Construction of the landscape
To construct the landscape of interactions between ATP-dependent chromatin remodeling and histone modifications in budding yeast, we used two data sets that provide a genome-wide measurement of enrichment levels of 25 histone modifications [
11,
12] and changes in gene expression accompanying the perturbation (mutation or deletion) of 33 ATP-dependent chromatin remodelers [
13]. We first identified cohort of genes for each modification and chromatin remodeler. Genes belong to one modification cohort if they display significantly high levels (Z score > 1.64,
P value < 0.05) of the corresponding modification at promoters (see Additional file
1). In this way, we obtained 25 sets of modification cohorts. The second data set is from a previously assembled expression compendium of chromatin modifiers [
13]. Genome-wide changes in gene expression were measured when various chromatin modifiers were deleted or mutated. We restricted the analysis to expression profiles for perturbation of ATP-dependent chromatin remodelers. This resulted in a refined data set consisting of 33 expression profiles. Considering that there are dual remodelers acting as both activators and repressors, we determined both positively and negatively regulated cohorts for each chromatin remodeler. Genes belong to positively (negatively) regulated cohort for one chromatin remodeler if they display significantly decreased (increased) changes (
Z score < -1.64 and
P value < 0.05 for decreased changes,
Z score > 1.64 and
P value < 0.05 for increased changes) in gene expression accompanying the perturbation of the corresponding remodeler (see Additional file
1). The details of data process in this paper are described in Methods section below.
Having identified the cohort for each modification and remodeler, we next determined whether the remodeler interacts with the modification or not using two criteria. First, we employed modification cohorts and the expression compendium of chromatin remodelers to identify the interactions. If the remodeler works in concert with the modification to regulate chromatin activities of a subset of genes, its perturbation should cause a differential change in expression of modification cohort genes because gene expression is linked with chromatin regulation. As in a previous study [
13], we used the Kolmogorov-Smirnov (K-S) statistical test. The K-S
P value provides a measure of the discrepancy in the distribution of gene expression values between the modification cohort and the rest of the genes. The K-S score indicates both the direction and significance of the discrepancy between the two distributions. A positive sign of the K-S score indicates positive regulation by the remodeler (i.e. the modification cohort genes tend to have lower expression values than the rest of the genes accompanying the perturbation of the remodeler), whereas a negative sign implies negative regulation. Second, we utilized remodeling cohorts and the modification data to detect the interactions between chromatin remodeling and modifications. If the chromatin remodeling interacts with the modification, its cohort should show significantly higher levels of the corresponding modification at promoters. We used the Mann-Whitney U-test to evaluate the difference in the medians of modification levels between the positively (negatively) regulated cohort and the rest of the genes. To avoid confusion in the following analysis, the
P value was set as 1 if the positively (negatively) regulated cohort genes show significantly lower modification levels than the rest of the genes. For each remodeler-modification pair, either positively or negatively regulated cohort was selected according to the sign of its K-S score determined above. We use this selection of remodeling cohort for each pair in the following analysis if there is no other statement.
We carried out the above two statistical tests for all chromatin remodeler-modification pairs and generated 61 significant pairs with
P < 0.05 (after Bonferroni correction for multiple testing) in both tests (Figure , see Additional file
2). However, the resulting interactions may be biased by transcriptional effect, as some modifications and remodeling both required for high gene activity may appear correlated, for no specific reason. To assay whether the results are biased by transcriptional effect, we performed principal component analysis for each chromatin remodeler-modification pair in terms of its modification and expression levels. We considered the first principal component to be transcription-related and represented each gene in the first principal component space. For each chromatin remodeler-modification pair, we calculated Pearson correlation coefficient between the above representation of each gene and its transcription rate. If the identified 61 significant pairs are biased by transcriptional effect, they should have significantly higher correlation coefficients. However, we found that there is no significant difference in the resulting correlation coefficients between the identified 61 pairs and the other pairs (
P = 0.11, t-test). Similar result could be reproduced when we represented genes in the second principal component space.
