Determinants of Heterochromatic siRNA Biogenesis and Function

SUMMARY

Endogenous small interfering RNAs (siRNAs) and other classes of small RNA provide the specificity signals for silencing of transposons and repeated DNA elements at the posttranscriptional and transcriptional levels. However, the determinants that define an siRNA-producing region or control the silencing function of siRNAs are poorly understood. Here we show that convergent antisense transcription and availability of the Dicer ribonuclease are the key determinants for primary siRNA generation. Surprisingly, Dicer makes dual contributions to heterochromatin formation, promoting histone H3 lysine 9 methylation independently of its catalytic activity, in addition to its well-known role in catalyzing siRNA generation. Furthermore, sequences in the 3’ UTR of an mRNA-coding gene inhibit the ability of siRNAs to promote heterochromatin formation, providing another layer of control that prevents the silencing of protein-coding RNAs. Our results reveal distinct mechanisms that limit siRNA generation to centromeric DNA repeats and prevent spurious siRNA-mediated silencing at euchromatic loci.

INTRODUCTION

RNA-based mechanisms silence the expression of transposons and foreign DNA sequences in eukaryotic organisms ranging from yeast to human (Ghildiyal and Zamore, 2009; Moazed, 2009; Sabin et al., 2013). This silencing is associated with the generation of small RNA (sRNA) molecules that act at both the posttranscriptional and transcriptional levels. The generation of the sRNA silencing trigger is associated with specific transcription events, but the mechanisms that distinguish between foreign and normal cellular transcription are poorly understood. These mechanisms are fundamentally important because inappropriate generation of sRNA could lead to spurious silencing of essential genes. Cells must therefore have evolved mechanisms that tightly regulate sRNA biogenesis and function.

In the fission yeast Schizosaccharomyces pombe, centromeres are surrounded by DNA repeats that are thought to be transposon remnants (Ekwall, 2007; Rhind et al., 2011). These repeats are assembled into heterochromatin, which contributes to de novo centromere assembly and proper chromosome segregation (Allshire et al., 1994). Transcription of the repeats gives rise to long noncoding RNAs, termed cenRNAs, which are processed into small interfering RNAs (siRNAs) by the RNAi machinery (Reinhart and Bartel, 2002; Verdel et al., 2004; Volpe et al., 2002). The siRNAs load onto the Ago1 subunit of the RNA-induced transcriptional silencing (RITS) complex, which also contains the GW domain protein Tas3 and the chromodomain protein Chp1 (Verdel et al., 2004). RITS then targets the centromeric repeats via base pairing between its siRNAs and nascent cenRNAs, leading to recruitment of the Clr4-Rik1-Cul4 (CLRC) complex, methylation of histone H3 lysine 9 (H3K9) by Clr4, and recruitment of HP1 proteins and histone deacetylases (Bayne et al., 2010; Gerace et al., 2010; Motamedi et al., 2004; Noma et al., 2004; Hong et al., 2005). H3K9 methylation also creates a binding pocket for Chp1 and stabilizes RITS association with chromatin (Noma et al., 2004). This step is critical for the recruitment of the RNA-dependent RNA polymerase complex (RDRC) and synthesis of the double-stranded RNA (dsRNA), which is subsequently processed into siRNA by the Dicer (Dcr1) ribonuclease (Motamedi et al., 2004; Sugiyama et al., 2005). Consistent with their role in mediating heterochromatin formation, the RITS complex and RDRC localize to foci in the nucleus that overlap heterochromatin proteins such as Swi6 (Colmenares et al., 2007; Noma et al., 2004; Sadaie et al., 2004). Moreover, although endogenous Dcr1 has not been detected by immunofluorescence, Dcr1 is physically associated with RDRC and, when moderately overexpressed, localizes to the nuclear periphery (Colmenares et al., 2007; Emmerth et al., 2010). It has remained unclear how localization of Dcr1 to the nuclear periphery contributes to its siRNA generation function.

Initiation of the cascade of events leading to RDRC-dependent siRNA amplification and heterochromatin formation requires a primary sRNA trigger. Analysis of centromeric transcription and sequencing of sRNA libraries has provided some clues into the origin of the primary sRNA. The outermost centromeric repeats are bidirectionally transcribed, producing overlapping complimentary cenRNAs that can potentially base pair to provide a source of dsRNA to be processed into primary siRNA by Dcr1 (Volpe et al., 2002). In addition, cenRNAs may form stem-loop structures that are cleaved by Dcr1 (Djupedal et al., 2009). If such primary siRNAs were produced by Dcr1, the levels of sRNAs should be higher in rdp1Δ than in dcr1Δ cells, since the fission yeast Rdp1 is only required for amplification or spreading of the siRNA signal. Previous high-throughput analyses of sRNA libraries have failed to detect any difference in sRNA levels between rdp1Δ and dcr1Δ cells, suggesting that in fission yeast, Dcr1 does not produce primary Rdp1-independent siRNAs (Halic and Moazed, 2010). On the other hand, analysis of the sRNA sequencing data revealed a population of Dcr1- and Rdp1-independent sRNA called primal RNA (priRNA) (Halic and Moazed, 2010). However, priRNAs map to the entire transcriptome, making it unlikely that their generation alone triggers the downstream events of siRNA amplification and heterochromatin formation.

siRNA-mediated heterochromatin formation appears to also be regulated by mechanisms that act downstream of siRNA generation. Studies involving long hairpin constructs show that Dcr1 efficiently processes the hairpin dsRNA into siRNAs, which load onto the RITS complex (Iida et al., 2008). However, hairpin siRNAs can only induce heterochromatin formation at a subset of targets. For example, hairpin-generated siRNAs fail to induce heterochromatin at a GFP reporter gene or the endogenous ura4⁺ locus but have variable silencing activity at the trp1⁺:: ura4⁺, arg3::ura4⁺-GFP, and ade6⁺-GFP loci (Iida et al., 2008; Sigova et al., 2004; Simmer et al., 2010). Although at the trp1⁺::ura4⁺ locus hairpin-mediated silencing correlates with antisense transcription, at the arg3::ura4⁺-GFP locus, hairpin-mediated silencing occurs in the apparent absence of antisense transcription. The mechanisms that determine whether primary siRNAs can trigger the amplification of secondary siRNAs and heterochromatin formation therefore remain unclear.

