Gene regulation differs greatly between related species, constituting a major source of phenotypic diversity. Recent studies characterized extensive differences in the gene expression programs of closely related species. In contrast, virtually nothing is known about the evolution of chromatin structure and how it influences the divergence of gene expression. Here, we compare the genome‐wide nucleosome positioning of two closely related yeast species and, by profiling their inter‐specific hybrid, trace the genetic basis of the observed differences into mutations affecting the local DNA sequences (cis effects) or the upstream regulators (trans effects). The majority (∼70%) of inter‐species differences is due to cis effects, leaving a significant contribution (30%) for trans factors. We show that cis effects are well explained by mutations in nucleosome‐disfavoring AT‐rich sequences, but are not associated with divergence of nucleosome‐favoring sequences. Differences in nucleosome positioning propagate to multiple adjacent nucleosomes, supporting the statistical positioning hypothesis, and we provide evidence that nucleosome‐free regions, but not the +1 nucleosome, serve as stable border elements. Surprisingly, although we find that differential nucleosome positioning among cell types is strongly correlated with differential expression, this does not seem to be the case for evolutionary changes: divergence of nucleosome positioning is excluded from regulatory elements and is not correlated with gene expression divergence, suggesting a primarily neutral mode of evolution. Our results provide evolutionary insights to the genetic determinants and regulatory function of nucleosome positioning.
Phenotypic diversity is often due to changes in gene regulation, and recent studies have characterized extensive differences between the gene expression programs of closely related species (Khaitovich et al, 2006; Tirosh et al, 2009). However, very little is known about the mechanisms that drive this divergence. Here, we analyze the evolution of nucleosome positioning, by comparing the patterns of nucleosomes between two yeast species, as well as generating the allele‐specific nucleosome profile in their hybrid. We ask two main questions: (1) what is the genetic basis of inter‐species differences in nucleosome positioning? and (2) what is the regulatory function of these differences?
Generally speaking, we can classify the genetic basis of the divergence in nucleosome positioning into two mechanisms. First, mutations in the local DNA sequence may influence the ability to bind nucleosomes at this region; we refer to these as cis effects. Second, mutations may affect the activity of various proteins that alter nucleosome positioning either actively (e.g. chromatin‐remodeling enzymes) or by simply competing with nucleosomes for binding to the same DNA sequence (e.g. transcription factors); we refer to these as trans effects.
To classify the observed inter‐species differences into cis versus trans effects, we measured allele‐specific nucleosome positions within the inter‐specific hybrid of the two species (Wittkopp et al, 2004; Tirosh et al, 2009). The hybrid contains the alleles of both species; hence, cis effects, which involve mutations that discriminate between the two alleles, will be maintained in the hybrid so that nucleosome positioning will be different between the alleles coming from the different species. Trans effects, in contrast, will not discriminate between the two hybrid alleles from the different species, as these two alleles reside together at the same trans environment (hybrid nucleus) and are thus regulated by the same set of proteins—the combination of proteins from the two species. Using this approach, we found that ∼70% of the inter‐species differences in nucleosome positioning are due to cis effects, whereas the rest is due to trans effects.
The local DNA sequence is indeed known to affect nucleosome positions, and many features of DNA sequences were proposed to influence nucleosome binding, either by rejecting nucleosomes, or by being favorable for nucleosome binding (Segal et al, 2006; Lee et al, 2007; Kaplan et al, 2009). We find, however, that nucleosome positions diverged primarily through changes in AT‐rich sequences, which exclude nucleosomes, whereas mutations in sequences that correlate with high‐nucleosome occupancy do not influence inter‐species divergence.
Nucleosomes restrict the access of proteins to the DNA and may thus affect DNA‐related processes such as transcription, recombination or replication. Indeed, promoters and regulatory sequences are often depleted of nucleosomes, and highly transcribed genes are associated with low occupancy of nucleosomes at their promoters (Lee et al, 2007). Several earlier studies also suggested that evolutionary divergence of gene expression is driven by changes in chromatin structure (Lee et al, 2006; Choi and Kim, 2008; Tirosh et al, 2008; Field et al, 2009). However, we find that nucleosome positions (or occupancy) at regulatory elements are largely conserved, and furthermore, that the inter‐species differences in nucleosome positions do not correlate with gene expression differences. These results suggest that nucleosome positioning is not a central mechanism for evolutionary changes in gene regulation and that most of the observed changes may be due to neutral drift.
