|Home | About | Journals | Submit | Contact Us | Français|
We assess the role of intrinsic histone-DNA interactions by mapping nucleosomes assembled in vitro on genomic DNA. Nucleosomes strongly prefer yeast DNA over E. coli DNA, indicating that the yeast genome evolved to favor nucleosome formation. Many yeast promoter and terminator regions intrinsically disfavor nucleosome formation, and nucleosomes assembled in vitro display strong rotational positioning. Nucleosome arrays generated by the ACF assembly factor display fewer nucleosome-free regions, reduced rotational positioning, and less translational positioning than obtained by intrinsic histone-DNA interactions. Importantly, in vitro assembled nucleosomes display only a limited preference for specific translational positions and do not show the pattern observed in vivo. Our results argue against a genomic code for nucleosome positioning, and they suggest that the nucleosomal pattern in coding regions arises primarily from statistical positioning from a barrier near the promoter that involves some aspect of transcriptional initiation by RNA polymerase II.
Eukaryotic genomes are packaged into regularly spaced arrays of nucleosomes, although the spacing between nucleosomes varies among species and cell types1. Despite this regularity, high-resolution, genome-wide analyses reveal a common nucleosomal pattern2-8. Nucleosomes are depleted at many (but not all) enhancer, promoter, and terminator regions, and they typically occupy preferred positions within mRNA coding regions and just upstream of the promoter. In yeast, the -1 and +1 nucleosomes flanking the promoter are strongly positioned, and the degree of nucleosome positioning gradually decreases from the 5′ to 3′ end of the coding region6,8.
There are three distinct mechanisms for nucleosome depletion in vivo. First, specific DNA-binding activator proteins can generate nucleosome-depleted regions by recruiting ATP-dependent remodeling complexes and histone acetylases9-11; this mechanism is independent of transcriptional activity12. Second, the process of transcriptional elongation by RNA polymerase II involves cycles of histone eviction and reassembly, and continuous high levels of Pol II elongation reduce histone density within coding regions13-15. Third, certain DNA sequences, notably poly(dA:dT) tracts, intrinsically disfavor nucleosome formation in vitro, increase chromatin accessibility in vivo, and are strongly over-represented in nucleosome-free regions16,17,18, indicating a role of intrinsic histone-DNA interactions.
The positioning of nucleosomes along the DNA is related to, but distinct from, the issue of nucleosome occupancy. Nucleosome occupancy or density reflects the average histone levels on a given region of DNA in a population of cells, but it does not address where an individual nucleosome is positioned with respect to a certain DNA sequence. Indeed, differently positioned nucleosomes within a given genomic region will all contribute to nucleosome density. Nucleosome positioning refers to two fundamental relationships between the histone octamer and the DNA wrapped around it. Rotational positioning defines the orientation of the DNA helix on the histone surface. Nucleosomes are rotationally positioned with a 10 bp helical periodicity, reflecting preferences for dinucleotides that face inwards or outwards with respect to the histones and optimize DNA bending19-21. The translational position of a nucleosome refers to the specific 146 bp sequence covered by the histone octamer, and it is often defined as the midpoint of this sequence. The degree of translational positioning can vary from perfect positioning, in which a nucleosome occupies a given 146 bp stretch in all DNA molecules in a population, to random positioning, in which nucleosomes occupy all possible genomic positions equally. In vivo, translational positioning is strongly influenced by relatively constant spacing between nucleosomes, which is presumably due to the action of nucleosome-remodeling complexes.
Nucleosomes can also be statistically positioned from a fixed barrier such as a DNA-binding protein22. Nucleosomes near the barrier are highly positioned, and the degree of positioning decreases in accord with the distance from the barrier due to variations in spacing between nucleosomes. A barrier model for statistical positioning can explain the location of nucleosomes in yeast genes8, but the molecular nature of the barrier is unknown.
Nucleosomes exhibit intrinsic DNA sequence preferences19,20,23, and many yeast promoter regions are nucleosome-depleted because the DNA sequence intrinsically disfavors nucleosome formation16,17. More generally, it has been proposed that there is a nucleosome code in which the pattern of nucleosome positioning in vivo is determined primarily by genomic DNA sequence and hence can be predicted16,24. Here, we use S. cerevisiae and E. coli DNA to examine the role of intrinsic histone-DNA interactions for establishing the nucleosomal pattern in vivo on a genome-wide basis.
