Raf1 Is a DCAF for the Rik1 DDB1-Like Protein and Has Separable Roles in siRNA Generation and Chromatin Modification

Non-coding transcription can trigger histone post-translational modifications forming specialized chromatin. In fission yeast, heterochromatin formation requires RNAi and the histone H3K9 methyltransferase complex CLRC, composed of Clr4, Raf1, Raf2, Cul4, and Rik1. CLRC mediates H3K9 methylation and siRNA production; it also displays E3-ubiquitin ligase activity in vitro. DCAFs act as substrate receptors for E3 ligases and may couple ubiquitination with histone methylation. Here, structural alignment and mutation of signature WDxR motifs in Raf1 indicate that it is a DCAF for CLRC. We demonstrate that Raf1 promotes H3K9 methylation and siRNA amplification via two distinct, separable functions. The association of the DCAF Raf1 with Cul4-Rik1 is critical for H3K9 methylation, but dispensable for processing of centromeric transcripts into siRNAs. Thus the association of a DCAF, Raf1, with its adaptor, Rik1, is required for histone methylation and to allow RNAi to signal to chromatin.


Introduction
Silencing mechanisms mediated by small RNAs occur in most eukaryotes. RNA interference (RNAi) can reduce gene expression post-transcriptionally by cleaving homologous transcripts or by inhibiting their translation [1]. Small RNAs can also act in the nucleus, inducing transcriptional gene silencing (TGS) [2,3]. Although siRNA-directed modification of homologous chromatin is widespread in eukaryotes, the details of how siRNAs mediate such events remain limited in most systems. [4,5]. In plants and fungi, the link between siRNA-directed DNA/chromatin modification and heterochromatin formation is well established [2,3]. The fission yeast, Schizosaccharomyces pombe, is a powerful system in which to study RNAi-directed heterochromatin formation in part because it contains single non-essential genes encoding each of the key RNAi components.
In fission yeast, siRNAs are important for heterochromatin formation on the centromeric outer repeats (composed of dg and dh elements) and other chromosomal locations. Although marker genes inserted at these loci are silenced [6], centromeric repeats are bi-directionally transcribed by RNAPII, producing doublestranded RNA (dsRNA) [2,3,7,8,9]. Dicer (Dcr1) cleaves these non-coding transcripts into siRNAs that guide the Argonaute (Ago1)-containing RITS complex to homologous nascent transcripts by sequence complementarity. Chromatin-bound-RITS recruits the histone methyltranferase Clr4 SuVar3-9 (a CLRC component; see below) to centromeric repeats via the linker protein Stc1 [10]. Clr4 methylates lysine 9 of histone H3 (H3K9me), providing binding sites for the chromodomain proteins Swi6, Chp1 (a RITS component) and Chp2 [11,12,13,14]. Two different non-mutually exclusive models have been proposed to explain how heterochromatin, once assembled, silences centromere repeat transcription. In the 'transcriptional gene silencing' (TGS) model, heterochromatin factors directly repress transcription [15]. In the 'co-transcriptional gene silencing' (CTGS) model the repeats are continuously transcribed and silencing is due to the efficient processing of transcripts to siRNA [16]. In wild-type cells, siRNA production and H3K9 methylation are coupled processes: deletion of genes encoding CLRC components results not only in loss of H3K9 methylation but also in loss of siRNA [17,18,19]. Similarly, deletion of genes encoding RNAi components abrogates siRNA production and reduces H3K9me [9]. However, cells expressing only mutant histone H3 (H3K9R) produce some detectable siRNAs, even though H3K9 can not be methylated [20,21]. Moreover, tethering the Rik1 CLRC component to ura4 RNA triggers silencing of the ura4 + gene independently of other CLRC subunits, but requires RNAi [21]. Thus CLRC itself, rather than its substrate H3K9, promotes siRNA production independently of H3K9. However, it is not known how CLRC integrates histone methylation with siRNA generation.