Does the apparent low influence of nucleosome positioning on gene expression divergence implies that nucleosome positions do not have a function in gene regulation? To address this, we examined two additional modes of gene regulation: transcriptional response to changes in growth conditions (glucose versus glycerol media), and the expression differences between different cell types (haploid versus diploid cells). Consistent with earlier studies, we found that the response to growth conditions is significantly, albeit weakly, associated with changes in nucleosome positioning. Interestingly, we also found a strikingly strong association between gene expression and nucleosomal changes in the two cell types. Taken together, these results suggest that nucleosome positioning is used preferentially for biological processes in which genes are turned on and off (e.g. different cell type), but less so during divergence of closely related species in which gradual changes accumulate over time.
Inter‐species hybrids can be used to dissect the relative contribution of cis and trans effects to the evolution of nucleosome positioning. Most (∼70%) differences in nucleosome positioning between two closely related yeast species are due to cis effects.
Cis effects are primarily due to divergence of AT‐rich nucleosome‐disfavoring sequences, but are not associated with divergence of nucleosome‐favoring sequences.
Differences in nucleosome positioning propagate to multiple adjacent nucleosomes, supporting the statistical positioning hypothesis.
Divergence of nucleosome positioning is excluded from regulatory elements and is not correlated with gene expression divergence, suggesting a neutral mode of evolution.
The large phenotypic differences between closely related species suggest that gene regulation is a major driving force of phenotypic diversity (King and Wilson, 1975). Widespread differences in gene expression were indeed identified when species with an almost identical set of genes were compared (Rifkin et al, 2003; Khaitovich et al, 2006; Tirosh et al, 2009). The genetic basis of most inter‐species variations, however, remains largely unknown. In particular, only a small fraction of the observed differences could be traced to sequence divergence at transcription‐factor binding sites (TFBSs) (Zhang et al, 2004; Tirosh et al, 2008). A compelling hypothesis is that expression divergence is due, at least partially, to changes in the chromatin environment and in particular the positioning of nucleosomes (Lee et al, 2006; Choi and Kim, 2008; Field et al, 2009).
The precise positioning of nucleosomes, in particular along gene promoters, can influence gene expression by restricting the accessibility of regulatory proteins for binding DNA (Yuan et al, 2005; Li et al, 2007). Consequently, promoter nucleosome occupancy was shown to correlate with the levels of gene expression (Lee et al, 2004, 2007; Liu et al, 2005; Pokholok et al, 2005; Hogan et al, 2006; Shivaswamy et al, 2008; Field et al, 2009; Zawadzki et al, 2009). The hypothesis that nucleosome positioning contributes also to the divergence of gene expression received some support from several analyses. In one study, inter‐species divergence of gene expression was correlated with predicted changes in promoter nucleosome occupancy (Tirosh et al, 2008). An association study reported that a large portion of inter‐strain expression variations are linked to a small number of chromatin regulators (Lee et al, 2006; Choi and Kim, 2008). Finally, differential expression of respiration genes in two distant yeast species was associated with differential nucleosome occupancy (Ihmels et al, 2005; Field et al, 2009). Yet, despite this potential role of nucleosomes, efforts to understand the genetic basis of regulatory evolution have focused on the divergence of TFBSs (Borneman et al, 2007; Doniger and Fay, 2007; Odom et al, 2007; Wray, 2007), whereas virtually nothing is known about the divergence of nucleosome positioning.
In this study, we used high‐throughput sequencing to map nucleosome positioning in two yeast species, Saccharomyces cerevisiae and Sacchormyces paradoxus, as well as their inter‐species hybrid. This data enabled us to characterize the divergence of nucleosome positioning between these closely related species, analyze the extent by which these changes can explain the evolutionary divergence of gene expression, and evaluate the relative contribution of different mechanisms that influence nucleosome positioning.
