly identical in regard to their repression state. Another important point in understanding the repression mechanism is that the telomeric silencer can lack the lacZ gene, i.e., the gene repressed by TSE in the target transgene. For example, a P-white-yellow transgene [66] inserted at 1A can repress the euchromatic Potu-lacZ-white (pCo) transgene used in this study (unpublished data). In this case, the homology between the telomeric and the euchromatic transgenes comes from the white transformation marker. This rules out the hypothesis that silencing of the target occurs via an interaction between rasiRNAs deriving from the telomeric silencer and the target transgene transcript; indeed, the targeted transcript in TSE corresponds to the lacZ sequence and no lacZ rasiRNAs can be produced by the silencer. Therefore, TSE would not occur by post-transcriptional gene silencing.
Target repression could thus occur via transcriptional gene silencing. This would fit with the variegated phenotype of repression observed. Such variegation strongly suggests that repressed targets undergo heterochromatinization. Two hypotheses can be proposed to explain such heterochromatinization. First, rasiRNAs produced by the telomeric silencer would interact with nascent transcripts on the target, and this interaction could provoke a local heterochromatinization of the target [64,65]. Heterochromatinization may then spread along the target as suggested by the repression of a P-otu-lacZwhite target by a telomeric P-white-yellow transgene. In this case heterochromatinization would start on the white sequence of the target and subsequently reach the lacZ sequence. HP1 and SU(VAR)3-7 could be involved in this spreading since centromeric heterochromatin spreading has been shown to be sensitive to the dose of these proteins [35,67]. An alternative hypothesis for target heterochromatinization is a DNA-DNA interaction between the telomeric and the euchromatic transgenes, leading to their pairing and to trans-heterochromatinization of the target following this pairing. Indeed, telomeric silencers are themselves in a heterochromatic state due to the heterochromatic nature of the TAS [68][69][70][71]. This is illustrated by a phenomenon referred to as telomeric position effect, in which a transgene inserted in TAS shows variegation in the eye for the transformation marker [72][73][74][75]. Again HP1 and SU(VAR)3-7 could play a role in this trans-heterochromatinization.

What Is the Role of the Telomere Cap in TSE?
Drosophila telomeres can be divided into three domains with respect to chromatin structure [71]: (1) the subtelomeric cluster of heterochromatic TAS repeats [68][69][70] in which the tested P elements or P transgenes are inserted; (2) distal to the TAS, a telomeric retrotransposon array tandemly repeated in the same orientation [76][77][78][79][80], the retrotransposon array is partially euchromatic [71], and transgenes located inside this domain do not show variegation for an eye marker [81]; and (3) at the extremity of the retrotransposon array, the telomere protein cap that prevents telomere fusion and regulates telomeric retrotransposon transcription and transposition [82][83][84][85]. Interestingly, HP1 (Su(var)205) and SU(VAR)3-7, both implicated in TSE, are part of the cap and therefore could be required for the telomeric locus to be a silencer. This hypothesis is supported by the analysis of the different Su(var)205 alleles. In addition to the two null alleles of Su(var)205 tested here, which showed a strong effect on TSE (Table 1), we also tested a deficiency and the Su(var)205 2 allele corresponding to a mutation in the chromodomain (involved in the histone H3 K9-methylated binding), which leaves the capping activity intact [85,86]. The deletion of Su(var)205 as expected has a strong effect on TSE but Su(var)205 2 has no effect (unpublished data). This suggests that HP1 effect can be mediated (at least partially) via its capping activity. Therefore, HP1 and SU(VAR)3-7 could affect the telomeric transgenes in TAS indirectly via such capping activity, for example, by affecting the localisation of the telomere inside the nucleus.
Finally, it could appear paradoxical that an amorphic mutant allele of the Su(var)3-9 HMT has no effect on TSE since RNA-dependent heterochromatin formation in most cases involves the histone H3 methylation transition from lysine 4 to lysine 9 to establish the link between short RNAs and formation of heterochromatin. However, this result can be explained by the fact that this HMT has been recently shown not to be responsible for histone H3 methylation on lysine 9 at Drosophila telomeres [85]; another unidentified HMT likely plays this role. We are therefore pursuing the candidate gene analysis in order to identify a HMT involved in TSE.

