Determinants of nucleosome organization in primary human cells.

Nucleosomes are the basic packaging units of chromatin, modulating accessibility of regulatory proteins to DNA and thus influencing eukaryotic gene regulation. Elaborate chromatin remodelling mechanisms have evolved that govern nucleosome organization at promoters, regulatory elements, and other functional regions in the genome. Analyses of chromatin landscape have uncovered a variety of mechanisms, including DNA sequence preferences, that can influence nucleosome positions. To identify major determinants of nucleosome organization in the human genome, we used deep sequencing to map nucleosome positions in three primary human cell types and in vitro. A majority of the genome showed substantial flexibility of nucleosome positions, whereas a small fraction showed reproducibly positioned nucleosomes. Certain sites that position in vitro can anchor the formation of nucleosomal arrays that have cell type-specific spacing in vivo. Our results unveil an interplay of sequence-based nucleosome preferences and non-nucleosomal factors in determining nucleosome organization within mammalian cells.

Previous studies in model organisms[3-7] as well as initial analyses in human cells[8] have identified fundamental aspects of nucleosome organization. We here focus on the dynamic relationships between sequence-based nucleosome preferences and chromatin regulatory function in primary human cells. We mapped tissue-specific and DNA-encoded nucleosome organization across granulocytes and two types of T-cells (CD4+ and CD8+) isolated from the blood of a single human donor, by isolating cellular chromatin and treating it with micrococcal nuclease (MNase) followed by deep sequencing of the resulting nucleosome-protected fragments (Methods, Supplementary Fig. 1). To provide sufficient depth for both local and global analyses, we used high-throughput SOLiD technology, generating 584, 342, and 343 million mapped reads for granulocytes, CD4+, and CD8+ T-cells, respectively. These are equivalent to 16x–28x genome coverage by 147 bp nucleosome footprints (cores; see Methods). The depth of sequence was critical for our subsequent analysis: while shallower coverage can illuminate features of nucleosome positions through statistical analysis (e.g. 6,8), any definitive map and thus comparison of static and dynamic positioning requires high sequence coverage throughout the genome.

To provide complementary data on purely sequence-driven nucleosome positioning in the absence of cellular influences, we reconstituted genomic DNA in vitro with recombinantly derived histone octamers to produce in vitro nucleosomes (Methods, Supplementary Fig. 2), and generated over 669 million mapped reads, representing 32x core coverage of the genome. To identify primary nucleosome positioning sites in DNA, the reconstitution was performed under conditions of DNA excess (see methods). We also generated a control dataset of 321 million mapped reads from MNase-digested naked DNA (Supplemental Materials). In the population of granulocytes (our deepest in vivo data set), over 99.5 % of the mappable genome is engaged by nucleosomes (Methods), and 50 percent of nucleosome-depleted bases occur in regions shorter than 160 bp.

We first focused on global patterns of nucleosome positioning and spacing by calculating fragment distograms and phasograms[6,7,9]. Distograms (histograms of distances between mapped reads’ start positions aligning in opposing orientation, Supplementary Fig. 3A) reveal the average core fragment size as a peak if there are many sites in the genome that contain consistently positioned nucleosomes. A positioning signal that is strongly amplified by conditioning the analysis on sites with 3 or more read starts (reflecting a positioning preference; 3-pile subset), is present not only in vivo (Fig. 1A), but also in vitro (Fig. 1B), demonstrating that many genomic sites bear intrinsic, sequence-driven, positioning signals. Phasograms (histograms of distances between mapped reads’ start positions aligning in the same orientation, Supplementary Fig. 3B) reveal consistent spacing of positioned nucleosomes by exhibiting a wave-like pattern with a period that represents genome-average internucleosome spacing. In granulocytes, the wave peaks are 193 bp apart (Fig. 1C, adjusted R2=1, p-value<10−15), which, given a core fragment length of 147 bp, indicates an internucleosome linker length of 46 bps. By contrast, the phasograms of both types of T-cells have spacing that is wider by 10 bp (Fig. 1D), equivalent to a 56 bp average linker length. These results are consistent with classical observations of varying nucleosome phases in different cell types[10,11]. Linker length differences have been tied to differences in linker histone gene expression[12,13], which we found to be 2.4 times higher in T-cells compared to granulocytes (84 RPKM[14] vs.35 RPKM). The in vitro phasogram (Fig. 1E) reveals no detectable stereotypic spacing of positioned nucleosomes, demonstrating a lack of intrinsic phasing among DNA-encoded nucleosome positioning sites.

