|Home | About | Journals | Submit | Contact Us | Français|
Despite technical advances, the future of chromatin mapping studies requires an ability to draw accurate comparisons between different chromatin states to enhance our understanding of genome biology. In this study, we used matched chromatin preparations to enable specific and accurate comparisons of Saccharomyces cerevisiae chromatin structures in the presence and absence of the co-repressor protein Tup1. Analysis of wild-type and tup1 Δ chromatin data sets revealed unique organizational themes relating to the function of Tup1. Regulatory regions bound by Tup1 assumed a distinct chromatin architecture composed of a wide nucleosome-depleted region, low occupancy/poorly positioned promoter nucleosomes, a larger number and wider distribution of transcription factor-binding sites and downstream genes with enhanced transcription plasticity. Regions of Tup1-dependent chromatin structure were defined for the first time across the entire yeast genome and are shown to strongly overlap with activity of the chromatin remodeler Isw2. Additionally, Tup1-dependent chromatin structures are shown to relate to distinct biological processes and transcriptional states of regulated genes, including Tup1 stabilization of Minus 1 and Minus 2 promoter nucleosomes at actively repressed genes. Together these results help to enhance our mechanistic understanding of Tup1 regulation of chromatin structure and gene expression.
The organization of eukaryotic DNA into chromatin is known to have far-reaching implications on the accessibility and functionality of genetic information stored in the nucleotide sequence (1–3). Eukaryotic genomes are packaged through the wrapping of 147-bp segments of DNA around histone proteins arranged in octamers known as nucleosomes (4,5). Multiple isoforms of the histone proteins can combine to form a number of distinct nucleosome octamers (6–8). Additionally, histones can undergo an assortment of covalent modifications to further diversify the chromatin landscape (9).
The coiling of each 147-bp DNA segment around a nucleosome spans approximately 1.7 helical turns and establishes important topological relationships between the nucleotide sequence and its local chromatin environment (5,10). Nucleosome positioning relative to underlying DNA sequences controls the access of regulatory factors to their genomic-binding locations and helps to coordinate expression programs in vivo (11–13). Consequently, the strength of each nucleosome's interaction with its associated DNA sequence can influence gene expression (14,15).
In yeast and other eukaryotes, nucleosome positioning is controlled by a combination of inputs. Rigid poly(A) tracts of DNA serve as nucleosome-excluding sequences and help direct the formation of nucleosome-depleted regions (NDRs) both in vivo and in vitro (15–19). Chromatin architecture is also believed to be influenced by statistical packing principles whereby well-positioned nucleosomes help organize the regular positioning of nucleosomes in adjacent large stretches of DNA (20–22). In addition to sequence-based considerations and packing rules, a number of protein complexes can actively position nucleosomes to regulate chromatin structure and transcription (2,23–25). The Tup1–Ssn6 corepressor in Saccharomyces cerevisiae is one such complex, soliciting several chromatin-mediated mechanisms to regulate gene expression in vivo (26).
In yeast, Tup1–Ssn6 is a global repressor of transcription, responsible for the repression of more than 180 genes involved in diverse signaling pathways (27,28). Tup1–Ssn6 was one of the first corepressor complexes to be identified (29), and has since served as a model for similarly structured corepressor proteins in other eukaryotes, including the Drosophila Groucho (Gro) protein, and the mammalian transducin β-like/transducin β-like related (TBL/TBLR) proteins and transducin-like enhancer of split (TLE) proteins (26,27,30). The Tup1–Ssn6 corepressor complex does not bind directly to DNA, but instead is recruited to promoters by sequence-specific DNA-binding proteins which coordinate the expression of specific subsets of genes (27). Upon recruitment to promoters, Tup1–Ssn6 is known to repress downstream genes through several mechanisms, including interfering with the recruitment of transcription machinery, histone deacetylase (HDAC) recruitment, and the establishment of nucleosome positioning (26). Chromatin-dependent mechanisms of Tup1 repression (nucleosome positioning and HDAC recruitment) are not mutually exclusive and redundant repression mechanisms are observed at many Tup1-regulated promoters (31). Moreover, unequal sensitivities of Tup1-regulated genes to the inactivation of different repression pathways in vivo suggests that responses to Tup1 repression may vary in a gene-specific fashion, whereby different groups of genes have evolved different strategies for utilizing Tup1–Ssn6 repression mechanisms. Genes that have developed a strong reliance on Tup1-dependent nucleosome positioning for repression in vivo have been well characterized by site-specific studies. Of the more than 100 genes regulated by Tup1 only a few have been characterized in detail including RNR3 (32,33), FLO1 (34), ANB1 (35), SUC2 (36,37) and other a-cell-specific genes (38–40). Tup1-dependent nucleosome positioning at the promoter regions of these genes is believed to repress transcription by limiting the accessibility of promoter elements to trans-acting factors (26).
In this study, we defined the genomic landscape for Tup1's regulation of chromatin structure, mapping chromatin in both wild-type and tup1Δ cells using a combination of matched micrococcal nuclease (MNase) digestions and high-throughput DNA sequencing methods. We identified Tup1-specific alterations in chromatin architecture at 96 genes across the genome and determined that Tup1 stabilized the positioning of the −1 and −2 promoter nucleosomes of these genes when Tup1 is actively repressing transcription. We determined that Tup1 regulates chromatin organization at the majority of its targets by cooperation with the ATP-dependent chromatin remodeler Isw2. In addition, we also identified distinct chromatin architecture at Tup1-bound promoters consisting of a wide NDR with low occupancy/poorly positioned promoter nucleosomes, and a larger number and wider distribution of transcription factor-binding sites (TFBS). These distinctive Tup1-bound promoters regulate genes which have high transcription plasticity, suggesting a functional role for the Tup1–Ssn6 complex in regulating genes with varied expression.
Wild-type (BY4741) and tup1Δ (BY4741/tup1Δ::KanMX) strains (Open Biosystems) were used for nucleosome mapping experiments. Tup1 deletion was confirmed by PCR analysis of genomic DNA. Chromatin mapping experiments sampled from the genomes of yeast strains growing at 30°C in exponential phase (0.6–0.8 OD600), as described previously (41).