In this study we use a combination of approaches to address how primary siRNAs are generated and what limits their ability to silence target sequences. We show that a class of Dcr1-dependent primary siRNAs can be detected in S. pombe and provide evidence that convergent overlapping transcription gives rise to genome-wide dsRNA that can be processed into primary siRNA by Dicer. The biogenesis and function of primary siRNAs is controlled at multiple levels. First, Dicer availability controls the processing of dsRNA to primary siRNA, as genome-wide siRNA production is observed only when Dicer is overexpressed. Second, the ability of endogenous primary siRNAs to induce heterochromatin formation is restricted to pericentromeric DNA regions, suggesting that features of cenRNA or centromeric DNA repeats contribute to the ability of primary siRNAs to induce heterochromatin. Third, we show that signals in the 3’ UTR of a euchromatic gene inhibit the ability of siRNAs to induce heterochromatin formation, suggesting that 3’ end processing signals may protect mRNA-coding genes from RNAi-mediated heterochromatin assembly. Finally, we show that Dcr1 physically associates with the CLRC methyltransferase complex and plays a structural role in centromeric H3K9 methylation, distinct from its commonly known functions in siRNA generation and silencing. Together, these results provide insight into the pathways that regulate siRNA biogenesis and siRNA-mediated silencing and demonstrate roles for Dicer in heterochromatin formation beyond siRNA generation.

RESULTS

Detection of Rdp1-Independent Primary siRNAs

Previous analysis of high-throughput sRNA sequencing did not reveal a difference in the distribution of Ago1-associated centromeric sRNAs between rdp1Δ and dcr1Δ cells (Halic and Moazed, 2010). However, improvements in the depth of Illumina sequencing technology have greatly increased genome coverage in sRNA libraries so that, whereas previous libraries contained between 1 and 6 million aligned reads, our current libraries contained between 20 and 45 million aligned reads. With this increased coverage, we found that Ago1-associated centromeric sRNA were 4-fold more abundant in rdp1Δ cells than dcr1Δ cells (Figures 1A, 1B, S1A, and S1B available online). We therefore conclude that a population of Dcr1-dependent, but Rdp1-independent, primary siRNAs is present in fission yeast.

Figure 1. Fission Yeast Pericentromeric Repeats Produce Primary siRNAs that Increase in Abundance with Overexpression of Dcr1, which Also Partially Restores Silencing in the Absence of the RNA-Dependent RNA Polymerase Complex.

(A) Ago1-associated sRNA reads obtained by high-throughput sequencing of libraries using the Illumina platform, mapped to the left half of the centromere of chromosome 1 (cen1) and normalized by total number of reads (in reads per million, y axis). Note the 100-fold difference in scale of y axis between wild-type and mutant libraries. Chromosome coordinates and the location of the centromere (central core, cnt), the innermost (imr) and outermost dg and dh repeats are indicated above and below the sRNA reads, respectively.

(B) Sum of normalized sRNA reads per library mapping to the dg or dh centromeric repeats of all three chromosomes.

(C and D) sRNA reads from sequencing of total small RNA libraries prepared from rdp1Δ cells transformed with a control (rdp1Δ + v) or with a Dcr1 over-expression plasmid (rdp1Δ + Dcr1OE) mapping to dg1L (C) and dh1L (D), respectively. Total genome-wide proportion of sRNAs mapping to dg and dh, respectively, are on the right.

(E) Diagram of the Dcr1 protein and the location of functional domains and catalytic site mutations in the RNase III domains used in this study.

(F) Cells spotted on EMMC nonselective, -URA, or 5FOA-containing medium, showing that Dcr1 overexpression restores silencing in rdp1δ cells. Mutations in one (Dcr1-1A) or both (Dcr1-2A) RNase III domains of Dcr1 diminish or abolish, respectively, the rescuing activity.

(G and H) qRT-PCR assays showing that the overexpression of wild-type, but not catalytically inactive, Dcr1 silenced the expression of the dg (G) or dh (H) transcripts. dg/dh transcript levels were normalized to ura4 transcript levels, and each sample was normalized to rdp1δ + vector. p values are based on a two-tailed, one sample Student's t test, testing that the mean is less than 1. Error bars reflect SD.

Dicer Overexpression Boosts the Levels of Primary siRNAs and Restores Silencing in the Absence of RDRC

Previous studies have indicated that Dcr1 is a limiting factor in siRNA generation. Dcr1 overexpression boosts the levels of hairpin- and centromere-derived siRNAs and partially rescues silencing and H3K9 methylation in rdp1Δ cells (Iida et al., 2008; Kawakami et al., 2012). To understand how the dsRNA substrate for overexpressed Dcr1 is generated, we performed deep sequencing of sRNA libraries from these cells. Dicer was expressed on a plasmid under the control of the strong nmt1 promoter, which is maximally induced in the absence of thiamine (Maundrell, 1990). When grown under nonrepressive conditions, overexpressed Dcr1 levels were about 100-fold higher than those of endogenous Dcr1 (Figure S1C). Overexpression of Dcr1 in rdp1Δ cells resulted in a 20-fold increase in centromeric primary siRNA levels (Figures 1C and 1D), suggesting that Dcr1 availability limits the production of primary siRNAs from dsRNA.

Consistent with its effect on siRNA levels and previous findings (Kawakami et al., 2012), overexpression of wild-type Dcr1 partially rescued the loss of silencing phenotype at a centromeric ura4⁺ transgene (otr1R::ura4⁺) in cells lacking RDRC subunits rdp1⁺, cid12⁺, or hrr1⁺ (Figures 1E, 1F, and S1D). Silencing was assessed by growth on the drug 5FOA, which is toxic to cells expressing ura4. To test whether this phenomenon requires Dcr1 catalytic activity, we also overexpressed Dcr1 with amino acid substitutions at either one (Dcr1-D937A, hereafter referred to as Dcr1-1A) or both (Dcr1-D937A-D1127A, hereafter referred to as Dcr1-2A) ribonuclease III (RNase III) catalytic sites (Figure 1E). While Dcr1-1A maintains in vitro dsRNA cleavage activity, no cleavage activity is detected in Dcr1-2A (Colmenares et al., 2007). In contrast to wild-type Dcr1, overexpression of the mutant Dcr1 enzymes in rdp1Δ either restored very weak silencing or did not restore silencing at all, respectively (Figure 1F). Quantitative RT-PCR (qRT-PCR) showed a partial restoration of silencing for endogenous cenRNAs in rdp1Δ cells overexpressing Dcr1, but with the exception of very weak silencing of dg transcripts in rdp1Δ+Dcr1-1A-OE cells, neither of the Dcr1 catalytic mutants restored noticeable dg or dh silencing (Figures 1G and 1H), suggesting that silencing was coupled to the increase in primary siRNA levels generated by Dcr1-mediated dsRNA cleavage.