The experimental design involves a comparison of nucleosome positions in samples prepared from yeast cells with those generated in vitro with purified histones and yeast genomic DNA, and it represents a genome-wide and high-resolution version of an approach we have used previously17. We assembled nucleosomes on a mixture of yeast and E. coli genomic DNAs either by salt dialysis or by ACF, an ATP-dependent chromatin assembly factor, using a 1:1 mass ratio of core histones:DNA to produce arrays of regularly-spaced nucleosomes (Fig. 1). Nucleosomes assembled by salt dialysis are formed solely by intrinsic histone-DNA interactions, and the linker regions between nucleosomes are very short. In contrast, ACF assembles nucleosomes into regularly spaced arrays with ~20 bp linker regions between nucleosomes that are similar to those observed in yeast cells. The comparison between nucleosomes assembled by salt dialysis or ACF addresses the issue of how chromatin assembly factors and nucleosome spacing affects nucleosome positioning. The assembled chromatin was treated with micrococcal nuclease under conditions in which mononucleosomes were the major product and linker regions were degraded. As a control, the same mixture of yeast and E. coli DNA was sonicated to generate DNA fragments of comparable size. Parallel DNA sequencing of the resulting samples generated 1-3 million uniquely mapped reads per sample, which corresponds to approximately 10-40 fold genome coverage.
Although we were primarily interested in the comparison between nucleosomes assembled in vitro and in vivo on yeast genomic DNA, we included E. coli DNA in the samples to address whether yeast cells evolutionarily selected for or against sequences that favor nucleosome formation. As E. coli does not have histones, we presumed that the DNA sequences in this organism are evolutionarily neutral with respect to nucleosome formation, such that preferred nucleosome-forming sequences will occur by chance. Interestingly, when compared to the sonication control, nucleosomes assembled in vitro by salt dialysis are 9 times more likely to form on yeast DNA than E. coli DNA (Fig. 1). A similar effect, albeit to a lesser extent (3-fold), is observed with ACF-assembled nucleosomes. These results strongly argue that the yeast genome has evolved to favor nucleosome formation.
For in vivo and in vitro samples, we generated a heat map of histone density at each nucleotide position and aligned these to promoters (Fig. 2a-c), terminators (Fig. 2d-f), and transcription factors (Fig. 2g, h) on a gene-by-gene basis. As observed in vivo, promoter and terminator regions are substantially nucleosome-depleted (visualized as blue in Fig. 2) in comparison to coding regions in chromatin assembled by salt dialysis or ACF. However, only a subset of nucleosome-depleted promoters in vivo are also depleted in chromatin assembled in vitro, and histone depletion at promoters is significantly more pronounced (darker blue) in vivo than in vitro. These observations indicate that intrinsic histone-DNA interactions are important for generating nucleosome-depleted promoter regions in vivo, but other mechanisms (e.g. activator-dependent recruitment of nucleosome-remodeling activities and perhaps transcriptional initiation) play a very significant (and perhaps greater) role at many promoters. Intrinsic histone-DNA interactions appear to be more important at terminator regions in vivo, because the correlation between in vitro and in vivo data is higher for terminators than promoters (Supplementary Fig. 1).
To address the role of intrinsic histone-DNA interactions on rotational and translational positioning, we used an approach that did not involve histone density, but rather treated each mapped read as an individual 146 bp nucleosome in the population. For chromatin assembled by salt dialysis, alignment of nucleosome 5′ ends reveals a striking 10-bp periodicity of AA/TT/AT dinucleotides both for yeast (Fig. 3a) and E. coli (Fig. 3b) DNA, indicating that intrinsic histone-DNA interactions play a major role in rotational positioning. Power spectrum analysis (a discrete Fourier transform that quantifies the amount vs. the frequency component of the data) shows that this 10-bp periodicity is significantly lower in nucleosomes assembled in vivo (Fig. 3c), suggesting that nucleosome assembly or remodeling factors decrease the effect of rotational positioning. Our results are consistent with the presumption that rotational positioning reflects the requirement of DNA to bend around the histone octamer, with more bendable sequences in contact with the histones, and less bendable sequences being solvent exposed. In this regard, the yeast genome shows preferential 10.2 bp periodicity of AA and TT dinucleotides, whereas the E. coli genome does not (Fig. 3d), and this likely contributes to the preferential assembly of nucleosomes on yeast DNA.
