An enhancer screen identifies new suppressors of small-RNA-mediated epigenetic gene silencing

Small non-protein coding RNAs are involved in pathways that control the genome at the level of chromatin. In Schizosaccharomyces pombe, small interfering RNAs (siRNAs) are required for the faithful propagation of heterochromatin that is found at peri-centromeric repeats. In contrast to repetitive DNA, protein-coding genes are refractory to siRNA-mediated heterochromatin formation, unless siRNAs are expressed in mutant cells. Here we report the identification of 20 novel mutant alleles that enable de novo formation of heterochromatin at a euchromatic protein-coding gene by using trans-acting siRNAs as triggers. For example, a single amino acid substitution in the pre-mRNA cleavage factor Yth1 enables siRNAs to trigger silent chromatin formation with unparalleled efficiency. Our results are consistent with a kinetic nascent transcript processing model for the inhibition of small-RNA-directed de novo formation of heterochromatin and lay a foundation for further mechanistic dissection of cellular activities that counteract epigenetic gene silencing.

Introduction Small RNAs are the common denominator of various RNA silencing pathways that regulate gene expression and protect the genome against mobile repetitive DNA sequences, retroelements, and transposons [1][2][3]. They function as specificity factors by guiding Argonaute protein-containing silencing complexes to their respective targets via base-pairing interactions [4,5]. In the fission yeast Schizosaccharomyces pombe, endogenous small interfering RNAs (siR-NAs) are indispensable for the maintenance of centromeric heterochromatin [6]. They originate from within centromeric heterochromatin and target the Argonaute-containing (Ago1) RNA-Induced Silencing Complex (RITS) in cis to nascent heterochromatic transcripts that are emanating from RNA polymerase II (RNA Pol II) transcribing the underlying repetitive DNA [7,8]. Besides binding to nascent RNAs, RITS also binds to methylated histone H3 lysine 9 (H3K9me) through its chromodomain-containing subunit Chp1 [9,10]. Constituting a positive feedback loop, the RITS complex recruits H3K9 methylation and RNA-dependent RNA polymerase activities to the locus it associates with [9,[11][12][13]. Dicer-mediated (Dcr1) processing of the resulting double-stranded RNAs leads to amplification of the siRNA pool and thereby reinforcement of the positive feedback loop [14].
Whereas siRNAs originating from heterochromatic repeats function well in cis to sustain H3K9 methylation, they do not act in trans to mediate de novo formation of heterochromatin at complementary protein-coding genes outside centromeric repeats in wild-type cells [15]. Similarly, synthetic siRNAs produced from RNA-hairpins are not sufficient to stably silence homologous protein-coding genes through the assembly of heterochromatin [16][17][18][19]. However, siRNAs have been shown to become potent mediators of RNAi-mediated epigenetic gene silencing in S. pombe cells that are mutant for mlo3 + , dss1 + , mst2 + , or genes encoding subunits of the Paf1 complex (Paf1C) [15,18,20,21]. Indicating potential evolutionary conservation, Paf1C also opposes PIWI/piRNA-directed silencing in Drosophila melanogaster [22].
Current understanding of the mechanisms that counteract small-RNA-mediated epigenetic gene silencing is in its infancy. Furthermore, rates at which silencing is initiated or maintained vary substantially between the different enabling mutations identified so far. For example, initiation of gene silencing was reported to occur in approximately 0.5-2% of mlo3Δ cells [15]. Silencing in Paf1C mutants is initiated in up to 20% of cells and is boosted to more than 80% upon additional deletion of the mst2 + gene [18,20]. Although initiation of heterochromatin assembly in mlo3Δ and Paf1C mutant cells is not efficient, the silent state is stably maintained once established, even when cells undergo meiosis. That is, Mendelian segregation of the silent allele was observed irrespective of the primary siRNA trigger and enabling mlo3 + deletion in one study [15]. Inheritance of the repressed state induced by hairpin-derived siRNAs remains dependent on the enabling mutations in Paf1C but ensues independently of the hairpin trigger as well [15,18,23]. Here, inheritance patterns of the silencing phenotype violate Mendel's laws, reminiscent of the paramutation phenomenon [18,23,24].
Whereas the role of the RNA export factor Mlo3 in counteracting small-RNA-mediated epigenetic gene silencing remains enigmatic, two non-mutually exclusive models have been put forward for Paf1C [25]. The first model suggests that dilution of K9 methylated H3 is lowered by the reduced histone H3 exchange rates in Paf1C mutants, stabilizing the H3K9 methylation state and hence the aforementioned positive feedback loop [26]. The second model suggests that mutations in Paf1C reduce the kinetics of nascent transcript release from chromatin, allowing sufficient time for RITS to recruit H3K9 methylation and RNA-dependent RNA polymerase activities that are necessary to initiate and propagate the positive feedback loop [18]. This model is supported by observations that alterations in the polyadenylation signal (PAS) of a target pre-mRNA enable siRNA-directed H3K9 methylation [27]. Yet, mutations in the pre-mRNA cleavage and polyadenylation machinery that would impair nascent transcript cleavage and hence potentially enable small-RNA-directed de novo formation of heterochromatin formation have not been identified. Thus, additional evidence supporting the second model is wanted.
In this study we have combined chemical mutagenesis with whole-genome sequencing in a sensitized reporter strain to obtain a more comprehensive list of putative suppressors of small-RNA-mediated epigenetic gene silencing. This revealed more than 20 novel silencing-enabling mutations in genes that are associated with RNA processing, regulation of transcription, or post-translational protein modification. Focusing on factors involved in pre-mRNA cleavage and polyadenylation, we show that single amino acid substitutions in Yth1, which is responsible for PAS recognition, lead to nearly 100% effective de novo formation of silent heterochromatin. Our work provides further support for a kinetic model for the inhibition of small-RNA directed de novo formation of heterochromatin and demonstrates that epigenetic gene silencing can be enabled by the acquisition of a plethora of mutant alleles in fission yeast.