We next performed chromatin immunoprecipitation sequencing (ChIP-seq) and ChIP-qPCR to determine whether H3K9 methylation was also restored by Dicer overexpression. At the pericentromeric DNA repeats, Dcr1 overexpression in rdp1Δ cells (rdp1Δ+ Dcr1OE) restored H3K9 dimethylation (H3K9me2) levels to levels near those of wild-type, but only had a modest, if any, effect on H3K9me2 levels at subtelomeric DNA regions (Figures 2A and 2B). As previously shown, when grown in the absence of thiamine, Dcr1 under the control of the nmt1 promoter was overexpressed greater than 100-fold (Figure S1C). However, even when Dcr1 overexpression was reduced to less than 20-fold over endogenous levels by growth in thiamine, there was still considerable restoration of centromeric H3K9me2 in rdp1δ cells (Figures S1C and S2A). Interestingly, overexpression of Dcr1-1A or Dcr1-2A in rdp1δ cells also resulted in increased H3K9me2, despite the fact that neither supported efficient silencing (Figures 2C and 1F–1H). Overexpression of the catalytically dead Dcr1-2A induced an ~4-fold increase in centromeric H3K9me2 in rdp1δ cells, close to the levels induced by overexpression of the wild-type protein (~5.7-fold increase) (Figure 2C). Notably, while wild-type Dcr1 was consistently able to induce H3K9me2 when overexpressed in rdp1δ cells, there was considerable clonal variation for either of the Dcr1 catalytic mutants, likely due to low levels of centromeric siRNAs in these backgrounds. This variation was reflected in assays testing for the sensitivity of cells to the microtubule-inhibiting drug thiobendazole (TBZ) (Figure 2D). These observations suggest that Dcr1 plays a noncatalytic structural role in mediating H3K9me2 and functional heterochromatin, which, under conditions of overexpression, can bypass the requirement for Rdp1. Importantly, overexpression of either Dcr1 catalytic mutant did not restore centromeric H3K9me2 in rdp1δ/dcr1Δ double-mutant cells, indicating that the low levels of siRNA produced by endogenous Dcr1 were required for the restoration of Rdp1-independent H3K9me2 (Figure 2E). Consistent with this observation, overexpression of Dcr1 in ago1Δ cells also did not result in increased centromeric H3K9me2 (Figures S2B and S2C). Together, these results demonstrate that the structural role of Dcr1, uncovered here, still requires Dcr1-generated siRNAs and Ago1.

Figure 2. Dcr1 Plays a Structural Role in Promoting H3K9 Methylation Independently of its Catalytic Activity.

(A and B) ChIP-seq experiments showing H3K9 dimethylation (H3K9me2) levels at the pericentromeric regions (A) and the left telomere of chromosome 1 (B) for the indicated wild-type and mutant cells, respectively. Libraries were sequenced on the Illumina platform and normalized to reads per million (y axis). Chromosome coordinates are indicated above the reads. See Figure 1 legend for centromere abbreviations. tlh1, telomeric RecQ-type helicase. (C) ChIP-qPCR assays showing that overexpression of wild-type and the indicated mutant Dcr1 enzymes restores H3K9me2 at the dg repeats in rdp1δ cells. Error bars indicate SD.

(D) Overexpression of wild-type or catalytically inactive Dcr1 proteins suppresses the thiabendazole (TBZ) sensitivity of rdp1δ cells. Note variable rescue of TBZ sensitivity in cells overexpressing mutant Dcr1. Cells were spotted in 5-fold serial dilutions on medium lacking leucine, with or without 17 mg/l TBZ.

(E) ChIP assays showing that endogenous wild-type Dcr1 is required for the rescue of H3K9me2 by overexpression of catalytically inactive Dcr1 in rdp1δ cells. Error bars indicate SD.

(F) Coimmunoprecipitation assays showing that overexpressed TAP-Dcr1 associates with Flag-tagged CLRC subunits Raf1 and Clr4.

(G) Coimmunoprecipitation assays showing that overexpressed TAP-Dcr1 interacts with Flag-Raf1 in the absence of Rdp1 (left) and that endogenously expressed TAP-Dcr1 interacts with Flag-Raf1 (right).

(H) Coimmunoprecipitation assays showing that overexpressed Flag-tagged Dcr1 and Dcr1-2A associate with myc-tagged CLRC subunit Rik1.

To further explore the noncatalytic role of Dcr1 in heterochromatin formation, we performed coimmunoprecipitation (coIP) experiments testing for interaction between Dcr1 and components of the CLRC methyltransferase complex. We found that overexpressed Dcr1 coimmunoprecipitated with CLRC subunits Clr4 and Raf1, and the interaction between overexpressed Dcr1 and Raf1 did not require Rdp1 (Figures 2F and 2G). Endogenous Dcr1 also interacts with Raf1, demonstrating that the physical association between Dcr1 and CLRC occurred without Dcr1 overexpression (Figure 2G, right side). Finally, we found that overexpressed Dcr1 and Dcr1-2A both interact with the CLRC subunit Rik1. These results support a model in which overexpressed Dcr1 localizes to centromeres in a siRNA- and Ago1-dependent manner and contributes to CLRC recruitment and H3K9 methylation independently of its dsRNA cleavage activity.

To determine whether the restoration of silencing upon Dcr1 overexpression in rdp1δ cells still required the RITS complex and whether restoration of silencing could occur by overexpression of other RNAi enzymes, we individually overexpressed each of Dcr1, Ago1, or Rdp1 in rdp1δ, dcr1Δ, and ago1Δ cells. No silencing of cenRNAs was observed, except for overexpression of Dcr1 in rdp1δ cells (Figures S2B, S2D, and S2E). In contrast, ChIP-seq analysis showed that overexpression of RNAi proteins in different RNAi mutant backgrounds commonly affected centromeric H3K9me2 levels to some degree (less than 2-fold), suggesting that the requirements for H3K9me2 levels are not as tightly restricted as they are for transcriptional silencing (Figures S2C and S2F–S2H).

Dicer Overexpression Results in Genome-wide Generation of Primary siRNAs from Convergent or Overlapping Transcription Units

In our analysis of the sRNA sequencing data in Dcr1 overexpression libraries, we observed a striking global upregulation of siRNAs mapping antisense to mRNA-coding genes and structural RNAs (i.e., rRNA, tRNA, and snRNA) (Figure 3A). Upon Dcr1 overexpression in rdp1δ cells, about 47% of convergent mRNA-coding gene pairs and 43% of mRNA-coding genes convergent with noncoding RNAs (ncRNAs) produced some antisense siRNAs, compared to 3.3% and 1.9%, respectively, in the absence of Dcr1 overexpression (Figure 3B). This association between siRNAs and convergent transcription suggests that overlapping sense and antisense transcripts from these loci produce dsRNA that is cleaved by Dcr1. Interestingly, anti-sense siRNAs also mapped to 21% of nonconvergent genes in rdp1δ cells, compared to 0.7% when Dcr1 was not overex-pressed. However, one-fourth (102) of nonconvergent genes that produced antisense siRNAs were tandem to a convergent gene that also expressed antisense siRNAs, suggesting the possibility that a long antisense read-through transcript produced dsRNA. This increase in genome-wide siRNAs may be due to increased Dcr1 availability upon overexpression. Fission yeast centromeres cluster near the spindle pole body, and moderately expressed Dcr1 associates mainly with the nuclear periphery (Funabiki et al., 1993; Emmerth et al., 2010). We propose that this colocalization contributes to the preferential centromerespecific generation of primary siRNAs. However, when Dcr1 is overexpressed, it is present throughout the nucleus and cytoplasm and is likely to gain access to dsRNAs that are produced genome wide.

