RNAe requires chromatin factors and correlates with H3K9me3

To ask if silencing is regulated transcriptionally or post-transcriptionally, we isolated total RNA from otherwise identical silent and active gfp::csr-1 strains and measured the abundance of pre-mRNAs and mRNAs by real-time quantitative PCR (qPCR). We found that both the pre-mRNA and mRNA levels were significantly reduced in the silent line compared to the active line (Figure 2B and D). Moreover, although a reduction at the pre-mRNA level appeared to account for the majority of silencing, a further reduction was evident at the mRNA level, suggesting that silencing is achieved at both transcriptional and post-transcriptional levels (Figure 2B and 2D).

Previous work has shown that the methylation of lysine 9 on histone H3 (H3K9me), a histone modification associated with silent chromatin, is enriched on high-copy number transgenes in the germ line (Bessler et al., 2010; Kelly et al., 2002). Furthermore, germ-line silencing of high-copy transgenes is dependent on a number of chromatin-associated factors, including the Polycomb-Group complex (MES-2/-3/-6), Trithorax-related (MES-4) and the heterochromatin proteins (HPL-1 and -2) (Couteau et al., 2002; Grishok et al., 2005; Kelly and Fire, 1998; Kelly et al., 1997). Consistent with these previous findings, we found that transgene sequences from a silent MosSCI allele, but not an active MosSCI allele, were enriched in Chromatin Immunoprecipitation (ChIP) experiments using antibodies specific for H3K9me3 (Figure 2C and D). Furthermore, we found that mes-3, mes-4, and hpl-2 mutants all de-silenced the gfp::csr-1 and gfp::cdk-1 transgenes (Table 2). These findings suggest that the maintenance of single-copy transgene silencing involves a chromatin component similar to that described for maintenance of high-copy transgene silencing.

Maintenance of silencing requires RNAi-related factors

The trans-acting nature of the silencing phenomenon suggested the possible involvement of an RNAi-related small RNA pathway. To explore this possibility we crossed a silent transgenic strain into strains bearing mutations in RNAi components. Two downstream factors in the exo-RNAi pathway, rde-3 and mut-7, which encode a beta-nucleotidyl transferase and a 3′–5′ exonuclease respectively (Chen et al., 2005; Ketting et al., 1999), are known to be required for the maintenance of transposon silencing and have been implicated in co-suppression (Dernburg et al., 2000; Ketting and Plasterk, 2000) and high-copy number transgene silencing (Tabara et al., 1999). Consistent with the involvement of these factors in the maintenance of RNAe we found that crossing a silent transgene into these mutant strains resulted in fully restored transgene expression (Table 2).

We also examined the consequences of crossing strains de-silenced in the rde-3 mutant background back into a wild-type rde-3(+) background. We found that for a gfp::csr-1 transgene de-silenced by rde-3, 27 % of rde-3(+) segregants (n=15) retained expression after outcross (Figure 2A and Supplementary figure 1). However, in contrast, strains bearing the gfp::cdk-1 transgene, also desilenced by rde-3, were always rapidly and fully re-silenced by reintroducing rde-3(+) (n>20).

Nuclear and Cytoplasmic WAGOs are required for silencing maintenance

Because RDE-3 and MUT-7 are required for the accumulation of RdRP-derived 22G-RNAs that engage WAGOs (Gu et al., 2009; Yigit et al., 2006), we asked whether WAGOs are required for the maintenance of single-copy transgene silencing by crossing silent lines with several different wago mutant strains. We found that a mutation in the predominantly cytoplasmic germ-line WAGO, wago-1(tm1414) (Gu et al., 2009), partially de-silenced a gfp::cdk-1 transgene but did not de-silence a gfp::csr-1 transgene (Table 2 and Figure 3A and C).