To compare low-resolution (0.5–1kb) ChIP-chip experiments of Hanlon et al. (42) to our high-resolution nucleosome mapping data, ChIP-chip data were converted into a continuous binding profile. Both continuous binding (log2 ratio) and P-value profiles were generated by ChIPOTle's sliding window approach (43). The probability of Tup1 binding to any gene's regulatory region was assigned as the minimal P-value for the ChIPOTle windows overlapping that region. In this approach, every gene is assigned a P-value for Tup1 binding. Regulatory regions with a Bonferroni corrected P<1×10−6 and mock uncorrected P>0.001 in any growth condition were designated as strongly bound. Weakly bound regions were also selected as regions with a Boniforroni corrected P<1×10−3 and mock uncorrected P>0.001 (Table 1).
Preparation of chromatin DNA samples for sequencing was conducted as described previously (16,25,44), with several key revisions (Supplementary Figure S1A). Cross-linked chromatin–DNA complexes were extracted by mechanical disruption (bead beating; 4×1min sessions, 2min on ice between sessions). MNase titrations were carried out as described previously (25); however, an added level of standardization was achieved by quantifying the input of extracted chromatin loaded into each digest reaction; 1mg total protein from whole-cell extracts per 200µl digest reaction. Total protein concentration for whole cell extracts (chromatin extracts) was quantified using a Bradford assay (OD595). Following MNase treatment and cross-link reversal, DNA was phenol:chloroform precipitated, resuspended in 50µl TE pH 8.0, and RNase treated. Small aliquots (20%, ~10µl) from each digest titration were then analyzed by gel electrophoresis (Supplementary Figure S1B). Gel intensity measurements for each lane were calculated using standard densitometry software provided by Biorad (Quantity OneTM) and exported to Microsoft ExcelTM for correlation analysis (Pearson). Correlation coefficients were calculated for digest titration levels no longer showing a visible di-nucleosome band (i.e. optimally digested samples) and spanning a region known to cover a size range of 0–400bp of DNA (Supplementary Figure S1C). A 100-bp standard ladder (Qiagen) was analyzed on the same gel to calculate a relative front (i.e. separation relative to standard fragment sizes) for the migrating DNA population. ‘Matched’ digests were selected as two optimally digested samples having a correlation coefficient (r>0.9). Correlation coefficients r>0.9 were selected because these showed the most reproducible change in MNase protection data on single site real-time quantitative PCR (RT-qPCR) analysis (See Supplementary Methods section). The remaining sample from each optimal digest was column purified (Zymo Research), bypassing the need to gel excise mono-nucleosome DNA. Avoidance of gel extraction to isolate chromatin DNA at this stage is a key step to assure sampling of similar chromatin populations. Matched DNA samples were prepared for sequencing using the standard Illumina protocol with additional care at the gel purification step to ensure that the same size range of DNA was selected.
Nucleosome DNA was sequenced by an Illumina Genome Analyzer IIx, as described previously (16,25,45). Sequencing reads were aligned to the March 2010 build of the S. cerevisiae genome (www.yeastgenome.org) using the ELAND algorithm, allowing only two possible mismatches. MNase protection was calculated for sequencing data and all comparison data sets by extending aligned reads by 120bp, dividing by the average read count per base pair and converting this ratio into continuous space (log2 ratio). Each data set was then standardized at single base pair resolution to have a mean occupancy of 0 and a standard deviation of 1. Individual nucleosome calls were made using the template filtering algorithm with default parameters, the seven standard defined templates and a minimum and maximum allowable nucleosome width of 80 and 220bp, respectively (25).
Yeast transcription start sites (TSSs) were taken from David et al. (46). Significant differences for Tup1-bound and unbound promoters between wild-type and tup1Δ average MNase protection plots were tested at 10-bp resolution for regulatory regions spanning TSS (−1000 to +200bp) using a two-tailed t-test. The resulting P-values were corrected for multiple comparisons by conservative Bonferroni correction. Tup1-bound promoters were sorted by downstream expression changes following Tup1 deletion using data from Green et al. (28). Genes with decreased downstream gene expression following Tup1 deletion were selected as being in the bottom 15% (15th percentile) of genes expressed in tup1Δ. Likewise, genes with increased downstream gene expression following Tup1 deletion were selected as being in the top 15% (85th percentile) of genes expressed in tup1Δ.
Nucleosomes called by template filtering were annotated to their closest promoter according to distance from TSS. The closest nucleosome upstream of each TSS, dyad within 500bp, was annotated as −1, with the second closest as −2, and third as −3. Similar annotations were made for downstream +1, +2 and +3 nucleosomes, respectively, for each promoter. Comparisons between called nucleosomes from unbound (Class NB) versus bound (Class SBU) promoters were made using 10000 permutation tests. Permutation testing randomly sampled the unbound promoter nucleosome data set for the parameters under examination. A similar analysis was conducted when comparing bound/wild-type versus bound/tup1Δ promoter nucleosomes.
To dissect the influence of Tup1 on chromatin structure in vivo, we employed a technical approach which minimizes the technical variation typically associated with MNase digestion steps utilized by genome-wide chromatin DNA preparations (see ‘Materials and Methods’ section; Supplementary Figure S1A) (25). Our approach allowed us to systematically select similarly digested chromatin preparations (i.e. matched MNase digestions) to control for differences in nuclease digestion extent and collect highly comparable chromatin populations. Matched digests were collected, similar to other chromatin-mapping protocols, but with the addition of several key steps (See ‘Materials and Methods’ section) to help ensure greater consistency of DNA sampling. Overall nuclease digestion profiles were compared by analyzing chromatin DNA populations (i.e. DNA remaining following MNase treatment) separated by gel electrophoresis. Matched samples were selected as having a highly correlated (r>0.9) distribution of chromatin DNA population size and abundance, as identified by ethidium bromide DNA population staining intensities (Supplementary Figure S1B and S1C). Matched samples were then deeply sequenced in parallel using an Illumina Genome Analyzer IIx.
