The specificity and topology of chromatin interaction pathways in yeast

The majority of chromatin regulators have specific effects on gene expression

Throughout the study a p-value of 0.05 in combination with a fold-change (FC) of 1.7 is applied as a threshold for calling a change in mRNA expression significant. The reported FC and p-value is based on the average of the four replicates versus the average of 200 wt profiles. Based on the variation observed within the wt cultures, a threshold can be applied to distinguish mutants with an expression-profile different from wt (Experimental Procedures). By these criteria, 80% of the mutants have altered gene expression. Strikingly, only 29% of kinase deletions and only 15% of phosphatase deletions exhibit expression-profiles different from wt (van Wageningen et al., 2010). This underscores the importance of chromatin complexes in regulation of gene expression.

A second general outcome of this study is that inactivation of chromatin factors result in quite specific effects on gene expression. A two-dimensional hierarchical cluster diagram summarizes all the changes (Fig. 1). In the 80% of mutants with an altered profile, on average 116 genes (2%) show significant changes (p<0.05, FC>1.7). The largest effect is for ssn6Δ with 973 differentially expressed genes (16%). The possibility that data normalization had masked global effects was ruled out by applying a spiked-in external RNA control strategy (van de Peppel et al., 2003), (Supplemental Experimental Procedures). Focus on viable deletion mutants results in coverage of an estimated 78% of known chromatin regulators (Table S1). Although inviable mutants likely have broader effects, the specificity observed here is nevertheless noteworthy since several of these factors are responsible for laying down general marks such as H3K4 (Set1), H3K36 (Set2) and H3K79 (Dot1) methylation. Genome-wide location maps of these histone marks show widespread location (Pokholok et al., 2005). The effects of their removal result in much more specific effects, with changes in expression of 55 genes in set1Δ, 72 genes in set2Δ and 2 genes in dot1Δ (Fig. 1). Disconnect between location and effect has been noted in individual cases before (Venkatasubrahmanyam et al., 2007; Rando and Chang, 2009). The specific nature of all the perturbation signatures (Fig. 1) suggests that this disconnect is a general feature of chromatin regulation. Lack of an effect does not necessarily imply that a factor is not operating on such a gene. Rather, the specific nature of the effects indicates that the function of the factor is important for only a specific subset of genes under this condition.

Open in a separate window

Fig. 1

Perturbation of chromatin regulators leads to specific effects on gene expression

Unsupervised hierarchical cluster diagram (cosine correlation) of all the mutants with profiles that differ from wt (Experimental Procedures) and all genes that change significantly (p<0.05, FC>1.7) in any mutant. The dendrograms indicate relationships between genes (top) and between mutants (right). The red part (right) depicts the relationship between Paf1C, Rad6/Bre and Set1C. Mutants are color-coded according to protein complex. Fold-change is indicated by the color-scale, with yellow for up-, blue for downregulation and black for no change, versus the average wt. dot1Δ is not included because the number of genes changing classifies it as similar to wt.

Assigning cellular roles to chromatin regulators

The specificity of the perturbation signatures (Fig. 1) can be exploited in several ways. In a first systematic analysis, cellular roles were assigned to chromatin factors by determining which functional categories of Gene Ontology (GO) are enriched in the signatures (Fig. S1, summarized in Fig. 2A). Many previously reported regulatory roles are confirmed. Regulation of middle-sporulation genes by Sum1/Rfm1/Hst1 (Xie et al., 1999) is reflected by enrichment for the GO Slim term ‘sporulation’ in these profiles (Fig. 2A). Signatures of the Rpd3L complex are enriched for ‘meiosis’ and signatures of the Sir complex are enriched for ‘conjugation’ (Fig. 2A), all as expected (Mallory and Strich, 2003; Rine and Herskowitz, 1987). New roles are also indicated. For example, eight complexes have two or more signatures enriched for GO terms associated with mating (Fig. S1, Fig. 2A “conjugation”). Reduced mating efficiency has been shown for four of these complexes: Sir, Mediator, Rad6/Bre1 and SAGA (Holstege et al., 1998; Rine and Herskowitz, 1987; Huang et al., 1997; Roberts and Winston, 1996). Deletion of two subunits in each of the remaining four complexes (Rpd3L, Isw2C, RSC, SWI/SNF) were tested. All show a significant reduction in mating (Fig. 2B) indicating involvement in this process. Other new roles are predicted (Fig. 2A, Fig. S1) illustrating how the data can be employed to link chromatin factors to specific cellular roles.