The Cullin-RING ligases (CRLs) are the largest family of multisubunit E3 ligases in eukaryotes and are formed by a neddylated member of the cullin family scaffold (CUL1-CUL5), a RING finger protein (RBX1 or RBX2) and an adaptor protein that bridges the CUL/RING scaffold to substrate receptors [28]. Amongst CRLs, CRL4 is known to ubiquitinate histones [29,30]. In the CRL4 complex, DDB1 acts as the adaptor. Affinity purification of DDB1 from human cells identified various WD-40 proteins as possible substrate receptors termed DCAFs (DDB1 and CUL4 Associated Factors) [31,32,33,34]. Several DCAFs (WDR5, RBP5 and EED) not only interact with DDB1 but are also members of histone methyltransferase complexes [31,32,33,34]. Although it is not known whether the interaction between these DCAFs and CUL4/DDB1 leads to ubiquitination of substrates, it might have important functional consequences since the knockdown of CUL4 or DDB1 reduces histone methylation [31]. Many DCAFs contain a specific WDxR motif that is important for their association with DDB1 [31,32,33,34].
In S. pombe, Cul4 associates with the canonical adaptor Ddb1 [35] and with Rik1 (which shares similarity with Ddb1; [36]) in CLRC [17,22,23,24]. Both Cul4-Ddb1 complex and CLRC affect heterochromatin integrity and neddylation of Cul4 is required for H3K9 methylation of heterochromatic regions [22,37]. This suggests that E3 ligase activity is involved in heterochromatin formation. However, heterochromatin defects observed in neddylation-defective Cul4 (cul4-K680R) could arise either from impaired Cul4-Ddb1 activity or from defective CLRC function. Furthermore, although Rik1 exhibits some similarity to Ddb1 and thus might act as an adaptor, recent analyses suggest that Rik1 is primarily an RNA binding protein which associates with centromeric transcripts via its CPSF-A domain [21]. Thus, it remains to be determined how CLRC contributes to heterochromatin integrity and whether CLRC acts as a DDB1-related E3 ligase which promotes heterochromatin formation in vivo.
Here we demonstrate that the CLRC components Rik1 and Raf1 can be structurally aligned with the human adaptor DDB1 and its DCAF DDB2, respectively. Our analyses are consistent with Raf1 acting as a DCAF for CLRC that contributes to siRNA amplification and H3K9 methylation by two distinct and separable routes. Our findings provide mechanistic insights into how siRNA production is integrated with chromatin modification via CLRC.

Raf1 is a chromatin associated component of the CLRC complex
Components involved in heterochromatin formation are typically localised in distinct chromatin-associated foci [14,38]. In contrast, previous studies overexpressing GFP-Raf1 from a strong promoter showed that it is a nuclear protein with no obvious chromatin localisation [23], however, this might not reflect its true subcellular localisation. Indeed, the genome-wide distribution of Raf1 suggests that it interacts predominantly with heterochromatic loci [13]. To reassess Raf1 localisation, we examined cells expressing GFP-Raf1 from its native promoter and found that it decorates several distinct chromatin-associated foci, the largest of which lies adjacent to the clustered kinetochores ( Figure 1A). ChIP confirmed that FLAG-Raf1 associates with centromeric repeats ( Figure 1B). We conclude that Raf1 is a chromatin-associated protein concentrated at heterochromatin.
Biochemical purification of CLRC components Rik1, Raf2 and Clr4 identified Raf1 as an interacting partner [17,22,24]. In contrast, two-step purification of TAP-Raf1 identified the histone demethylase Lid2 plus Cul4 and Rik1, but apparently not Raf2 and Clr4, known CLRC subunits [39]. This raises the possibility that Raf1 is a component of two distinct complexes: CLRC (Cul4/ Rik1/Raf1/Raf2/Clr4) and an alternative Lid2/Cul4/Rik1/Raf1 complex. To further investigate this, we affinity selected FLAG-Raf1 from cell lysates and identified associated proteins by mass spectrometry. Silencing assays indicate that FLAG-tagged Raf1 is functional ( Figure S1A). Single step FLAG affinity purification is less stringent than the two-step TAP affinity purification and was performed in mild conditions to identify as many Raf1 interacting proteins as possible. Although we detect all CLRC components, Lid2 peptides were absent ( Figure 1C). Moreover, while Rik1 coimmunoprecipitated with Raf1 (co-IP; Figure 1D), Rik1 could not be detected in Lid2-TAP IPs ( Figure S1B). We conclude that Raf1 mainly associates with the known CLRC components.

