Thus, consistent with its role as a chromatin remodeler, deletion of ISW1 led to extensive changes in nucleosome positioning and occupancy. Notably, these effects were primarily within coding regions (Figure 3). First, most of the changes in nucleosome occupancy observed upon deletion of ISW1 were localized at nucleosomes +2 to +4 within the coding regions, and typically reduced nucleosome occupancy at this region (Figure 3a). Second, the positioning of nucleosomes at coding regions, but not at intergenic regions, became fuzzier upon deletion of ISW1 (Figure 3d, e). For example, only 25% of the reads at the HOL1 coding region mapped to within 20 bp of the estimated nucleosome positions in the Δisw1 strain, compared to 45 to 49% of the reads in each of the other strains (Figure 3d). Fuzziness increased in the Δisw1 strain for approximately 1,000 genes (Figure 3e), whereas decreased fuzziness was observed for only 44 genes (Figure S4 in Additional file 1). Similar results were obtained for S. paradoxus (Figure 3e) and for the hybrid (not shown).

Third, shifts of nucleosome positions were particularly enriched at the mid-coding region of genes (Figure 3b, c). Notably, the shifted positions, as observed in the Δisw1 strain, were typically more consistent with sequence-based predictions than the positions observed in the wild-type strain (Figure S5 in Additional file 1). This indicates that Isw1 normally slides nucleosomes into energetically less-favorable positions. Thus, the observed shifts most likely reflect the direct ATP-dependent remodeling activity of Isw1 2930, although we cannot exclude the possibility that some of these changes are due to indirect effects. We therefore focused our subsequent analysis on Isw1-dependent shifts in nucleosome positions. These shifts are widespread and are comparable in magnitude to those found upon RNA polymerase (PolII) inactivation (see below).

In principle, the enrichment of Isw1-dependent shifts at mid-coding regions could be explained by statistical positioning: if nucleosome positions are primarily determined by border elements positioned at the two ends of the coding region, then nucleosomes at the middle of genes, where shifts in Δisw1 are mostly observed, would be less constrained and more susceptible to regulation 313233. However, as described below, the patterns of Isw1-dependent shifts argue against this interpretation and instead support an active mechanism that directs Isw1 activity to mid-coding regions.

First, the presence of Isw1-dependent shifts at mid-coding regions is not correlated with the presence of nucleosome-free regions, or with strong positioning sequences at the ends of genes 31 (not shown). Second, these shifts display a strong direction bias: almost exclusively, the shifts occur in the direction opposite to that of elongation (Figure 4a) - in 85% of the cases, mid-coding nucleosomes were shifted upstream in Δisw1, towards the start codon. This highly significant directionality (P < 10^-16) is not expected by models of statistical positioning, but suggests instead that Isw1-dependent shifts reflect its function during elongation 15. Third, although the shifts propagate to flanking nucleosomes, as expected from statistical positioning models, this propagation is again biased, with downstream nucleosomes affected significantly more than upstream nucleosomes (Figure 4b). For example, the +4 nucleosome of ATP23 is shifted upstream by 34 bp, its downstream nucleosome (+5) is shifted by 22 bp, but its upstream nucleosome (+3) is not shifted at all (Figure 2a). As a result, the linker region between the +3 and +4 nucleosomes is practically abolished. More generally, the distance between the predicted centers of the Isw1-shifted nucleosomes and their upstream flanking nucleosomes drops from a median of 165 bp in the wild type to only 150 bp in Δisw1 (Figure 4c). Given the expected nucleosome length of 147 bp, this suggests that there are virtually no linker regions between these nucleosome pairs in Δisw1.

How is the specificity of Isw1 to mid-coding nucleosomes of particular genes established? Previous studies have shown that chromatin remodelers, including Isw1 and Isw2, interact with histone modifications, suggesting that Isw1 might be recruited through specific interactions with histone marks that characterize mid-coding regions 3435. Indeed, we find that genes with Isw1-dependent shifts are enriched with several histone modifications and depleted of other modifications (Figure 4d-f). Furthermore, modifications that are enriched at genes with Isw1 shifts tend to peak at mid-coding regions, while modifications that are depleted at these genes tend to peak around the TSS. Hence, Isw1-shifts are correlated with histone modifications, both across genes and within genes (Figure 4e).

Combined analysis of these modifications, together with other features (mRNA levels, gene length and cryptic initiation), shows that the most significant effect is from trimethylation of H3K79 (Figure S6 in Additional file 1). This modification peaks at the mid-coding region and is the most strongly correlated with Isw1 shifts, both before and after controlling for the other features. For example, while the average Isw1 shift of +5 nucleosomes is approximately 10 bp over all genes, it is only approximately 1 bp for genes with low H3K79me3 and approximately 17 bp for genes with high levels of this modification (Figure 4d). Other modifications had only minor effects in the combined analysis, although we cannot exclude the possibility that they directly influence Isw1.

