The yeast ISW1b ATP-dependent chromatin remodeler is critical for nucleosome spacing and dinucleosome resolution

Isw1 and Chd1 are ATP-dependent nucleosome-spacing enzymes required to establish regular arrays of phased nucleosomes near transcription start sites of yeast genes. Cells lacking both Isw1 and Chd1 have extremely disrupted chromatin, with weak phasing, irregular spacing and a propensity to form close-packed dinucleosomes. The Isw1 ATPase subunit occurs in two different remodeling complexes: ISW1a (composed of Isw1 and Ioc3) and ISW1b (composed of Isw1, Ioc2 and Ioc4). The Ioc4 subunit of ISW1b binds preferentially to the H3-K36me3 mark. Here we show that ISW1b is primarily responsible for setting nucleosome spacing and resolving close-packed dinucleosomes, whereas ISW1a plays only a minor role. ISW1b and Chd1 make additive contributions to dinucleosome resolution, such that neither enzyme is capable of resolving all dinucleosomes on its own. Loss of the Set2 H3-K36 methyltransferase partly phenocopies loss of Ioc4, resulting in increased dinucleosome levels with only a weak effect on nucleosome spacing, suggesting that Set2-mediated H3-K36 trimethylation contributes to ISW1b-mediated dinucleosome separation. The H4 tail domain is required for normal nucleosome spacing but not for dinucleosome resolution. We conclude that the nucleosome spacing and dinucleosome resolving activities of ISW1b and Chd1 are critical for normal global chromatin organisation.


Results
The ISW1b complex is the major nucleosome spacing enzyme in yeast. We have used MNaseseq to determine the chromatin organisation at yeast genes. Briefly, this technique involves micrococcal nuclease (MNase) digestion of the chromatin in purified nuclei to predominantly mononucleosomes. Suitably digested DNA samples are then used to construct libraries for paired-end sequencing. Alignment to the yeast genome indicates the location of each sequenced nucleosome. The midpoint of each sequence is assumed to represent the midpoint of the nucleosome (also called the dyad). An average plot for all 5770 yeast genes is obtained by aligning all genes on their transcription start site (TSS) or on their major + 1 nucleosome position, summing all of the nucleosome dyads on every gene, and normalizing to the genomic average (set at 1). A typical plot for wild type cells is shown in Fig. 1a. We define nucleosome spacing as the average distance between the dyad peaks for the + 1 to + 5 nucleosomes (i.e. the first five nucleosomes on the average gene), measured by the slope of the regression line. However, in some mutants, the phasing is so poor that the spacing cannot be measured accurately (see below). The degree of phasing is indicated by the height of the nucleosome peaks: higher peaks indicate more coincident dyads and therefore better nucleosome positioning.
We have shown previously 15 that cells lacking Isw1 have weaker nucleosome phasing and reduced spacing relative to wild type cells (Fig. 1b). However, isw1Δ cells lack both the ISW1a and the ISW1b remodeling complexes. To assess the relative contributions of ISW1a and ISW1b to phasing and spacing, we examined the chromatin organisation in various ioc mutants. Cells lacking the ISW1a complex (ioc3Δ) show a slight increase in spacing (replicates: 166, 167 bp) with no change in phasing relative to wild type (165, 165 bp) (Fig. 1c). Cells lacking an ISW1b subunit (ioc2Δ or ioc4Δ) have intermediate nucleosome spacing (161-163 bp) (Fig. 1d,e): less than in wild type (165, 165 bp), but not as low as in isw1Δ cells (159, 160 bp), although the spacing is shorter on genes over 2 kb (see below). Small spacing differences are reproducible (compare biological replicate experiments; Fig. 1). The change in spacing becomes increasingly obvious downstream, because the nucleosome position shift is additive relative to the first (+ 1) nucleosome. For example, if the average spacing increases from 160 to 165 bp, then the downstream shift in average position for the + 2 nucleosome is 5 bp, it is 10 bp for the + 3 nucleosome, and 15 bp for the + 4 nucleosome, and so on. Phasing is weaker in the ioc2Δ mutant (Fig. 1d), but hardly affected in the ioc4Δ mutant (Fig. 1e). Cells lacking both ISW1b ancillary subunits (the ioc2Δ ioc4Δ double mutant) have similar chromatin organisation to the ioc2Δ single mutant, exhibiting the weaker phasing observed in ioc2Δ cells as well as the shorter spacing observed in both the ioc2Δ and ioc4Δ single mutants (Fig. 1f). The similar spacing in the ioc2Δ and ioc4Δ single mutants and the ioc2Δ ioc4Δ double mutant (~ 162 bp) suggests that both Ioc2 and Ioc4 are required for ISW1b spacing activity. Chromatin organisation in cells lacking all three ancillary subunits (the ioc2Δ ioc3Δ ioc4Δ triple mutant) is very similar to that in isw1Δ cells (Fig. 1g). This is expected, because both complexes should be inactive in both mutants, given that the Isw1 subunit by itself has no remodeling activity in vitro 26 .