Author Summary
Heterochromatin is a specialized form of chromatin which is frequently assembled on DNA sequences with little or no coding potential. Heterochromatin formation involves specific post-translational modifications of histone tails (e.g. methylation of histone H3 on lysine 9). In fission yeast, Schizosaccharomyces pombe, heterochromatin is found at centromeres, telomeres, and the mating type locus. Heterochromatin integrity at centromeres is important for normal chromosome segregation. The heterochromatin associated repeats at fission yeast centromeres are known to be transcribed, and these non-coding transcripts are processed into siRNAs. siRNA production is required for establishment and maintenance of H3K9 methylation. But H3K9 methylation itself is required for siRNA production. It is not known how these two processes are coupled. In this study we use structural modelling and genetic analyses to demonstrate that the heterochromatin component Raf1 plays an essential role in coupling H3K9 methylation and siRNA production. Our analyses show that the heterochromatin factors Rik1 and Raf1 can be structurally aligned with Cul4-E3 ubiquitin ligase components DDB1 and DDB2, respectively. We show that specific mutations impair the association of Raf1 with Rik1 and prevent H3K9 methylation but not siRNA production. These functional studies provide mechanistic insights into how siRNA production and chromatin modification are integrated.
might represent the adaptor and the DCAF, respectively, for CLRC. To test these possibilities, we performed structural alignments of Rik1 and Raf1 and found that they can be modelled on DDB1 and on the DCAF DDB2 (Rik1: DOPE score 211385; Raf1: GA431 score 0.6 respectively) (Figure 2A and 2B). This model reveals that, like human DDB1, Rik1 is composed of three WD-40 b-propeller domains (BPA, BPB and BPC) composed of 216 WD-40 repeats and a C-terminal helical domain ( Figure 2B and S2A). Raf1, like DDB2, is predicted to contain an N-terminal helical domain and a 7-bladed WD-40 b-propeller domain forming a ring (Figure 2A and 2B). Unlike DDB2, the helical and WD-40 domains of Raf1 are separated by an additional domain of unknown function ( Figure S2B). As in the DDB1/ DDB2 complex, our model predicts that the Raf1 helical domain can be inserted into the Rik1 BPA-BPC cleft and the 'top' of the bpropeller ring interacts with the 'bottom' surface of Rik1 ( Figure 2B).
Alignment of the Raf1 WD-40 repeats with those of other known DCAFs revealed that Raf1 contains two WDxR motifs (1: aa515-518 and 2: aa573-576) ( Figure S3A). WDxR motifs represent a 'DCAF signature' and are important for docking DCAFs to DDB1 [31,32,33,34]. Our Raf1 model indicates that these two WDxR motifs are located on the surface of the bpropeller ring and might provide specific sites for interactions with other factors ( Figure S3E). To test the importance of Raf1 WDxR motifs, we mutated the endogenous raf1 gene in S. pombe to express FLAG-Raf1-R518A or FLAG-Raf1-R576A. Western analysis indicated that both mutant proteins are expressed at levels similar to wild-type FLAG-Raf1 ( Figure S3B). All components of CLRC are required for heterochromatin integrity and consequently for the transcriptional silencing of marker genes placed within centromeric heterochromatin, and centromere function [6]. Like clr4D cells, heterochromatin-mediated silencing of cen1:ade6 + is disrupted in raf1-R518A and raf1-R576A mutants, as indicated by white/expressing rather than red/silent colonies ( Figure 2C). Moreover, both mutations impair heterochromatin dependent silencing at the mating type locus ( Figure S3D).