We next asked whether remodeling by Isw1 influences the regulation of gene expression. To examine the genome-wide correlation between the effects of Isw1 on nucleosome positions and on gene expression, we compared the expression profiles of wild-type and Δisw1 strains, as well as Δhtz1 and Δgcn5 control strains (Figure S7 in Additional file 1). Although Δisw1 displayed the most extensive differences in nucleosome positioning, changes in gene expression in this strain were minor, with only approximately 1% of the genes altered by at least 2-fold and approximately 4% of the genes by at least 1.5-fold. At some genes, changes in gene expression correlated with Isw1-dependent nucleosome remodeling. For example, the -2 nucleosome of the TMA10 gene is evicted in all strains, except for Δisw1, where it covers multiple transcription factor binding sites (Figure 5a). Consistent with this, the expression level of TMA10 decreased in Δisw1 (Figure 5b).

However, in contrast to TMA10, the nucleosome occupancy of most promoter binding sites was not altered by deletion of ISW1, as the majority of Isw1-dependent nucleosome changes occur within coding regions. Furthermore, altered gene expression was not enriched at genes whose nucleosome positions or occupancies were affected by ISW1 deletion (Figure 5c; Figure S7 in Additional file 1). Similarly, expression differences between the two species were not correlated with species-specific effects of ISW1 deletion (Figure 5c; Additional file 3). These results are consistent with recent work that demonstrated that, for the MET16 gene, nucleosome remodeling and transcription regulation reflect distinct functions of Isw1 36. Similarly, expression changes were only weakly associated with differences in nucleosome positioning for Δhtz1 and Δgcn5(Figure S7 in Additional file 1).

Thus, changes in nucleosome positioning in Δisw1 are generally not associated with regulation of transcription levels, and are highly enriched at mid-coding regions. These results may indicate that Isw1-dependent remodeling is required primarily for maintaining normal chromatin structure at coding-regions during PolII elongation. In the absence of Isw1, coding-region nucleosomes may be perturbed during transcription elongation, resulting in the observed shifts, as well as fuzziness of nucleosome positioning and decreased occupancy. We reasoned that such perturbed chromatin structure may allow aberrant transcription initiation from cryptic sites within coding regions, as previously shown for defects in various elongation factors 37383940414243. Consistent with this, we found that coding-regions with Isw1-dependent shifts were enriched with cryptic initiation sites, as mapped in strains with defects in Spt6, Spt16 37 and Set2 44 (Figure 5d). This suggests that genes that are prone to defects in chromatin structure that permit cryptic initiation are also more sensitive to deletion of Isw1, linking Isw1 to suppression of cryptic initiation. Indeed, Isw1 was found as one of the 50 factors whose deletion promotes cryptic initiation at the FLO8 gene 37.

Finally, we compared the nucleosome shifts in Δisw1 to the nucleosome shifts found upon inactivation of PolII 45. Inactivation of PolII shifts nucleosomes downstream of their native positions (towards the 3' end), as opposed to the upstream shifts in Δisw1. Thus, some nucleosomes can adopt at least three stable positions: the native position occurring in the wild type; an upstream position when the activity of Isw1 is compromised; and a downstream position when PolII is inactivated. However, although some nucleosomes are shifted both by deletion of Isw1 and inactivation of PolII, we could not detect a consistent association between the two (r = -0.02), suggesting that different factors determine the susceptibility of nucleosomes to Isw1 and to PolII. Furthermore, Isw1-dependent shifts are localized to mid-coding regions while PolII shifts are also observed at the 5' ends of coding regions (Figure 6a).

Importantly, the extent of shifts in nucleosome positioning appears to be comparable for Isw1 and PolII, and, if anything, is even larger for Isw1 (limiting the comparison to upstream shifts in Δisw1 and downstream shifts for PolII inactivation). First, in both cases approximately 40% of the genes have shifts larger than 15 bp. Second, assuming that nucleosomes are only shifted upstream in Δisw1 and therefore that downstream shifts in Δisw1 reflect the extent of errors in calling nucleosome positions, we estimate that approximately half of the +5 nucleosomes are shifted upstream in Δisw1 (Figure 6b). Similar analysis of PolII inactivation (assuming that nucleosomes are only shifted downstream and that upstream shifts reflect the extent of errors) suggests that only a third of the +4 nucleosomes are shifted downstream (Figure 6b).