Sequencing of matched chromatin samples generated an extensive sequencing data set for the two strains with 20.6 and 23.7 million mapped sequencing reads for wild-type (BY4741) and tup1Δ genomes, respectively. Given that the yeast genome is packaged by an estimated 70000 nucleosomes (47), our data represents greater than 250× average coverage for nucleosomal DNA. To our knowledge, the magnitude of this coverage provides the most extensive mononucleosome sequencing data set available for S. cerevisiae to date. Sequencing data was processed in three ways depending on the downstream analysis employed. First, we examined the ends of the sequence reads on individual strands (Figure 1A). This representation of the data is completely unprocessed and is an exact representation of the raw aligned sequencing output. Second, we extended each 36-bp sequence read to a total length of 120bp. Although nucleosomes are 147bp, we found that extrapolation to 120bp provided better demarcation of nucleosome ends than did extrapolation to 147bp with no effect on occupancy comparisons. We then counted the number of times each unique position in the genome was included within a 120bp extended sequence and standardized between experiments to a total of 1 million sequence reads, to correct for variable sequencing coverage (Figure 1B). Lastly, extended read counts were converted to a ‘MNase protection’ score by dividing by the average tag count per base pair, and converting into continuous data-space by calculating a log2 ratio (Figure 1C), similar to previous analysis (16,45). We prefer to use the term ‘MNase protection’ to describe this data rather than ‘nucleosome occupancy’ because MNase protection may alternatively be conferred by non-nucleosomal protein complexes. MNase protection reflects the relative level of DNA protection from MNase digestion and at most locations represent the occupancy/density of nucleosomes at each genomic location in relation to the entire genome sequenced in that chromatin preparation (48).
To test the homology between our two MNase protection data sets, a Pearson correlation coefficient was calculated comparing wild-type and tup1Δ chromatin maps across the entire genome. As expected, the correlations between our matched samples was extremely high (r=0.934). To compare this homology to other data sets, the same correlation analysis was repeated for other published chromatin data sets of wild-type yeast of the same background genotype also grown in rich media. Across eight experiments from two different groups, only two experiments showed a correlation similar to ours (Supplementary Table S1). Interestingly, our matched samples, derived from different strains, were more similar than the majority of published replicates from the same strain. This result highlights how differences in technical preparation can introduce deviations into chromatin maps and how our matched preparation helps to control this influence.
To better visualize local versus global homology between our chromatin data sets, we calculated correlation coefficients and average changes in MNase protection for all 1-kb windows covering the yeast genome. As expected from our almost identical data sets, >90% of 1-kb windows had a correlation >0.9 and the majority of windows showed little change in average MNase protection (90% were within ±0.1) (Supplementary Figure S2). Overall, results from both approaches (correlation and average change) indicate that only a small fraction of the genome’s chromatin changes in the absence of Tup1.
As an alternative approach to assess the similarity of our samples, we utilized a computational algorithm for characterizing nucleosome DNA signals from high-throughput DNA sequencing methods, known as template filtering (25). We employed this approach with the seven available templates to determine locations and characteristics for all nucleosomes in both data sets, mapping 57469 and 57327 nucleosomes each for wild-type and tup1Δ chromatin, respectively (Supplementary Table S2). Importantly, analysis of nucleosomes called by template filtering confirmed that wild-type and tup1Δ chromatin samples were of similar size populations and had similar template frequencies, as expected of a matched digest (Supplementary Figure S3).
Importantly, our sequencing results also recapitulated results of published site-specific studies of nucleosome positioning at the PHO5 promoter (Supplementary Figure S4) (13). Nucleosomes at the PHO5 promoter have not been shown to be influenced by Tup1 regulatory mechanisms and, as expected, homologous levels of MNase protection are seen at this locus (between wild-type and tup1Δ chromatin) compared to known Tup1-regulated regions including FLO1, SUC2 and RNR3, which showed significant dissimilarity that was specific to the loss of Tup1 (See next section).
Having successfully prepared highly correlated chromatin data sets, we hypothesized that the high degree of homology between our samples would enhance our ability of detecting Tup1-specific differences in chromatin structure, when comparing our wild-type and tup1Δ data sets. To test this hypothesis, we explored the relationship between Tup1 binding and chromatin structure differences between our data sets by integrating our sequencing data with a recently published ChIP-chip data set on Tup1 binding (42). Specific genomic fragments were scored for Tup1 binding with a P-value and/or Tup1 occupancy (binding) score (see ‘Materials and Methods’ section). Depending on the analysis, 100-bp genomic segments or 1-kb regulatory regions were scored. A gene's regulatory region was defined as extending 1-kb upstream of the open reading frame, and each such regulatory region was classified with respect to Tup1 binding by P-value (Table 1). Each bound regulatory region was further classified as bidirectional or unidirectional depending on if it was upstream of two divergently transcribed genes (list of all regulatory regions available in Supplementary Table S3). For the majority of our analysis, we compared regulatory regions classified as not bound by Tup1 (NB) to strongly bound unidirectional (SBU) and/or weakly bound unidirectional (WBU) (Table 1). Importantly, integration of genome-wide Tup1-binding data with our chromatin sequencing data sets revealed that our approach accurately reproduced the results of many high-resolution single-site studies on the perturbation of chromatin by Tup1. For example, at the well-studied Tup1-regulated locus, FLO1, peaks of Tup1 enrichment aligned with close proximity to regions of chromatin dissimilarity between matched-digest wild-type and tup1Δ samples, consistent with the effects of a known direct-acting Tup1–Ssn6 repression mechanism at that location (Figure 1). Moreover, previously mapped differences in FLO1 chromatin structure between wild-type and ssn6Δ strains were captured by our analysis with high fidelity and sensitivity, including changes in nucleosome occupancy and positioning directly upstream from transcription start (−1000 to −1bp) and long-range differences (−5000 to −4000bp) (34). In addition to FLO1, the well-documented role of Tup1 in establishing nucleosome positioning and occupancy at the SUC2 and RNR3 promoters was also evident in our data (Supplementary Figure S5 and S6) (32,33).