Nucleosome phasing on a gene can be influenced by interference from a phasing signal emanating from downstream elements, such as a downstream promoter 40 . This effect can be eliminated by restricting the analysis to promoter-proximal nucleosomes on long genes 40 , defined here as genes with distances of > 2 kb between the TSS and the transcript termination site (1616 genes; Supplementary Fig. S1). We find that gene length has no effect on nucleosome spacing in wild type cells (165 bp on all genes and on long genes) or in ioc3Δ cells, which have slightly longer spacing than wild type cells (167 bp on all genes and on long genes). However, the ioc2Δ and ioc4Δ single mutants, the ioc2Δ ioc4Δ double mutant and the ioc2Δ ioc3Δ ioc4Δ triple mutant all exhibit shorter spacing on long genes than on all genes (see Fig. 1). In fact, the spacing on long genes is about the same in ioc2Δ (159 bp), ioc4Δ (160 bp), ioc2Δ ioc4Δ (159 bp), ioc2Δ ioc3Δ ioc4Δ (158 bp) and isw1Δ (159 bp) cells ( Supplementary Fig. S1). Thus, long genes do not exhibit the intermediate spacing observed for all genes in the ioc mutants (Fig. 1d, e, f; Supplementary Fig. S1). These data suggest that the phasing potential of downstream elements is enhanced in the absence of ISW1b, resulting in phasing interference and altered spacing on short genes, but not on long genes. Taken together, our observations support the conclusion that the ISW1b complex is primarily responsible for wild type nucleosome spacing, whereas ISW1a plays a very minor role.  (Ioc4) and Isw1 are enriched on different sets of genes, suggesting that these genes might be differentially affected by ISW1a or ISW1b 27 . We determined whether the chromatin organisation of these sets of genes is differentially affected by loss of ISW1a or ISW1b, as might be expected. In the case of ISW1a, we found that Ioc3-enriched genes show very slightly increased spacing (166 vs. 165 bp) in wild type cells and in ioc3Δ cells (167 vs. 166 bp), but these differences are probably negligible ( Supplementary Fig. S2). Thus, ISW1a has little or no differential effect on the chromatin organisation of its putative target genes. Moreover, the spacing on Ioc3-bound genes is affected in ioc4Δ cells, indicating that ISW1a-enriched genes are affected by ISW1b (Ioc4) ( Supplementary  Fig. S2). In the case of ISW1b, Ioc4-enriched and non-enriched genes have the same spacing and phasing in wild type cells, suggesting that they are not differentially affected by ISW1b. In ioc4Δ cells, both sets of genes have shorter spacing than wild type ( Supplementary Fig. S2). Ioc4-enriched genes may have slightly shorter spacing than non-enriched genes, but the effect is subtle (Supplementary Fig. S2). As expected, Ioc4-enriched genes are not affected in ioc3Δ cells ( Supplementary Fig. S2). Thus, ISW1b affects the chromatin organisation of both sets of genes. Finally, Isw1-enriched genes have the same spacing as non-enriched genes in wild type cells and, although the phasing is slightly better on the Isw1-enriched genes, this effect is also observed in the absence of Isw1 ( Supplementary Fig. S2). Isw1-enriched and non-enriched genes show similar changes in isw1Δ cells (short spacing and weaker phasing), although there is a slight difference in spacing of 1-2 bp between the two sets of genes ( Supplementary Fig. S2). Overall, the Isw1 complexes affect the chromatin organisation of both enriched and non-enriched genes similarly, with only subtle differences at most. It is unclear why we do not observe obvious differences in chromatin organisation for putative target and non-target genes defined by Ioc3, Ioc4 or Isw1 enrichment; perhaps all genes have some bound ISW1b, such that the putative target genes are only modestly enriched relative to non-target genes. Alternatively, loss of one ISW1 complex might affect the distribution and/ or activity of the other ISW1 complex.