Centromeric heterochromatin mediates robust sister-centromere cohesion and is therefore required for accurate chromosome segregation during mitosis [40,41]. Defective heterochromatin causes a quantifiable increase in the frequency of lagging chromosomes on late anaphase spindles [42]. Cells bearing the raf1-R518A or raf1-R576A mutation exhibit a frequency of lagging chromosomes equivalent to clr4D cells, indicating that this centromeric function of heterochromatin is disrupted ( Figure 2D).
Heterochromatin formation requires processing of centromeric transcripts into siRNA and methylation of H3K9 (H3K9me). Deletion of any gene encoding a CLRC component results in loss of H3K9me, accumulation of centromeric transcripts and a dramatic reduction of siRNA levels [17,18,19]. ChIP analyses indicate that H3K9me2 is reduced to background levels in raf1-R518A and raf1-R576A cells ( Figure 2E). Importantly, Clr4 levels are equivalent to wild-type in both mutants ( Figure S3C). Consistent with this loss of H3K9me2 and as observed in clr4D cells, high levels of centromeric transcripts also accumulate ( Figure 2F). This is due to increased transcription given that higher levels of RNAPII are detected on centromeric repeats in both mutants compared to wild-type cells ( Figure S5E). However, raf1-R576A cells contain low levels of centromeric siRNAs whereas raf1-R518A cells retain high levels of these siRNAs ( Figure 2G and S3F). Our structural model predicts that R518 resides on the surface of Raf1 that interacts with Rik1 and predicts that the R518A mutation may specifically impair Raf1-Rik1 interactions ( Figure S3G). In contrast, R576 is more distantly located from Rik1 and may have more profound structural effects.
Together our findings demonstrate that the CLRC complex is likely to adopt a canonical Cul4-E3 ligase architecture and that Raf1 is a DCAF for CLRC. Moreover, we show that WDxR motifs are required for heterochromatin integrity and that the raf1-R518A mutation separates the function of CLRC in methylating H3K9 from its role in generating siRNAs.
The raf1-1 mutation conditionally disrupts heterochromatin without affecting siRNA generation To dissect mechanisms governing heterochromatin assembly and maintenance, we screened for temperature-sensitive mutants that disrupt heterochromatin at the restrictive temperature (36uC) but not the permissive temperature (25uC) and isolated the raf1-1 mutation. raf1-1 produces a protein with a missense mutation (T495I) in the third WD-40 repeat; this residue is conserved in fungi suggesting that it is important for Raf1 function ( Figure  S3A). Similar levels of mutant FLAG-Raf1-1 and wild-type FLAG-Raf1 proteins are detected at 36uC thus phenotypes are not due to Raf1-1 degradation ( Figure S4A). Marker gene silencing within centromeric repeats (cen1:ade6 + or cen1:ura4 + ), the mating type locus (mat3-M:ura4 + ) and at telomeres (tel1L:his3 + ) is alleviated in raf1-1 cells at 36uC, but not 25uC ( Figure 3A and Figure S4B, S4C). Consistent with defective centromeric heterochromatin integrity, a high frequency of lagging chromosomes is observed in late anaphase raf1-1 cells at 36uC ( Figure 3B). Moreover, H3K9me2 levels on centromeric repeats in raf1-1 cells are similar to wild-type cells at 25uC but negligible at 36uC ( Figure 3C). Consistent with reduced H3K9me2, Swi6 localisation and unprocessed centromere repeat transcript accumulation is temperature dependent in raf1-1 cells ( Figure 4A and 4B). Importantly, Clr4 levels are unaffected by the raf1-1 mutation ( Figure S4B).