Open in a separate window

Fig. 2

Specific cellular roles for chromatin machinery components

(A) Enrichment for Gene Ontology Slim terms (top) depicted for each mutant (right), by plotting the significance (hypergeometric test). Yellow indicates enrichment in upregulated genes, blue indicates enrichment in downregulated genes, with a threshold of p<0.01. Fig. S1 depicts all Gene Ontology terms. (B) Histogram showing the mean mating efficiency for mutants (bottom), two from each complex with no previously reported mating defect, with ste7Δ as control. Error bars reflect standard deviations of duplicates. Significance was calculated by two-sided T-test. *p<0.05, **p<0.01, ***p<0.005.

Structure-function analyses of protein complexes that regulate chromatin

The specificity of the signatures also facilitates investigation of functional relationships between protein complex subunits. This second systematic analysis is inspired by the structure of the dendrogram that summarizes similarities between signatures (Fig. 1). The dendrogram is automatically generated upon hierarchical clustering, with no prior knowledge of protein complex composition. Each signature is color-coded according to protein complex (Fig. 1, right), revealing that complex composition is a primary determinant of the dendrogram structure. This indicates that subunits derived from the same protein complex often have similar perturbation signatures. More detailed analysis reveals that this holds especially for complexes that consist of four or less different subunits (Fig. 3, Fig. S2 for all 32 complexes). Throughout the manuscript, for each set of mutants grouped together, the results for all genes with a significant change (p<0.05, FC>1.7) in any single mutant are depicted, rather than selecting for similarly behaving genes. As an example, the HIR complex is a histone chaperone involved in replication-independent deposition of histones throughout the cell-cycle (Green et al., 2005). HIR contributes to setting up a remodeler-resistant, repressive chromatin structure (Prochasson et al., 2005). Deletion of the different HIR subunits results in upregulation of the same set of 9–10 genes (Fig. 3C). Besides exemplifying data consistency, the HIR profiles show that each component is equally important. This agrees with the finding that all HIR subunits are required for interaction with the histone chaperone Asf1, essential for the role in nucleosome assembly (Green et al., 2005). Only a small group of genes show strong mRNA changes in the HIR signatures (Fig. 3C), likely reflecting the function of redundant mechanisms of histone deposition (Kaufman et al., 1998; Formosa et al., 2002). Nearly identical subunit signatures are also observed for the HDAC, SAS and Sir complexes (Fig. 3A, B, D), as well as for the Sum1/Rfm1/Hst1, CAF-1 and Ku complexes (Fig. S2). Although signatures derived from the same complex are generally similar, there are also numerous interesting exceptions, especially in complexes consisting of more than four subunits. This reveals the presence of auxiliary, peripheral and shared subunits, as well as the submodular organization of larger complexes (Fig. 3E–H).

Open in a separate window

Fig. 3

Structure-function analysis of chromatin regulatory complexes

For each complex (AH), all genes with significant changes (p<0.05, FC>1.7) in any single mutant are shown, yellow indicating up-, blue indicating downregulation and black indicating no change, relative to the average wt and scaled as in Fig. 1. If a gene is not depicted, then no change in expression is observed outside the thresholds p<0.05, FC>1.7. Fig. S2 depicts all 32 complexes. (A) HDAC (B) SAS (C) HIR (D) Sir (E) SWR1, without Bdf1 as different isolates of bdf1Δ gave different results. The apparent lack of “downregulation” of the deleted gene in swc7Δ and swc3Δ (right hand box) is due to very low expression levels. All such mutants were PCR-verified for carrying the deletion (Supplemental Experimental Procedures). (F) Isw1a and Isw1b (G) Rpd3S and Rpd3L. Ume6 is not shown due to aneuploidy. (H) SAGA. The dendrogram was generated by hierarchical clustering. The grey line indicates that no probe for SUS1 is present on the array. (I) A network of SAGA subunit relationships generated using an edge-weighted, spring-embedded network algorithm (Shannon et al., 2003), revealing connections between submodules.