Figure 3. Dcr1 Overexpression Results in the Generation of Sense and Antisense Primary siRNAs at a Subset of Euchromatic Transcription Units.

(A) Summary of sRNA sequencing results, showing the effect of Dcr1 overexpression in rdp1δ cells on the percentage of sRNA reads mapping complementary to protein-coding RNA (mRNA), noncoding RNA (ncRNA), ribosomal RNA (rRNA), transfer RNA (tRNA), and small nuclear RNA (snRNA).

(B) Summary of sRNA sequencing results showing the effect of Dcr1 overexpression on the percentage of convergent or nonconvergent transcription units with antisense sRNAs.Genes with lessthan 1 kbbetween stopcodonswere defined asconvergent, and the minimum cutoff for antisense sRNAreaddensity was 1 rpmpkb.

(C) Elevated sense and antisense primary siRNA levels associated with the snu66-ptb1 convergent transcription units upon Dcr1 overexpression.

(D) Elevated sense and antisense primary siRNA levels associated with the atf1-isp4 convergent transcription units upon Dcr1 overexpression. Note siRNA spreading into the adjacent ncRNA transcription units.

(E) Dcr1-induced elevated sense and antisense primary siRNA levels associated with overlapping antisense transcription.

(F) Elevated sense and antisense primary siRNA levels at the lys3-SPAC2F7.02c convergent transcription units upon Dcr1 overexpression. Blue peaks in (C), (D), (E), and (F) indicate siRNA sequencing reads, and green peaks indicate H3K9me2 ChIP-seq reads.

Upon inspecting the distribution of reads at specific genomic loci, we found numerous examples of convergent or overlapping transcription units to which sense and antisense siRNAs mapped in approximately 1:1 ratios (Figures 3C–3F and S3A–S3E). Further analysis did not reveal any significant increase in H3K9me2 or reduction in RNA levels upon Dcr1 overexpression (Figures 3C–3F and S3, data not shown). Therefore, unlike centromeric primary siRNAs, euchromatic siRNAs, which are present at levels comparable to centromeric primary siRNAs, are unable to induce silencing or heterochromatin formation.

The Presence of mRNA 3’ Processing Signals Inhibits the Ability of Hairpin siRNAs to Promote Silencing and Secondary siRNA Synthesis

As shown above (Figures 1 and 3), Dcr1 overexpression promotes primary siRNA generation genome wide. However, silencing and Rdp1-dependent siRNA amplification (secondary siRNA synthesis) only occur within the centromeric repeats. This suggests that some feature of mRNA coding genes protects them from RNAi silencing. In this regard, we have previously shown that ura4⁺ at its endogenous locus is refractory to silencing by abundant hairpin-produced siRNAs but that ura4⁺ could be silenced by hairpin siRNAs at a different locus where we observed antisense transcription (Iida et al., 2008). However, Simmer et al. (2010) showed that hairpin-mediated silencing of a different reporter gene did not correlate with antisense transcription, suggesting that antisense transcription is not a required feature. We speculated that antisense transcription might make a locus more susceptible to siRNA-mediated silencing because antisense transcription through the 3’ UTR of a gene could disrupt proper 3’ end processing and termination of the sense transcript. mRNA-coding genes contain specific sequences at their 3’ UTRs that may inhibit the ability of siRNAs to use them as scaffolds for the recruitment of H3K9 methylation and siRNA amplification machineries. We therefore asked whether disrupting the endogenous ura4⁺ 3’ UTR sequence, which might interfere with 3’ end processing of the transcript, would make the locus more amenable to hairpin-mediated silencing.

We generated strains with different modifications to the ura4⁺ endogenous locus: the ura4-UTR^DIS locus, where a kanMX gene was inserted between the ura4⁺ stop codon and its 3’ UTR; the ura4-UTR^D locus, where the 3’ UTR was deleted and replaced with the kanMX gene; and the ura4-UTR-kanMX locus, where the kanMX gene was inserted downstream of the 3’ UTR (Figure 4A). Consistent with previous results (Iida et al., 2008), the ura4⁺ hairpin did not induce detectable silencing at the unmodified ura4⁺ or the ura4-UTR-kanMX loci (Figures 4B and S4A). However, the hairpin did induce weak silencing of the ura4-UTR^DIS allele and strong silencing of the ura4-UTR^D allele (Figure 4B). ura4-UTR^D showed weak growth on 5FOA medium even in the absence of hairpin expression, possibly because the heterologous 3’ UTR from the distantly related budding yeast Ashbya gossypii at the kanMX gene is not fully functional in fission yeast (Goldstein and McCusker, 1999) (Figure 4B). Hairpin-induced silencing in ura4-UTR^DIS cells was accompanied by a reduction in the levels of ura4⁺ sense RNA (Figure S4B).

Figure 4. The Polyadenylation and Cleavage Signals in the ura4⁺ 3’ UTR Inhibit Hairpin siRNA-Induced Silencing and Heterochromatin Formation.

(A) Diagrams of the wild-type ura4⁺ locus and ura4 alleles with modified 3’ UTRs. On the left two diagrams, four downstream polyadenylation and termination signals are indicated. These sites were mutated as in Birse et al. (1997) to generate strains ura4-UTR¹²³, ura4-UTR⁴, and ura4-UTR¹²³⁴, all of which have a downstream KanMX gene as in ura4-UTR-kanMX.

(B) Silencing assays showing the effect of 3’ UTR displacement or deletion on hairpin-induced silencing. otr1R::ura4⁺ serves as a positive control. hp, ura4⁺ hairpin.

(C) ChIP experiments showing that the ura4⁺ hairpin preferentially induces H3K9me2 at the ura4-UTR^D locus. H3K9me2 was lost upon the deletion of either clr4⁺ (clr4Δ) or the ura4⁺ hairpin (hpΔ). Error bars indicate SD.

(D) Northern blot showing the expression of ura4 RNA from indicated wild-type or mutant alleles.

(E) Silencing assays as in (B) showing the effect of specific polyadenylation or termination signal mutations on hairpin-induced silencing.

Despite the considerable hairpin-dependent increase in silencing of the ura4-UTR^D cells on 5FOA-containing medium, we did not observe a corresponding decrease in ura4⁺ sense transcript levels. This may be due to the fact that cells were grown in nonselective medium for RNA isolation. However, high levels of antisense transcripts in ura4-UTR^D were reduced more than 5-fold upon expression of the hairpin, suggesting that the hairpin does mediate transcriptional silencing (Figure S4B). All modified ura4 alleles showed levels of antisense transcription much higher than the unmodified locus, likely due to a cryptic reverse promoter in the kanMX gene (Figure S4B). Northern analysis showed that the ura4-UTR^DIS and ura4-UTR^D transcripts were longer than wild-type ura4, likely due to fusion with the kanMX transcript (Figure S4C). Finally, ChIP experiments showed that hairpin-induced silencing of the ura4-UTR^D allele was accompanied by H3K9me2, which was lost upon deletion of the hairpin or RNAi factors (Figures 4C and S4D). These results indicate that sequences within the 3’ UTR of ura4⁺, possibly cleavage or polyadenylation (polyA) signals, interfere with RNAi-mediated transcriptional gene silencing and provide an explanation for the inability of hairpin-produced siRNAs to silence euchromatic reporter genes at the transcriptional level.