ISW1b and Chd1 account for the extreme chromatin disruption in cells lacking both Isw1 and
Chd1. Cells lacking Chd1 have a mild chromatin organisation defect 14,15 , characterized by somewhat shorter spacing than wild type (164, 163 bp vs. 165, 165 bp) and weaker phasing (Fig. 2a). This effect is slightly stronger on long genes (chd1Δ: 162 bp; wild type: 165 bp; Supplementary Fig. S1). In contrast, chromatin organisation in cells lacking both Chd1 and Isw1 (the chd1Δ isw1Δ double mutant) is extremely disrupted 14,15 (Fig. 2b). The chd1Δ ioc2Δ ioc4Δ triple mutant ( Fig. 2c) also has extremely disrupted chromatin, whereas the chd1Δ ioc3Δ double mutant is very similar to the chd1Δ single mutant (Fig. 2d). This observation is consistent with our conclusion that the ISW1b complex is much more important than the ISW1a complex for global nucleosome spacing ( Fig. 1). However, chromatin organisation in the chd1Δ ioc2Δ ioc4Δ triple mutant is not quite as severely disrupted as in the chd1Δ isw1Δ double mutant: the phasing in the triple mutant is very poor, but somewhat stronger than in the double mutant (Fig. 2b,c). Measurement of the spacing in these two mutants is not appropriate because most of the peaks are too weak to measure accurately. The difference between the chd1Δ ioc2Δ ioc4Δ triple mutant and the chd1Δ isw1Δ double mutant is that the ISW1a complex (Isw1-Ioc3) is still present in the former mutant. This suggests that ISW1a contributes some residual order to genic chromatin in the absence of both Chd1 and ISW1b.
ISW1a and ISW1b have little effect on the chromatin of very active genes. Since ISW1b preferentially associates with transcriptionally active genes 39 , we examined whether ISW1a or ISW1b mediate specific effects at highly active genes. We defined highly active genes as those which have > 4 times the genomic average signal using our published ChIP-seq data for the Rpb3 subunit of Pol II 15 . Although this is a somewhat arbitrary threshold, it is clear from heat map analysis that relatively few genes have high levels of Pol II; most genes have relatively low Pol II levels ( Supplementary Fig. S3). On average, the 300 most active genes have ~ eightfold higher Rpb3 density than the remaining genes. Accordingly, we compared the average chromatin structure of the most active genes with that of the remaining, much less active, 5470 genes ( Supplementary Fig. S3). In wild type cells, the chromatin of the highly active genes is poorly organised, characterized by much reduced and irregular spacing, weak phasing and a much wider nucleosome-depleted region (NDR) at the promoter, which extends upstream 41,42 . Although the average spacing on the highly active genes cannot be measured accurately due to poor phasing, it is clearly shorter than the spacing on the other genes (Supplementary Fig. S3; compare nucleosome peak locations). The chromatin organisation of the highly active genes is not strongly differentially affected by any of the ioc∆ mutations or isw1∆ ( Supplementary Fig. S3), whereas the less active genes have the spacing observed for all genes, as expected. We conclude that ISW1a and ISW1b have no differential effect on the chromatin organisation of highly active genes, with the caveat that their general state of disruption might obscure subtle effects. On the other hand, nucleosome phasing on the top 300 active genes is weaker in cells lacking Chd1 than in wild type or in any of the ioc mutants ( Supplementary Fig. S3), suggesting that Chd1 is the most important spacing enzyme for highly active genes, and consistent with a direct association of Chd1 with transcript elongation factors 43 .
