By using deep sequencing technology, we previously identified 4,897 open chromatin loci in yeast 
based on the FAIRE assay 
. In this work, we profiled in vivo
nucleosomes by means of MNase-mediated purification of mononucleosomes (see Methods
). Unexpectedly, we discovered the presence of boundary nucleosomes just inside of open chromatin (black curve in ), a pattern which also appeared with 46,080 open chromatin regions identified in the GM12878 human lymphoblastoid cells by the ENCODE project (black curve in ). This evolutionarily conserved feature was commonly found for promoter and non-promoter regulatory regions.
Positioning of boundary nucleosomes within open chromatin.
nucleosomes that were reconstituted purely based on naked DNA 
also peaked within open chromatin in both yeast and human (gray shade in ). In yeast, the corresponding DNA sequences displayed an increase in the C/G dinucleotide frequency (red dots in Figure S1
) and a decrease in the A/T dinucleotide frequency (blue dots in Figure S1
), exhibiting nucleosome-favouring features near the boundaries of accessible chromatin. In yeast, >60.8% of open chromatin regions had sequence-directed (in vitro
) nucleosome positioning whereas >25.6% had nucleosome positioning in vivo
). In human, the fraction of nucleosome-possessing chromatin sites is lower than in yeast but the same tendency (higher in vitro
than in vivo
occupancy) is maintained (Table S1
). Although there was a difference in the peak position between the in vivo
and in vitro
nucleosomes particularly in human, the relative distance was consistent between promoter and non-promoter regions. Therefore, we propose that nucleosome-encoding sequences are more associated with the boundary in vivo
nucleosomes rather than the center of regulatory regions as previously observed 
. The in vitro
nucleosomes in non-promoter regions appeared to be positioned at the center of open chromatin because the average size of non-promoter regions, as estimated by the location of the inside FAIRE peak (blue curve in ), was smaller than that of promoters. Indeed, the in vitro
nucleosomes peaked at the center of small-sized (<500 bp) open chromatin regions while forming a bimodal peak in longer regions (>1 kb) (Figure S2
). On the other hand, the in vivo
nucleosomes formed a bimodal peak regardless of the size of the region (Figure S2
When examined according to TF binding sites (TFBSs) in the human cells, two strongly positioned nucleosomes were found 200 bp away on average from empirical TFBSs (based on the ChIP-seq of ~90 TFs), and periodic nucleosome phasing was observed in the surrounding regions (see black curve in ). A less stable positioning of the flanking nucleosomes and less distinct phasing of the surrounding nucleosomes were obtained when sequence-predicted TFBSs (based on the Transfac database) were used (gray curve in ). Intriguingly, sequence tags from DNase I hypersensitive sites (DHSs) were confined within the 400 bp region centered on the TFBS (black curve in ). The coincidence between the position of the two flanking nucleosomes (yellow lines in ) and the edges of the DHS tag cluster (yellow lines in ) was not observed when DHS tags were aligned by Transfac sequence motifs (gray curve in ). This implies that the boundary nucleosome positioning and the nucleosome phasing may be dependent on in vivo TF binding events.
Boundary nucleosome positioning within open chromatin.
We then sought to examine nucleosome organization across defined open chromatin domains. As illustrated in , the nucleosomes positioned within open chromatin near the boundaries may carry specific histone modifications while DNA-binding factors may bind in between the flanking nucleosomes. Maintaining nucleosome signatures at the borders may help to prevent occlusion of regulatory elements by histones. The boundary positioning of nucleosomes was confirmed by the genome-wide average patterns (black solid lines in ). Notably, different histone modifications showed different patterns across open chromatin (coloured lines) and H2A.Z-containing nucleosomes (black dotted line) were observed in between the boundary nucleosomes. TF binding was concentrated in between the two flanking boundary nucleosomes (Figure S3
Histone modifications and H2A.Z occupancy across open chromatin in human.