To examine our MNase protection data genome-wide in relation to Tup1 binding, we plotted the MNase protection profiles (i.e. nucleosome occupancy) for both data sets against each other for consecutive non-overlapping 100-bp windows covering the entire yeast genome (Figure 2A). As expected from the high correlation coefficient for our data sets, the plot illustrated clearly that the majority of the genome's 100-bp segments maintained similar MNase protection values regardless of the presence of Tup1 (R2=0.86). Interestingly, however, regions with dissimilar MNase protection profiles between wild-type and tup1Δ data sets occurred in a directional fashion, whereby decreased nuclease protection was only seen for tup1Δ data relative to wild-type. The loss of MNase protection (and underlying nucleosome occupancy) when Tup1 is deleted is consistent with Tup1–Ssn6's role in stabilizing nucleosomes in vivo (26) (41). The specificity of this result can be discerned when plotting only regions of the genome containing the lowest Tup1-binding scores (Bottom 10%) and comparing to regions with the highest Tup1-binding scores (Top 10%) (Figure 2B and C, respectively). When the bottom 10% of Tup1-binding sites are plotted there is an absence of locations showing a loss in MNase protection when Tup1 is deleted, and the similarity between chromatin data sets improves (R2=0.92). On the other hand, when examining the top 10% of Tup1-binding sites there is a large directional bulge and the chromatin data sets show increased dissimilarity (R2=0.72). These data demonstrate that differences in MNase protection between our chromatin data sets, across the yeast genome, are specifically related to neighboring Tup1 binding and occur in a directional fashion consistent with Tup1's role in stabilizing nucleosome occupancy. It is important to remember that the Tup1 ChIP-chip data were obtained at low resolution (0.5–1kb), and many of the sites with high Tup1-binding scores would be caused by proximity to true binding locations. Given this difference in resolution, it is likely that the specificity of the plot in Figure 2C is underestimated.
As an alternative approach to investigating the specificity of Tup1 binding to chromatin structure differences between our data sets, regions of chromatin dissimilarity (1-kb windows or larger) were classified as either having a poor correlation between MNase protection data (r<0.5) or a large average change in MNase protection across a given region of interest (tupΔ-wt<−0.5). The average Tup1-binding score was calculated for these dissimilar regions and compared to a random sampling of similar-sized windows across the yeast genome (Figure 2D and E). Overall, regions of high dissimilarity in chromatin structure were enriched for Tup1 binding to a significantly greater extent than randomly sampled regions, indicating that high Tup1 enrichment was a specific characteristic of dissimilar chromatin regions between our data sets, again consistent with Tup1's role in modifying chromatin structure (26). Interestingly, the specificity of Tup1 enrichment to chromatin dissimilarity between wild-type and tup1Δ data sets was eroded when unmatched wild-type chromatin preparations were instead compared to our tup1Δ data (Supplementary Figure S7). The decrease in average Tup1 enrichment seen at regions of dissimilarity between wild-type and tup1Δ chromatin structures when unmatched chromatin samples are compared highlights the utility of collecting highly correlated chromatin DNA populations.
Comparing wild-type and tup1Δ chromatin maps allowed us to dissect the genomic landscape of Tup1's influence on chromatin structure. In total, the dissimilar windows identified by analysis for Figure 2D and E cover <4% of the budding yeast genome sequence, consistent with Tup1's presumed role modifying select regions of chromatin, rather than more broadly perturbing the genome at-large. Of dissimilar regions, 78% of dissimilar regions were identified by both correlation and average change analysis, suggesting that the loss of Tup1 regulation influences both nucleosome occupancy and positioning in vivo.
Localization of Tup1–Ssn6 is known to vary from gene to gene and its influence on chromatin structure has been documented at both intergenic regions and overlying coding sequences (26). Of the 238 unidirectional regulatory regions bound strongly by Tup1 (Class SBU) in our experiments, 92 (38.7%) were identified as having at least 500bp of dissimilarity (by correlation, average change, or both) between wild-type and tup1Δ with 31 (13%) coding regions identified as being dissimilar. In total, 14 of the 31 dissimilar coding regions identified were also associated with a dissimilar regulatory region, indicating that Tup1's activity on chromatin structure is predominantly based at promoter regions, a feature seen throughout our analysis.
To assess differences in promoter packaging within and between our chromatin data sets, genes were aligned by TSS and averaged MNase protection was plotted at 10-bp resolution for a 1-kb window up and downstream of the TSS, similar to previous studies (16,45). As expected, average aligned MNase protection data were extremely similar between wild-type and tup1Δ strains for all yeast TSS overall (Figure 3A). Like typical yeast promoters, both wild-type and tup1Δ averaged MNase protection (nucleosome occupancy) plots displayed a NDR directly upstream of the TSS flanked by well-positioned nucleosomes, especially over the coding region. The same characteristic pattern of MNase protection was also seen for all promoters not bound by Tup1 (Figure 3B). However, when promoters bound by Tup1 were examined alone, a markedly different pattern of chromatin structure was seen. As illustrated by Figure 3C, genes with promoters bound by Tup1 (Class SBU) exhibited an increased NDR width upstream of the TSS, and a significantly different pattern of nucleosome occupancy and positioning surrounded that region compared to unbound (Class NB) promoters (Figure 3B). The unique chromatin structure profile associated with Tup1-occupied promoters was not a product of our data normalization and transformation procedures, since the same result was seen when our data was viewed in non-continuous count space (Supplementary Figure SF8).
As an alternate analysis of promoter structures, we also analyzed our chromatin data by individually calling promoter nucleosomes using template filtering (25). This analysis revealed that the −2, −1, +1, +2 and +3 promoter nucleosomes all showed significantly lower nucleosome occupancy values and significantly more deviance in positioning at Tup1-bound promoters, compared to their unbound counterparts (Figure 3D). As suggested by the broader NDR, the internucleosomal distance between the −1 and +1 nucleosomes was significantly larger (40-bp increase; P=0.0001). Additionally, a significantly increased inter-nucleosomal distance between the −1 and −2 nucleosomes (29-bp increase, P=0.0029) was also evident compared to unbound promoters. Distances between +1, +2 and +3 nucleosomes remained unchanged.
In addition to examining our MNase protection data in relation to Tup1 binding, we also preformed unsupervised clustering to categorize promoters across the genome in relation to our two data sets (Supplementary Figure S9). K-medoids clustering with values of K>3 consistently produced a discrete cluster with a wide NDR region, similar to that seen for Tup1-bound promoters in Figure 3C. Hypergeometric distribution analysis for this wide-NDR cluster consistently revealed an enrichment for Tup1-bound promoters and also enriched for promoters with a GO Slim annotation for genes involved in translation. These results suggest a possible link between the wide NDR structure we are seeing, Tup1 binding, and specific biological pathways.