We mapped the genome‐wide nucleosome positioning of the budding yeast S. cerevisiae and its close relative S. paradoxus (∼85% genome identity) by subjecting genomic DNA to MNase digestion, followed by high‐throughput sequencing (Figure 1A). For each species, we profiled a wild‐type strain and, in addition, five mutant strains each deleted of a key chromatin regulator (Figure 1B). The influence of these chromatin regulators on nucleosome positioning will be discussed elsewhere, whereas here we focus on the inter‐species comparison of nucleosome positioning. Analyzing ∼4000 aligned promoters and coding regions, we identified reliable inter‐species differences at ∼10% of the nucleosomes (Figure 1C; Supplementary Figures S1 and S2). Differences that appear to be due to MNase digestion bias were excluded from further analysis (Supplementary Figure S1) and the remaining differences were classified into three classes (see Materials and methods): nucleosomes whose occupancy was changed by at least twofold (∼2000 cases), nucleosomes that were shifted in their positions (∼300 cases), and nucleosomes that were completely lost or gained in one of the species (∼170 cases).
Distinguishing cis versus trans effects on differential nucleosome positioning
Nucleosome positioning is determined in cis by the local DNA sequences (Ioshikhes et al, 2006; Segal et al, 2006; Kaplan et al, 2009), and in trans through the activity of various factors such as chromatin‐remodeling enzymes and the transcriptional machinery (Whitehouse et al, 2007; Hartley and Madhani, 2009; Kaplan et al, 2009; Zhang et al, 2009; Weiner et al, 2010). The relative contribution of cis and trans factors in determining nucleosome positioning is debated (Kaplan et al, 2009; Zhang et al, 2009). To classify the inter‐specific differences in nucleosome positioning into those generated by mutations in cis or in trans, we mapped the allele‐specific nucleosome positioning in the inter‐species hybrid (Wittkopp et al, 2004; Tirosh et al, 2009). Within the hybrid, two orthologous alleles are regulated by the same trans environment; differences between the alleles must therefore be due to cis effects. In contrast, inter‐species differences that disappear in the hybrid reflect trans effects (Figure 2A and B). This analysis classified ∼70% of the inter‐species differences as cis (a similar difference observed also between the hybrid alleles) and the remaining ∼30% as trans (differences were not observed between the hybrid alleles) (Figure 2C).
Cis effects depend on the local DNA sequence and are thus expected to be relatively independent of genetic or environmental perturbations. In contrast, trans effects depend on protein activity, which is more likely to vary when conditions are changed. Predicted cis effects were indeed consistent between wild type and mutant backgrounds, whereas trans effects were variable (Supplementary Figure S3). Notably, only few of the trans effects could be attributed to one of the five regulators examined, suggesting a complex regulation by multiple factors (Supplementary Figure S3). A compelling hypothesis is that trans divergence is dominated by differences in environmental sensing and signal transduction, as we recently demonstrated for gene expression divergence (Tirosh et al, 2009).
To further examine the genetic basis for nucleosome divergence, we used a recent estimation for the contribution of each DNA 5‐mer to nucleosome binding (Field et al, 2008). Most cis‐dependent nucleosome gains/losses and changes in occupancy were correctly predicted by this model (Figure 2D; Supplementary Figure S4), whereas the model had no predictive power for trans‐dependent nucleosome gains/losses. Notably, the predictive power of the model was due almost exclusively to the nucleosome‐disfavoring property of AT‐rich sequences, particularly poly(dA/dT) tracts (Iyer and Struhl, 1995; Segal and Widom, 2009) (Figure 2E; Supplementary Figure S4). In contrast, changes in other 5‐mers that code for nucleosome‐free DNA, including the binding sites for the RSC nucleosome‐remodeling complex (Badis et al, 2008) did not contribute significantly to nucleosome divergence. Moreover, changes in sequences implicated as having a nucleosome‐favoring activity were only weakly associated with diverged positions. Similarly, changes in dinucleotide patterns that are also implicated in the control of nucleosome positioning (Segal et al, 2006; Field et al, 2008) were not predictive of nucleosome gains/losses or reduced occupancy (Supplementary Figure S5).
Shift propagation through statistical nucleosome positioning
In contrast to the success of the sequence‐based model in predicting nucleosome gains/losses or occupancy changes, it performed quite poorly in predicting shifts in nucleosome positioning (Supplementary Figure S4). Specific cases suggested that shifts propagate through several nucleosomes, thus explaining the poor predictive power of local DNA sequences (Figure 3A and B; Supplementary Figure S6). Shift propagation is indeed predicted by models of statistical positioning (Kornberg and Stryer, 1988; Mavrich et al, 2008; Zhang et al, 2009), which assumes that nucleosomes bind DNA mostly at random, but an apparent ordering is generated by some fixed border that creates a statistical preference for specific arrangements of adjacent nucleosomes. A shift in the position of the border elements will disrupt the preferred arrangement of multiple adjacent nucleosomes.