The finding that wago-1 mutants failed to de-silence gfp::csr-1 and only partially de-silenced gfp::cdk-1 suggested that additional WAGOs contribute to RNAe (Figure 3I). Furthermore, because RNAe involves a chromatin component, we suspected that nuclear WAGOs might be important for RNAe. The nuclear WAGO, NRDE-3/WAGO-12, is required for nuclear RNAi and transcriptional silencing in somatic tissues (Burton et al., 2011; Guang et al., 2008), and nrde-3 mutants failed to de-silence a gfp::csr-1 transgene in the germ line (Table 2). However, within the WAGO sub-clade that includes NRDE-3 (Figure 3I), we identified WAGO-9 (HRDE-1/C16C10.3) as a nuclear WAGO that is restricted to the germ line (Figure 3G). Furthermore, we found that wago-9(tm1200) mutants fully de-silenced a gfp::csr-1 transgene and partially de-silenced a gfp::cdk-1 transgene (Figure 3B and D), the converse of the relationship between wago-1(tm1414) and these RNAe lines. The de-silencing of gfp::cdk-1 was increased in a wago-1; wago-9 double mutant (Figure 3E).

Because gfp::cdk-1 was not completely desilenced by these wago mutant combinations, we asked if additional members of the nuclear WAGO sub-clade play a role in gfp::cdk-1 silencing. Indeed, gfp::cdk-1 was strongly de-silenced in a wago-9; wago-10 (t22h9.3); wago-11(f49f6a.1); nrde-3 quadruple mutant, as well as in a wago-9; wago-10 double mutants (Table 2 and Figure 3F). Taken together, these findings indicate that cytoplasmic and nuclear WAGOs contribute to RNAe in parallel and that the input from cytoplasmic and nuclear WAGOs varies between individual RNAe lines.

The small RNAs that associate with WAGO-1 were previously identified by immunoprecipitation (IP) of FLAG::WAGO-1 followed by deep sequencing of associated small RNAs (Gu et al., 2009). We performed similar studies using a flag::wago-9 transgene. We found that the targets of WAGO-9 are largely overlapping with those of WAGO-1 (Figure 3H). These observations suggest that nuclear and cytoplasmic WAGOs share targets and are likely to share a common 22G biogenesis pathway.

Silencing correlates with accumulation of 22Gs targeting GFP

To examine the small RNA profile associated with germ-line silencing, we dissected gonads from different transgenic lines, including active, silent, and converted lines (e.g. active to silent and silent to active lines), and prepared small RNA libraries for deep sequencing (Figure 3J and Supplementary Figure 1). Strikingly, each silenced line exhibited a marked accumulation of 22G-RNAs that were restricted to the gfp portion of the transgene sequence (Figure 3J and Supplementary Figure 1). Consistent with the idea that these 22Gs are WAGO-pathway dependent, we found that 22G-RNA levels targeting gfp were significantly reduced in lines converted from silent to active by crossing through an rde-3 mutant background (Supplementary Figure 1).

Native germ-line-expressed genes are recognized by low levels of 22G-RNAs that engage CSR-1 (CSR-1-22Gs) (Claycomb et al., 2009). We found that the transgene sequences corresponding to endogenous germ-line expressed mRNA sequences, always exhibited low 22G-RNA levels similar to those observed for the endogenous sequences in wild-type non-transgenic animals (Figure 3J and Supplementary Figure 1). These findings suggest that the WAGO-mediated silencing signal only targets the foreign sequences of the transgene.

Initiation of silencing requires the Piwi Argonaute PRG-1

Despite interacting with distinct small RNA species, both PRG-1 and RDE-1 function as primary AGOs upstream of WAGO-22G mediated silencing (Batista et al., 2008; Das et al., 2008; Pak and Fire, 2007; Sijen et al., 2007; Yigit et al., 2006). However, we found that neither prg-1 nor rde-1 mutants could activate an already established silent transgene (Table 2). To explore the possibility that either PRG-1 or RDE-1 is involved in the initiation of RNAe, we generated new transgenic lines by directly injecting into prg-1 and rde-1 mutants. We chose to inject the gfp::cdk-1 construct, because 100 % of MosSCI lines were silent when established in the wild-type background (n=21) (Table 1). In an rde-1(ne300) mutant strain, we found that the cdk-1::gfp transgene was silenced in all three newly isolated lines (Table 1). Strikingly, however, when we repeated the same experiments with prg-1(tm872) mutants, the cdk-1::gfp transgene was fully active in all five independently generated transgenic lines (Figure 4 and Table 1). Taken together, these findings suggest that PRG-1 and piRNAs are involved in the initiation of transgene silencing, whereas dsRNA (e.g. from bi-directional transcription of the transgene) is not involved.