Conclusions
Telomeric TSE appears to be a complex repression mechanism that requires genes involved in heterochromatin formation and in RNA silencing. In Drosophila, interaction between small RNA silencing pathways and transcriptional repression was previously shown to exist in somatic tissues for cosuppression phenomena or for variegation phenotypes with the white marker in the eye [49,59]. TSE shows that interaction between RNA silencing and heterochromatin formation can also occur in the germline. This type of silencing appears to be the basic mechanism for P-element repression [28], although some P-encoded repressors encoded by euchromatic copies may contribute in some cases to P repression [10,87]. TSE has the same complex inheritance as the P cytotype [10,88,89]. The subtelomeric heterochromatin thus represents a piRNA producing ''platform'' [4] available for recent invaders of the genomes to establish their own repression. The study of TSE illustrates the genetic properties of such a platform.

Materials and Methods
Experimental conditions. All crosses were performed at 25 8C and involved three to five couples in most cases. All ovary lacZ expression assays were carried out using X-gal overnight staining as described in Lemaitre et al. 1993 [90], except that ovaries were fixed for 6 min.
Transgenes and strains. P-lacZ fusion enhancer trap transgenes (P-1152, BQ16, and BC69) contain an in-frame translational fusion of the Escherichia coli lacZ gene to the second exon of the P transposase gene and contain rosy þ as a transformation marker [91]. The P-1152 insertion comes from stock number 11152 of the Bloomington Stock Center (http://flystocks.bio.indiana.edu) and was mapped at the telomere of the X chromosome (site 1A); this stock was previously described to carry a single P-lacZ insertion in TAS [26]. However, in our number 11152 stock, we have mapped two P-lacZ insertions in the same TAS unit and in the same orientation that might have resulted from an unequal recombination event duplicating the P-lacZ transgene. P-1152 is homozygous viable and fertile. BQ16 is located at 64C in euchromatin of the third chromosome and is homozygous viable and fertile. BC69 is inserted on Chromosome 2 in the first exon of the vasa gene and results in a vasa loss of function; it is consequently homozygous female sterile. P-1152 shows no lacZ expression in the ovary ( Figure 1A), whereas BQ16 and BC69 are strongly expressed in the nurse cells and in the oocyte ( Figure 1B and 1D, respectively). P-Co1 is an insertion of the pCo transgene (P-otu-lacZ) on the third chromosome (87A-B), which is homozygous viable and fertile. bgalactosidase expression of the pCo transgene is driven by the otu promoter and is therefore strongly detected in both nurse cells and the mature oocyte [27]. This transgene contains a white gene as a transformation marker.
Mutations used for the candidate gene analysis. Su(var)205, aub, piwi, r2d2, and loqs genes are located on Chromosome 2, whereas Su(var)3-7, Su(var)3-9, hls (synonym of spindle-E), and armi are located on Chromosome 3. Loss of function is lethal in the case of Su(var)205 and loqs, female sterile in the case of aub, piwi, hls, r2d2, and armi, and is viable and fertile in the case of Su(var)3-9. Loss of function of Su(var)3-7 is viable when the mother is heterozygous mutant but lethal when the mother is homozygous mutant [92].
Su(var)2-5 05 was x-ray induced and corresponds to a null allele since it only encodes the first ten amino acids of the HP1 protein [67]. Su(var)2-5 04 was EMS induced and encodes a truncated HP1 protein, ;85% of wild type size [67], missing a domain necessary for targeting of HP1 to the nucleus [93]. Lines carrying these mutations were kindly provided by Gunter Reuter.
Su(var)3-9 06 was generated by x-ray and corresponds to an amorphic allele [37,95]. Line carrying this mutation was kindly provided by Gunter Reuter.