A recent study has measured occupancy at every yeast promoter region for seven chromatin remodelers (Isw1a, Isw1b, Isw2, Swi/Snf, Rsc, Ino80, and Swr-c) [
14], yielding opportunity for examining direct targets of the remodeling activities. Using this small dataset along with histone modification data, we carried out similar analysis to derive the interactions between chromatin remodeling and histone modifications. Genes belong to targets of one remodeler (i.e. cohort) if they display significantly high occupancy levels (
Z score > 1.64,
P value < 0.05) of the corresponding remodeler at promoters (see Additional file
1). A remodeler-modification pair is determined to be associated if the modification cohort genes exhibit significantly higher occupancy levels of the corresponding remodeler and the remodeling cohort genes have significantly higher corresponding modification levels. Our analysis generated 13 significant pairs with
P < 0.05 (after Bonferroni correction for multiple testing) in both Mann-Whitney U-tests (see Additional file
3), 8 of which are included in the 61 significant pairs generated above. Moreover, for 12 of the 13 significant pairs, the modification cohort genes exhibit significantly different expression levels (
P < 0.01, K-S test) accompanying the perturbation of the corresponding remodeler and the remodeling cohort genes have significantly higher corresponding modification levels (
P < 0.01, Mann-Whitney U-test). These results show that our method is robust to the choice of dataset.
Based on the 61 identified pairs above, we provided the first global picture of interactions between ATP-dependent chromatin remodeling and histone modifications in a eukaryote (Figure ). In the following analysis, we focused on the 61 significant remodeler-modification pairs.
Specificities of the landscape
The landscape demonstrates the characteristic interactions between ATP-dependent chromatin remodeling and histone modifications in regulating chromatin activity. We found that there is selectivity of histone modifications and chromatin remodelers involved in the interactions: 12 of the 25 histone modifications show interactions with chromatin remodelers, and 14 of the 33 remodelers are connected with histone modifications (Figure ). Furthermore, some specific remodelers work in concert with more histone modifications than the other remodelers. There are six remodelers (Snf2, Swr1, Isw2 & Ume6, Rsc30, Rsc8, and Mot1) that interact with five or more histone modifications. Interestingly, these six remodelers show a preference for histone acetylation. We next investigated whether there is known experimental evidence in previous studies supporting our identified interactions.
Rsc3, Rsc8 and Rsc30 are components of the RSC chromatin remodeling complex which contains almost half of the known bromodomains in the yeast genome for binding acetylated lysines [
15]. A recent study has demonstrated that the ATP-dependent remodelling complex RSC shows a striking preference for H3 but not H4 acetylated chromatin [
5]. It has also been reported that H3K14 (i.e. histone H3 lysine 14) acetylation acts to increase recruitment of the RSC to nucleosomes [
16]. Our results are consistent with these observations: the three components interact with H3 acetylation, but not H4 acetylation. Rsc3 interacts with H3K9, H3K14, and H3Nterm acetylation, Rsc30 is associated with H3K9, H3K14, H3K18, and H3Nterm acetylation, and Rsc8 is connected with H3K14, H3K18, and H3Nterm acetylation. In addition to these known interactions, our results show that Rsc3 interacts with H2A.ZK14 acetylation, Rsc30 is associated with H2A.ZK14 acetylation, H3K4 and H3K36 trimethylation, and Rsc8 is connected with H2A.ZK14 acetylation and H3K4 trimethylation. Interestingly, our results demonstrate that Rsc3 and Rsc30 regulate the associated modification cohort in distinct ways (positive and negative regulation, respectively, Figure ), consistent with experimental evidence that they have different roles in regulation, although they interact physically [
17].
Swr1 is Swi2/Snf2-related ATPase that is the structural component of the SWR1 complex. Bdf1, a member of SWR1 complex, contains two bromodomains that recruit SWR1 to acetylated histones [
18]. Consistent with this, Swr1 interacts with H2A.ZK14, H3K9, H3K56, H3Nterm, and H4Nterm acetylation. Unlike most other remodelers, Swr1 regulates the associated modification cohort negatively (Figure ). We found that the Swr-c cohort genes (defined by the remodeler occupancy data) show lower transcription rates than the other genes (
P < 10
-6, Mann-Whitney U-test), although SWR1 is required for deposition of histone H2A.Z which is linked to gene activation [
19].