An enhancer screen identifies 20 novel mutant alleles that enable small-RNA-mediated epigenetic gene silencing
To identify novel factors that suppress the susceptibility of protein-coding genes for epigenetic silencing via siRNAs that are acting in trans, we employed an ade6 + -based reporter system and whole-genome sequencing pipeline of a previous screen that had revealed Paf1C as a potent inhibitor of siRNA-directed heterochromatin formation [18]. ade6 + is a suitable reporter because Ade6-deficient cells form red colonies on limiting adenine indicator plates, whereas ade6 + expressing cells appear white. This allows simple assessment and quantification of the initiation, maintenance, and inheritance of siRNA-mediated silencing [28].
Because deletion of the mst2 + gene substantially increases the rate at which silencing is established de novo in Paf1C mutant cells [20], we created a sensitized reporter strain in which the mst2 + gene was deleted, and a RNA hairpin (ade6-hp) complementary to 250 nt of the trp1 + ::ade6 + reporter was expressed from the nmt1+ locus on chromosome I (Fig 1A). In the absence of additional enabling mutations, the siRNAs generated from the ade6-hp do not stably silence the complementary trp1 + ::ade6 + reporter gene in trans in a mst2 Δ background [20] ( Fig 1A). To screen for mutants that would enable ade6 siRNAs to establish and maintain robust trp1 + ::ade6 + silencing, we mutagenized our sensitized reporter strain with ethylmethansulfonate (EMS). This was followed by several selection and triaging steps before the introduced mutations were finally mapped by whole-genome sequencing (Fig 1B).
Of roughly 280'000 EMS treated single cell derived colonies, about 700 colonies showed red or red/white variegating phenotypes. To select against loss-of-function mutations in the adenine biosynthesis pathway, these colonies were grown in the absence of adenine. Clones that were still able to grow were subsequently shifted to 37˚C. At this temperature, the repressed state of a heterochromatinized ade6 + reporter gene is reversed, i.e. white instead of red colonies are formed. At this step, we ended up with 103 colonies in which the trp1 + ::ade6 + reporter gene was silenced epigenetically (showing white and red phenotypes at 37˚C and 30˚C, Schematic representation of the tester strains (mst2 + or mst2 Δ ) used in this study. Primary ade6 siRNAs are produced from the nmt1 + locus on chromosome I (green). They can only induce silencing of the ade6 + reporter gene on chromosome II, if the tester strain acquires an enabling mutation (white color of the colonies shown indicates expression of the ade6 + reporter gene despite the presence of ade6 siRNAs in these cells). (B) Workflow of the EMS mutagenesis screen. For the initial screening, a mst2 Δ tester strain was used. Final hit validation was performed in respectively). After these were backcrossed four times, mutations that segregated with an ade6 + repression phenotype were mapped by whole-genome next-generation sequencing (S1 Table). Validating our screen, four clones had acquired mutations in subunits of the Paf1C complex. In another clone we found a nonsense mutation in the res2 + gene, which we had previously shown to display a weak silencing phenotype when deleted (Table 1) [18]. Thus, our screen reliably identifies mutants that enable small-RNA-mediated epigenetic gene silencing, even if initiation rates are poor.
To validate the mapped sequence alterations (S1 Table) as the causative mutations, and to test if they could also function independently of impaired Mst2 activity, we reconstituted the candidate point mutations in our original tester strain (mst2 + ) [18]. This revealed 20 novel silencing-enabling point mutations, which reliably recapitulated the ade6 + silencing phenotype ( Fig 1C). Consistent with small-RNA-mediated epigenetic silencing responses, the ade6 + repression phenotypes were reversible and depended on a functional dcr1 + allele in all 20 strains (S1 and S2 Figs). Eight of these novel enabling mutations were found in genes associated with RNA processing, four in genes encoding regulators of transcription, and three in genes that have been implicated in post-translational modification of histones. Another five mutations were found either in genes of unknown function or in genes related to lipids ( Table 2). In conclusion, our enhancer screen has identified 20 novel high confidence alleles that enable siRNAs to induce gene silencing in trans.