Both ISW1b and Chd1 are important for resolution of close-packed dinucleosomes. We reported previously that a major contributing factor to chromatin disruption in the chd1Δ isw1Δ double mutant is the presence of close-packed dinucleosomes, primarily involving the + 2 nucleosome (i.e. dinucleosomes containing either the + 1 and + 2 nucleosomes or the + 2 and + 3 nucleosomes) 22 (Fig. 3a) www.nature.com/scientificreports/ above (Figs. 1 and 2) include only mononucleosome data (the analysis was limited to DNA fragments of 120-180 bp). A dyad plot is not appropriate for dinucleosomes, since the midpoint of a dinucleosome would be located between the two nucleosomes. Instead, we used nucleosome occupancy (coverage) plots, in which the number of times each genomic base pair appears in either mononucleosomes or dinucleosomes is counted and normalized to the genomic average. In the chd1Δ isw1Δ double mutant, there is a strong dinucleosome peak at the + 2 position and a depressed + 2 mononucleosome peak, consistent with the presence of a significant fraction of all + 2 nucleosomes in dinucleosomes 22 (Fig. 3a; Supplementary Fig. S4). We determined the relative contributions of Chd1 and Isw1 to dinucleosome resolution. A dinucleosome peak is observed in both the chd1Δ and isw1Δ single mutants, but it is weaker than in the chd1Δ isw1Δ double mutant (Fig. 3a). The + 2 mononucleosome peak is also weaker in the single mutants relative to wild type. These data indicate that the Chd1 and Isw1 remodeling enzymes both www.nature.com/scientificreports/ contribute independently and make additive contributions to dinucleosome resolution, such that neither enzyme is capable of resolving all dinucleosomes on its own. Next, we determined the separate contributions of the ISW1a and ISW1b complexes to Isw1-dependent dinucleosome resolution (Fig. 3b). The ioc2Δ and ioc4Δ single mutants have an enhanced dinucleosome peak and depressed mononucleosome peak, as observed for the isw1Δ single mutant, suggesting that both ancillary subunits of the ISW1b complex are required for Isw1-dependent dinucleosome resolution. Consistent with www.nature.com/scientificreports/ this conclusion, the ioc2Δ ioc4Δ double mutant has a dinucleosome peak similar to that of the single mutants. In contrast, there are fewer dinucleosomes in the ioc3Δ mutant, which has a slightly lower + 2 dinucleosome peak than wild type and somewhat higher levels of the + 2 and + 3 mononucleosomes ( Fig. 3b; Supplementary  Fig. S4), indicating that ISW1a is not important for dinucleosome resolution. These conclusions are supported by high levels of dinucleosomes in the chd1Δ ioc2Δ ioc4Δ triple mutant versus wild type, resulting from loss of both ISW1b and Chd1 (Fig. 3c). Similarly, the chd1Δ ioc3Δ double mutant has more dinucleosomes than wild type, but less than the chd1Δ ioc2Δ ioc4Δ triple mutant, which can be accounted for by the absence of Chd1 with little contribution from ISW1a (Fig. 3c). Thus, the ISW1b complex accounts quite well for the role of Isw1 in dinucleosome resolution.