To more systematically examine the propagation of shifts, we focused on nucleosomes whose position shifted to a large extent between the species, and examined the flanking nucleosomes. Shifts propagated for approximately four flanking nucleosomes, decreasing in size from a maximum of ∼50 to ∼20 bps in the nearest nucleosomes. Propagation of shifts was similar for nucleosomes in both directions (upstream and downstream), indicating that it is not due to steric hindrance (Supplementary Figure S7). Notably, although sequence‐based analysis had some power to predict shifts of the central nucleosomes (largest shift), it had no power to predict shifts in adjacent upstream (U) or downstream (D) nucleosomes (Figure 3C), supporting the notion that this propagation is indeed due to statistical positioning rather than changes in local sequences.
Shifts propagated away from nucleosome‐free regions (NFRs) but did not propagate across NFRs (Figure 3D). For example, shifts of the −1 nucleosome typically propagate further upstream to the −2 and −3 nucleosomes but not through the NFR to the +1 and +2 nucleosomes (Figure 3A). In contrast, shifts adjacent to the +1 nucleosome, that is at the +2 nucleosome or at a −1 nucleosome in genes without NFR, did propagate across this nucleosome, even for +1 nucleosomes associated with strong nucleosome‐favoring sequences (Figure 3D; Supplementary Figure S7). Thus, NFRs, but not +1 nucleosomes, seem to serve as stable border elements that establish the statistical positioning of adjacent nucleosomes.
Divergence of nucleosome positioning at regulatory elements
Our analysis so far focused on the genetic basis of nucleosome divergence. Next, we wished to examine the functional consequences of this divergence. Recent studies proposed that chromatin structure has an important role in generating gene expression divergence between closely related species (Lee et al, 2006; Choi and Kim, 2008), but this suggestion was not examined experimentally. Specific cases supported this proposal. For example, the −1 nucleosome of ERG26 covers two conserved TFBSs in S. cerevisiae but is absent in S. paradoxus (Figure 4A). This nucleosomal difference is also observed within the hybrid, and is associated with increased AT‐richness in S. paradoxus (Figure 4A), suggesting a cis difference through exclusion of this nucleosome only in S. paradoxus. Consistent with the increased accessibility of the two S. paradoxus binding sites, expression of ERG26 is ∼2‐fold higher in S. paradoxus than in S. cerevisiae, and this expression difference is also observed within the hybrid alleles, indicating a cis effect (Figure 4A).
We performed a systematic analysis of nucleosome positioning at TFBSs (MacIsaac et al, 2006), where differences are expected to be most influential. Although nucleosome changes are enriched at promoters, they are significantly underrepresented at conserved TFBSs (Figure 4B), and their occurrence gradually increases away from the positions of TFBSs (Figure 4C). Hence, divergence of nucleosome positioning is generally excluded from functional elements, which might indicate that it is mostly non‐functional.
At positions of TFBSs that diverged between the species, differences in promoter nucleosome occupancy were more abundant than in positions of conserved TFBSs, but less abundant than in promoter positions without TFBSs. Divergence in nucleosome occupancy and in the sequences of TFBSs were generally not correlated (Supplementary Figure S8). However, at some promoters, we find that nucleosomal changes are coupled to the divergence of TFBSs, such that one species has a strong binding site that is covered by nucleosome whereas the other species has a much weaker binding site that is not covered by nucleosome (Figure 4D). These suggest compensatory evolution of nucleosome positioning and TFBSs, and thus also in these cases we expect that differences in nucleosome positioning did not drive expression divergence.