Arginine to cysteine substitution in the Yth1 cleavage/polyadenylation factor enables efficient initiation of heterochromatin-mediated gene silencing
Among all mutants tested, cells with an arginine to cysteine mutation at position 59 in the yth1 + gene (yth1-R59C) displayed the strongest silencing phenotype (Figs 1C, 2A and 2B). The Yth1 protein is a subunit of the cleavage and polyadenylation factor complex (CPF) and is responsible for the recognition of the AAUAAA polyadenylation signal in pre-mRNAs [29]. This is interesting because a previous study highlighted the importance of the PAS in preventing the formation of heterochromatin [27].
Stability of a heterochromatin-mediated silencing phenotype depends on the rate at which heterochromatin is established, or on the robustness of the mechanisms that preserve the silent chromatin state through mitosis, or both. For example, in Paf1C mutant cells, silencing is established rather inefficiently, but it is very stably propagated through subsequent cell divisions [18,20] (Fig 2C and 2D). In yth1-R59C cells, we observed an initiation rate of silencing that was close to 100% ( Fig 2C). Silencing was maintained stably but appeared more mst2 + cells. (C) Representative images of validated point-mutations that are sufficient to enable RNAi-directed ade6 + silencing, which is indicated by the red color. Colonies showing a silencing phenotype on YE-NAT (100ug/ml nourseothricin) plates were selected and spread on YE plates to monitor stability of ade6 + repression. Cells are mst2 + but harbor mutations in individual genes as indicated. variegating than the silencing phenotype observed in Paf1C mutants (Fig 2A and 2D). Furthermore, we also observed silencing in yth1-R59C cells that express synthetic ura4-hp siRNAs instead of ade6-hp siRNAs, and a trp1 + ::ura4 + instead of a trp1 + ::ade6 + reporter (S3A Fig).
Thus, we conclude that the yth1-R59C allele enables trans-acting siRNAs to effectively initiate the formation of heterochromatin at their target locus. The silencing phenotypes described above, together with its temperature-sensitivity, strongly imply RNAi-mediated heterochromatin formation at the target locus. To formally demonstrate this, we assessed initiation of silencing of the trp1 + ::ade6 + reporter gene in yth1-R59C cells lacking either a functional dcr1 + gene or the primary siRNA producing nmt1 + ::ade6-hp + locus. We observed silencing in neither of these strains (Fig 2B), demonstrating the necessity of siRNA biogenesis. To confirm the formation of a heterochromatic structure at the trp1 + ::ade6 + locus, we performed chromatin immunoprecipitation (ChIP) experiments using an antibody specifically recognizing di-methylated lysine 9 on the N-terminal tail of histone H3 (H3K9me2). As predicted, the H3K9me2 mark was significantly enriched in yth1-R59C cells. yth1 + cells were not different from cells lacking Clr4, which is the sole H3K9 methyltransferase in S. pombe ( Fig 2E). Finally, assembly of heterochromatin at the trp1 + ::ade6 + reporter gene was accompanied by the production of secondary ade6 + siRNAs that are not encoded in the ade6-hp ( Fig 2F).
These results demonstrate that replacing arginine at position 59 of Yth1 with a cysteine does not critically affect expression of the trp1 + ::ade6 + reporter gene ( Fig 2B). However, it enables highly efficient initiation of heterochromatin-mediated gene silencing upon expression of primary siRNAs that are complementary to the ade6 + pre-mRNA (S3B and S3C Fig).