Histone marks associated with active gene transcription such as H3K9ac, H3K27ac, H3K4me2, and H3K4me3 coincided with H2A.Z distribution across open chromatin (). While the acetylation patterns (red and orange lines) were well overlapping with H2A.Z positioning, there was a slight dip on the methylation levels (violet and blue lines). By using comprehensive chromatin data in human T cells, encompassing H2A.Z occupancy, histone methylation and acetylation marks, and MNase-digested nucleosomes 
, we calculated relative H2A.Z levels across the genome and compared them with histone modification levels. H2A.Z incorporation positively correlated with most histone acetylations, in particular with H3K9ac and H3K27ac, but not with histone methylations except H3K4me3 and H3K4me2 (Figure S4
). Those active histone marks are expected to decrease nucleosome stability and this may explain the low occupancy of the H2A.Z-enriched central nucleosomes. Nucleosome purification in low salt conditions revealed the enrichment of H2A.Z nucleosomes at the nucleosome-free region of the promoter as defined in high salt conditions 
Histone methylations such as H3K4me1, H3K9me3, H4K20me1, and H3K79me2 were absent on the central H2A.Z nucleosomes but present on the flanking nucleosomes (). Enhancer elements marked by H3K4me1 alone are inactive or poised until they turn into active enhancers in the wake of H3K27ac modifications 
. H3K9me3 is also associated with poised enhancers. High levels of H3K9me3 are found in enhancers that are inactive in one cell type but become active in another under the control of the stimulus-induced demethylase Jmjd2d 
. H4K20me1 was found to be associated with transcription activation in the context of canonical Wnt signaling 
and with specific classes of enhancers that are deprived of H2A.Z: certain classes of enhancers are enriched in H2A.Z but not H4K20me1 while others are enriched in H4K20me1 but not H2A.Z 
. Promoter H3K79me2 was linked to active transcription in flies 
and in humans 
but in another study it did not show any preference toward either active or silent genes 
. A role for H3K79me2 in enhancer regulation remains to be elucidated. Taken together, histone modifications related to inactive or poised enhancers or other regulatory states occur on the nucleosomes at the borders of open chromatin.
Unlike the above histone modifications, H3K27me3 and H3K36me3 are not concentrated in specific regions but spreading across multiple nucleosomes 
. H3K36me3 forms a broad domain of enrichment across the body of genes as a regulator of alternative splicing 
. While H3K27me3 typically shows a domain-like profile similarly to H3K36me3, it can also form a peak around the transcription start site of bivalent genes 
or appear at poised enhancers 
. Both marks (red and green line in ) were present on nucleosomes (black solid line in ) that were distant from open chromatin, as opposed to the other marks that were absent on these nucleosomes (). A higher level of H3K27me3 (red line) was observed on the boundary nucleosomes as compared with H3K36me3 (green line), maybe indicating the association of H3K27me3 with poised promoters or enhancers.
To examine the positional changes in the borders of open chromatin according to genetic variation, we identified open chromatin loci in 96 different yeast strains 
consisting of the parental strains (BY4716 and RM11_1a) and the descendants resulted from their crossing 
. We aligned all open chromatin sites in the laboratory strain (BY4716) by the 5′ boundary, center, and 3′ boundary, and then mapped the relative locations of nearby open chromatin loci in the other strains, resulting in the cluster of homologous regions falling within a certain distance ().
Local changes of yeast open chromatin upon genetic perturbation.
While the central location changes within 25 bp upstream or downstream, the border shifts by ~75 bp away probably giving rise to changes in the size of the region (). The effect of technical variation or inherent data structure could be ruled out in general (Figure S5
). Importantly, the borders with a higher intrinsic propensity for nucleosome positioning showed a higher degree of deviation, clearly separating those with the in vitro
occupancy score 
<0 and >0.5 (). We used the score of 0.5 as the threshold for a positioned in vitro
To identify genetic determinants of the local boundary shifts, we carried out quantitative trait locus (QTL) mapping for the end-to-end distances of the open chromatin boundaries that were identified in BY4746 and were <100 bp away from their homologous sites in all the other strains. At a false discover rate (FDR) of 0.01, 39.2% of the boundary shifts were significantly associated with at least one genetic marker in trans. About 5.4% were associated with cis-acting elements located within 100 kb. In terms of the number of associations, the trans- and cis-associations accounted for 84.3% and 15.7%, respectively.