The presence of auxiliary subunits is exemplified by the SWR1 complex, involved in deposition of the histone variant H2A.Z (Kobor et al., 2004; Krogan et al., 2003; Mizuguchi et al., 2004). In contrast to the other subunits, deletion of SWC7 yields a profile similar to wt (Fig. 3E). This agrees with recent biochemical analyses showing that Swc7 is not required for complex integrity or H2A.Z replacement in vitro (Wu et al., 2009). Swc3 has also been classified as an auxiliary subunit, with little contribution to H2A.Z deposition (Wu et al., 2005, 2009). Deletion of SWC3 results in the same hallmark signature observed for the other components (Fig. 3E), indicating a role not captured by the in vitro analyses. Fig. S2 contains further examples of auxiliary function as well as examples of peripheral roles. The signature of a peripheral subunit is a subset of the effects observed in the others, indicating requirement in only one aspect of the function of a complex (Fig. 3G).

Other deviating subunit signatures result from subunits shared between different complexes. A shared subunit may functionally contribute to both complexes. This is illustrated by ISW1, a component common to the chromatin remodeling complexes Isw1a and Isw1b (Vary et al., 2003). ISW1 deletion yields a signature that largely combines the effects observed upon deletion of the two complexes that it is part of (Fig. 3F). Similarly, Rpd3S and Rpd3L are two related histone deacetylase complexes (Carrozza et al., 2005b). Deletion of the shared components RPD3 or SIN3, results in a signature that is mostly a combination of the profiles from the complex-specific subunits (Fig. 3G). A practical outcome is the identification of model genes suitable for distinguishing between the function of the two complexes. Deletions of the Rpd3L associated proteins, Ash1 and Cti6 (Carrozza et al., 2005a; Puig et al., 2004), show only a subset of the changes observed in other Rpd3L deletion mutants (Fig. 3G), exemplifying peripheral roles, required for particular genes only.

Larger complexes can often be subdivided into distinct submodules, also amenable to analysis by expression profiling. Analysis of the coregulator SAGA indicates four submodules (Fig. 3H and I). SAGA has previously been subdivided into a TATA-binding module, an acetylation module and a structural integrity module (Fig. 3I) (Sterner et al., 1999). The different signatures indicate that the recently characterized deubiquitination components form a functionally distinct submodule, with Sgf73 and Sgf29 connecting this module to the rest of the complex (Fig. 3I) (Rodriguez-Navarro, 2009; Bonnet et al., 2010). As with the other examples (Fig. 3, Fig. S2), this illustrates various ways in which the signatures can provide information on the function of individual subunits within complexes.

A correlation network of chromatin machinery interactions

The data was next analyzed to uncover global properties, focusing on how different complexes work together. The structure of the hierarchical cluster dendrogram is determined by similarities between subunits, but also by similarities between different complexes as a whole (Fig. 1). This is illustrated by the red part of the dendrogram, indicating relationships between Paf1C, Rad6/Bre1 and Set1C. To systematically capture such relationships, a correlation network was derived from the expression data (Fig. 4). In this network, each node represents a mutated component, color-coded according to protein complex. Edges connecting two nodes are drawn if the correlation between the profiles is higher than a specific threshold. Distances between nodes are dependent on the similarity of all connections. The threshold was set to capture those relationships resulting in identical, similar, or broadly overlapping signatures (Suppl. Experimental Procedures).

Open in a separate window

Fig. 4

A correlation network of perturbation signatures reveals chromatin machinery interactions

Nodes represent mutated components, color-coded according to complex composition. Edges between nodes are drawn if the correlation between the two signatures is more than 0.4, capturing all identical, similar or broadly overlapping profiles. Fig S3A depicts a heatmap with all correlations. The network was generated with an edge-weighted, spring-embedded algorithm, so that the distances between nodes represent the similarity of connections. Interactions between several complexes are marked by the Roman numerals I (SAGA and SWI/SNF), II (Rad6/Bre1 and Set1C), III (Tup1 and HDAC) and IV (Paf1C and Set3C).