To more specifically test the idea that polyadenylation and cleavage signals in the 3’ UTR inhibit siRNA-mediated silencing, we generated a series of directed mutations or deletions in four termination signals downstream of the ura4 open reading frame (ORF): two site-determinant elements (SDE1 and SDE2), one efficiency element (EE), and one downstream sequence element (DSE) (Birse et al., 1997) (Figure 4A). The elements were numbered 1–4, respectively, with the following mutations introduced at each: in ura4-UTR¹²³, a 34 bp deletion in SDE1, a 4 bp substitution of SDE2, and a substitution of 11 bp of the efficiency element with an 8 bp linker sequence; in ura4-UTR⁴, deletion of the 31 bp DSE; and in ura4-UTR¹²³⁴, a combination of all the above mutations and deletions. All strains included a KanMX cassette inserted downstream of the DSE (Figure 4A). Northern analysis showed a slightly longer transcript in ura4-UTR¹²³ and a much longer run-on transcript in ura4-UTR¹²³⁴ (Figure 4D). Growth assays on 5FOA showed hairpin-mediated silencing at both ura4-UTR¹²³ and ura4-UTR¹²³⁴, but not ura4-UTR⁴, demonstrating that trans silencing by ectopic siRNAs correlates with read-through transcription (Figure 4E).

The above results suggest that signals within the 3’ UTR region of ura4⁺ interfere with siRNA-mediated steps that may involve the recruitment of RDRC/Dcr1 and secondary siRNA generation, the recruitment of the CLRC complex and H3K9 methylation, or both. In order to distinguish between these possibilities, we tested whether 3’ UTR sequences affected secondary siRNA generation by sequencing siRNA libraries of either wild-type ura4⁺, ura4-UTR^DIS, or ura4-UTR^D cells that expressed the ura4⁺ hairpin (Figures 5A–5D). Secondary siRNA synthesis is expected to result from Rdp1-dependent dsRNA synthesis at the targeted locus and the spreading of siRNAs to regions flanking the siRNA-targeted domain. As shown in Figures 5B–5E, hairpin expression induced a low amount of secondary siRNA synthesis, indicated by the spreading of siRNA reads along the ura4 transcript. Secondary siRNAs were more abundant at the ura4-UTR^DIS locus and by far the most abundant at the ura4-UTR^D locus. These results indicate that signals within the ura4⁺ 3’ UTR region interfere with primary siRNA-mediated recruitment of the machinery that promotes secondary siRNA synthesis.

Figure 5. The 3’ UTR of ura4 Inhibits the Generation of Secondary siRNAs.

(A) Diagram of the ura4 hairpin RNA containing the cox4 intron. Blue arrows, double-stranded region complementary to ura4⁺; gold loop, cox4 intron.

(B–D) siRNA-seq of Ago1-associated siRNAs at the wild-type ura4⁺ (B), ura4-UTR^DIS (C), or ura4-UTR^D (D) loci, shown in log scale. Note the spreading of siRNAs to the tam14 gene at the ura4-UTR^D locus.

(E) Sum of normalized siRNA-seq read densities mapping to the indicated loci in reads per million per kilobase (rpmpkb). As noted in the Experimental Procedures, total reads mapping to ura4-UTR^D are approximately 30%–40% of those in ura4⁺ or ura4-UTR^DIS due to an unusually high number of reads mapping to one siRNA, located in rDNA, in the ura4-UTR^D sample.

(F) Hairpin-induced secondary siRNA reads mapping to the cox4 locus (left). The levels of intronic cox4 siRNAs are comparable to the levels of siRNAs mapping to the ura4 hairpin shown in (B)–(D).

(G) Northern blot showing hairpin-dependent isoforms containing the cox4 intron.

We next tested the effect of mutations in RNA processing or heterochromatin proteins on siRNA synthesis at the ura4-UTR^D and cox4⁺ loci. At the ura4-UTR^D locus, deletion of rdp1⁺ resulted in the loss of secondary siRNAs, whereas deletion of clr4⁺ resulted only in an 2-fold decrease in secondary siRNA synthesis (Figures S5A and S5B). Consistent with previous northern blot analysis (Iida et al., 2008), deletion of either rdp1⁺ or clr4⁺ did not diminish primary hairpin siRNA levels (Figure S5A). Furthermore, deletion of TRAMP component cid14⁺, which results in a decrease in the levels of centromeric siRNAs (Bühler et al., 2007), resulted in a modest decrease in the levels of hairpin-derived and secondary siRNAs (Figures S5A and S5B).

The ura4⁺ hairpin construct contains a 355 nt intron from the cox4⁺ gene that forms the loop in the stem-loop hairpin structure (Figure 5A) (Iida et al., 2008). We observed a high proportion of reads mapping sense and antisense to the cox4 intron at a density (in reads per million per kilobase [rpmpkb]) comparable to that of the siRNAs from the dsRNA stem of the ura4 hairpin (Figures 5E and 5F). Northern analysis showed the presence of hairpin-dependent RNAs containing the cox4 intron, suggesting that the intron was not efficiently spliced from the hairpin, thus becoming a substrate for RDRC (Figure 5G). Furthermore, at the endogenous cox4⁺ locus, we detected low levels of anti-sense siRNAs that mapped to the mature spliced cox4 RNA regions flanking the intron, which is consistent with the low levels of secondary siRNAs at the unmodified ura4⁺ locus (Figures S5A and S5C). As expected, the cox4 tertiary siRNAs were lost in rdp1δ cells and decreased in cid14Δ cells but, surprisingly, became more abundant in clr4Δ cells (Figures S5A and S5C). One likely explanation for this observation is that in cells lacking clr4⁺, the RNAi machinery is released from the centromere and becomes more available at the cox4 locus. The detection of siRNAs that map to the cox4 mature mRNA indicates that abundant intronic cox4 secondary siRNAs can target the cox4 premRNA prior to splicing, most likely cotranscriptionally, and initiate tertiary siRNA synthesis.