Having characterized a distinct chromatin structure associated with Tup1-bound promoters, we next sought to identify additional mechanisms that could best explain the observed pattern of MNase protection (and underlying nucleosome positioning/ occupancy) at these loci. Previous work has shown that both TFBS locations and active (occupied) binding sites in vivo are preferentially enriched at the characteristic NDR region upstream from each promoter’s TSS, consistent with the model that nucleosome occupancy has a strong influence on transcription factor (TF) occupancy (49–53). Given these characteristic distributions and their relationship to chromatin structure at promoter regions, we tested whether differences in TFBS locations underscored the distinct chromatin structure profiles associated with Tup1-bound and unbound promoters. Using the locations of highly conserved TFBSs identified by MacIsacc et al. (51), we plotted the distribution of conserved TFBS locations in relation to downstream-coding regions (ATG) and overlying chromatin data (Figure 4A and B). As illustrated by Figure 4A and B, promoters not bound by Tup1 showed a characteristic distribution of TFBS locations, with locations correlating to overlying occupancy profiles and significantly enriching at the characteristic NDR region (−10- to −200-bp upstream of ATG) upstream of coding regions, similar to previous analysis (49). In contrast, Tup1-bound promoters showed a wider distribution of conserved TFBS locations, lacking the peak of enrichment of TFBS locations at the characteristic NDR region (−100 to −200bp). Conserved TFBS locations at Tup1 promoters were instead more spread out in their distribution across the promoter region, consistent with the distinct wide-NDR architecture of these genes. Importantly, the unique distribution of TFBS locations was also seen for conserved TFBSs active in the condition tested (bound in YPD), was observed when compared to wide unbound promoters, and was not dependent on promoter type (single versus double) or Tup1-binding class (Data not shown).
Knowing that sequence-specific DNA-binding TFs must compete with and/or interact with nucleosomes to bind DNA and regulate transcription, we hypothesized that an increased number of TFs and/or TFBSs could be present at Tup1-bound promoters, either causing or resulting from the unique chromatin architecture and underlying TFBS distributions we identified. To test this hypothesis, we calculated the average number of conserved TFBSs and TFs present for each promoter class in Table 1. As illustrated by Figure 4C and D, Tup1-bound promoters showed a significantly larger number of both conserved TFBSs and bound TFs compared to unbound promoters, on average. Importantly, significant differences were seen regardless of Tup1-binding class and an even larger difference was evident for both TFBS number and number of bound TFs when we compared Tup1-bound single promoters versus unbound single promoters (data not shown).
Having uncovered a distinct structure for Tup1-bound genes, we hypothesized that the wider NDR with the higher number and spreading of conserved TFBSs would allow Tup1-regulated genes to respond more readily to various environmental signals. Accordingly, we proposed that Tup1 binding would be associated with the capacity to modulate gene expression upon changing environmental conditions, i.e. transcriptional plasticity (54). To examine transcriptional plasticity we used an extensive microarray data set for yeast grown in 13 diverse environmental transitions (55) and calculated transcriptional plasticity as the standard deviation for a gene's expression across all growth conditions. As illustrated in Figure 5A, our analysis found that Tup1 binding was significantly correlated (P=4.0×10−55) to transcriptional plasticity. Importantly, we also tested the specificity of the relationship between Tup1 binding and downstream expression plasticity by conducting the same analysis for all regulatory factors bound in rich media. Interestingly, we found that the association of Tup1-binding with downstream plasticity is stronger than that seen for all other factors tested, implying that plastic gene expression is a specific property of Tup1-regulated genes (Figure 5B). Overall, the strong association of Tup1 binding with downstream expression plasticity suggests that the distinctive packaging of Tup1-bound promoters may reflect a shared ability to broadly alter gene expression programs via Tup1 regulation to coordinate appropriate responses to environmental signals.
Having identified several structural features unique to Tup1-bound promoters, we next sought to examine how the deletion of Tup1 influenced this chromatin structure, hoping to cast light onto the mechanism by which Tup1 regulates chromatin. Analysis of our TSS plot of Tup1-bound promoters (Figure 3C) revealed that deletion of Tup1 caused a decrease in MNase protection across the entire intergenic region with significant decreases at −1000 to −450-bp upstream of the TSS and directly upstream of the TSS; −70 to −10bp (gray bars; Bonferroni corrected P<0.05). Significant decreases in intergenic nuclease protection levels overlied the positions of the −2 and −1 promoter nucleosomes, upstream of the wider NDR, and suggest that Tup1-dependent nucleosome occupancy at these locations may play a key role in its regulatory mechanism. Additionally, when looking at individual nucleosomes called by template filtering, there was also a significant loss of called −1 and −2 promoter nucleosomes in the tup1Δ strain (Figure 3D).
Significant changes in the average MNase protection profile in Figure 3C may appear slight because this plot encompasses all Tup1-bound promoters, including loci where Tup1 is not regulating through chromatin. Therefore, to test the reproducibility of changes in MNase protection driving the significant differences identified in Figure 3C, changes in MNase protection and underlying histone H3 and H4 occupancy were assayed using RT-qPCR on three additional biological replicates at fifteen selected sites. As illustrated by Supplementary Figures SF10–SF12, strongly dissimilar regions between our data sets (r<0.5; YMR319C, YLR295C, YBR157C) showed large reproducible changes in MNase protection on RT-qPCR analysis and weakly dissimilar regions (0.5<r<0.7; YIL056W) showed smaller changes also with high consistency. Importantly, changes in MNase protection were not seen at the well-characterized PHO5 locus, an expected result since this locus is not regulated by Tup1 (Supplementary Figure S10). Additionally, significant changes in nucleosome occupancy were also shown to underlie changes in nuclease protection profiles at all dissimilar regions tested thus demonstrating that at the majority of sites changes in MNase protection are caused by changes in nucleosome occupancy (Supplementary Figures S10–S12). Given these results, we were confident that strongly dissimilarity regions defined by an r<0.5 were specific and reproducible, therefore, the remainder of our analysis focused on strongly dissimilar regions between our two data sets.