PAS recognition controls small-RNA-mediated epigenetic gene silencing
CPF is a large multisubunit protein complex possessing ATP-polynucleotide adenylyltransferase, phosphatase, and nuclease activities that are required for the cleavage and polyadenylation of pre-mRNA transcripts [29]. Yth1 is part of the poly(A)polymerase module of CPF and is essential for cellular viability [30]. Because 3' end processing of pre-mRNAs has previously been implicated in the control of RNA silencing pathways in yeast, flies, and plants [18, 27,  . Multiple individual originator colonies (white or red in C or D, respectively) were spread to single cell density on YE plates to assess initiation/maintenance of the silencing phenotype: n = 6 for paf1-Q264Stop white originator (C, total counted number of colonies = 2547), n = 6 for paf1-Q264Stop red originator (D, total counted number of colonies = 3016), n = 12 for yth1-R59C white originator (C, total counted number of colonies = 4063), n = 9 for yth1-R59C red originator (D, total counted number of colonies = 9468). (E) ChIP analysis of H3K9me2 in the strains indicated (trp1 + ::ade6 + , nmt1 + ::ade6-hp + ). Fold enrichments were normalized to adh1 + and are shown relative to background levels measured in clr4 Δ cells. Error bars indicate standard deviation, n = 3 independent biological replicates, p-values were calculated with a two-tailed Student's t-test. The ade6 + primer pairs do not discriminate between endogenous ade6-704 and trp1 + ::ade6 + genes. (F) Small RNA sequencing was performed with yth1 + and yth1-R59C cells to assess secondary ade6 + siRNA production. Read counts were normalized to library size. The part of ade6 + that is complementary to the primary siRNAs encoded by the ade6-hp is denoted.
Because protein structures of S. pombe CPF have not yet been determined, we selected the structure of human CPSF-30 (homolog of S. pombe Yth1) in complex with CPSF-160 (homolog of S. pombe Cft1), WDR33 (homolog of S. pombe Pfs2), and PAS RNA as a homology model to infer the functional consequences of the enabling mutations that we have identified (PDB ID 6DNH) [37]. This revealed that mutations in E73 and Y74 are likely to disturb the Yth1-Cft1 interaction and the C91R mutation the Yth1-Pfs2 interaction, whereas the other mutations are predicted to weaken the interaction with the PAS in the pre-mRNA (Fig 3D). Interestingly, K55, R59, C91, and Y99 contribute to the binding of Yth1 to the adenosine at position 4 (A 4 ) of the PAS, suggesting that an adenosine at position 4 is critical for the prevention of silencing. Indeed, mutating A 4 in the trp1 + ::ade6 + reporter to either U, C, or G enabled siRNAs to initiate silencing (Figs 3E and S3B). Further supporting the importance of PAS recognition, we observed siRNA-dependent trp1 + ::ade6 + reporter silencing upon mutation of F115 in Pfs2 (Figs 3F and S3D), which stacks with A 6 of the PAS (Fig 3D).
Interestingly, initiation frequency of the silencing response was lower in pfs2-F115H cells than observed with A 4 PAS mutants (Fig 3E and 3G). Yet, once established the silent state was remarkably stably maintained (Fig 3G). We note that residue C91 is part of the second CCCH zinc finger motif in Yth1. Correct folding of this zinc finger is important for binding A 4 /A 5 in the PAS as well as for the interaction of Yth1 with Pfs2 ( Fig 3D). Thus, Yth1-C91 mutations are likely to stimulate both initiation and maintenance of silencing, providing a potential explanation why we have mapped C91 five times in this screen.