We determined the degree of translational positioning by defining the central position of each nucleosome and counting the number of hits within 20 bp windows (to allow for the imprecision of MNase cleavage). In principle, if a nucleosome is perfectly positioned (i.e. 100% of the DNA molecules contain a nucleosome center within the window), the expected number of hits is determined by the sequencing coverage, and we define this value as 1. Thus, the degree of nucleosome positioning is defined as the observed number of hits divided by the expected number for a perfectly positioned nucleosome. As expected, chromatin from yeast cells shows a great deal of translational positioning, with many nucleosomes being highly positioned (Fig. 4a; Supplementary Fig. 2). Nucleosomes assembled by either in vitro method show significantly less translational positioning than observed in vivo. Nevertheless, translational positioning of these in vitro assembled nucleosomes is above the level observed in randomly positioned nucleosomes.
To determine the contribution of intrinsic histone-DNA interactions for the nucleosome positioning pattern in vivo, we defined the peak 20-bp window for the +1, +2, etc. nucleosomes on an individual gene basis using in vivo nucleosome data16. We then determined the number of nucleosomes within each of these defined windows in the various nucleosome preparations (Fig. 4b-d). For nucleosomes generated by salt dialysis or ACF, nucleosome positioning is above that expected from randomly positioned nucleosomes (Supplementary Fig. 3). However, the degree of translational positioning is far below that observed with an independent sample of nucleosomes from yeast cells (compare Figs. 4b,c with 4d). These observations indicate that intrinsic histone-DNA interactions make only a modest contribution to the in vivo pattern of translational positioning, and we estimate that they account for ~20% of the in vivo positions (Supplementary Fig. 3). More importantly, unlike the situation in vivo (Fig. 4d), this modest contribution of intrinsic positioning is not significantly affected by the location of the positioned nucleosome with respect to the mRNA initiation site (i.e. nucleosomes defined as +1 through +10 nucleosomes), indicating that other features underlie this positioning pattern.
We assessed how a chromatin assembly factor affects intrinsic nucleosome positioning by comparing the samples generated in vitro by ACF and salt dialysis. First, both the fraction of nucleosome-depleted promoters and the degree of histone depletion are lower in chromatin assembled by ACF as compared to salt dialysis (Fig. 2). Second, the 10-bp periodicity is significantly lower in nucleosomes assembled by ACF (Fig. 3b), indicating that nucleosome assembly decreases the effect of rotational positioning. Third, ACF reduces the degree of translational positioning beyond that due to intrinsic histone-DNA interactions (Fig. 4a). These observations indicate that ACF diminishes the effect of intrinsic histone-DNA interactions, presumably because nucleosome spacing constraints force nucleosomes to occupy positions that are not preferred solely based on DNA sequence. These observations are consistent with the function of the Isw2 nucleosome-remodeling complex in yeast cells25 and with the reduced rotational positioning in vivo (Fig. 3b). Further, they suggest that nucleosome assembly factors that govern nucleosome spacing in vivo will diminish the effects of intrinsic histone-DNA interactions on histone density as well as rotational and translational positioning.