As an alternative approach to investigate Tup1 stabilization of promoter nucleosomes in relation to transcriptional regulation, we again performed an analysis of TSS-aligned data for Tup1-bound promoters, but additionally sorted the chromatin data into three groupings according to downstream expression in tup1Δ strains (28) (Figure 6A–C). Genes repressed by Tup1 and subsequently expressed (de-repressed) upon its deletion, again exhibited a similar decreases in MNase protection across the regulatory region, significantly at the location of the −2 nucleosome (Figure 6B, gray bars; Bonferroni corrected P<0.05). Interestingly, an opposing result was obtained at genes where Tup1 may be playing a role in activation of expression (i.e. repressed in tup1Δ). In place of losing nuclease protection at activated promoters of these regions, we instead saw a slight increase in protection over the NDR region (Figure 6C), indicating that Tup1's influence on the chromatin at these regions is different from that at genes it actively represses. Additionally, when examining genes with expression profiles unchanged by Tup1 deletion in the condition being examined (rich medium; Figure 6A), no change in chromatin structure was seen, suggesting that the differences we see in chromatin structure for Tup1-bound genes are due to its role in repressing transcription.
In addition to sorting MNase protection data for Tup1-bound promoters by downstream expression, we clustered change in MNase protection profiles (tup1Δ—wt) for all Tup1-bound promoters for a 1-kb window surrounding TSS (Figure 6D). Clustering of bound promoters yielded only two significantly distinct patterns of change in MNase protection, regardless of the number of clusters generated. Cluster 1 had a broad increase in MNase protection across Tup1-bound promoters and consistently enriched for genes with associated GO slim process of translation, the same functional term associated with Tup1-bound promoters in our initial clustering analysis (Supplementary Figure S9). The broad increase in MNase protection seen at these promoters also parallels the increase seen for genes believed to be activated by Tup1 (Figure 6C). Cluster 2 showed a broad decrease in MNase protection across promoter regions and significantly enriched for downstream genes de-repressed in a tup1Δ strain, similar to the result in Figure 6B. The same changes in structural patterns for Tup1-bound genes were also seen when we clustered change in MNase protection data for all available TSSs (including unbound promoters). As illustrated in Supplementary Figure S13, broad increases or decreases in MNase protection are again seen for wide-NDR promoters that enrich for Tup1 binding (Clusters 1 and 2). Similar to results in Figure 6D–F, Tup1-associated promoters with a broad increase in MNase protection following deletion of Tup1 again enrich for genes associated with translation and ribosome biogenesis and assembly. In contrast, Tup1-associated promoters with a broad decrease in nucleosome occupancy following deletion of Tup1 enrich for genes involved in response to chemical stimuli and unknown biological processes. Together, results from Figure 6, Supplementary Figures S9 and S13 illustrate how Tup1 can have different effects on chromatin structure, depending on both the associated biological processes and downstream expression events involved at the genes it regulates.
Chromatin-dependent mechanisms of Tup1 repression are believed to repress transcription by limiting the accessibility of promoter elements to trans-acting factors (26). To explore the relationship between Tup1 regulation, chromatin structure and trans-activating factors, we examined the chromatin structure surrounding all annotated TFBSs (51). The average change in MNase protection (tup1Δ-wt) for each TF was determined, and change-in-protection data (tup1Δ-wt) were aligned by TFBS center. To identify relationships among the TFs, TFBS change-in-protection data were sorted by the average change-in-protection for the middle 100-bp centering on the TFBS (Figure 7A). Three distinct patterns of chromatin structure emerged (Figure 7B–D) whereby the average change in MNase protection profiles increased, decreased or did not change following Tup1 deletion (tup1Δ–wt). Overall, the majority of chromatin at TFBSs was not altered by Tup1 deletion, consistent with the fact that Tup1 regulates just a small fraction of the genome.
TFBSs which are centered on locations where MNase protection was decreased, suggesting nucleosome loss in a tup1Δ, could represent TFs whose binding is blocked directly by Tup1-stabilized nucleosomes or blocked by Tup1 itself. Indeed, several of the TFs (Mig1, Sko1, Nrg1, Phd1, Sut1, Aft1, Cin5, Skn7, Yap6, Rcs1, Rox1) that showed a dramatic loss of MNase protection in the tup1Δ strain were known co-factors of the Tup1–Ssn6 complex. These TFs have been shown to physically interact with Tup1 or Ssn6 and are likely DNA-binding recruiting proteins for the Tup1–Ssn6 complex (42). Identification of these factors further confirms the specificity of our assay at detecting Tup1-specific changes in chromatin structure. Additionally, other TFs with TFBSs present at MNase protected regions dependent on Tup1 may represent still uncharacterized Tup1–Ssn6 cofactors or TFs whose binding may be regulated by Tup1. The general stress TFs Msn2 and Msn4 fall in this category and showed decreased nuclease protection (i.e. nucleosome loss) in tup1Δ. Interestingly, Msn2/Msn4 have been shown to regulate many of the genes required for the environmental stress response (56–58), and our findings suggest that their binding may by regulated by Tup1-dependent nucleosome occupancy/positioning.
Surprisingly, there were a few TFBSs located at regions showing an increase in MNase protection following Tup1 deletion, including Rap1, Fhl1, Sfp1 and Abf1. All four of these TFs target protein biosynthesis genes, some of the most highly expressed genes in rapid growing yeast. Since Tup1 has been shown by ChIP-chip analysis to bind at many protein biosynthesis genes, this change in nuclease protection could be directly related to Tup1 regulation (42). To test if this increase in MNase protection was a direct or indirect effect of Tup1 regulation, we examined all genes annotated with the protein biosynthesis GO-slim term and compared between Tup1-bound and unbound regulatory regions (Supplementary Figure S14). Protein biosynthesis genes showed an increase of similar magnitude in MNase protection in the NDR upstream of the TSS, regardless of Tup1-binding class, suggesting that the change in MNase protection at these sites was not directly caused by Tup1. While this increase in MNase protection appears indirect, the overall architecture of Tup1-bound protein biosynthesis genes was significantly different from that of unbound protein biosynthesis genes. As we demonstrated earlier for all Tup1-bound genes, the NDR for Tup1-bound protein biosynthesis genes was significantly wider than for unbound protein biosynthesis genes and promoter nucleosomes showed variable positioning. This result suggests that there is a unique subset of protein biosysnthesis genes which is regulated by Tup1.