Deletion strains were constructed on the background of S. cerevisiae (BY4743) and S. paradoxus (CBS 432 ho::nat MATα) using standard techniques, introducing G418 and Kan resistance in S. cerevisiae and S. paradoxus, respectively. We verified that these deletions did not cause cell-cycle defects (Figure S9 in Additional file 1). Mono-nucleosomal DNA was isolated from cells grown to log-phase in rich media (YPD medium, 30°C) by digestion with MNase (see Supplementary Methods and Figure S1 in Additional file 1 for full details). Mono-nucleosomal DNA from the two species was pooled and subjected to Illumina high-throughput sequencing with one lane for wild-type strains and two lanes (biological repeats) for each of the mutant strains. Similarly, one lane was used to sequence the wild-type hybrid and two lanes for each of the mutant hybrids formed by mating the respective mutants from the two species. Data for biological repeats was averaged.

Reads of 34 to 40 bp were mapped to the genomic sequences of S. cerevisiae and S. paradoxus with Eland, allowing up to two mismatches within the first 32 bp; approximately 50% of the reads were mapped to a single location in one of the genomes, or were mapped to single locations in both genomes but with at least two more mismatches to one genome. These reads could thus be confidently mapped to a specific location in one of the genomes and the remaining reads were excluded. The genomic similarity between the two yeast species is approximately 85%, with only 13% of the aligned sequences having less than two mismatches for a single read length (36 bp). Thus, our approach of sequencing the two species together excludes approximately 13% of the genome, in which no reads are unambiguously mapped to either species, but does not affect the majority of the genome. Since we look for differences between wild-type and mutant strains, and use the same methods for mapping reads in both cases, this approach should have no effect on the observed differences but only hinders the detection of differences at highly conserved regions, which are excluded from the analysis.

Since reads of approximately 36 bp corresponded to the ends of approximately 150-bp fragments, the location of each mapped read was converted into the expected center position of the original DNA fragment. This was done by assuming a constant fragment length for each lane and each species. This length was estimated as the median distance between peaks of read-density in the forward strand and consecutive peaks of reads from the reverse strand (Table S1 in Additional file 1).

We obtained the number of reads that mapped to each base pair and transformed it to 'nucleosome occupancy', that is, the number of reads that cover each base pair, assuming that reads correspond to mono-nucleosome fragments of 147 bp. For prediction of center nucleosome positions we also defined 'nucleosome scores' by Gaussian filtering of the number of reads at each base pair, with a window of 50 bp and standard deviation of 25 bp 19. This transformation produces sharper peaks and allows a better estimation of nucleosome center positions. We estimated the positions of nucleosomes as peaks of nucleosome scores that were (i) not among the 10% peaks with lowest scores, and (ii) not within 100 bp of another peak with higher score. The number of nucleosomes defined by these criteria corresponded to approximately 80% of nucleosomal DNA, as estimated by previous studies 24.

For comparative analyses, nucleosome scores from all samples were normalized to the same distribution using percentile normalization. The raw data of mapped reads and the normalized nucleosome scores are available at the Sequence Read Archive and Gene Expression Omnibus databases (accession number GSE18939).

We compared nucleosome positioning at genes and promoters (1 kb) for each gene between different strains. If two nucleosomes from one strain paired with the same nucleosome from the other strain, then the one that is more distant from the single nucleosome was regarded as a possible nucleosome loss/gain. Nucleosomes whose positions differed by at least 15 bp between strains and that had a t-test P-value lower than 0.05 were regarded as a possible nucleosome shift. The t-test was performed by comparing the distribution of read positions of the two strains around the center positions of the respective nucleosome (taking all reads that map to at most 30 bp from the center position of one of the strains). Nucleosomes whose occupancy level differed by at least two-fold (after correcting for the overall difference in occupancy levels between the corresponding samples) were regarded as a possible occupancy change.

Each potential nucleosome loss/gain was also required to have at least two-fold higher occupancy in the strain with the nucleosome (compared with the strain without the nucleosome) and that this nucleosome will be supported by at least eight reads. To further increase the confidence of the predicted nucleosomal changes, we repeated the analysis above only for the reads that mapped to the forward strand and (separately) only for the reads that mapped to the reverse strand. We required that potential changes would pass all of the above thresholds in each of the strands, and that nucleosomes at positions of potential changes are mapped in the forward and reverse analyses to within 30 bp of their positions in the combined analysis (10 bp for shifts).

We defined Htz1 bound and unbound genes as the highest 20% and lowest 40% ChIP ratios, respectively 27; Gcn5 bound and unbound genes were defined as those with P-values lower than 0.05 and higher than 0.4, respectively. Isw1, Ioc2 and Ioc3 bound genes were defined as in Venters et al. 26 and non-bound genes were defined as those that were identified as bound by at least one other factor but not by these particular factors.

Genome-wide expression levels of the wild-type and mutant strains were measured for the two species, with a multi-species array, as described previously 51. Differential expression was defined as at least 1.5-fold differences, although the use of other thresholds did not significantly alter the results (not shown).