A Transcript Lacking 3’ Processing Signals Is Targeted by RNAi

As noted above, cells carrying a ura4⁺ gene lacking a 3’ UTR grow on medium containing 5FOA, indicating that even without being targeted by hairpin ura4⁺ siRNAs, aberrant ura4-UTR^D RNAs are partially silenced. We wished to test whether the RNAi pathway contributed to this silencing, perhaps along with the TRAMP/exosome pathway. Toward this goal, we made deletions of RNAi genes ago1⁺, rdp1⁺, and dcr1⁺, as well as clr4⁺ and cid14⁺, which encodes the noncanonical polyA polymerase in the TRAMP complex that targets aberrant transcripts for degra dation by the exosome. Deletion of any of these genes resulted in an immediate loss of silencing at the ura4-UTR^D locus, suggesting that both RNAi and the TRAMP/exosome pathways were required for silencing of the defective ura4-UTR^D RNA (Figure 6A). However, we noted that extended passaging of RNAiD or clr4Δ (but not cid14Δ) cells resulted in a regaining of ura4-UTR^D silencing, suggesting that the TRAMP/exosome pathway compensated for the loss of the RNAi-mediated mechanism of silencing.

Figure 6. RNAi-Mediated Silencing Correlates with Low or Unusual Polyadenylation Signals.

(A) Silencing assays showing that RNAi is required for the silencing of a ura4 allele that lacks 3’ end processing signals (ura4-UTR^D). ura4-UTR^D silencing is lost upon deletion of RNAi components (dcr1, rdp1, ago1), the H3K9 methyltransferase clr4, or the TRAMP component cid14.

(B) Silencing assay showing that Dcr1 overexpression induces stronger silencing at ura4-UTR^D than does expression of the siRNA-producing ura4 hairpin.

(C) ChIP experiments showing that Dcr1 overexpression induces H3K9me2 at the ura4-UTR^D locus 15-fold higher than does a ura4 hairpin. Note that H3K9me2 at ura4-UTR^D + Dcr1OE was comparable to that of a ura4 gene inserted at an outer centromeric repeat. Error bars indicate SD.

(D) Scatterplot comparing reads mapping downstream of the indicated transcription units in a WT polyA sequencing library to reads mapped by expression microarray (Dutrow et al., 2008). For dg/dh, reads mapping within the entire repeat were counted. Note that expression levels are relatively lower for dg/dh compared to most genes or other categories of transcription units and that reads mapping to act1 or rpl4101 are above the 97^th percentile on both axes.

(E) Example of polyA sequencing reads mapping downstream of open reading frames. Note that reads mainly map to one or two predominant peaks.

(F) PolyA and siRNA sequencing reads mapping to the outer centromeric repeats of chromosome 1. Note the extremely low level of polyA reads mapping in wild-type and the existence of many low polyA peaks, which are dispersed throughout large regions, mapping to the forward and reverse strands of dg and dh RNAs in rdp1δ.

We next tested whether Dcr1 overexpression could potentiate the RNAi-dependent, but ura4⁺ hairpin-independent, silencing of ura4 alleles with defective 3’ UTRs. We observed strong silencing at ura4-UTR^D, but none of the other alleles, and this silencing was maintained even when overexpression of the thiamine-repressible nmt1-dcr1⁺ was reduced by growth in thiamine-containing medium (Figures 6B, S6A, and S6B; see Figure S1C for Dcr1 protein levels when nmt1-dcr1⁺ cells were grown in thiamine-containing medium). In addition, Dcr1 over-expression resulted in strong Ago1- and Rdp1-dependent H3K9me2 at the ura4-UTR^D locus, comparable to H3K9me2 levels at centromeric reporter gene otr1R::ura4 and much stronger than that observed with the ura4 hairpin (Figures 6C and S6C). Weak H3K9me2 at the ura4-UTR^D locus was also observed when the catalytically inactive Dcr1-2A was over-expressed (Figure S6C). Together, these results suggest that a transcript lacking proper 3’ end processing signals is targeted for silencing and heterochromatin formation by the RNAi machinery.

We propose that silencing of the ura4-UTR^D locus, described above (Figures 6A and 6B), and silencing of the outer centromeric repeats occur via similar RNAi-mediated mechanisms. Both the ura4-UTR^D locus and centromeric repeats possess sense and antisense transcripts, and as described below, both possess weak or abnormal polyadenylation and termination signals. We performed direct RNA sequencing of total polyadenylated transcripts in either wild-type or rdp1δ cells and used a strategy that mapped the last nucleotide before the polyA tail. The polyA sequencing (polyA-seq) results revealed notable differences in the polyadenylation pattern of transcripts arising from the outer centromeric repeats compared to mRNAs. We summed all reads mapping either downstream of annotated transcription units or within the dg/dh repeats of each chromosome in a wild-type strain. In general, we observed a good correlation between transcript abundance, based on microarray expression data (Dutrow et al., 2008), and the number of polyA-seq reads (Figure 6D). As expected, the centromeric dg and dh transcripts ranked low compared to genic transcription units and relative to their expression levels, although they contain higher polyA levels than structural RNAs, such as rRNAs and tRNAs (Figure 6D). Moreover, in contrast to mRNAs, which contained one or two major polyA peaks, the centromeric dg and dh transcripts contained multiple weak polyA peaks that generally mapped throughout the transcribed region that gives rise to siRNAs (Figures 6E and 6F). These observations suggest that centromeric transcripts, like the ura4-UTR^D locus, possess weak or abnormal 3’ end processing signals, which may contribute to their susceptibility to RNAi-mediated silencing.

DISCUSSION

The findings presented here provide insight into the mechanisms that determine which genomic loci produce primary siRNAs and the factors that determine whether the primary siRNA signal is amplified and mediates heterochromatic gene silencing. Our results indicate that properties of transcribed RNAs and Dicer availability together determine whether a locus produces siRNA and whether the siRNA can induce heterochromatin formation. Below, we discuss the implications of these findings for control of RNAi-mediated heterochromatin formation.

Biogenesis of Primary siRNA

The mechanisms that determine which transcripts give rise to siRNA play a fundamental role in defining how RNAi regulates genome organization and function. In S. pombe, transcripts from the centromeric dg and dh repeats, and related sequences at the mating type locus and subtelomeric DNA regions, are precursors for siRNA synthesis (Bühler et al., 2008; Cam et al., 2005; Halic and Moazed, 2010; Reinhart and Bartel, 2002; Verdel et al., 2004). In addition to Dicer, siRNA generation from these loci requires several complexes that include the RDRC, the Ago1-containing RITS complex, and the CLRC complex (Bühler et al., 2006; Motamedi et al., 2004; Noma et al., 2004; Verdel et al., 2004). Models of siRNA amplification propose that the RITS complex targets the noncoding dg and dh RNAs and recruits RDRC to initiate siRNA amplification. However, this targeting requires the presence of a trigger sRNA bound to Ago1 in the RITS complex that must base pair with the target RNA (Halic and Moazed, 2010). We previously described a population of Dcr1-independent sRNAs, called priRNAs, which could potentially perform this trigger function (Halic and Moazed, 2010). In this study, we achieved a deeper coverage of Ago1-associated sRNAs and were able to detect a population of Dcr1-dependent, but Rdp1-independent, sRNAs. We conclude that these sRNAs define a population of primary siRNAs that results from Dcr1-mediated cleavage of dsRNAs produced from the intermolecular base pairing of forward and reverse centromeric dg and dh RNAs. Although an alternative model, involving Dcr1-mediated cleavage of intramolecular dg or dh stem-loop structures (Djupedal et al., 2009; Kawakami et al., 2012), cannot be ruled out, the even distribution of primary siRNAs along both strands of the transcribed regions of dg and dh and the similarity of this distribution to that of secondary Rdp1-amplified siRNAs in wild-type cells lead us to favor the former model (Figures 1 and S1). Thus, S. pombe contains two distinct populations of Dcr1-dependent and Dcr1-independent sRNAs that can act as triggers for further siRNA amplification and heterochromatin formation (Figure S7).