Set2 contributes to dinucleosome separation. The Ioc4 PWWP domain binds preferentially to H3-K36me3 relative to unmethylated H3-K36 in vitro, suggesting that ISW1b may be regulated by H3-K36me3 38,39 . The only H3-K36 methyltransferase in yeast is encoded by SET2. We examined whether ISW1bmediated changes in nucleosome spacing and dinucleosome resolution depend on Set2. Nucleosome spacing in a set2Δ mutant (replicates: 165 and 163 bp) is slightly lower than wild type (165 and 165 bp), although this difference is probably not significant (Fig. 4a). Loss of Ioc4 has a stronger effect on nucleosome spacing ( Fig. 1e; ioc4Δ replicates: 163 and 162 bp) than loss of Set2 (Fig. 4b), which is also true for long genes ( Supplementary  Fig. S1). Like ioc4Δ cells, set2Δ cells have higher dinucleosome levels than wild type cells (Fig. 4c). Interestingly, there are more + 3 and + 4 dinucleosomes in set2Δ cells than in ioc4Δ cells (Figs. 4c; 3b; Supplementary Fig. S4). However, there is no corresponding depression in the + 2, + 3 and + 4 mononucleosome peaks in set2Δ cells (Fig. 4c); instead, these mononucleosomes are higher than in wild type. There is an apparently compensatory reduction in NDR occupancy and the − 1 nucleosome (Fig. 4c), which is also observed in ioc4Δ cells but not in the other ioc mutants (Fig. 3b). Overall, these data suggest that Set2 and H3-K36me3 may be more important for the dinucleosome resolving activity of ISW1b than for its spacing activity.
Loss of Set1 has no effect on global chromatin organisation. The binding of ISW1 to chromatin is mediated indirectly through H3-K4 trimethylation 37 . On the other hand, a direct interaction might be possible, mediated by the putative PHD domain in the Ioc2 subunit of ISW1b 38 . Since Set1 is the only H3-K4 methyltransferase in yeast, we tested this possibility by examining set1Δ cells. However, chromatin organisation in set1Δ cells is essentially identical to wild type at the global level ( Supplementary Fig. S5) and quite different from that in ioc2Δ cells (Fig. 1d). Dinucleosomes involving the + 3 and + 4 nucleosomes may be somewhat elevated in set1Δ cells, although this is unclear because the replicates are not consistent in this respect (Supplementary Fig. S4; Supplementary Fig. S5). In conclusion, Set1 and therefore H3-K4me3 are not required for ISW1-dependent nucleosome spacing.

Deletion of the H4 N-terminal tail domain results in shorter nucleosome spacing but does not increase dinucleosome levels.
Nucleosome mobilisation in vitro by ISWI complexes isolated from different organisms requires the H4 N-terminal tail domain, specifically the "basic patch" (residues [16][17][18][19] 33,44,45 . In addition, genetic interactions between isw1 mutations and H4 mutations (point mutations and a tail deletion) suggest involvement in a common pathway 19 . We reasoned that deletion of the H4 N-tail might result in a chromatin organisation similar to that observed in the isw1Δ single mutant: reduced spacing, weaker phasing and increased dinucleosome formation (Figs. 1b, 3a). We constructed a yeast strain in which both H3/H4 gene loci (HHT1-HHF1 and HHT2-HHF2) were deleted and covered by a plasmid carrying wild type HHT1-HHF1 or HHT1-HHF1Δ21, in which the first 21 amino acid residues of the H4 N-tail had been deleted. This strain displayed a clear growth phenotype, with a doubling time of ~ 2.6 h, compared with ~ 1.8 h for the wild type strain, in synthetic complete (SC) medium. We find that nucleosome spacing is shorter in cells lacking the H4 N-tail (Fig. 5a), though not as short as in isw1Δ cells (Fig. 1b), and both mutants have poor phasing. These observations are consistent with the requirement of Isw1 for the H4 N-tail domain. However, removal of the H4 N-tail resulted in only a very slight increase in dinucleosome prevalence ( Fig. 5b; Supplementary Fig. S4), unlike loss of Isw1 (Fig. 3a), indicating that loss of Isw1 and loss of the H4 N-tail are not equivalent.