Deletion of non-essential subunits of the phosphatase module of the cleavage and polyadenylation factor complex enables small-RNA-mediated epigenetic gene silencing
The foregoing results implicate the fission yeast CPF in the control of small-RNA-mediated epigenetic gene silencing. Because we have mapped enabling mutations in the poly(A) polymerase-module of CPF only, we asked whether mutations in the other modules would similarly enable silencing. Unfortunately, the genes encoding subunits of the nuclease module are  [30], preventing us from testing those. However, we were able to delete the dis2 + , ppn1 + , swd22 + , and ssu72 + genes, which encode subunits of the phosphatase module ( Fig 4A). Like the poly(A) polymerase module mutants, the four phosphatase module knockout strains enabled small-RNA-mediated gene silencing (Figs 4B and 4C and S4A-S4C). Interestingly, Swd22 and Ssu72 have recently been shown to be required for RNAi-independent assembly of facultative heterochromatin [32]. Thus, the phosphatase module of CPF may have opposing roles depending on the pathway that leads to H3K9 methylation.
A remarkable feature of CPF mutants that we have investigated in this study is that they by and large phenocopy wild-type cells, i.e. neither growth nor global gene expression is largely affected (S5 Fig). For example, we did not observe any major differences in steady-state ade6 + mRNA levels in CPF or PAS mutant cells in the absence of ade6 siRNAs (Figs 5A and 5B and S5A). Also, polyadenylation of ade6 + mRNA in wild-type and mutant cells was indistinguishable in our assay (Fig 5B). However, we observed compromised cleavage of the ade6 + pre-mRNA, which was prominent in many of the mutants investigated (Fig 5A upper panel and  5C). This strongly implies reduced kinetics of the 3' end processing reaction.
In conclusion, our results are consistent with previous works that have implicated pre-mRNA processing factors in small-RNA-mediated silencing responses [15,18,31]. Our detailed analyses of S. pombe CPF mutants reinforce the importance of an efficient 3' end processing reaction to avoid an unwanted gene silencing response.