Although we detect no H3K9me2 on centromere repeats in raf1-1 at 36uC, it is possible that residual levels of H3K9me2 remain. To determine if heterochromatin is erased at 36uC, a heterochromatin establishment assay was performed. At the mating type locus, active RNAi is required to establish heterochromatin, but once assembled, RNAi is not required to maintain heterochromatin [43]. Hence, loss of dcr1/RNAi has no impact on mating type locus silencing (i.e. mat3-M:ura4 + ). However, when presented with a naive template, dcr1D cells fail to assemble heterochromatin de novo and are unable to silence mat3-M:ura4 + . To test whether the raf1-1 mutant erases heterochromatin structures, a raf1-1 dcr1D double mutant was generated at 36uC and mat3-M:ura4 + silencing assessed after shifting cells to 25uC. Silencing of mat3-M:ura4 + could not be established in this double mutant following a shift down from 36u to 25uC, but remained intact in the raf1-1 dcr1D double mutant generated at 25uC ( Figure 4C). This demonstrates that raf1-1 cells are unable to form heterochromatin without RNAi and indicates that heterochromatin is completely erased by the raf1-1 mutation at 36uC.
Surprisingly, unlike clr4D and raf1D mutants, centromeric siRNAs are produced at wild-type levels in raf1-1 cells at both temperatures ( Figure 4D, Figure S4E and S5D). It is possible that siRNA levels remain high because raf1-1 inhibits the degradation of pre-existing siRNAs. However, high siRNA levels remain in raf1-1 cells which have undergone 144 divisions at 36uC; preexisting siRNA would be diluted out ( Figure S4E). Thus, the continual synthesis of siRNAs from centromere repeat transcripts must be unaffected by the defect in raf1-1. Moreover, these siRNAs are loaded into Ago1/RITS as indicated by their association with FLAG-Ago1 ( Figure 4E). We conclude that as with raf1-R518A, the raf1-1 mutation uncouples siRNA production from H3K9 methylation.

Methylation of H3K9, but not siRNA production, depends on an intact CLRC complex
The interaction between the adaptor DDB1 and its DCAFs is known to require intact WDxR motifs in the DCAF [32,33,34]. Our structural alignment predicts that the raf1-1 T495I mutation is located on the 'top' surface of the WD-40 ring that interacts with Rik1 ( Figure S5A). It is therefore possible that our specific raf1 mutations impair the ability of Raf1 to associate with Rik1 and that this interaction is essential for establishing and maintaining H3K9 methylation on heterochromatic loci. Indeed two-hybrid assays reveal that the binding of Raf1 to Rik1 [23] is disrupted in both WDxR motif mutants and the raf1-1 mutation ( Figure 5A). All three mutations also impair co-immunoprecipitation of Rik1 with Raf1 from S. pombe extracts ( Figure 5B). In addition, the association of the Clr4 H3K9 methyltransferase with Raf1 and Rik1 is lost in raf1-R518A, raf1-R576A and raf1-1 mutants ( Figure 5C and Figure S5B). Recently we identified Stc1 as protein that links the RNAi machinery to CLRC [10]. Consistent with this, these same three specific raf1 mutations abolish the association of the Rik1 CLRC component with Stc1 ( Figure S5C). Furthermore, ChIP analyses indicate that FLAG-Raf1-R518A, FLAG-Raf1-R576A and FLAG-Raf1-1 do not associate with centromeric repeats ( Figure 5D). We conclude that binding of Raf1 to Rik1 is required to allow Raf1 to associate with centromeric repeats. This interaction is required to assemble an active CLRC complex in order to establish and maintain H3K9 methylation and hence functional heterochromatin.
The coupling of H3K9 methylation and siRNA production could be achieved by bringing the distinct activities required for both processes together in the same complex. Since the raf1D and raf1-R576A mutations results in loss of siRNAs, Raf1 is clearly required for siRNA production ( Figure S5D and Figure 2G). However, the raf1-R518A and raf1-1 mutants lose H3K9 methylation without affecting siRNA synthesis ( Figure 2E, 2G; Figure 3C; Figure 4D; Figures S4E and S5D). This suggests that specific interactions between Raf1 and individual proteins in CLRC and/or other unknown factors are sufficient to allow siRNA generation in the absence of H3K9 methylation. An intact CLRC complex may not be required. Indeed, co-immunoprecipitation experiments show that a weak but detectable interaction between Raf1 and Raf2 remains in raf1-1 and raf1-R518A, but not raf1-R576A cells ( Figure 5E). We conclude that although an intact CLRC complex is dispensable for siRNA synthesis, a minimal Raf1-Raf2 interaction might be sufficient to allow the processing of centromere repeat transcripts to siRNAs.