Three properties indicate network reliability. First, as expected, subunits of complexes are often grouped together. Second, the chromatin perturbation network is scale-free, distinct from a network that has a random topology (Fig. S3B) (Barabasi and Oltvai, 2004). Third, the network recapitulates many previously established interactions. For example, SAGA and SWI/SNF are tightly connected (indicated by I in Fig. 4), reflecting that SWI/SNF preferentially remodels nucleosomes acetylated by SAGA (Chandy et al., 2006). Similarly, there are many connections between Rad6/Bre1 and Set1C (Fig. 4, II). This agrees with the requirement for Rad6/Bre1 to ubiquitinate H2BK123 prior to H3K4 methylation by Set1C (Dover et al., 2002; Sun and Allis, 2002). The network indicates that like SAGA-SWI/SNF, this is also a general interaction, with functional consequences on the many genes shared between the corresponding signatures. The network also evaluates interactions between coregulators and chromatin complexes. Tup1-Ssn6 is a global coregulator (Smith and Johnson, 2000). Various repressive mechanisms have been proposed for Tup1 (Zhang and Reese, 2004b). The network indicates that repression through recruitment of the Hda1 deacetylase complex (HDAC) is most prevalent (Fig. 4, III). Evidence for another, less general mechanism, is presented below. The network reflects many individual interactions, facilitating evaluation of those that have previously been characterized on individual genes only.

The network also shows how different chromatin regulators are functionally connected and indicates that almost all regulators are linked to others. This shows that chromatin interactions do not form separate pathways that occur in isolation on different gene sets, but are interconnected. This agrees with the many genetic interactions found between chromatin regulators (Collins et al., 2007). Strictly linear, one-to-one pathway relationships between different regulators are sparse. Perturbation of linear signaling pathways, such as the mating pheromone or High Osmolarity Glycerol (HOG) MAPK cascades, do yield one-to-one relationships. In such pathways, essentially identical mRNA changes are observed upon perturbation of any component (Fig. S4). Here, only the interaction between the Set2 H3K36 methyltransferase and the Rpd3S histone deacetylase complexes (Carrozza et al., 2005b; Joshi and Struhl, 2005; Keogh et al., 2005) shows such a linear, one-to-one relationship (Fig. 5A). All other pathway relationships between chromatin regulators branch (Fig. 4). In other words and as is exemplified below (Fig. 5B and Fig. 6), the signature of one pathway component only partially overlaps or forms a subset of the signature of another pathway component. The varying degrees of overlap in the perturbation signatures are likely caused by the participation of complexes in different interactions, important for different genes.

Open in a separate window

Fig. 5

The Leo1 subunit of Paf1C contributes to Set3C recruitment

(A) Identical signatures of Set2 and Rpd3S, indicating a linear pathway. (B) Hierarchical cluster diagram of expression signatures of Bur2, Paf1C, Rad6/Bre1C, COMPASS and Set3C. The genes are colored and scaled as in Fig. 1. (C) and (D) ChIP of Hos2-TAP in wt, paf1Δ, leo1Δ and cdc73Δ. PCR was carried out on RPS13 (C) and PYK1 (D). Numbered primer locations are shown schematically (top). The left panel shows PCR products and the right panel shows quantification of replicates, with error bars for standard deviation. Signals are normalized to input. (E) and (G) ChIP from wt and set3Δ with anti-H3 or anti-acetyl H4 on RPS13 (E) and PYK1 (G). Results are normalized to total H3 signal. (F) and (H) ChIP from wt or leo1Δ with anti-H3, anti-acetyl H4, anti-H3K4me3 or anti-H3K4me2 on RPS13 (F) and PYK1 (H). (I) Previously described (black arrows) and proposed (blue arrow) pathways connecting Bur1/2 and Set3C. The pathways branch at Paf1C and regulate Set3C in two ways: via dimethylation of H3K4 by COMPASS (black arrows) and via the proposed H3K4 methylation-independent mechanism (blue arrow).