Discussion
In this study we have identified novel mutant alleles that make S. pombe susceptible for RNAimediated de novo assembly of silent chromatin. Though we have focussed on a functional dissection of mutations in the pre-mRNA cleavage and polyadenylation machinery in this study (Fig 4A), it will be equally exciting to dissect the role of the other RNA processing factors that our screen has revealed. Likewise, further investigating the many alleles linked to chromatin biology or transcription regulation promises to further improve our understanding of the intricate mechanisms that keep RNA-mediated epigenetic processes in check ( Table 2).
Our work on CPF presented here is consistent with our previously proposed kinetic model for the inhibition of de novo formation of heterochromatin that is mediated by trans-acting primary siRNAs. In this model, the rate at which the nascent transcript is released from the DNA template is predicted to constitute a rate limiting step for the initiation of heterochromatin assembly and eventually gene silencing [18]. We find it striking that ade6 + pre-mRNA cleavage in yth1-R59C cells is similarly affected as in cells harbouring ade6 + genes with A 4 mutated PASs (Fig 4B and 4C). As predicted by the kinetic model, mutations that we have mapped in CPF are thus likely to result in decelerated cleavage and release of the nascent transcript from the site of transcription, opening up a window of opportunity for the siRNAguided RITS complex to base-pair with pre-mRNA and recruit the histone methylation machinery. Such a model nicely explains why the initiation rates that we have observed in yth1-R59C cells are so remarkably high (Fig 2C).
Although we could not investigate factors of the CPF nuclease module, we deleted four genes that are encoding subunits of the phosphatase module. While to seemingly various degrees, all four mutants enabled silencing. This is intriguing because Ssu72 and Dis2 are active phosphatases, letting us to speculate that inhibition of RNAi-mediated heterochromatin formation could be regulated by kinase signalling pathways. This is of particular interest in light of a recent report that described the isolation of heterochromatin-dependent epimutants that are resistant to caffeine, which is abolished in RNAi mutants [38]. A similar phenomenon had been described earlier in Mucor circinelloides, which can cause deadly fungal infections in humans. Similar to S. pombe responding to low doses of caffeine, M. circinelloides can become resistant against antifungal drugs through RNAi-mediated epigenetic gene silencing [39,40]. It is tempting to speculate that sensing of toxic substances in the environment could be  Error bars indicate standard deviation, n = 3 or 4 independent biological replicates for lys4 + or ade6 + , respectively. (A -C) cDNA was prepared from RNA that was isolated from cells with the indicated genotype. These cells did not express ade6 + siRNAs.
https://doi.org/10.1371/journal.pgen.1009645.g005 signalled to the RNAi-inhibiting modules that we have identified in our works. Potential modulation of such signalling by the CPF phosphatases is an attractive hypothesis that will be worthwhile further investigation.
Although regulation of RNAi-mediated epigenetic gene silencing through direct environmental sensing is appealing, we do not exclude the possibility that RNAi-mediated de novo formation of heterochromatin on protein-coding genes strictly depends on the acquisition of an enabling mutation. In this model, an epigenetic gene silencing response would always be preceded by a genetic change. This is supported by the many different enabling genetic mutations that we and others have identified so far [15,18,20], and by our unsuccessful efforts to trigger small-RNA-mediated gene silencing under various environmental conditions in wild-type cells. Thus, we urge the community to consider the possibility that acquisition of a genetic mutation had preceded the establishment of the observed siRNA-triggered epigenetic silencing phenotype, especially when "on" and "off" expression states segregate with a 2:2 Mendelian ratio in seemingly wild-type cells [15]. As we have already discussed elsewhere, prior acquisition of RNAi-enabling genetic mutations could also explain why virulent isolates of M. circinelloides have an enhanced ability to develop drug resistance through epimutations [25].
The latter model would be fully consistent with the concept of biological bet-hedging, as enabling stochastic RNAi-mediated epigenetic silencing might help the microbe to adapt to an ever-changing environment [25]. Therefore, it could be advantageous for a yeast to hedge its bets by acquiring an enabling mutation in case its environment keeps changing. Because this comes along with decreased fitness in stable conditions, such mutations would be expected to disappear in a laboratory strain. This may explain why S. pombe cells that we are growing in our labs are refractory to RNAi-mediated gene silencing.

Yeast strains
S. pombe strains were generated following a PCR-based protocol [41] or by standard mating and sporulation. For a list of strains generated in this study see S2 Table. As a general procedure to validate the newly identified enabling alleles, the mapped mutations were introduced in the mst2 + tester strain by transformation of the mutated ORF, which was marked with URA3 from Candida albicans. After successful integration, URA3 was removed by FOA counter selection. To mutate dis32 + , the ORF was deleted with URA3, which was subsequently replaced by transformation of the mutated dis32 + ORF and FOA counter selection. The SPBPB21E7.10-Q2Stop early ORF truncation was generated by the insertion of an hphMX cassette.

Plasmids
Plasmids were cloned by standard molecular biology techniques. For a list of plasmids generated in this study see S3 Table. The plasmid expressing yth1 + and ura4 + (pMB1869) was constructed by cloning yth1 + , including 978bp upstream and 530bp downstream sequence, into KpnI/PstI digested pFY20, which was a kind gift from Mari K. Davidson. To generate the hph + marked plasmids, the ura4 + marker of pFY20 was first replaced with hph + (pMB1867).