Discussion
ISW1b is the primary nucleosome spacing enzyme in yeast. We and others have shown previously that Isw1 and Chd1 are both needed for normal global chromatin organisation in yeast 14,15 . We proposed that these two enzymes compete to set nucleosome spacing in wild type cells, with Isw1 being dominant, setting wild type spacing, and Chd1 directing shorter spacing 15 . In the absence of Isw1, the spacing is short, which we attributed to Chd1 activity. An important complication, which we have addressed here, is that there are two complexes containing the Isw1 ATPase subunit 26 . We have determined their respective roles in chromatin organisation in vivo: ISW1b and Chd1 are primarily responsible for global nucleosome spacing in wild type cells, whereas ISW1a has little effect on spacing (Fig. 6a). We note that Ino80C also affects global spacing [27][28][29] , but it is not yet clear how its activity meshes with those of ISW1b and Chd1. A difficulty for our competition model is that, in the absence of Chd1, ISW1b is expected to win the competition, resulting in wild type spacing. Instead, we observe that spacing is somewhat shorter in cells lacking Chd1 15 ( Fig. 2a; Supplementary Fig. S1). One possible explanation is that the dominance of ISW1b depends on Chd1 and so, in its absence, we see an increased contribution from ISW1a, such that the spacing represents the average of ISW1a and ISW1b activities. If so, cells having only ISW1b (the chd1Δ ioc3Δ double mutant) are expected to have wild type spacing. However, the result is intermediate spacing, very similar to the chd1Δ single mutant (Fig. 2) www.nature.com/scientificreports/ www.nature.com/scientificreports/ www.nature.com/scientificreports/ has already made arrays with short spacing, although this model is inconsistent with in vitro data showing that purified ISW1b can space nucleosomes by itself 26 . A more satisfying explanation involves dinucleosomes (Fig. 6b). Consider a single gene. If there are regularly spaced nucleosomes downstream of a close-packed + 1/+ 2 dinucleosome (i.e. no linker), these nucleosomes will be out of phase by one linker length with nucleosomes in cells with no dinucleosome on this gene. Other cells may have a + 2/+ 3 dinucleosome instead, which will alter the positions of regularly spaced downstream nucleosomes to give a different phasing. The result is generally weaker phasing (peak flattening) and a shift to shorter average spacing. Moreover, since the dinucleosomes are not counted in the mononucleosome phasing pattern, there will be missing occupancy around the + 2 nucleosome, resulting in more pattern disruption (primarily a depressed + 2 nucleosome peak). These effects will increase as the fraction of genes having a dinucleosome increases: the chd1Δ isw1Δ double mutant and the chd1Δ ioc2Δ ioc4Δ triple mutant both have high dinucleosome levels and extremely poor phasing. The chd1Δ single mutant has fewer dinucleosomes, resulting in a smaller shift in the peaks to shorter spacing. These considerations can account for the shift to shorter spacing in the absence of Chd1 and for the major disruption of chromatin in the absence of both Chd1 and ISW1b. Thus, we propose that the spacing is shorter in cells lacking Chd1 because ISW1b cannot resolve all of the dinucleosomes.
Previously, we showed that Isw1 and/or Chd1 is required to resolve dinucleosomes, or to prevent their formation 22 . Here we show that both enzymes are important for separating dinucleosomes and that, as observed for spacing activity, ISW1b is much more important than ISW1a. In vitro, both ISW1a and ISW1b can move nucleosomes reconstituted on a pair of 601 nucleosome positioning sequences farther apart, even if separated by a linker of only 4 bp 46 , indicating that both ISW1 complexes probably have close-packed dinucleosome resolving activity.
The Set2 methyltransferase contributes to dinucleosome separation. We observe that the ancillary subunits of ISW1b, Ioc2 and Ioc4, are both necessary for ISW1b-dependent chromatin organisation, although Ioc2 makes a greater contribution, since it affects phasing as well as spacing. We suggest that Ioc2 and Ioc4 cooperate to influence the remodeling activity of the Isw1 subunit. Both may be linked to specific histone modifications: Ioc4 binds to reconstituted nucleosomes carrying H3-K36me3 with higher affinity than to unmethylated nucleosomes, suggesting that Set2-mediated H3-K36me3 might be an important regulator of ISW1b 38,39,47 . Our data provide some evidence to support this proposal, because although loss of Set2 has only a subtle effect on nucleosome spacing, it does result in increased dinucleosome levels, similar to that observed for loss of Ioc4. We propose that Set2-mediated trimethylation of H3-K36 is important for ISW1b-mediated dinucleosome resolution, through interaction of H3-K36me3 with its Ioc4 subunit.