Open in a separate window

Fig. 6

Specific subsets of genes are sensitive to particular interactions

Previously reported interactions not represented in the network of global correlations, but showing significant overlap on specific subsets of genes. All genes with significant changes (p<0.05, FC>1.7) in any single mutant are shown, except for B. Genes are colored and scaled as in Fig. 1. (A) Tup1, Isw2C and HDAC. (B) RNR3 expression and overlap between Isw2C and Tup1. Here all genes are depicted with significant upregulation in tup1Δ and in any Isw2C signature. (C) SAS and SWR1C. Genes marked by an asterisk (*) are subtelomerically located (<50 kb from chromosome end).

The Leo1 subunit of Paf1C contributes to a distinct branch of Set3C recruitment

To investigate branching and interconnectivity, we first focused on the connections observed between the Paf1C, Rad6/Bre1C, COMPASS/Set1C and Set3C complexes (Fig. 4II and IV). A single chromatin interaction pathway can be assembled from previous studies. Paf1C associates with RNA polymerase II and regulates ubiquitination of H2BK123 by Rad6/Bre1C (Ng et al., 2003; Wood et al., 2003; Robzyk et al., 2000). These processes are dependent on the Bur1/2 cyclin dependent kinase complex (Wood et al., 2005). Subsequent to H2B ubiquitination, COMPASS/Set1 methylates H3K4 (Krogan et al., 2003, 2002; Dover et al., 2002). It has recently been shown that dimethylated H3K4 is bound by Set3C (Kim and Buratowski, 2009). These findings can be summarized in a linear pathway (Fig. 5I, black lines). Many aspects of this pathway are reflected in the network (Fig. 4, II and IV) and in the underlying perturbation signatures (Fig. 5B). BUR2 deletion clusters with Paf1C (Fig. 5B), highlighting its role in recruitment of Paf1C and regulation of Rad6/Bre1 (Wood et al., 2005). Going further down the pathway, the signatures of downstream complexes (e.g. Set1C, Set3C) are subsets of upstream ones (e.g. Paf1C, Rad6/Bre1, Fig 5B). Interestingly, one Paf1C subunit, rtf1Δ, clusters closer to the Rad6/Bre1complex than to members of its own complex, indicating that the sole role of Rtf1 is to promote Rad6/Bre1C function (Ng et al., 2003). Since Set3C recognizes H3K4 dimethylation by Set1C (Kim and Buratowski, 2009), and is furthest downstream, it is surprising that Set3C signatures are more similar to two Paf1C subunits (leo1Δ and cdc73Δ) than to Set1C (Fig. 5B). This is also reflected in the network (Fig. 4, IV). Similarity between leo1Δ and Set3C is even more surprising given that loss of the Paf1C subunits Rtf1, Ctr9, Paf1, all result in defective H3K4 dimethylation, required for Set3C recruitment, while loss of Leo1 has no effect on this mark (Xiao et al., 2005). Together, this suggests that the Leo1 subunit of Paf1C may also contribute to Set3C function in a manner distinct from the pathway through Set1C and Rad6/Bre1 (Fig. 5I, blue arrow).

A functional link between Paf1C and Set3C, separate from the previously established pathway, was tested by chromatin immunoprecipitation (ChIP, Fig. 5C–H). Loss of Leo1 or Cdc73 results in reduced recruitment of Set3C (Fig 5C, D). Set3C is a histone deacetylase and acetylation levels are increased in leo1Δ (Fig. 5F, H). To ensure that the reduced Set3C presence observed in leo1Δ is not simply a reflection of the role of Paf1C in recruitment of Set1C, H3K4 dimethylation patterns were determined. As reported previously (Xiao et al., 2005), leo1Δ has no effect on H3K4 dimethylation (Fig. 5F, H). This agrees with a contribution towards recruiting Set3C that is distinct from the Rad6/Bre1-Set1C pathway. PAF1 deletion leads to a larger decrease in Set3C recruitment (Fig 5C, D), suggesting possible loss of both modes of Set3C recruitment in paf1Δ. The data therefore support a model in which the pathways branch at the level of Paf1C. Paf1C contributes to Set3C recruitment in two ways (Fig. 5I), through the previously described pathway via H3K4 methylation and in a separate manner dependent on Leo1 (Fig. 5I, blue arrow). The latter was predicted by analysis of the chromatin perturbation signatures (Fig. 4, ​,5B)5B) and perhaps reflects the role of Leo1 in RNA binding (Dermody and Buratowski, 2010). Besides illustrating branching and interconnectivity of a chromatin interaction pathway, this also exemplifies ways in which the signatures can be applied to investigate interaction pathways further.