Open in a separate window

Figure 4

PRG-1 is required for the initiation of RNAe

prg-1(tm872) mutant worms injected with the gfp::cdk-1 construct (top right) give rise to MosSCI lines that express GFP::CDK-1 (P0, top left). The micrographs show the expression status of GFP::CDK-1 in oocyte nuclei (arrowheads) before (P0) and after outcrossing to wild type (F1 and F2 panels), and after segregating homozygous prg-1(+) and prg-1(−) strains for several generations (F3–F10 panels). More than 10 worms were examined per generation. Results are detailed in the text.

When established in the wild-type background, the epigenetic state of a transgene, whether active and silent, is stably maintained over many generations. If PRG-1 is only required for the initiation of silencing, then we expected that active transgenes established in the prg-1 mutant background would remain stably active even after outcrossing to a wild-type strain. However, we found that the active transgenes established in the prg-1 mutant rapidly became silent after outcrossing. We found that 96 % of the heterozygous F1 progeny (n=24) exhibited active expression of gfp::cdk-1. However, by the F3 generation, only 9 % of heterozygous or homozygous wild type exhibited active expression (n=66). Conversely, among the F3 that were once again homozygous for prg-1, 77 % (n=30) maintained the active state. These findings raise the possibility that PRG-1 may not be the most upstream factor in the initiation of silencing. Alternatively, PRG-1 may be upstream of competing pathways: one that initiates silencing and one that initiates anti-silencing (see DISCUSSION).

Repetitive and single copy transgenes exhibit distinct but overlapping silencing mechanisms

The silencing of high-copy and single-copy transgenes share several features, including chromatin and WAGO 22G pathway requirements. However, several observations suggest that high-copy transgenes are subject to distinct modes of recognition and silencing. First, high-copy transgenes were at best only partially de-silenced in the mutant contexts described above (data not shown) whereas single copy transgenes were fully desilenced and in some cases even maintained their expression after outcrossing to wild type. Second, high-copy transgenes were fully and rapidly silenced in the germ-line of prg-1 mutant animals (data not shown), indicating that a distinct initiation step is involved in silencing. Third, high-copy number silencing was observed even when only the native germ-line gene sequences were present in the transgene (data not shown), whereas silencing of the single-copy transgene was correlated with the presence of foreign sequences within the germ-line-expressed portion of the transgene construct. Finally, unlike the single-copy silencing described here, where trans-silencing remains focused on foreign sequences, high-copy transgenes were found to elicit co-suppression of the endogenous gene (Dernburg et al., 2000; Ketting and Plasterk, 2000). Taken together these observations are consistent with the existence of at least two distinct modes of silencing that act on transgenes, one that depends on high-copy number and can spread throughout the transgene, and a second that requires PRG-1 and is restricted to portions of the transgene composed of foreign sequences.

21U-RNAs complementary to gfp are correlated with 22G biogenesis

Our findings suggest that transgene silencing is initiated by PRG-1 and depends on the presence of foreign gfp sequences in the transgene. In a parallel study, PRG-1 was shown to initiate silencing of synthetic reporters containing sites perfectly complementary to 21U-RNAs (Lee et al., co-submitted and Miska E., personal communication). Mismatched pairing was also correlated with silencing both on transgenes (Miska E., personal communication) and on presumptive endogenous targets (Lee et al., co-submitted and Miska E., personal communication). We have not identified 21U-RNAs that are perfectly complementary to gfp, however, there are potentially dozens of high affinity 21U-RNA-GFP target sites (data not shown). Our recent studies (Lee et al., co-submitted) suggest that PRG-1/21U-RNA targeting initiates 22G-RNA biogenesis within a +/− 40 nt window around the site of 21U-RNA complementarity on the target RNA. We found 8 regions on gfp where 22G-RNAs were detected at greater than 75 reads per million in a silent strain (Figure 6A). We identified potential high-affinity 21U-RNA interactions in all 8 regions. These base-pairing interactions and the proximal 22G-RNAs found in a silent transgenic strain are shown at single-nucleotide resolution in Figure 6A (also see EXPERIMENTAL PROCEDURE). Validation of these candidate 21U-RNA target sites and the general rules that govern piRNA targeting remain to be elucidated in the future.