In yeast cells, the +1 nucleosome is highly positioned, and the degree of positioning decreases progressively at more downstream positions within the coding region8, a hallmark of statistical positioning. Thus, the mechanism by which the +1 nucleosome becomes highly positioned is the key to understanding the nucleosomal pattern in vivo. Our results in Figs. 2 and and44 demonstrate conclusively that the strong positioning of the +1 nucleosome is not due to intrinsic histone-DNA interactions. Furthermore, several observations indicate that nucleosome-depleted regions per se are insufficient to constitute the barrier needed for statistical positioning. First, statistical positioning occurs to a much lesser extent for nucleosomes upstream of nucleosome-free promoter regions (Fig. 2c), and hence is largely directional. Second, statistical positioning is very limited in either direction from the nucleosome-depleted terminator regions (Fig 2f). Third, strong statistical positioning is observed in both directions from Abf1 and Reb1 sites in promoter regions (Fig. 2g,h), indicating that these DNA-binding proteins (with associated factors) can serve as barrier elements for statistical positioning. Rap1, and (to a lesser extent) Mbp1 and Cbp1 also serve as barrier elements, but this is not the case for many other transcription factors (Supplementary Fig. 4). Lastly, the spacing between nucleosome ends (i.e. tag position relationships) for the entire data set reveals strong statistical positioning in vivo, whereas this is very limited in the salt-assembled chromatin and not detectable in ACF-assembled chromatin (Fig. 4e).
The observation that statistical positioning occurs directionally from promoter regions strongly argues that the transcription initiation process is critical for establishing the strong translational position of the +1 nucleosome. This view is reinforced by the striking relationship between position of the +1 nucleosome and the location of the mRNA initiation site (Fig. 4f). We therefore propose that, although intrinsically nucleosome-depleted regions facilitate assembly of the preinitiation complex, an early step(s) in the transcription process (i.e. those preceding extensive elongation) is the major determinant for a strongly positioned +1 nucleosome (see Discussion).
Our results strongly argue against the idea of a nucleosome code in which nucleosome positions in vivo are determined primarily by DNA sequence16,24. Most significantly, the in vivo pattern of statistical positioning (the strong positioning of the +1 and to a progressively lesser extent more downstream nucleosomes) is not observed in vitro, and instead is linked to the process of transcriptional initiation. In addition, a nucleosome code does not explain why an S. cerevisiae genomic region shows different nucleosomal positions when present in S. cerevisiae or S. pombe cells17, and it does not easily account for species- and cell-type-specific differences in nucleosome spacing1. Our genome-wide analysis is consistent with the well established ideas that histones have significant DNA sequence preferences for nucleosome formation and rotational positioning19-21, and that intrinsic histone-DNA interactions play an important role in generating nucleosome-depleted promoter regions17,18. However, the fact that histones have DNA sequence preferences for nucleosome formation is conceptually different from a nucleosome code that determines where nucleosomes are located in vivo. By analogy, DNA sequence motifs for a transcription factor do not constitute a code that is sufficient to specify where a protein binds in vivo26,27.
Our central conclusion disagrees with that of a related study16 that appeared after our work was completed and also involved large-scale sequencing of nucleosomes assembled on yeast genomic DNA by salt dialysis. However, our analysis of this independent dataset (and a related analysis of this data; Fig. 4a of Ref. 9) confirms, and indeed provides additional support for, the conclusion that the in vivo pattern of statistically positioned nucleosomes is not due to intrinsic histone-DNA interactions (Supplementary Fig. 5). The conflicting conclusions of these two studies are largely explained by the difference between the concepts of histone density and translational positioning. Although these concepts are related, histone density measures the average amount of histones on a given region of DNA in a population, and this cannot be used to determine translational positioning, which refers to the precise position of an individual nucleosome with regard to a given DNA sequence. Our central conclusion is based primarily on translational positioning analysis, whereas the major conclusion of the other study was based on the high correlation coefficient of histone densities between the in vivo and in vitro samples (see below) rather than on direct measurements of translational positioning. Thus, while low nucleosome density at many promoters and terminators in vivo is largely due to intrinsically weak histone-DNA interactions, the translational positions of where nucleosomes are actually located in vivo are not determined primarily by the underlying DNA sequence.
The correlation of histone densities between the in vivo and in vitro samples is lower in our experiments than observed in the independent dataset (0.74 vs. 0.54; Supplementary Fig. 6). However, correlation coefficients are strongly affected by methodological issues. Chromatin reconstitution in our study involved equal amounts of histones and DNA and the generation of regularly spaced nucleosome arrays. In contrast, reconstitution in the other study involved a histone:DNA mass ratio of 0.4, therefore resulting in isolated nucleosomes (~1 nucleosome/400 bp) with most of the DNA not in the form of chromatin. The use of limiting histone concentrations is advantageous for measuring intrinsic affinities of different genomic regions, but it also represents an artificial competitive situation between DNA sequences that does not reflect the conditions in vivo where histones are not limiting and the majority of the genomic DNA is nucleosomal.