We previously showed that Dcr1 overexpression results in increased siRNA generation from a ura4⁺ hairpin construct, suggesting that Dcr1 is limiting for siRNA generation even when an abundant dsRNA substrate is present (Iida et al., 2008). The results presented here show that Dcr1 overexpression boosts primary siRNA levels by about 20-fold and partially suppresses the requirement for all three subunits of the RDRC in centromeric silencing. Consistent with these results, a recent report showed that Dcr1 overexpression restored silencing in rdp1δ and dsh1Δ cells (Kawakami et al., 2012). Dsh1 is a putative inner nuclear membrane protein that may help localize Dcr1 to cenRNAs by a mechanism that remains to be defined. These observations suggest that the transcription of centromeric repeats provides a source of dsRNA that can be directly processed into primary siRNAs by Dcr1, providing a trigger for secondary siRNA amplification (Figure S7). However, as discussed below, our results also suggest that Dcr1 plays distinct catalytic and noncatalytic roles in promoting RNAi-dependent heterochromatin formation.

Further support for the idea that overlapping transcription provides a source of dsRNA for Dicer comes from the observation that Dcr1 overexpression results in transcriptome-wide generation of siRNAs from convergent or overlapping transcription units. In Dcr1-overexpressing cells, in which Dcr1 is not restricted to the nuclear periphery (Emmerth et al., 2010), we observe antisense primary siRNAs that map to approximately half of all protein-coding genes that either are convergent with other protein coding genes or contain annotated overlapping antisense RNAs (Figure 3B). These results strongly suggest that pervasive dsRNA, resulting from base pairing with overlapping complimentary transcripts, is formed in S. pombe but normally not processed into siRNA because Dcr1 levels and localization are strictly regulated. The S. pombe Dcr1 activity is regulated at multiple levels via its association with heterochromatin through interactions with the RITS complex and RDRC (Colmenares et al., 2007), its localization to the nucleus and nuclear periphery (Emmerth et al., 2010), and its direct or indirect interaction with the nuclear pore complex and the Dsh1 protein (Emmerth et al., 2010; Kawakami et al., 2012). However, the significance of these interactions for siRNA generation has been unclear. Since endogenous fission yeast Dcr1 preferentially processes centromeric RNAs into primary siRNAs (Figure 1), we propose that the localization of both Dicer and centromeric repeats to the nuclear periphery limits siRNA generation and amplification to the subset of overlapping RNAs that are transcribed from centromeric repeats. The mechanisms that mediate the localization of centromeres to the nuclear periphery remain to be determined but are likely to involve interactions with inner nuclear membrane proteins.

Catalytic and Structural Functions of Dicer in Centromeric Silencing

Our analysis of the effects of Dcr1 overexpression on histone H3K9 methylation and silencing reveal unexpected catalytic and structural roles for Dcr1 in promoting silencing and H3K9 methylation. Although only wild-type Dcr1 overexpression restores silencing in rdp1δ cells (Figures 1A–1H), the overexpression of a catalytically dead Dcr1 in rdp1δ cells also results in a substantial increase in H3K9 methylation levels (Figure 2C). These results give rise to two important conclusions. First, they suggest that Dcr1 can play a structural role in promoting H3K9 methylation independent of its siRNA-producing activity and Rdp1. We note that the ability of the catalytically dead Dcr1 to promote H3K9 methylation in rdp1δ cells relies on Ago1 and endogenous Dcr1; therefore, Dcr1 is likely to contribute to the recruitment of the CLRC complex to nascent cenRNAs in steps downstream of siRNA generation (Figure S7). Second, they suggest that the catalytic activity of Dcr1 plays an important role in silencing beyond promoting H3K9 methylation. We propose that this role involves the cotranscriptional degradation of cenRNAs, the byproduct of which is siRNAs.

Inhibition of siRNA-Mediated Silencing by 3’ Processing Signals

Most protein-coding genes contain signals in their 3’ UTRs that mediate efficient cleavage, polyadenylation, and export of their mRNA (Proudfoot, 2011; Richard and Manley, 2009). Our findings suggest that in addition to these critical roles in mRNA metabolism, signals in 3’ UTRs protect protein-coding genes from siRNA-mediated silencing (Figure 7A). The deletion or displacement of the 3’ UTR makes the euchromatic ura4⁺ gene, which is normally refractory to siRNA-mediated silencing, a target for siRNA-mediated silencing. We propose that proximity to polyadenylation signals prevents siRNA-mediated silencing by promoting the release and export of the RNA before it can be targeted by the siRNA-containing RITS complex (Figure 7). This targeting is required for both the recruitment of RDRC, which synthesizes the dsRNA precursor for the generation of secondary siRNA by Dcr1, and recruitment of the CLRC complex, which methylates histone H3. Perhaps more strikingly, the partial silencing of a ura4 allele lacking 3’ processing signals requires both the RNAi and the TRAMP/exo-some pathways, suggesting that RNAi may play a global role in silencing aberrant RNAs without proper 3’ end processing signals (Figure S7B).

Figure 7. Control of siRNA Biogenesis and Function in S. pombe.

(A) Convergent transcription at euchromatic loci results in dsRNA generation, but because the concentration of Dicer (Dcr1) is limiting, this dsRNA is not processed into siRNA unless Dcr1 is overexpressed. Signals in the 3’ UTR inhibit siRNA-mediated heterochromatin formation by promoting polyadenylation, which is coupled to mRNA release from the site of transcription and export from the nucleus.

(B) Centromeric RNAs contain weak polyadenylation and transcription termination signals, and their slower release from the site of transcription may allow them to act as scaffolds for siRNA-mediated heterochromatin formation. Dcr1* denotes a noncatalytic role for Dcr1 in promoting heterochromatin formation at steps downstream of siRNA generation.