Discussion
The interaction between the E3 ligase adaptor Rik1 and the DCAF Raf1 correlates with methylation of lysine 9 on histone H3 The CLRC complex, harbouring the histone methyltransferase Clr4 Suvar3-9 , performs an essential role in the establishment and maintenance of heterochromatic structures. In addition to Clr4, the complex contains the components Cul4, Rik1, Raf1 and Raf2.
Rik1 is a WD-40 repeat protein that shares homology with the E3 ligase adaptor DDB1 [36]. DDB1 contains 21 WD-40 repeats that form the seven blades of three b-propellers that mediate association with Cul4 and DCAFs. This supports the possibility that Cul4-Rik1 form the core of an E3 ligase [24]. However, the C-terminus of Rik1 also shows homology with the b-propeller domain of cleavage and polyadenylation factor CSPF-A a wellknown RNA binding protein [18]. Based on this CPSF-A homology, it has been suggested that Rik1 might bind RNA through this domain [21]. Although we cannot exclude that Rik1 is a bifunctional protein, our manual alignment detects 21 WD-40 repeats within Rik1 allowing it to be structurally aligned along its entire length with DDB1 ( Figure 2B and Figure S2A). In our model, the CSPF-A homology domain corresponds to the bpropeller BPC involved in the interaction with the DCAF Raf1 ( Figure 2B and Figure S2A). Moreover, we also show that Raf1 can be structurally aligned to the DCAF DDB2, the partner of DDB1 (Figure 2A, 2B and Figure S2B). As expected for a bona fide DCAF, Raf1 associates directly with the putative substrate adaptor Rik1 and contains two signature WDxR motifs which we show are required for the association of Raf1 with Rik1 and for H3K9 methylation and heterochromatin formation ( Figure 2B and Figure S3). This suggests that the CLRC complex can adopt the architecture of a Cul4 E3 ligase in which Rik1 is the adaptor protein bridging the interaction between Cul4 and the DCAF Raf1 ( Figure 6A). In addition, we show that the association of the DCAF Raf1 with Rik1 within an intact CLRC complex is critical for H3K9 methylation but it is dispensable for the processing of centromeric transcripts into siRNAs (Figure 2, Figure 4, and Figure 5).
We envisage two alternative scenarios that can explain how the Cul4-Rik1-Raf1 complex might mediate H3K9 methylation ( Figure 6A). First, it is possible that the CLRC is an active E3 ligase and that mono-or poly-ubiquitination of specific factors must occur to allow methylation of H3K9. Thus the DCAF Raf1 would be essential for the recognition and ubiquitination of key specific substrate(s) (Figure 6A top panel). In other systems, Cul4-  Percentage of lagging chromosomes in anaphase of wt, raf1D and raf1-1 cells at 25uC or 36uC. Cell lacking heterochromatin display higher rates of chromosome missegregation at lower temperature [42]. (C) H3K9me2 ChIP: levels associated with cen-dg relative to act1 + in clr4D and raf1-1 cells normalised to wild-type at 25u or 36uC. Error bars: SD. doi:10.1371/journal.pgen.1002499.g003  [29,30], it is possible that a specific histone residue needs to be ubiquitinated to allow methylation of H3K9. E3 ligase activity has been shown to associate with affinity selected CLRC in vitro [24], however, we have been unable to detect this activity in similar experiments. This suggests that CLRC E3 ligase activity may be particularly inefficient and/or the ubiquitination events that it mediates are transient.