Divergence of nucleosome positioning does not explain differential expression
To directly examine whether inter‐species divergence in nucleosome positioning can explain the divergence of gene expression, we used gene expression data generated for the same strains under the same conditions (Tirosh et al, 2009). This analysis showed that nucleosomal changes are generally not associated with inter‐species expression changes, consistent with their exclusion from TFBSs. First, genes that are associated with diverged nucleosome positions are not more likely to diverge in expression (Supplementary Figure S9). We also considered the genes whose overall nucleosome occupancy along the promoter (or the ORF) diverged between the species, but these genes did not diverge in expression more than genes with conserved nucleosome occupancy (Supplementary Figure S9). Second, we found that genes that are differentially expressed are not more likely to diverge in nucleosome positioning (Figure 5A). Furthermore, the average nucleosome occupancy of genes with either higher or lower expression in S. cerevisiae, compared with S. paradoxus, was indistinguishable between the promoters, coding regions and 3′‐UTRs of the two species (Figure 5B; Supplementary Figure S9). Finally, these results persisted when we tried to distinguish between genes that are activated or repressed by promoter regulatory elements (Supplementary Figure S9). Thus, our data indicate that most evolutionary changes in nucleosome positioning did not affect expression levels and are thus more likely to reflect neutral drift.
This lack of correlation between the divergence of nucleosome positioning and that of gene expression contrasts the situation within the same species: recent studies found significant, although modest, correlations between changes in nucleosome positioning and changes in gene expression when comparing S. cerevisiae at different growth conditions (Shivaswamy et al, 2008; Zawadzki et al, 2009) or cell‐cycle stages (Hogan et al, 2006). We observed a similarly significant correlation when S. cerevisiae was transferred from growth on glucose to glycerol (Figure 5C and D).
Interestingly, in contrast to the lack of correlation between differential positioning and expression observed for the two species and the weak correlation observed for different growth conditions, we found a striking correlation for comparison of haploid and diploid cells. Nucleosome positioning in haploids differed from that of diploids in only a minority of the genes, but these were highly enriched (P<10−16) with haploid‐specific genes (Figure 5E; Supplementary Figure S10). In fact, most (10 of 18) genes annotated as haploid specific (Galgoczy et al, 2004) displayed dramatic differences in nucleosome positioning between haploids and diploids, in a manner that was mostly correlated with the gene expression differences. For example, the mating gene STE4 has two promoter nucleosomes that cover multiple binding sites, which are specifically evicted in haploids, partially through the activity of ISW1 (Figure 5F). Furthermore, the +1 nucleosome is shifted downstream in the haploids thus exposing the transcription‐start site (Figure 5F).
Comparison of nucleosome positioning between two closely related yeast species allowed us to examine different models of nucleosome positioning from an evolutionary perspective. First, by analyzing the allele‐specific nucleosome profiles in a yeast hybrid, we estimated that the majority (∼70%) of the inter‐species differences in nucleosome patterns are caused by changes in the local DNA sequences (cis). Second, we find that cis differences that involve nucleosome gain/loss, as well as changes in occupancy are well explained by local divergence of AT‐rich sequences, which are known to be disfavored for nucleosome binding (Iyer and Struhl, 1995; Segal and Widom, 2009). In contrast, sequences that are enriched at positions of bound nucleosomes have a marginal effect on the divergence pattern, suggesting that their enrichment might not be causal for nucleosome binding. Third, the pattern of shift propagation that we observe supports a significant role of statistical nucleosome positioning. Our results are consistent with a model in which nucleosome positioning is primarily determined by exclusion at specific regions such as NFRs and statistical positioning at adjacent sequences (Kornberg and Stryer, 1988; Mavrich et al, 2008).
Earlier studies have reported significant, albeit weak, correlations between changes in nucleosome positioning and gene expression in S. cerevisiae (Lee et al, 2004, 2007; Liu et al, 2005; Pokholok et al, 2005; Hogan et al, 2006; Shivaswamy et al, 2008; Field et al, 2009; Zawadzki et al, 2009), and we found a similarly weak correlation following changes in carbon source. In contrast, we describe a strong association for ploidy‐dependent changes, in which most haploid‐specific genes have dramatic changes in nucleosome positioning. These results may indicate that chromatin remodeling is used preferentially to prevent transcription of genes that are switched off (e.g. haploid‐specific genes in a diploid), rather than fine‐tune the levels of expressed genes. Consistent with this possibility, nucleosome remodeling by ISW2 has recently been shown to suppress transcription from cryptic sites (Whitehouse et al, 2007).
In contrast to differential positioning within S. cerevisiae, differences between the two species are, overall, not correlated with divergence of gene expression and are excluded from regulatory elements. These results suggest that transcription regulation is robust to many variations in nucleosome positioning, as long as they do not influence critical regulatory elements, and therefore that these neutral variations could accumulate during evolution with little influence of natural selection.