CSR-1 as an anti-silencing Argonaute

At least three mechanisms must work together to explain the all-or-none nature (expressed or silent) of the epigenetic states observed, and the stable heritability of these states once established (Figure 6B). The genetic studies, thus far, have implicated PRG-1 in the initiation of silencing and the WAGO pathway in the maintenance of silencing. The third pathway required is a “maintenance of expression” or “anti-silencing pathway”. Such a pathway is necessary to explain why, once established, active transgenes are stably transmitted from one generation to the next without undergoing spontaneous silencing. An anti-silencing pathway could also explain how certain active transgenes are able to dominantly activate silent transgenes (Figure 6B).

The CSR-1 22G pathway targets endogenous germ-line-expressed mRNAs (Claycomb et al., 2009), and is an ideal candidate for an anti-silencing pathway. In vitro, CSR-1 is catalytically active and capable of cleaving a target (Aoki et al., 2007), whereas the all WAGOs lack key catalytic residues (Yigit et al., 2006). Perhaps CSR-1 can compete by selectively destroying RNAs on which RdRP is bound, thus preventing or attenuating the production of WAGO 22G-RNAs. It is not known how CSR-1 targeting is first established. However, all of the transgenes that we analyzed contain endogenous germ-line expressed sequences known to be targeted by CSR-1 22Gs. Perhaps CSR-1 22Gs can spread in trans along a target transcript as has been shown for the transitive RNAi mediated by WAGOs after dsRNA targeting (Pak and Fire, 2007; Sijen et al., 2007; Yigit et al., 2006). If so, then stable expression of a transgene may reflect the spread of CSR-1 targeting to the foreign portion of the transgene prior to PRG-1 recognition.

Interestingly, although the anti-silencing signal initially appears to be sufficient to prevent PRG-1 driven silencing, it is not sufficient to prevent silencing initiated in crosses with a silent transgene or when dsRNA is used to stimulate gene silencing. If CSR-1 22G-RNAs represent the anti-silencing signal, then it will be interesting to explore whether the levels of CSR-1 22G-RNAs build up over generations. If so then, the older transgenes, which were able to activate a silent transgene, may show relatively high levels of CSR-1 22G-RNAs targeting gfp when compared to newly established lines. However, it is also possible that as yet unknown features of the chromatin environments of the different transgenes drives their different sensitivity to trans-silencing and their differing abilities to trans-activate or to recover from silencing spontaneously.

Finally, it is worth noting that PRG-1 may function upstream of RdRP recruitment for both the CSR-1 and WAGO pathways. If so, then the decision to express or silence a new transgene may represent the result of a competition between the CSR-1 and WAGO pathways for RdRP loading, downstream of this initial recruitment. An expectation for such a model would be that both the maintenance-of-silencing (non-self) and maintenance-of-expression (self) pathways should fail to initiate when PRG-1 is absent. To further explore this question it will be important to analyze the behavior of additional transgenes established in the prg-1 mutant background.

MosSCI by direct injection

MosSCI lines were generated by the direct insertion method using strain EG4322 and EG5003 as described (Frokjaer-Jensen et al., 2008). Targeting vectors are described in Supplemental Information.

MosSCI by heat-shock and ivermectin selection

Strain WM189 was injected with a DNA mixture containing 50 ng/μl each of pRF4::rol-6(su1006), pCCM416::Pmyo-2::avr-15, and pJL44::Phsp-16.48::MosTase::glh-2utr (Frokjaer-Jensen et al., 2008), and either 1 ng/μl or 50 ng/μl of targeting vector. MosSCI was performed using the heat-shock method (Frokjaer-Jensen et al., 2008) and single-copy insertion lines were selected on ivermectin to select against animals carrying the extrachromosomal array. Additional details are provided in Supplemental Information.