Having identified distinct effects of Tup1 on chromatin structure that relate to the associated biological processes and expression state of regulated genes, we next asked if Tup1's effect on chromatin structure was also associated with the activity of other regulatory factors. Tup1–Ssn6 has been shown to require the ATP-dependent chromatin remodeling enzyme Isw2 to position nucleosomes for gene repression at the well-characterized RNR3 locus, both upstream of and across the RNR3-coding region (32,33). Given the well-established collaboration between Isw2 and Tup1 nucleosome-positioning activities in vivo at RNR3 and the highly similar effects that Tup1 and Isw2 yield on upstream nucleosome occupancy (59–61), we hypothesized that the association of Isw2-remodeling with Tup1-remodeling events extended beyond individual loci and represented a more frequent collaboration of remodeling activities across the genome. To test this hypothesis, association testing was performed comparing Isw2-remodeled genes defined by Whitehouse et al. and Tup1-remodeled promoters identified by our analysis. Only 5′ Isw2 remodeling events were considered for this analysis, since the regions of significant Tup1 remodeling, we identified were at the 5′-end of genes (i.e. within 0- to −1000-bp upstream of TSS, Figure 3C). As illustrated in Figure 8A, a greater proportion of Tup1-remodeled sites were also remodeled by Isw2 (52%), when compared to non-Tup1-remodeled genes (21%). Importantly, with a relaxed stringency for defining Tup1 remodeling (r<0.7) and inclusion of all Tup1-bound regulatory regions (SBU, WBU, SBB and WBB), there were a total of 96 Tup1 bound and remodeled genes of which 48 or 50% were also Isw2-remodeled. A significant association of Tup1 remodeling with Isw2 remodeling was also seen at unbound (Class NB) promoters and likely represent direct effects whose Tup1-binding state was not identified by previous low-resolution ChIP-chip experiments (42). Overall, the significant association of Tup1 remodeling with Isw2 remodeling indicates that collaborative actives of Tup1 and Isw2 seen at the RNR3 promoter extend beyond individual loci and represent a genome-wide regulatory collaboration.
Decreased histone acetylation has been shown to track with Tup1–Ssn6 localization (62,63) and it is believed that Tup1 may repress gene expression by creating a self-reinforcing state through its interactions with hypoacetylated histones and histone deacetylase complexes (HDACs) in vivo (26,63–66). We hypothesized that Tup1 remodeling could be associated with a distinct histone modification profile at remodeled promoters, reflecting potential collaborative activities of Tup1 chromatin regulation with HDAC activity to regulate transcription. To test this hypothesis, we aligned average chromatin modification data from Pokholok et al. (67) for a 1-kb window surrounding TSS and compared modification profiles by Tup1-binding class. Interestingly, no significant differences in overall chromatin modification profiles were evident when comparing Tup1-bound promoters to unbound promoters (Figure 8B). Instead, only when Tup1-bound promoters were separated into Tup1 repressed genes versus non-repressed was hypoacetylation shown to track with Tup1 targets which are actively being repressed (Figure 8B). Additionally, a similar modification profile was also seen for Tup1-bound promoters that were either remodeled by Tup1 or remodeled by Isw2 (Supplementary Figure S15). This result suggests that Tup1's interactions with hypoacetylated histones and HDACs in vivo is a feature of actively repressed Tup1-bound loci and not Tup1 localization alone. Additionally, that the distinct chromatin modification profiles associated with gene repression mirrored those of Tup1- and Isw2-remodeled loci, further suggests that collaboration of HDAC complexes and other chromatin modifiers is a general feature of active Tup1 repression through chromatin.
Genome-wide chromatin DNA isolation protocols typically utilize a combination of nuclease digestion and gel electrophoresis to isolate mono-nucleosomal DNA fragments (44). However, since MNase digestion does not release all nucleosomes at the same rate (68), the resulting mononucleosomal DNA population has a variable composition which depends largely on the extent of nuclease digestion (25). This variation presents a challenge of distinguishing genuine heterogeneity in nucleosome positions and occupancy/density between two chromatin populations from those created as an artifact of technical preparation, a dilemma increasingly encountered by chromatin biologists (69) and one made significantly more difficult when attempting comparisons between data sets from different laboratory groups, strains or organisms. Indeed, the brief comparison of genome-wide data sets in Supplementary Table S1 highlights the influence that subtle technical variations alone can exert between individual biological replicates from the same strain. While comparison of our tup1Δ data to samples from the Weiner et al. digestion series made in Supplementary Figure S7 represent intentional exaggerations of the influence that technical variation can yield on signal specificity, the low Tup1 enrichment (i.e. poor signal specificity) seen for all Weiner samples indicates that genome-wide data set correlations up to, and possibly exceeding, r=0.77 are insufficient to achieve specific comparisons of chromatin structures.
Ultimately, the future of chromatin mapping studies requires an ability to make sensitive and specific comparisons between different chromatin states, be that strains, cell-types or environmental conditions, etc. to enhance our understanding of genome biology. Our matched MNase digestion approach was designed to accomplish this goal by reducing the technical variation introduced by the MNase digestion extent and gel extraction steps common in genome-wide chromatin mapping protocols. The procedure was successful in generating nucleosome maps that were essentially identical for the wild-type and tup1Δ strains sequenced, except for Tup1-specific differences in chromatin structure (Figure 2). More importantly, the specific differences in MNase protection our analysis identified were highly reproducible and related to changes in underlying nucleosome occupancy (Supplementary Figures S10–S12). Together, this data underscores the utility of matching the extent of MNase digestion as an approach to reduce technical variance in chromatin preparations and to enable accurate and reproducible downstream comparisons of chromatin structures.
Comparison of TSS-aligned promoter MNase protection data indicates that Tup1's influence on chromatin structure is primarily localized at upstream regulatory regions, specifically those associated with the Minus 1 and Minus 2 promoter nucleosomes (Figure 3C and D). Analysis of gene expression data also suggests that Tup1 promotes increased MNase protection at upstream locations to facilitate the repression of gene expression (Figure 6B). Additionally, analysis of clustering data indicates that decreases in MNase protection in a tup1Δ are associated with stress response pathways whereas increases in MNase protection are instead associated with translation and ribosome biogenesis pathways (Figure 6D–F and Supplementary Figure S13). Consistent with this, regulatory factors associated with both stress response and protein biosynthesis pathways show distinct Tup1-dependent chromatin structures overlying their binding sites in vivo (Figure 7). The effects of Tup1 regulation on chromatin structure at protein biosynthesis genes appears to be indirect (Supplementary Figure S14), however, multiple lines of evidence are presented which indicate that Tup1 regulates a distinct subset of these genes, as the characteristic Tup1 promoter structure is present for only the Tup1-bound fraction of protein biosynthesis genes (Supplementary Figures S13 and S14).