The central role of AT‐rich sequences in excluding nucleosomes, the statistical positioning of nucleosomes, and the low correlation between changes in nucleosome positioning and in gene expression are consistent with earlier analyses in S. cerevisiae as well as in other organisms (Ozsolak et al, 2007; Field et al, 2008; Mavrich et al, 2008; Zawadzki et al, 2009). However, these might not be universally conserved, as recent reports suggest that in the fission yeast, Schizosaccharomyces pombe, AT‐rich sequences are not enriched at NFRs, inter‐nucleosomal distances are smaller (∼154 bps) and the correlation with gene expression is higher than in S. cerevisiae (Kristell et al, 2010; Lantermann et al, 2010). It would be interesting to examine how our results would generalize to other organisms such as S. pombe and multicellular organisms.
Materials and methods
Yeast strains and growth conditions
S. cerevisiae (BY4743), S. paradoxus (CBS 432 ho∷nat MATα) and their hybrid (BY4741 X CBS432 ho∷nat MATα) were grown to a log‐phase growth at 30°C at rich media (YPD medium) or glycerol (YPglycerol). For each of the species and the hybrids, we profiled nucleosome positioning in the wild‐type backgrounds and in five mutant strains (only the first three for the hybrid) deleted of the following chromatin regulators: ISW1, HTZ1, GCN5, BRE1 and RAD6.
High‐throughput sequencing of mono‐nucleosomal DNA fragments
Mono‐nucleosomes were isolated from the two species (or from the hybrids) by standard protocols, pooled, and subjected to Illumina high‐throughput sequencing (see Supplementary Information for more details). Reads that were unambiguously mapped to only one of the species were retained for further analysis. Note that our analysis is restricted to ∼87% of the genomic positions that differ between the species, which could induce a slight over‐estimation of cis effects (see Supplementary Information). As reads are mapped to the ends of ∼150 bp fragments, we converted the mapped positions into the center positions of the ∼150 bp fragments, by adding or subtracting (based on the direction of each read) half the length of a typical fragment. This fragment length was estimated for each sample as the median distance between consecutive peaks of read density and was between 135 and 160 for all samples.
Estimation of nucleosome occupancy and positions
Nucleosome occupancy was generated by averaging the number of reads over windows of 150 bps. In addition, we defined nucleosome scores by smoothing the same data with a Gaussian filter with windows of 50 bps and s.d. of 25 bps. Nucleosome center positions were identified as peaks of nucleosome scores, with the lowest 10% of the peaks excluded. The raw data and normalized nucleosome scores are available at the GEO database (Accession number GSE18939).
Comparison of nucleosome positioning
At each aligned region of the two species, we identified nucleosomes that are gained/lost, shifted in position or changed in occupancy using strict criteria. These include at least twofold difference for occupancy changes, minimal number of reads (8) for gains/losses and shifts of at least 30 bp and statistical significance; in addition, each difference was evaluated based only on reads that mapped to the Watson strand and, separately, based only on reads that mapped to the Crick strand and only differences that were consistent in both analyses were retained (see Supplementary Information for full details).
Control experiments with naked DNA
To control for MNase and other biases, we conducted similar experiments with naked DNA from the two species and the hybrid (Supplementary Figure S1). Inter‐species differences that were also observed in the control experiment were excluded from further analysis, as they might originate from MNase bias and not from evolutionary differences. This analysis also eliminates other systematic errors that generate differential occupancy in the two species, such as sequencing errors or alignment errors. Note that this control excludes ∼30% of the nucleosomes from our analysis, which could bias our results. Our analysis was also restricted to genomic positions that differ between the species (to allow species‐specific read mapping, as described above), to genes that have one‐to‐one orthology relationships and complete sequence alignments (Kellis et al, 2003), and was focused on promoters and 2 kb of each gene. Relaxing these criteria did not have a major effect on our main conclusions (Supplementary Figure S11).
Classification to cis versus trans
The extent of differences between the two species (Δparents) was compared with that between the two hybrid alleles (Δhybrid). We expect to have Δhybrid approximately equal to Δparents in cis differences (i.e. ∣Δhybrid–Δparents∣ is low) and approximately zero in trans differences. We therefore classified cis changes as those where ∣Δhybrid–Δparents∣<∣Δhybrid∣ and trans changes as those where ∣Δhybrid–Δparents∣>∣Δhybrid∣. Intermediate cases, in which the difference between these values is small, were excluded from this analysis (see Supplementary Method).