Snf2 and Swi1 belong to the SWI/SNF chromatin remodeling complex that contains bromodomains. Specifically, it has been shown that the Swi2/Snf2 bromodomain has a higher affinity for acetylated H3K9 and H3K14 peptide compared with unmodified H3 peptide [
20]. Our results not only reproduce this observation, but also show interactions of Snf2 with H2A.ZK14, H3K18, and H3Nterm acetylation in rich media. However, Snf2 and Swi1 interact with only H2A.ZK14 and H3K9 acetylation in minimal media. These results suggest that their regulatory manner differs between the two growth conditions.
Mot1, a member of the Snf2/Swi2 protein family of ATPases, functions by removing TATA-binding protein (TBP) from DNA. In addition, Mot1 is required for nucleosome remodeling independently of TBP recruitment [
21]. To our knowledge, there is no experiment exploring the interactions between Mot1 and histone modifications. We showed that different mutations of Mot1 (
mot-14 and
mot-42) affect its interactions with different histone modifications. Together, Mot1 interacts with H2A.ZK14, H3K9, H3K14, H3K18, H3Nterm, H4 acetylation, and H3K4 trimethylation. Its preference for histone acetylation may be due to the bromodomains of the Snf2/Swi2 protein family it belongs to [
7].
Arp8, Ino80, and Ies6 are subunits of INO80, a chromatin remodeling complex that is involved in regulation of transcription and in DNA damage response [
22]. Our results show that Ies6 interacts with H3K9 and H3Nterm acetylation in rich media. In the presence of DNA damage, Ino80 interacts with H2A.ZK14 and H3K14 acetylation, and Arp8 is associated with H2A.ZK14, H3K9, H3K14 acetylation, and H3K4 trimethylation. These results demonstrate that different components of INO80 act to interact with histone modifications in different growth conditions. INO80 and NuA4 histone acetyltransferase complexes share the protein Arp4 [
22], which may account for the interaction between INO80 and histone acetylation.
Isw2, one ATP-dependent chromatin remodeling enzyme, is involved in gene activation and repression [
23]. We found that there is no interaction between Isw2 alone and histone modification. We thus reasoned that the nine interactions between Isw2 & Ume6 and histone modifications are mainly attributable to transcription factor Ume6. To test this possibility, we examined the transcriptional effect of Ume6 deletion on the nine associated modification cohorts [
24]. Seven of the nine cohorts (except H3K56 and H4Nterm acetylation) show significant changes (
P < 0.05, K-S test) in gene expression upon the deletion of Ume6.
Taken together, we validated our approach to demonstrate that it accurately predicts experimentally determined interactions between ATP-dependent chromatin remodeling and histone modifications (see Additional file
2). In addition to the known interactions, our approach also uncovers many new interactions between these two activities, giving the first global landscape in yeast.
On the other hand, we analyzed the number of remodelers to which each histone modification is linked. Interestingly, only H3, H4 and H2A.Z modifications display interactions with chromatin remodeling. As a previous analysis has indicated that H3–H4 tetramers are ~20 times more stable than H2A-H2B dimers [
25], more ATP-dependent chromatin remodelers should be required to modulate H3–H4 tetramers. We also calculated the Pearson correlation coefficient between transcriptional activity [
26] and modification enrichment at promoters for each histone modification (Figure ). Our result shows that the modifications, which display higher positive correlation with transcriptional activity, tend to work with more ATP-dependent chromatin remodelers. This result suggests that the regulation of chromatin structure at active promoters involves more chromatin remodelers. Recruitment of more chromatin remodelers is expected to make nucleosomes more dynamic. As expected, the cohort promoters for modifications working with at least four chromatin remodelers have significantly higher rates of histone H3 turnover [
27] than those for modifications working with at most one chromatin remodeler (
P < 10
-92, Mann-Whitney U-test).
Mechanisms of the interactions between the two activities
We sought to understand the mechanisms of how ATP-dependent chromatin remodeling interacts with histone modifications. One possible mechanism is through interactions between chromatin remodeling complexes and histone-modifying complexes [
28,
29]. Another mechanism involves links between histone residues and remodelers [
30]. We examined the genome-wide prevalence of these two mechanisms. To test the first possibility, we first determined histone-modifying enzymes for each histone modification using modification cohorts of genes and expression profiles accompanying the perturbation of histone acetyltransferases (HATs) and methyltransferases (HMTs) [
13]. If the histone-modifying enzyme directs the histone modification, its perturbation should cause a differential change in expression of modification cohort genes. As the first strategy above for detecting remodeling-modification interaction, we used the K-S statistical test to derive histone-modifying enzymes for each histone modification (
P < 0.05 and positive regulation, after Bonferroni correction for multiple testing, see Additional file
2). For each remodeling-modification interaction identified above, we determined interaction state between chromatin remodeler and histone-modifying enzymes of the histone modification using a general repository of experimentally determined protein-protein interactions [
31]. Chromatin remodelers in most remodeling-modification interactions (44 of 61) are shown to interact with at least one histone-modifying enzyme of their connected modifications (see Additional file
2).