Whole-genome sequencing
Genomic DNA was isolated from overnight cultures using the MasterPure yeast DNA isolation kit (Epicentre). Genomic DNA libraries for next-generation-sequencing were prepared from 50ng of sonicated DNA, using the NEBnext Ultra kit (NEB) following the manufacturer's protocol. Libraries were sequenced 50bp single-end on the Illumina HiSeq2500 platform. Basecalling and quality scoring was performed using RTA v1.18.64, and demultiplexing using bcl2fastq2 v2.17 (Illumina). For SNP calling we adapted a previously described pipeline [18]: For each strain, between 6.4 and 17.6 million (mean of 11.5 million) 50-nucleotide reads were generated and aligned to the Schizosaccharomyces pombe 972h-genome assembly (obtained on 17 September 2008 from http://www.broad.mit.edu/annotation/genome/schizosaccharo-myces_group/MultiDownloads.html) using 'bwa' (version 0.7.15) with default parameters, but only retaining single-hit alignments ('bwa samse -n 1' and selecting alignments with 'X0:i:1'), resulting in a genome coverage between 26 and 71-fold (mean of 47-fold). The alignments were converted to BAM format, sorted and indexed using 'samtools' (version 1.3.1). Potential PCR duplicates were removed using 'MarkDuplicates' from 'Picard' (http://picard. sourceforge.net/, version 2.7.1). Sequence variants were identified using GATK (version 3.6) indel realignment and base quality score recalibration. A set of high confidence variants was identified in an initial step as known variants, followed by single nucleotide polymorphism (SNP) and INDEL discovery and genotyping for each individual strain using standard hard filtering parameters, resulting in a total of 14-103 sequence variants (mean of 68) in each strain compared to the reference genome. Finally, variants were filtered to retain only high quality single nucleotide variants (QUAL > = 50) of EMS type (G|C to A|T) with an allelic balance > = 0.9 (homozygous) that were not also identified in the parental strain (sms0), reducing the number of variants per strain to a number between 1 and 8 (mean of 3.6).

Selection of silencing-enabling yth1 mutants
To rescue growth of yth1Δ S. pombe cells, they were first transformed with an yth1 + expressing plasmid (pMB1869) before the endogenous yth1 + gene was deleted with a kanMX cassette, resulting in the strain SPB3646. SPB3646 was transformed with the mutant p-yth1 � /hph + library and grown on YES plates for 2 days at 30˚C. Cells were then replica plated on YE plates supplemented with 0.226g/l leucine, 0.226g/l lysine, 0.226g/l histidine, 0.226g/l uracil, 2g/l FOA and 100mg/l Hygromycin B (YE4S+FOA+Hygromycin B). Plasmids were recovered from yeast colonies with a silencing phenotype and subsequently sequenced by Sangersequencing. To confirm the silencing phenotype, the isolated plasmids were transformed in SPB3646, followed by growth on YE4S+FOA+Hygromycin B plates to force loss of the yth1 + rescue plasmid and to score for colony color.

Silencing assays
To assess ade6 + expression, serial five-fold dilutions of the respective strains were plated on yeast extract (YE) plates and incubated at 30˚C for 3-4 days. Plates were stored at 4˚C overnight before pictures were taken.
To quantify initiation and maintenance rates of silencing, either single-cell-derived white (for initiation) or red (for maintenance) colonies were selected from a YE-NAT plate. A single colony was resuspended in H 2 O and 50-500 cells were seeded on YE plates, which were incubated at 30˚C for 4 days. Colonies were categorized and counted, after an additional overnight incubation at 4˚C, using a deep learning pipeline for high-throughput colony segmentation and classification [28].

Chromatin immunoprecipitation (ChIP)
ChIP experiments were performed as described previously [18] with a histone H3K9me2-specific mouse monoclonal antibody from Wako (clone no. MABI0307).

RNA isolation and cDNA synthesis
Total RNA was isolated using the MasterPure Yeast RNA Purification Kit (Epicentre). cDNA was synthesized using the PrimeScript RT Master Mix (Takara).