The alternative scenario is that the CLRC complex is not an active E3 ligase in vivo and that Cul4, Rik1, Raf1 and Raf2 just form a protein scaffold that acts to target Clr4 methyltransferase to heterochromatic repeats, independently of ubiquitination ( Figure 6A bottom panel). In this case the specific mutations which impair the Rik1-Raf1 interaction may just disrupt the scaffold so that Raf1 is not targeted to the heterochromatic repeats and Clr4 no longer associates with Rik1 and Raf1. Interestingly, the DCAF DDB2 has been shown to bind DNA with the 'bottom' surface of its WD-40 ring [44]. Similarly, the WD-40 ring of the DCAF WDR5 specifically binds the histone H3 N-terminal tail methylated on lysine 4 [45,46,47]. Thus, it is possible that the WD-40 ring of the DCAF Raf1 also binds DNA or histones to allow the correct targeting of the histone methyltransferase Clr4 to heterochromatic repeats.

Raf1 function in H3K9 methylation and RNAi are separable
CLRC plays a dual role in heterochromatin formation: it harbours the histone methyltransferase Clr4 and hence it is responsible for H3K9 methylation, it also mediates siRNA production since cells lacking any single CLRC subunit have low centromeric siRNA levels ( Figure S5D and [17,18]). Therefore, in wild-type fission yeast H3K9 methylation and siRNAs synthesis are coupled. This ensures formation of heterochromatin at centromeres, telomeres, and the mating type locus, and prevents promiscuous silencing at other chromosomal regions. However, cells expressing only mutant histone H3 (H3K9R) have been shown to produce some detectable siRNAs, even though H3K9 can not be methylated [20,21]. Such analyses implicate CLRC itself, rather than its known substrate H3K9, in promoting siRNA production independently of H3K9.
Here we have isolated two mutations in the raf1 gene (raf1-1 and raf1R518A) that destroy CLRC integrity without affecting siRNA levels. Although we cannot exclude that a minimal CLRC complex is still present in these mutants, our results indicate that individual CLRC components can stimulate processing of centromeric transcripts into siRNAs independently of CLRC complex integrity ( Figure 6B). It remains to be determined how CLRC components promote this H3K9 methylation-independent siRNA production. Our analyses of specific raf1 mutants suggests that the putative E3 ligase activity of the CLRC complex is not required for siRNA synthesis given that in raf1-1 and raf1-R518A mutants (predicted to impair the E3 ligase activity of the complex) siRNA levels remain high ( Figure 2G and Figure 4D). One possibility is that Clr4 can methylate specific substrates independently of CLRC integrity and that this is a key event in siRNA production [21]. In accordance with this, Clr4 was recently shown to methylate the RNA processing factor Mlo3 and this methylation correlates with high siRNA levels [48]. However, Clr4-mediated methylation of specific substrates cannot be the only event required to trigger siRNA production since the loss of any CLRC component results in dramatic reduction of siRNA levels. We find that high siRNA levels correlate with a low but detectable Raf1-Raf2 interaction. Complete disruption of this Raf1-Raf2 interaction (as observed in raf1-R576A) cuts siRNA production to undetectable levels. Interestingly, centromeric repeats are transcribed preferentially in S-phase and Raf2 has been recently shown to interact with Cdc20 (the catalytic subunit of DNApolymerase e) [49,50,51]. Coordination of DNA replication, siRNA generation and methylation of specific substrates may be essential for heterochromatin establishment and maintenance of heterochromatic structures.
Our analyses suggest that integration of DNA replication, siRNA production and methylation of H3K9 is achieved by bringing the distinct activities required for these processes together in the same protein complex (CLRC). This ensures the assembly of robust heterochromatin structures.