Alternatively, the different correlation coefficients might arise from differences in the extent of micrococcal nuclease cleavage of the in vitro-generated samples used in the two studies (data from the same in vivo samples were used in both analyses). This can occur because the extent of enzymatic cleavage affects the ratio of cleavage within linker or mononucleosomal DNA, and the considerable sequence specificity of micrococcal nuclease28 results in both differential cleavage of linker regions and differential cleavage of mononucleomes. In addition, correlation coefficients, and hence apparent similarities, of histone densities between in vitro and in vivo samples are likely to be inflated by the DNA sequence specificity of micrococcal nuclease and biases in DNA sequencing that apply to the analysis of all mononucleosomal samples. Indeed, when contributions of micrococcal nuclease and sequencing bias are removed by measuring nucleosome affinities of a collection of 150-bp regions, correlation coefficients are significantly reduced16.
Most importantly, these methodological issues do not affect our conclusion that intrinsic histone-DNA interactions are not the major determinant of nucleosome positioning in vivo. Indeed, our conclusions are robust to different datasets done under different experimental conditions by different laboratories.
The mechanism by which the +1 nucleosome becomes highly positioned is the key to understanding how statistical positioning within mRNA coding regions is achieved. By definition, strong positioning of the +1 nucleosome must be determined by specific DNA sequences. In contrast to the prediction from the nucleosome code hypothesis, we demonstrate that strong positioning of the +1 nucleosome is not due to intrinsic histone-DNA interactions, and that an intrinsic nucleosome-depleted region is insufficient to strongly position a nucleosome. Instead, the position of the +1 nucleosome is highly related to the transcriptional initiation site, and hence depends on DNA sequences linked to the process of transcriptional initiation.
In support of a transcription-based mechanism, Drosophila genes also show a highly preferential spacing between the transcriptional initiation site and the +1 nucleosome4, although the spacing is larger. Interestingly, because the spacing between the preinitiation complex and mRNA initiation site is larger in yeast than in Drosophila29, the distance between the preinitiation complex and +1 nucleosome is roughly similar in both species. This suggests the possibility that the location of the preinitiation complex may be an important determinant for the location of the +1 nucleosome. However, it is also possible that the difference in spacing might be related to the fact that, unlike the case for yeast genes, many Drosophila genes have a paused RNA polymerase located downstream of the initiation site30.
We propose the following transcription-based model for positioning the +1 nucleosome. First, as discussed previously17,18, intrinsically nucleosome-depleted promoter regions facilitate association of the transcription machinery. Second, some component(s) of the transcriptional initiation machinery interacts with a nucleosome-remodeling complex and/or histones to position the +1 nucleosome. Third, once positioned, the +1 nucleosome might spatially constrain subsequent initiation events, thereby reinforcing the position of the +1 nucleosome. An appealing feature of this model is that it can explain why statistical positioning occurs on genes that are poorly transcribed. In particular, Pol II is detected at nearly all yeast genes31,32, and nucleosome replacement is inversely correlated with transcription33. Hence, it is likely that, once formed via an initiation event, a positioned +1 nucleosome will be relatively stable at genes whose transcription is initiated infrequently. In any event, whatever the precise molecular mechanism, the ultimate DNA sequence determinant for positioning the +1 nucleosome is an event linked to transcriptional initiation, not the intrinsic sequence preferences of histones to form nucleosomes.