Small RNA cloning from isolated germ lines

Ten gonads from each strain were dissected in 1x PBS containing 0.1 mM EDTA, 1 mM Aurin tricarboxylate, 0.1% Tween 20, and 0.2 mM levamisole (Wang et al., 2009). Total RNAs were extracted with 5 volumes of TRI Reagent (MRC). Small RNAs were gel-purified and cloned as described (Gu et al., 2009). gfp::csr-1 small RNAs were pre-treated with Tobacco Acid Phosphatase (TAP, Epicenter Biotechnologies). gfp::cdk-1 and cdk-1::gfp small RNAs were pre-treated with CIP/PNK (NEB). Libraries were sequenced in the UMass Deep Sequencing Core using an Illumina HiSeq instrument.

Small RNA cloning from FLAG::WAGO-9 immune complexes

Synchronous adult flag::wago-9 worms were dounced in a stainless steel homogenizer. FLAG::WAGO-9 was immunoprecipitated from 20 mg of lysate essentially as described (Gu et al., 2009). Small RNAs were extracted from WAGO-9 immune complexes as well as a portion of the input lysate, gel-purified, pre-treated with TAP, cloned and sequenced as above.

Computational analysis of small RNAs

Deep sequencing data were processed and analyzed using custom Perl scripts (Gu et al., 2009). Definition of WAGO and CSR-1 22Gs are described in (Gu et al., 2009). Candidate 21U-RNAs that target gfp were identified by searching for seed sequences (nts 2–8th) that base-pair with at most two G:U wobbles, and at most 3 unpaired non-seed residues (nts 9–21st). Additional details are provided in Supplemental Information. Perl scripts are available on request.

Chromatin Immunoprecipitation

ChIP was performed essentially as described (Claycomb et al., 2009) except that synchronized adult neSi8 gfp::csr-1 (RNAe) and neSi9 gfp::csr-1 worms were dounced in a stainless steel homogenizer (30 strokes) prior to cross-linking with 2.6 % formaldehyde. Immunoprecipitations were performed in a total volume of 1 mL (5 mg) with 10 μg of anti-Histone H3 (ab1791, Abcam) or anti-H3K9me3 (ab8898, Abcam) antibodies. Immune complexes were recovered with 50 μL of Protein A Dynabeads (Invitrogen). Three independent ChIP experiments were performed and analyzed by quantitative PCR.

Quantitative PCR

Quantitative PCR was performed as described (Claycomb et al., 2009) using an ABI 7500 Fast Real-Time PCR instrument. For RNA analysis, cDNA was generated from 1 μg of total RNA using random hexamers and Superscript III Reverse Transcriptase (Invitrogen). gfp::csr-1 expression was measured relative to clp-3 mRNA levels. H3K9me3 ChIP was first normalized to Histone H3 ChIP, and fold enrichment was then determined relative to an H3K9me3 negative control locus. Primer sequences are provided in Supplemental Information.

Transgenerational RNAi phenotype

A single neSi9 gfp::csr-1, neSi12 cdk-1::gfp, tsIs1 oma-1::gfp or neIs2 gfp::wrm-1 adult worm was placed onto each of 10 plates seeded with gfp(RNAi) food. A single F1 worm from each plate was transferred to OP50 (control) or gfp(RNAi) food and each line was maintained for 10 generations by transferring a single worm from each plate to the corresponding food source, OP50 or gfp(RNAi). In each generation, 10 progeny from each plate were scored for gfp expression (100 total for each condition).

Western blot analysis

Antibodies used for Western blotting are anti-CSR-1 (Claycomb et al., 2009), GFP (A01704, Genscript) and α-Tubulin (MCA78A, Serotec) antibodies.

We thank Kirsten Hagstrom and Anna Zinovyeva for primer information and technical advice on ChIP experiment and helpful discussion. We are grateful to the members of the Mello lab for their input and discussion, especially Sandra Vergara for her help with genetic studies, William Stanney III for his technical help, and Elaine Youngman and Colin Conine for their advices on figure presentation. Some strains used in this study were obtained from the Caenorhabditis Genetic Center. H.C.L. is supported by Ruth L. Kirschstein National Research Service Awards fellowship (GM099372-2). This work is supported by NIH grant (GM058800) to C. C. M. and Howard Hughes Medical Institute.