We note that neither ISW1 complex requires H3-K36me3 to space nucleosomes in vitro, since they are active on nucleosomes containing unmodified recombinant histones. On the other hand, in vivo, H3-K36me3 may have local effects on ISW1b activity, rather than global effects, which we have not detected. These are probably not occurring at the most active genes because their chromatin organisation in set2Δ cells is similar to wild type ( Supplementary Fig. S3). H3-K36me3 recognition may also affect other functions of ISW1b, such as gene repression 20 or mRNP quality control 23 .
The fact that Ioc2 has a putative PHD domain suggests that ISW1b may interact with an additional methylated histone residue, although there is no evidence for this at present 38 . It is unlikely to be H3-K4me3, because our data indicate that the nucleosome spacing activity of the ISW1b complex is independent of Set1 and therefore of H3-K4me3. H3-K4me3 may also bind to one of the chromodomains of Chd1 48 , although this is controversial 49 . If Chd1 does indeed bind to H3-K4me3, this binding is not critical for its nucleosome spacing function, since we find that loss of Set1 does not result in global chromatin defects, unlike loss of Chd1.
The H4 N-terminal tail domain is required for normal nucleosome spacing. The H4 tail domain is generally required by ISWI-like enzymes for activity 4,5,44,50 . The ISWI ATPase subunit has an inhibitory AutoN domain that resembles the H4 tail, which is displaced by the H4 tail when ISWI binds to a nucleosome 4,5 . The isolated yeast Isw1 subunit is inactive without Ioc subunits unless the AutoN domain is inactivated by mutation 33 , implying that the Ioc subunits may regulate Isw1 activity through the AutoN domain. We find that the H4 tail deletion mutant exhibits shorter spacing, consistent with inactivation of ISW1b, but there is little increase in dinucleosomes. The most likely explanation for the non-equivalence of the H4Δ21 and isw1Δ mutations is that the H4 tail domain is involved in multiple processes, including interactions with other remodelers and proteins. This possibility is supported by the fact that the H4Δ21 mutant has a strong growth defect, unlike the various ioc mutants and the isw1Δ mutant, which have no obvious growth defect. Since Chd1 also makes contacts with the basic patch in the H4 tail 51,52 , its remodeling activity may also be compromised, although H4Δ21 chromatin is clearly not as disrupted as chd1Δ isw1Δ chromatin. The H4 N-tail is also required by ISW2 in vitro 53 , although global chromatin organisation is unaffected in isw2Δ cells 14,15 . The low prevalence of dinucleosomes in the H4Δ21 mutant suggests that the H4 tail might be required for dinucleosome formation, possibly because the increased negative charge on nucleosomes lacking the H4 tail might limit how closely nucleosomes can approach one another 54,55 . Deletion of the H4 tail also results in the loss of multiple post-translational modification sites, preventing interaction with many chromatin factors.
MNase-seq. Nuclei were prepared as described 61 . MNase digestion of nuclei and construction of pairedend libraries was as described 62 , except that mononucleosomal DNA was not gel-purified, in order to retain mononucleosomes and dinucleosomes in the correct proportions 22 . Two biological replicate experiments were performed for all strains (i.e., the replicate experiments for each strain were performed entirely independently). MNase-seq data were analysed using scripts originally described by 22 ; modified code is provided (Supplementary Code).

Data availability
The datasets generated during the current study are available in the NCBI Gene Expression Omnibus (GEO) repository [GSE156224]. Published data sets used in the current study are available in the GEO repository (MNase-seq data for YJO484: GSE117514 22 ; Rpb3 ChIP-seq data: GSE69400 15 ).