Prediction by sequence models
To examine how much of the differences could be predicted from sequence analysis, nucleosome occupancy at aligned positions of the two species were predicted using a method described earlier (Field et al, 2008), which is primarily based on the contribution of each 5‐mer to nucleosome positioning. The predictive power of this model was evaluated by comparing the differences in predicted nucleosome occupancy at locations of observed nucleosomal changes to the differences at randomly selected positions (control regions). For each difference in scores, we examined the frequency of control regions that have higher difference (false positive rate) and the frequency of nucleosome changes with higher score difference (true positive rate). For positions of shifts, the prediction score was defined as:
(Ocer,cer−Opar,cer)+(Opar,par−Ocer,par), where Oi,j is the predicted occupancy of species i at the location of the nucleosome of species j. We performed similar analysis with other sequence‐based models and got similar results (not shown). Individual 5‐mers were defined as nucleosome‐enriched or linker‐enriched if they have at least a twofold enrichment in the wild‐type S. cerevisiae sample.
Carbon source‐dependent changes in nucleosome positioning
Nucleosome positioning was measured for the wild‐type strains of the two species (pooled) both in glucose and glycerol and processed as described above. We compared nucleosome positioning within S. cerevisiae among the two conditions by focusing on ∼4000 S. cerevisiae genes where inter‐species expression differences were measured earlier (Tirosh et al, 2009) and where genomic differences allowed the species‐specific mapping of at least 100 reads to the S. cerevisiae alleles. This analysis was not affected by MNase bias, as we are comparing the same species in different conditions. We compared the average nucleosome occupancy for genes upregulated or downregulated at glycerol (compared with glucose) by at least 1.5‐fold, based on earlier microarray analysis (Tirosh et al, 2009). Similar results were obtained when we examined S. paradoxus or when we used different cutoffs for differential expression (not shown).
Ploidy‐dependent changes in nucleosome positioning
To identify the differences in nucleosome positioning between haploids and diploids, we compared the nucleosome scores of four haploid samples from each species (WT, Δhtz1, Δgcn5 and Δisw1) to the corresponding alleles of the four hybrid dipolids (WT, Δhtz1, Δgcn5 and Δisw1), or to WT diploids of each species (see below). The haploid strains of each species were compared only with the corresponding alleles in the hybrid (the alleles of the same species). A measure of haploid–diploid nucleosome difference was defined as the Euclidean distance between nucleosome scores at −300 to +200, averaged across the four different sample pairs and normalized by the average Euclidean distance between sample pairs from the same group (haploid–haploid or diploid–diploid). This analysis was performed separately for S. cerevisiae (a versus a/α) and S. paradoxus (α versus a/α), and the average value of the two species was used for further analysis to identify all haploid–diploid differences, including those specific to a or to α cells. Similar results were obtained when we avoided the use of hybrid diploids and compared only WT haploids with the WT diploids of each species (not shown); we present the data from comparison of haploids to hybrid diploids (Figure 5E), as it is based on more samples and higher coverage than the analysis of the WT diploids. Half of the genes with the lowest haploid–diploid difference were considered as having similar nucleosome positioning, and the 28 genes with the highest haploid–diploid difference were considered as having different nucleosome positioning, as this number gave the most significant enrichment of haploid‐specific genes (Galgoczy et al, 2004). Differences in nucleosome positioning were compared with differences in gene expression, which we previously measured for haploids and diploids of the two species (S. cerevisiae a versus a/α and S. paradoxus α versus a/α) (Tirosh et al, 2009).
This work was supported by the Helen and Martin Kimmel Award for Innovative Investigations, the Crown Human Genome Center, the Israeli Ministry of Science and the European Research Council (Ideas).
Conflict of Interest
The authors declare that they have no conflict of interest.
Supplementary Methods, Supplementary Figures S1–S11
This is an open‐access article distributed under the terms of the Creative Commons Attribution License, which permits distribution, and reproduction in any medium, provided the original author and source are credited. This license does not permit commercial exploitation without specific permission.
- Copyright © 2010 EMBO and Macmillan Publishers Limited