Small and total RNA sequencing and analysis
Small RNA libraries were prepared with the QIAseq miRNA Library Kit (QIAGEN, Cat. No: 331505) according to the manufacturer's instructions and sequenced with an Illumina Next-Seq500 (75bp single-end). 3' adapter sequences were trimmed using cutadapt [42](version 1.18) (cutadapt -a 'adapter'-discard-untrimmed -m 18) and untrimmed or <18nt long reads were discarded. The remaining reads were aligned to the S. pombe genome (ASM294 version 2.24) using bowtie [43] (version 1.2.2) (bowtie -f -M 10000 -v 0 -S-best-strata). Displayed are UCSC genome browser [44] tracks of uniquely mapped reads that are normalized to one million reads (RPMs).
Small RNA and total RNA sequencing data have been deposited at the NCBI Gene Expression Omnibus (GEO) database and are accessible through GEO series number GSE173837.

Quantitative real-time PCR
Real-time PCR on cDNA samples and chromatin immunoprecipitation (ChIP) DNA was performed as described using a Bio-Rad CFX96 real-time system using SsoAdvanced SYBR Green supermix (Bio-Rad) [48]. For primer sequences see S4 Table. Qualitative RT-PCR PCR on cDNA was performed using the fast-cycling PCR kit (Qiagen). PCR products were analyzed by agarose gel electrophoresis. Primer sequences are listed in S4 Table. 3' RACE 3' RACE to assess PolyA tail length was performed as described elsewhere [49]. Briefly, first strand cDNA was synthesized from total RNA with an anchored oligo-d(T) 17 VN primer using ProtScript reverse transcriptase (NEB). A first round of amplification was performed with primers mb7234 and mb13037. A second round of amplification was performed with primers mb135 and mb13038.
This suggested adequate structural conservation that allowed the analysis of mutations in S. pombe in the context of the human complex. A model for these point mutations was generated by mutating mapped residues in HsCPSF-30 (6DNH) using PyMOL (The PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC). RNAi-directed silencing in yth1-R59C cells is not unique to the trp1 + ::ade6 + silencing reporter. The yth1-R59C mutation was introduced into cells that express synthetic ura4-hp siRNAs instead of ade6-hp siRNAs, and a trp1 + ::ura4 + instead of a trp1 + ::ade6 + reporter. ura4DS/E denotes a partial deletion of the endogenous ura4 + gene. Silencing of the ura4 + reporter was assessed by growth in the presence or absence of 5-FOA (which is toxic to ura4 + expressing cells). Note that 5-FOA resistant colonies did only form in the presence of ura4-hp siRNAs and simultaneous mutation of yth1 + . (B) Silencing assays showing the degree of ade6 + silencing in cells harboring mutations in yth1 + (yth1-R59C) or at the 4 th position in the PAS of the ade6 + reporter gene. See Fig 3E for a quantification of the initiation of silencing rates in the PAS mutants. (C) Silencing assay demonstrating that trp1 + ::ade6 + silencing in yth1-R59C cells depends on RNAi (dcr1 Δ , w/o ade6-hairpin) and H3K9 methylation (clr4 Δ ). (D) Silencing assay demonstrating that trp1 + ::ade6 + silencing in pfs2-F115H cells depends on RNAi (dcr1 Δ , w/o ade6-hairpin) and H3K9 methylation (clr4 Δ ). (TIF)

S5 Fig. Steady-state mRNA levels remain largely unaffected in cells harboring RNAienabling CPF mutations. (A)
Quantitative RT-PCR with primer pairs amplifying ade6 + or act1 + mRNAs in the respective mutant strains, which do not express any ade6 + siRNA. mRNA levels were normalized to U6snRNA and are shown relative to the levels measured in wildtype cells. Error bars indicate standard deviation, n = 3 independent biological replicates. (B) Pairwise comparisons of gene expression (RNA-seq) between wild-type and CPF or trp1 + :: ade6 + PAS mutant strains. Cells did not express primary ade6-hairpin siRNAs. Fragments per kilobase million (FPKMs) were calculated and averaged over three replicates. Each scatterplot depicts log2-transformed FPKM values of the wild-type against one of the mutants. (TIF) S1