Tup1 has also been shown to regulate TF-binding specificity previously, whereby a nucleosomal hair-trigger blocks the conditional binding of a TF until the appropriate environmental signal is presented (41). Therefore, we attempted to identify the most likely TFs that could respond to Tup1-regulation through chromatin by examining the change in MNase protection centered on TFBSs (Figure 7). While our approach appears promising and identified the stress induced factor Xbp1 and the general stress response factors Msn2 and Msn4, this analysis is limited because the majority of the data for the bound TFBS were obtained in experiments preformed during growth in rich media (50). Further experiments with the appropriate environment for conditional activity of suspected TFs will be required to identify conditional TFBSs whose binding may be regulated by Tup1-dependent nucleosome occupancy. Additionally, Tup1 regulation of binding through chromatin will need to be confirmed by alternate approaches. However, from this analysis, it is clear that Tup1-dependent nucleosome occupancy specifically overlies the binding sites of many stress response factors and its own recruiters, presenting another link between chromatin structure regulation by Tup1 and stress–response pathways.
Tup1–Ssn6-mediated repression is believed to operate through a multistep mechanism involving HDAC recruitment, interactions with hypoacetylated histones and transcription machinery, and cooperation with other chromatin remodeling complexes such as Isw2 (26). Combined mutations for genes encoding each of these factors have demonstrated larger effects on repression at individual loci in vivo, compared to single mutations, suggesting that cooperation between repressive factors enables separate, synergistic contributions to the regulation of genes (31). Moreover, unequal sensitivities of Tup1-regulated genes to the inactivation of different repression pathways in vivo also indicates that responses to Tup1 repression may vary in a gene-specific fashion, whereby different groups of genes have evolved different strategies for utilizing Tup1–Ssn6 repression mechanisms (28). Here, we show that a majority (52%) of Tup1-remodeled promoters are also Isw2-remodeled (Figure 8A), indicating that the majority of Tup1-remodeling may be caused by its interaction with Isw2. Results extend individual loci studies defining a Tup1 gene repression mechanism through Isw2-dependent nucleosome positioning (33,60) and demonstrate that this mechanism is functioning at many Tup1-regulated sites across the genome. Interestingly, an analysis of the association of site-specific Tup1 recruiting cofactors binding with Tup1-remodeling events demonstrated that nearly all Tup1 recruiting cofactors are associated with Tup1's activity on chromatin and that no single factor is solely responsible for directing Tup1's chromatin remodeling efforts. Additionally, the binding of Tup1, which follows the aforementioned recruiting efforts, is also not significantly predictive of which regions of chromatin Tup1 selects to remodel. Therefore, it is clear from our analysis that the specificity of Tup1 chromatin remodeling is not defined by Tup1 recruitment or binding alone and requires additional cellular inputs. Further examination, focusing on post-translational modifications of Tup1-associated cofactors and their corresponding signaling pathways should shed light on how certain sites in the genome can have their chromatin remodeled by Tup1 while other sites remain unaffected.
Despite the variable roles Tup1 can play in regulating gene expression, Tup1-bound genes assume a distinct promoter packaging, consisting of low nucleosome occupancy/poorly positioned nucleosomes and a significantly wider NDR region (Figure 3C). Both the low nucleosome occupancy and wide NDR seen at Tup1-bound promoters in our study may reflect an abundance of easily digested nucleosomes at these loci that were digested away by our chromatin preparation or could reflect a more open chromatin structure overall. Regardless of the possibilities, it is clear from our analysis that Tup1-bound genes are structured significantly differently than the rest of the yeast genome. This distinct promoter packaging also includes a wider distribution of conserved TF-binding sites and these loci are targeted by a larger number of trans-activating factors in vivo. Together, this data demonstrates that Tup1 regulation requires inputs from more regulatory factors at distinct promoter locations compared to other transcriptional regulatory mechanisms active in S. cerevisiae.
Overall, the characteristic chromatin signature of Tup1-regulated genes is consistent with previous studies of chromatin structure identifying distinct nucleosome occupancy signatures in relation to transcriptional processes (49), but what does this mean? Why does Tup1 selectively target open regulatory regions? From analysis in Figure 3C, we can see that Tup1 is not responsible for creating the distinct architecture at the genes it targets because deletion of Tup1 did not revert the Tup1-bound promoters to a narrower NDR promoter with increased nucleosome occupancy and improved positioning. Deletion of Tup1 instead caused the opposite effect, producing a significant decrease in MNase protection at the −1 and −2 promoter nucleosomes. Therefore, knowing that this distinctive chromatin structure is not reliant on Tup1 activity, we instead propose that Tup1 is targeted based on its unique ability to enhance downstream expression plasticity of these distinctly structured genes (Figure 5). We propose that the ability of Tup1 to bind to both active and repressed genes (26), the diverse DNA-binding partners solicited by Tup1 (42), the different regions of the Tup1–Ssn6 complex capable of interacting with DNA and protein partners (70), as well as the multiple mechanisms of transcriptional repression that Tup1 can elicit, all enable Tup1 to take advantage of these open promoters to promote downstream transcriptional plasticity. From our analysis, it seems likely that Tup1-dependent nucleosome occupancy is a key part of these transcriptional responses, as distinct changes in chromatin structure are seen at Tup1-regulated promoters depending on the associated gene expression program (Figure 6A–C), biological process (Figure 6D–F), or underlying regulatory factor-binding sites (Figure 7) involved.
Supplementary Data are available at NAR Online.
National Science Foundation (grant IIS1016929 to M.J.B.); PhRMA predoctoral fellowship in Informatics (to J.M.R.) Funding for open access charge: SUNY at Buffalo Department of Biochemistry.
Conflict of interest statement. None declared.