Our analysis of siRNAs in cells that express the ura4 hairpin uncovered targets for ura4 hairpin siRNAs that provide further support for the cotranscriptional targeting of nascent transcripts by RNAi. First, the hairpin itself becomes a target for Rdp1-dependent dsRNA synthesis, resulting in high levels of siRNAs that correspond to the cox4 intron in the hairpin loop (Figure 5F). Splicing of the cox4 intron may be inefficient (see Figure 5G), allowing dsRNA synthesis by RDRC prior to intron removal. The density of reads originating from the cox4 intron is comparable to the density of reads mapping to the double-stranded region in the ura4 hairpin, implying that a substantial portion of the ura4 reads originating from the hairpin are actually Rdp1 derived. Second, we observe tertiary siRNAs at the endogenous cox4 locus at regions flanking its intron. Here, Rdp1 association must occur prior to splicing of the cox4 intron at its endogenous transcript. We note that the hairpin itself is a much better target for secondary siRNA generation than ura4 mRNAs, even when the 3’ end processing signals are deleted (e.g., ura4-UTR^D). A possible explanation for this observation is that the presence of an intron in the hairpin RNA and its targeting by the splicing machinery may improve the ability of RNAi to target the transcript. In this regard, previous work has uncovered links between splicing and RNAi. In S. pombe, a subunit of RDRC associates with the spliceosome, and mutations in some splicing factors have defects in RNAi-mediated silencing (Bayne et al., 2008; Motamedi et al., 2004). Moreover, in C. neoformans, transcripts containing suboptimal splicing signals become RNAi targets (Dumesic et al., 2013).

EXPERIMENTAL PROCEDURES

Plasmid and Strain Constructions

S. pombe strains and plasmids used in this study are described in Tables S1 and S2, respectively.

Growth Assays

Cells were grown in 5 ml of complete minimal medium (EMMC, Sunrise Science Products) at 30° C overnight, diluted into fresh EMMC medium, and harvested in log phase (10⁶–10⁷ cells/ml). Cells were washed with water and then resuspended in water to a concentration of 2 × 10⁷ cells/ml. For silencing assays, 5 μl of serial 10-fold dilutions were then spotted on appropriate plates (EMMC, EMMC-URA, and EMMC + 5FOA, with or without leucine). For TBZ assays, 5 μl of serial 5-fold dilutions were spotted on nonselective plates or plates containing 17 mg/l TBZ. Plates were incubated at 30° C for 3–7 days and photographed.

sRNA Libraries

To purify total sRNAs, cells were grown in 100 ml yeast extract with supplements (YES) (Moreno et al., 1991) to a concentration of ~2 × 10⁷ cells/ml. Pellets were processed using the mirVana miRNA Isolation kit (Ambion), and the resulting RNA was used for library construction. Total and Ago1-associated small RNA libraries were constructed as previously described (Halic and Moazed, 2010). Briefly, 21–24 nt RNA was size selected on a 17.5% polyacrylamide/7 M urea gel and ligated to a 3’ adaptor. The ligated species were size selected on a 17.5% polyacrylamide/7 M urea gel and ligated to a 5’ adaptor. RNA was then reverse transcribed into cDNA and PCR amplified in a two-step process. Amplified cDNA was gel purified and sequenced on an Illumina GAIIx or HiSeq platform.

Analysis of PolyA and sRNA Sequences

Sequencing data were analyzed using customized Python scripts (R.Y., unpublished data), which are available upon request. For sRNA libraries, reads with a maximum 1 nt mismatch were aligned to the S. pombe genome using Maq (http://maq.sourceforge.net/), normalized for reads per million, and visualized using the Integrative Genomics Viewer (IGV) (http://www.broadinstitute.org/igv/). Reads mapping to more than one location were randomly assigned. For wild-type Ago1-associated sRNA libraries, approximately 30% of reads generally map to rDNA (as for ura4⁺ and ura4-UTR^DIS with ura4-hp libraries), though this proportion may vary. For unclear reasons, in the ura4-UTR^D + ura4-hp library, 75% of reads mapped to rDNA, resulting in read values (in reads per million) that are 2.5-fold lower for the remaining reads mapping genome wide, including to regions of interest such as dg and dh repeats, ura4, and cox4.

Total RNA Purification

Total RNA was purified using the hot phenol method (Leeds et al., 1991). A more detailed protocol is provided in the Supplemental Experimental Procedures. Total RNA preparations were further purified using RNeasy Mini kits (QIAGEN) following the RNeasy Mini Protocol for RNA Cleanup provided in the manufacturer's handbook.

PolyA Sequencing

Purified total RNA was submitted to Helicos for direct RNA sequencing of immobilized polyadenylated RNA on oligo dT chips (Ozsolak et al., 2010). Reads with a minimum 20 nt length and maximum 1 mismatch were aligned using the basic pipeline of the Helisphere software. Reads mapping to greater than one location were randomly assigned. Files were normalized for reads per million and converted to IGV-viewable formats.

Reverse Transcription

cDNA was prepared using transcript-specific oligonucleotide primers (Table S3) and SuperScript III Reverse Transcriptase from Invitrogen.

Quantitative PCR

DNA or cDNA was amplified with the Taq polymerase using primers described in Table S3 in the presence of SYBR Green. For ChIP-qPCR, reported values are the percent of input using the DC_T method. For qRT-PCR, relative RNA levels were quantified using the DC_T method and normalized to either act1 or ura4 levels. Enrichment relative to the indicated RNA was calculated after normalization. Error bars in all figures indicate SD.

Northern Blot

Northern blot was performed as described previously (Bühler et al., 2006).

Chromatin Immunoprecipitation

ChIP experiments were performed as previously described (Huang and Moazed, 2003). A description is available in the Supplemental Experimental Procedures.

Sample Preparation for Multiplex ChIP-Seq

Libraries for Illumina sequencing were constructed following the manufacturer's protocols, starting with ~5 ng of immunoprecipitated DNA fragments. Each library was generated with custom-made adapters carrying unique barcode sequences at the ligating end (Wong and Struhl, 2011). Barcoded libraries were mixed and sequenced with Illumina HiSeq 2000. Raw reads were separated according to their barcodes and mapped to the S. pombe genome using Bowtie. Mapped reads were normalized to reads per million and visualized in IGV.

Coimmunoprecipitation Assay and Western Blotting

Immunoprecipitations were carried out as described in the Supplemental Experimental Procedures. For western blotting, 2 ml of cells grown to optical density 600 (OD₆₀₀) = 1 were pelleted. Cells were resuspended in 450 μl cold 2 M NaOH + 7.5% b-mercaptoethanol and incubated on ice for 15 min. We added 450 μl of 55% trichloroacetic acid, followed by incubation on ice for 15 min and centrifugation at 13,000 rpm for 10 min (Eppendorf 5415R). Pellets were washed once in acetone, dried, and resuspended in high-urea sample buffer (8 M urea, 5% SDS, 2 M Tris [pH 6.8], 1.5% dithiothreitol [DTT], bromophenol blue). For blotting, anti-Flag-HRP (Sigma) and anti-actin (Abcam 8224) were used.