Genetic markers with >5 trans
-linkages included chromatin remodelers and transcription regulators (Table S2
). The largest number of associations was found for IES6
, which encodes a protein that associates with the INO80 chromatin remodelling complex. INO80 is an ATP-dependent nucleosome spacing factor that is involved in nucleosome positioning and mobilization with a role in transcription and DNA repair 
. Not only general transcription factors such as SRB2
, a subunit of the RNA polymerase II mediator complex, but also several sequence-specific transcription factors were identified (Table S1
). Three of the subunits of the MCM2-7 complex, which is involved in DNA replication, were also associated with multiple regulatory regions (Table S1
). While 42% of boundary shifts were associated with genetic variation, perturbation in cellular environment caused by combinatorial or secondary effects of multiple genetic alterations may underlie other local changes.
We then compared the results of the boundary QTL mapping with those of the QTL mapping for chromatin accessibility as previously performed for the same dataset 
. The fraction of the cis
-associations in the boundary QTL mapping (15.7%) was two times higher than that in the accessibility QTL mapping, implying that underlying DNA sequences play a significant role in the regulation of open chromatin boundaries. Sixty-six boundary shifts were associated in cis
with 226 genetic markers while 853 boundaries were in trans
with 431 genetic markers. Interestingly, only for 4.5% of the 66 cis
-associated boundaries and 5.0% of the 853 trans
-associated boundaries, the relevant chromatin region was also identified in the accessibility QTL mapping. This supports that the variation in boundary locations does not simply reflect the variation in chromatin accessibility despite a possible mechanistic correlation between peak size and peak width. While different target chromatin regions were identified in the two QTL mappings, there was a considerable overlap of responsible regulatory loci. Among the 431 regulatory loci that were associated in trans
with boundary variations, 52.4% were also responsible for chromatin accessibility in trans
, and 58.0% of these dual chromatin QTLs were trans
-expression QTLs as well. On the other hand, 15.0% of cis
-QTLs for boundary variations were cis
-QTLs for chromatin accessibility. The overlapping fraction is low because a single marker cannot usually cover multiple different chromatin regions in cis
. However, 97.1% of these dual chromatin QTLs were cis
-expression QTLs. This cross-confirmation suggests that the regulatory loci identified in each QTL mapping may be functional with many of them exerting effects on transcription regulation.
To investigate the functional effect of boundary shifts on gene transcription, we examined the pattern of boundary variations in relationship with the transcription pattern of the gene whose expression level is associated with the same genetic marker and whose tss is located within 1 kb from the open chromatin of question. For example, in the locus illustrated in , the expression level of TAT1 () and the boundary location of the upstream open chromatin peak () are both associated with common local genetic markers. In this case, the gene is transcribed from right to left, and the left boundary (orange box in ), but not the right boundary, of the chromatin peak was genetically associated. The strains with the RM genotype at this locus tend to have the left boundary farther from that in the BY strain and closer to the tss () and have higher expression levels of the gene (). In fact, the distance of the left boundary to the tss was correlated with the expression level ().
Effects of chromatin border regulation on nearby gene expression.
We found that in all cases in which a boundary location is associated with a local or distant genetic marker in common with the expression level of a gene located within 1 kb from the chromatin peak, only the boundary that faces the tss, but not the boundary on the other side, has been identified in the QTL mapping. Therefore, the example provided in is a general feature of the relationship between chromatin border regulation and gene expression regulation. This is a novel finding and it is currently unclear by what mechanism the border of accessible chromatin can affect or be affected by the transcription of the gene it faces. Active histone modifications on the boundary nucleosome or an active physical interaction of TFs and RNA polymerase II may result in an extension of chromatin borders towards the tss.
Our results reveal an evolutionarily conserved feature of nucleosome positioning within accessible chromatin. The nucleosomes residing at the boundaries of open chromatin seems to play a role in demarcating functional regulatory regions such that DNA binding events take place in between these flanking nucleosomes in the middle of the accessible chromatin area. We also found that the positioning of these demarcating nucleosomes is coupled with in vivo TF binding events and that the sequence preferences of the underlying DNA for nucleosome formation are proportional to genetic variation in the size of the accessible region. Therefore, the variation in the width of accessible chromatin regions caused by the locational changes of the open chromatin borders may arise in concert with the modulation of the boundary nucleosomes by post-translational histone modifications and by chromatin regulators and in association with the activity of nearby gene transcription.