Specific subsets of genes are sensitive to particular interactions

The threshold applied to generate the correlation network (Fig. 4) only identifies interactions with broad overlaps in the respective signatures. The branched network topology reveals that different combinations of factors are important for different gene sets. Functionally relevant interactions will not necessarily result in strongly overlapping signatures. We therefore also investigated previously reported interactions not represented in the network of global correlations. Besides functioning through the Hda1 deacetylase complex (Fig 4, III), the corepressor Tup1 has also been shown to interact with the nucleosome remodeler Isw2C on RNR3 (Zhang and Reese, 2004a, 2004b). Although the overall correlation is lower than the threshold applied to generate the network, tup1Δ and isw2Δ signatures do overlap significantly (Fig. 6A, p=9*10⁻⁷, hypergeometric test). In agreement with previous observations (Zhang and Reese, 2004a, 2004b), upregulation of RNR3 in isw2Δ is very weak (1.3-fold, p<1*10⁻⁶). The analysis indicates that the interaction between Tup1 and Isw2C, previously characterized on RNR3, has functional implications for a specific subset of genes. Compared to RNR3, several other genes show a much stronger response in isw2Δ (Fig. 6B). Besides providing a further illustration of branching (Fig. 6A), a practical outcome is therefore the ability to pinpoint model genes that are most sensitive to a particular interaction.

Regulation in the context of telomeric location

Many of the individual interactions captured in the global network represent histone code “writer” – “reader” relationships (Fig. 4). The results only loosely fit the histone code hypothesis however. There is a large amount of branching between different regulators. This is due to the partial nature of the overlaps between signatures of different complexes. Combinatorial effects are part of the original histone code theory (Strahl and Allis, 2000), but would only agree with the branched network topology if different histone marks also exhibit partially overlapping location. Analyses of histone marks across the yeast genome have not yet resulted in the discovery of gene set specific patterns (Pokholok et al., 2005; Rando and Chang, 2009). We therefore focused on recent proposals suggesting that binding of chromatin factors to specific marks is dependent on context (Berger, 2007; Lee et al., 2010; Campos and Reinberg, 2009).

The nature of such contexts is not well understood and the data offers the possibility to explore these further. One context may be formed by location close to telomeres. Telomere-proximal genes are silenced through the telomere position effect, mediated by the Sir proteins (Perrod and Gasser, 2003). Boundary factors, such as H4K16 acetylation and H2A.Z, function synergistically to prevent Sir spreading on a specific subset of genes located at subtelomeric regions (Shia et al., 2006). Overlapping function, but only in the specific context of subtelomeric location, is indeed recapitulated in the expression signatures of the SAS H4K16 acetyltransferase complex and SWR1-C (Fig. 6C).

To investigate this context further, all signatures were tested for significant enrichment of subtelomerically positioned genes (Suppl. Experimental Procedures). For subtelomerically enriched signatures, the average expression in a 1500 bp sliding window from the chromosome end is depicted (Fig. 7A). The contributions of SAS, SWR1C and Rpd3L (Ehrentraut et al., 2010), as well as of Sir proteins, are as expected. In addition to established factors, several others have subtelomerically enriched signatures (Fig. 7A), including the coregulator Mediator. Members of different Mediator submodules (Guglielmi et al., 2004) affect subtelomeric genes differently. Disruption of the Tail (med2Δ, med3Δ, med15Δ) results in upregulation of subtelomeric genes (p = 7*10⁻⁴, p = 6*10⁻⁵ and p = 2*10⁻¹⁵ respectively, chi-squared test), while disruption of the Middle (med1Δ, med5Δ, med9Δ) results in downregulation (Fig. 7A, p = 2*10⁻²⁰, p = 5*10⁻¹³ and p = 2*10⁻¹⁰, respectively). Although possible explanations include indirect effects, a more direct mechanism is supported by binding of Med5 to telomeres (Esnault et al., 2008). To investigate these observations, the genome-wide location of Sir3 was assessed in two Mediator deletion mutants, med15Δ from the Tail and med1Δ from the Middle (Fig. 7B and Fig. S5). Loss of MED15 causes decreased Sir3 occupancy at several subtelomeric ends (Fig. 7B and Fig. S5). In contrast, loss of MED1 results in increased Sir3 occupancy, coincident with increased silencing (Fig. 7A). Sir mRNA and protein levels are not affected in the mutants (data not shown). In both mutants, Sir3 occupancy is changed, but no general spreading of Sir3 is observed. This suggests that Mediator does not form a boundary element similar to H4K16 acetylation, but instead influences the magnitude of Sir protein binding. The effect of Mediator mutants varies at individual telomeres, in agreement with varying degrees of Sir protein repression at different telomeres (Pryde and Louis, 1999). The changes on subtelomeric Sir3 occupancy are widespread, not just restricted to the vicinity of genes with changed expression in Mediator mutants (Fig. S5). It is therefore unlikely that the changes in Sir occupancy are indirectly reflecting the coregulatory role of Mediator in gene expression. These results agree with a report that Mediator binds directly to H4K16 deacetylated nucleosomes, competing with Sir for nucleosome binding (C. Gustafsson, under review). Together this indicates that Mediator has different roles in different genomic locations.