We next examined the role of links between histone residues and remodelers in remodeling-modification interactions. Previous studies have measured genome-wide expression when some specific histone residues (H3K4, H3K27, H3N-terminal, H4K8, H4K12, H4K16, and H4N-terminal) were mutated [
32,
33]. If the histone residue interacts with the remodeler, its mutation should cause a significantly different change in gene expression between the remodeler cohort genes and the rest of the genes. We used the K-S statistical test as above on the remodeler cohorts of genes and expression profiles accompanying the mutation of histone residues. We restricted analysis to the remodeling-modification interactions whose expression profiles accompanying the mutation of the corresponding histone residues are available. Remodeler cohort genes in ~94% of remodeling-modification interactions show significantly different expression changes (
P < 0.01) accompanying the mutation of the corresponding histone residues (see Additional file
2). This result suggests that histone residues play important roles in remodeling-modification interactions.
Taken together, we showed that most remodeling-modification interactions act via interactions of remodelers with both histone-modifying enzymes and histone residues. The prevalent dual interactions of remodelers also validate our landscape since our remodeling-modification interactions are not trained on interactions of remodelers with both histone-modifying enzymes and histone residues at all. As the two modes of interaction are not mutually exclusive, they might together guarantee the proper interactions between remodeling and modifications.
The effect of cooperativity between remodeling and modifications
We investigated into the effects of cooperativity between remodeling and modifications versus independent remodeling or modifications on genome-wide properties. To this end, we first identified three gene cohorts of independent modification, independent remodeling, and both modification and remodeling. As mentioned above, genes belong to one modification cohort if they display significantly high levels of the corresponding modification at promoters. Genes belong to regulated cohort for one chromatin remodeler if they display significantly different changes in gene expression accompanying the perturbation of the remodeler. For each of the 61 significant remodeler-modification pairs, the cohort of independent modification includes genes that belong to the corresponding modification cohort but not any remodeling cohort. The cohort of independent remodeling includes genes that belong to the corresponding remodeling cohort but not any modification cohort. The cohort of both modification and remodeling is the intersection between the corresponding modification cohort and the corresponding remodeling cohort. We compiled the three types of cohorts from all identified remodeling-modification interactions, respectively. This yields three sets of genes: modification-independent cohort, remodeling-independent cohort, and modification and remodeling cohort (see Additional file
4). By applying these strict criteria we ensured a low level of false positives for the three distinct cohorts (see Additional file
5). In the following analysis, we focused on the three cohorts.
We first analyzed the three gene cohorts in terms of nucleosome occupancy. Recent studies have measured high-resolution nucleosome occupancy across the yeast genome [
34,
35]. These valuable data allow for a direct examination of the effect of different activities on nucleosome occupancy. Modification and remodeling cohort promoters have significantly lower nucleosome occupancy [
34] than the other two cohorts (Figure ). It is known that genomic DNA sequence is an important determinant of nucleosome positioning [
36]. However, there is no significant difference in sequence preferences for nucleosomes [
34,
36] among the three cohorts (data not shown), indicating that the differences in nucleosome occupancy among the three cohorts are not due to the differences in sequence preferences for nucleosomes. These results imply that a combination of ATP-dependent chromatin remodeling and histone modifications causes lower nucleosome occupancy.