Cotranscriptional gene silencing plays a subsidiary role in repression of heterochromatic repeats
Two different models have been proposed to explain how heterochromatin could silence centromeric repeats. In the 'TGS' model, heterochromatin factors have been proposed to repress transcription at centromeres [15] whereas the 'CTGS' model suggests that centromere repeats are continuously transcribed and silencing is caused by the efficient cleavage of centromeric transcripts by RNAi into siRNA [16]. Our analyses demonstrate that the RNAi machinery is active in raf1-1 cells, so that centromeric transcripts are processed into siRNAs independently of heterochromatin integrity. raf1-1 cells produce siRNAs at levels similar to wild-type cells and they are loaded into Ago1. However, despite this, high levels of unprocessed centromeric transcripts persist, indicating that RNAi fails to destroy them. The fact that heterochromatin is absent in raf1-1 and raf1-R518A cells and transcripts remain high, even though high levels of homologous Ago1-associated centromeric siRNAs are present, is more compatible with a TGS model where heterochromatin directly represses RNAPII transcription of centromere repeats.
In agreement with this model, RNAPII ChIP clearly shows an increase of RNAPII occupancy in the raf1 mutants compared to wild-type cells ( Figure S5E). This observation is also consistent with the fact that tethered Clr4 can generate heterochromatin and silence marker genes independently of RNAi [52].
The coupling of non-coding transcription with chromatin modification to establish and maintain distinct epigenetic states is a general mechanism of gene regulation in eukaryotes. Links between Cul4-dependent ubiquitination and histone methylation are continuing to emerge in different systems. Further analyses will determine how these activities are integrated to regulate specific chromatin states.

Structural modelling
The homology model of Raf1 was created with the program Modeller (9v8) using the structure of DDB2 (PDB code 3EI4) [44]. Iterative rounds of alignment adjustment of the two protein sequences and model building were attempted until the Modeller scores (diagnostic of model quality) were optimised. The homology model of Rik1 was created with the program Modeller (9v8) using 5 templates (PDB code: 2B5L; 317N; 3189; 318C; 3E0C). The generated model (85.2) had the following model scores: RMSD: 1.536; MolPDF: 453.98; DOPE: 211385.

Chromatin immunoprecipitation
Chromatin immunoprecipitation (ChIP) was performed as described [54] with the following modifications. Cells were fixed in 1% PFA/15 min for H3K9me2 ChIP or in 1% PFA/20 min for RNA Polymerase II ChIP. For ChIP analyses of raf1-R518A and raf1-R576A strains, cells were grown at 32uC. For ChIP analyses of raf1-1 strain, cells were kept at the indicated temperature (25uC, 32uC or 36uC) for at least 96 hours. One microliter of monoclonal H3K9me2 antibody (m5.1.1), two microliter of anti-FLAG M2 monoclonal antibody (Sigma, F1804) or five microliters of RNA Polymerase II 8WG16 antibody (COVANCE, MMS-126R) was used per ChIP. Duplex PCR was performed to analyse ChIP samples using oligonucleotides specific to the regions of interest and to the control gene fbp1 (Table S2). Real-time PCR (qPCR) was performed using the LightCycler 480 SYBR Green I Master (Roche) on a LightCycler 480 Instrument (Roche). qPCR analysis primers are in Table S2. Relative enrichments were calculated as the ratio of product of interest to control product (act1 + or tRNA) in IP over input. Histograms represent data from three biological replicates analysed in parallel.
Co-IPs for Western analysis were performed on 2 g of cells as above but for 1 hr. IP'd material was washed four times with icecold lysis buffer (50 mM Hepes pH7.5, 150 mM KCl, 0.1% NP40), resuspended in SDS sample buffer and analysed by SDS-PAGE. For tandem affinity purification (TAP)-tagged strains, Dynabeads coupled to IgG were used (gift from K. Hardwick). For Western analysis the following antibodies were used: anti-FLAG M2 (Sigma, F1804), anti-HA 12CA5 (gift from K. Samejima) and anti-myc (A14) (Santa Cruz sc-789), all at 1:1000. Further details of IP protocol are provided in Text S1.

RNA analysis
Northern analysis of long non-coding centromeric transcripts and centromeric siRNAs were performed as described previously [56]. RNA probes are listed in Table S2.