Open in a separate window

Fig. 7

Analysis of subtelomeric expression in chromatin mutants indicates that Mediator is a regulator of subtelomeric silencing

(A) For subtelomerically enriched signatures (p<0.05, chi-squared test), the average expression in a 1500 bp sliding window (steps of 250 bp) is plotted according to the distance from the chromosome end. (B) Sir3 occupancy measured by ChIP-on-Chip in wt, med1Δ or med15Δ at the right arms of chromosome 2 and 3. The ORFs are positioned above the X-axis for Watson genes and below the axis for Crick genes. Sir3 occupancy for all chromosome ends is shown in Fig. S5.

Expression-profiling and deletion strains

Each mutant strain, BY4742 (Table S2), was profiled four times from two independently inoculated cultures and harvested in early mid-log phase in synthetic complete medium with 2% glucose. Sets of mutants were grown alongside wt cultures and processed in parallel. Dual-channel 70-mer oligonucleotide arrays were employed with a common reference wt RNA. All steps after RNA isolation were automated using robotic liquid handlers. These procedures were first optimized for accuracy (correct fold-change) and precision (reproducible result), using spiked-in RNA calibration (van Bakel and Holstege, 2004). After quality control, normalization and dye-bias correction (Margaritis et al., 2009), statistical analysis was performed for each mutant versus the collection of 200 wt cultures. The reported fold-change is an average of the 4 replicate mutant profiles versus the average of all wts. 58 genes showed stochastic changes in wt profiles (wt variable genes) and were excluded from all analyses. Incorrect strains from the collection (13%) as indicated by aneuploidy (6%), incorrect deletion (5%) or additional spurious mutation affecting the profile (2%) were remade and reprofiled (Table S2). Less than 1% of the wt profiles had more than three genes changing compared to the average wt as determined by the same criteria as for the mutants (p<0.05, FC>1.7) and after exclusion of the wt variable genes. This threshold was therefore applied to determine whether a mutant had a profile different from wt.

Mating

The tester strain was CSHL3-10 (trp1-239, ura3-52, mat a). Cells were allowed to mate on a filter for 45 min. Serial dilutions were plated on –TRP and –TRP –HIS –URA plates. After 2 days of growth, colonies were counted. The percentage of diploids versus haploids plus diploids was determined. Experiments were normalized to the total mating efficiency of that experiment and calculated relative to wt.

Chromatin Immunoprecipitation (Fig 5)

Chromatin immunoprecipitations were carried out as described (Kim and Buratowski, 2009), see Supplemental experimental procedures. The sequences of oligonucleotides used are listed in Table S3.

We thank V. Geli and T. van Welsen for strains, H.Th.M. Timmers, M. Vermeulen, F. van Leeuwen and J.A. Lenstra for critical reading. Supported by the Netherlands Bioinformatics Centre (NBIC) and the Netherlands Organization of Scientific Research (NWO), grants 016108607, 81702015, 05071057, 91106009, 021002035 (TLL), 86307007 (PK), 70057407 (JJB).