We next sought to understand why modification and remodeling cohort promoters have significantly lower nucleosome occupancy than remodeling-independent cohorts. Histone modifications could cooperate with transcription factors (TFs) to regulate DNA-templated processes [
10] and TFs could compete with nucleosomes for occupancy along the genome [
36]. We asked whether TF binding information contributes to the significant difference in nucleosome occupancy. Indeed, modification and remodeling cohort promoters are highly enriched with TF binding sites [
37] compared with the other two cohorts (Figure ). Moreover, binding sites are highly localized in linker DNA [
34] at modification and remodeling cohort promoters and modification-independent cohort promoters (Figure ). We asked whether histone modifications facilitate TF binding or occur as a consequence of TF binding. Experiments on individual TFs and genes revealed that TFs recruit HATs for specific acetylation [
38]. The generation of genome-wide expression profiles that correspond to the deletion of various TFs [
24] allows us to address this question on a genome scale. If the TF recruits the HAT, its deletion should cause a differential change in expression of the corresponding acetylation cohort as histone acetylation is thought to play an important role in modulation of gene expression [
39]. We performed the K-S statistical test on the TF-acetylation associations identified in a previous study [
38]. We found only ~47% of pairs whose acetylation cohort show significant change (
P < 0.01) in gene expression accompanying the deletion of the corresponding TF compared with the rest of the genes (see Additional file
6). One possible explanation for this observation is that multiple TFs may work in a redundant fashion, providing robustness to the regulatory system. Another explanation is that the recruitment of HATs by TFs may not be a universal mechanism of the relationship between TFs and histone acetylation. The study mentioned above has showed that H3K14 and H3K18 acetylation levels decrease in TF hir3 mutant at the YDR224C promoter, and H3K18 acetylation level decreases in TF yml081W mutant at the YDR525W promoter [
38]. Our identified significant TF-acetylation pairs include Hir3-H3K18 acetylation and YML081W-H3K18 acetylation pairs. However, we found no significantly different change in expression between the H3K14 acetylation cohort and the rest of the genes when Hir3 was deleted. Indeed, the deletion of Hir3 causes significant decrease (> 2-fold) in expression level of YDR224C. This result indicates that the recruitment of H3K14 acetyltransferases by Hir3 may be specific for individual genes.
Analysis of gene activity revealed that modification and remodeling cohort genes have significantly higher transcription rates and absolute mRNA abundance [
26], whereas of the other two cohort genes are comparable to genome-wide levels (Figure ). This result is consistent with the general observation that the level of nucleosome occupancy in promoter is inversely proportional to the corresponding gene transcription rate [
40]. Furthermore, modification and remodeling cohort promoters have significantly higher rates of histone H3 turnover [
27] than those of the other two cohort promoters (Figure ). We further analyzed gene activity in various conditions for the three cohorts. We compiled gene expression data from 1,082 published microarray experiments under various cellular conditions. For each gene, we calculated the proportion of experiments in which it displayed significantly up-regulated expression changes, and defined the normalized resulting value as openING rate. The openING rate reflects the general gene activity in various conditions. Modification and remodeling cohort genes also show significantly higher openING rates (Figure ), indicating that the higher gene activities of modification and remodeling cohort genes are conserved among various conditions.
We examined whether there is significant difference in other properties among the three cohorts. Modification and remodeling cohort and remodeling-independent cohort promoters are highly enriched with TATA boxes [
41], whereas modification-independent cohort promoters are depleted of TATA boxes (Figure ). A previous study has shown that nucleosome-inhibited genes tend to have TATA boxes [
41]. Chromatin remodeling is thus required to overcome the nucleosomal barrier. H2A.Z nucleosomes help to establish nucleosome-free region (NFR) directly upstream of the transcription start site (TSS) in yeast [
42]. Modification and remodeling cohort and modification-independent cohort promoters are depleted of H2A.Z nucleosomes [
42] (Figure ). Experimental evidence has also shown that H2A.Z nucleosomes tend to appear in hypoacetylation regions [
43]. Interestingly, modification and remodeling cohort and modification-independent cohort promoters also have significantly lower RNAP II occupancy [
44] at the 200 bp upstream regions (Figure ). Our result suggests a potential link between histone modifications and depletion of RNAP II. We further examined whether any of the three cohorts were enriched with genes that are annotated by Gene Ontology (see Additional file
7). Modification-independent cohort is highly enriched with genes that are involved in RNA-related processes. Remodeling-independent cohort genes tend to participate in biogenesis processes. Modification and remodeling cohort genes tend to play housekeeping roles.
1 Qian Xiang,1 Jihua Feng,1 Jiang Wang,1 Yangyang Deng,1 and Caisheng He1