Three strong mutant alleles of aub induced by EMS were used. All of them are homozygous female sterile. aub QC42 [96] comes from the Bloomington Stock Center (stock number 4968). aub HN2 [96] has an amino acid substitution. aub N11 [97] has a 154-bp deletion, resulting in a frameshift that is predicted to add 16 novel amino acids after residue 740 [98]. aub HN2 and aub N11 were kindly provided by Paul Macdonald.
piwi 1 and piwi 2 are P-element-induced mutations; piwi 1 contains a P transgene inserted in the first exon [46]. The orientation of the transgene is opposite to that of the gene. piwi 2 contains a P transgene inserted in exon 4 in the same orientation as the gene. In both cases, homozygous females are sterile. Lines carrying these mutations were kindly provided by Alain Bucheton. hls E616 (synonym of spn-E 1 ) was generated by EMS [43,99]. hls D125 (synonym of spn-E hlsD125 ) was generated by P-element excision resulting in deletion of coding sequences and adjacent sequences. Deletion may extend into other genes [43]. Lines carrying these mutations were kindly provided by Utpal Bhadra. armi 1 and armi 72.1 are P-element-induced mutations; armi 1 has an insertion of a P transgene in the 59 UTR of the armitage gene [44]. armi 72.1 derives from armi 1 by an imprecise excision of the P element, resulting in a deletion of armi sequences in the 59 untranslated region [39]. These two alleles are female sterile and come from the Bloomington Stock Center (stock numbers 8513 and 8544). r2d2 1 results from a deletion induced by the remobilization of a P transgene located in 59 of the r2d2 gene [50]. The 4.9-kb deletion removes the entire r2d2 open reading frame [52]. This allele is an amorphic allele and homozygous females are sterile due to abnormal ovaries. The siRNA pathway is severely affected as shown by a dsRNAinitiated RISC assay in ovaries of homozygous mutant females [52]. This line comes from the Bloomington Stock Center (stock number 8518).
loqs f00791 results from a piggyBac transgene insertion 57 bp upstream of the transcription start site of loqs [54], a gene necessary for miRNAdirected silencing. loqs f00791 is a hypomorphic allele that is homozygous viable. At the homozygous state, this allele has a strong effect on miRNA processing in ovaries. It also has a negative effect on siRNA silencing [54]. This allele comes from the Bloomington Stock Center (number 18371).
All the alleles described above are maintained over a balancer chromosome. Additional information about mutants and stocks are available at FlyBase (http://flybase.bio.indiana.edu). Figure S1. Detailed Mating Scheme to Detect Epigenetic Transmission of TSE M5 indicates Muller5, a balancer of the X chromosome devoid of P sequence, Sb indicates Stubble, a dominant marker of the TM3 balancer Chromosome 3, and Xa means Xasta, a dominant marker of the attached Chromosomes 2 and 3 (T(2;3)ap Xa ). In each case, the maternally inherited chromosomal complement is written above the bar. G 1 females carrying the telomeric silencer P-1152 and the P-otu-lacZ transgene (P-Co1 located on Chromosome 3) used as a TSE target were generated by two reciprocal crosses; in the maternal lineage of TSE (left), P-1152 was maternally introduced, whereas it was paternally introduced in the paternal lineage (right). G 1 females were crossed with P-1152; P-Co1 males to establish, in the two lineages, G 2 females having the same genotype (P-1152/ M5; P-Co1/ Sb) but differing in the cytoplasm transmitted by the G 0 females. The same procedure was repeated over six generations. At each generation, females were dissected and TSE was quantified in ovaries. The results are shown on Figure 2. Found at doi:10.1371/journal.pgen.0030158.sg001 (51 KB PPT). Figure S2. r2d2 Homozygous Mutant Context Allows TSE Quantification Ovaries from a homozygous r2d2 1 female that carries a P-1152 silencer locus and a BQ16 target transgene. Found at doi:10.1371/journal.pgen.0030158.sg002 (2.4 MB PPT).