We purified S. cerevisiae and E. coli genomic DNAs separately, lightly sonicated them to generate fragments 5-10 kb in length, and then combined them in a 3:1 mass ratio. We assembled the resulting DNA mixture into chromatin either by salt dialysis or by using a purified system containing recombinant Drosophila NAP-1 and ACF as well as purified native histones from Drosophila embryos as described previously34. We performed three independent chromatin assembly reactions on the same DNA mixture. We extensively digested chromatin with micrococcal nuclease to yield core particles, purified the resulting mononucleosomal DNA, and then performed massively parallel sequencing on an Illumina Genome Analyzer. We aligned sequence tags to the S. cerevisiae (Stanford Genome Database Apr 2008 build) and Escherichia coli K12 MG1655 (U00096) genomes, allowing 2 mismatches per mapped read. Depending on the sample, we generated 1-3 million uniquely mapped sequence tags. As a control, we sonicated the mixture of yeast and E. coli DNA to fragments of comparable size to mononucleosomal DNA.
We generated nucleosome density profiles by extending the 5′ end of each sequence tag 146 bp and then piling up all extended tags. We obtained heat maps of nucleosome density profiles (normalized to enrichment ratios) for 1752 isolated promoters aligned by transcriptional start sites (TSS), 1548 isolated terminator regions aligned by termination sites (TTS), and promoter regions aligned by sequence motifs of DNA-binding transcription factors. To avoid complications arising from genomic regions with multiple functions, we generated heat maps on genes with isolated promoter or terminator regions, which are defined as 1 kb regions upstream or downstream of a given gene that do not overlap with other genes.
We defined rotational positioning by the periodicity of AA/TT/AT dinucleotides in the mononucleosomal DNA. We determined dinucleotide periodicity and tag position relationships by aligning the 5′ ends of sequence tags as described previously7. For power spectrum analysis of the mononucleosomal samples, we performed discrete Fourier transform on the +11 to +160 interval of AA/TT/AT fraction pattern to measure the power of 10 bp periodicity. To apply power spectrum analysis to yeast and E. coli DNA, we converted these sequences to binary sequence according to whether AA/TT/AT was present at each dinucleotide position, split this binary sequence into 1024 bp fragments, and then applied the Fourier transform.
The degree of translational positioning is defined as the number of nucleosome centers (the position 73 bp downstream from the 5′ end of the sequence tag) within 20 bp windows centered at each nucleotide position divided by the number of nucleosome centers with a 160 bp window centered at the same position. The maximum positioning degree within a 160 bp window is the positioning degree of most “positioned” nucleosomes contained in this region; hence, the percentage of genomic regions with maximum positioning degree larger than certain threshold is a global measurement of nucleosome positioning status. To compare the nucleosome positioning status among samples, we randomly sampled the same number of tags along with a control comprising the same number of random tags in mappable genomic regions
We defined the nucleosome patterns of 3774 non-overlapping genes, on a gene by gene basis, as the peak 20-bp window for the +1 through +10 nucleosome centers based on chromatin from yeast cells grown in YP ethanol16. We first generated the average positions of the +1 to +10 nucleosomes for these 3774 genes, and then for each gene, identified the peak 20-bp window within the 100 bp region defined by the average position. We then determined the number of nucleosome centers within these defined 20-bp windows in nucleosomes generated by salt dialysis, ACF, or in cells grown in YPD medium.
We thank Istvan Albert, Zhenhai Zhang, and Frank Pugh for analyses and commentary during the early stages of this work, Ying Lei and Hyunjin Shin for help with the heat maps and power spectrum analysis, respectively. This work was supported by grants to J.T.K. (GM 58272), X.S.L. (HG 4069), K.S. (GM 30186) from the National Institutes of Health.
Data Accession Numbers: All the DNA sequencing data in this work is deposited at GEO with an accession number of GSE15188.
Author Contributions: Y.Z. performed all the bioinformatic analyses and contributed to writing of the manuscript; Z.M. prepared the genomic DNAs and the sonicated DNA sample, prepared the micrococcal nuclease-treated samples for DNA sequencing, and contributed to the experimental design, data analysis, and writing of the manuscript; B.P.R. assembled nucleosomes on genomic DNAs; J.T.K. contributed to experimental design and analysis of the nucleosome samples; G.E. sequenced the DNA samples with contribution from M.S.; X.S.L. contributed to the design and interpretation of the bioinformatic analyses as well as the writing of the manuscript; K.S. conceived the project, contributed to the data analysis, and wrote most of the manuscript.