Figure 1

Global parameters of cell-specific nucleosome phasing and positioning in human

a, In vivo granulocyte distogram (calculation explained in Supplementary Fig. 3A). X axis represents the range of recorded distances. Y axis represents frequencies of observed distances within 1-pile (blue) and 3-pile (red) subsets. 1-pile subset represents the entire dataset, 3-pile subset represents a subset of sites containing 3 or more coincident read starts. b, Distogram of the in vitro reconstituted nucleosomes showing 1-and 3-pile subsets as in (a). c, In vivo granulocyte phasogram (calculation explained in Supplementary Fig. 3B). X axis shows the range of recorded phases. Y axis shows frequencies of corresponding phases. Plotted are phasograms of 1-, 3-, and 5-pile subsets. Inset, linear fit to the positions of the phase peaks within 3-pile subsets (slope = 193 bp). d, Phasograms of blood cell types. Inset, linear fits in CD4+ T-cells (203 bp) and granulocytes (193 bp). e, Phasograms of 1-, 3-and 5-pile subsets in the in vitro data.

Using a positioning stringency metric (Methods; Supplementary Fig. 4) that quantifies the fraction of defined nucleosome positions within a given segment, we calculated the fraction of the genome that is occupied by preferentially positioned nucleosomes at different stringency thresholds. The maximum number of sites at which some positioning preference can be detected statistically is 120 M, covering just over 20% of the genome (Supplementary Fig. 5)at the low stringency of 23%. Thus, the majority of nucleosome positioning preferences is weak, and nucleosomes across the majority of the human genome are not preferentially positioned, either by sequence or by cellular function.

We next focused on how transcription and chromatin functions affect nucleosome organization regionally. For each cell type, we generated deep RNA-seq data and binned genes into groups according to their expression levels. The average spacing of nucleosomes was greatest within silent genes (CD4+ T-cells, 206 bp, Fig. 2A) and decreased by as much as 11 bp as the expression levels went up (t-statistic p-value 6.5×10−34). This suggests that transcription-induced cycles of nucleosome eviction and reoccupation cause denser packing of nucleosomes and slight reduction in nucleosome occupancy (Supplementary Fig. 6). On the basis of this result, we hypothesized that higher-order chromatin organization as implied by specific chromatin modifications might be associated with specific spacing patterns. Using previously published ChIP-seq data, we identified regions of enrichment[15] for histone modifications that are found within heterochromatin (H3K27me3, H3K9me3)[16], gene-body euchromatin (H4K20me1, H3K27me1)[16], or euchromatin associated with promoters and enhancers (H3K4me1, H3K27ac, H3K36ac)[17], and estimated spacing of nucleosomes for each of these epigenetic domains. We found that active promoter-associated domains contained the shortest spacing of 178–187 bp, followed by a larger spacing of 190–195 bp within the body of active genes, while heterochromatin spacing was largest at 205 bp (Fig. 2B). These results reveal striking heterogeneity in nucleosome organization across the genome that depends on global cellular identity, metabolic state, regional regulatory state, and local gene activity.

Figure 2

Transcription and chromatin modification-dependent nucleosome spacing

a, Nucleosome spacing as a function of transcriptional activity. X axis represents gene expression values binned according to RPKM values. Internucleosome spacing is plotted along the Y axis. Dashed lines represent genome-wide average spacing for each cell type. b, Nucleosome spacing within genomic regions marked by specific histone marks in CD4+ T-cells. Bar height plots estimated nucleosome spacing for each histone modification. Bar colors differentiate chromatin types (euchromatin vs heterochromatin).

To characterize DNA signals responsible for consistent positioning of nucleosomes, we identified 0.3 million sites occupied in vitro by nucleosomes at high stringency (> 0.5; Methods). The region occupied by the center of the nucleosome (dyad) exhibits a significant increase in G/Cusage(Poisson p-value < 10−100; Fig. 3A). Flanking regions increase in A/T usage as the positioning strength increases (Fig. 3B). A subset of in vitro positioned nucleosomes (stringency > 0.5) which are also strongly positioned in vivo (stringency > 0.4) revealed increased A/T usage within the flanks (Fig. 3C) compared to in vitro-only positioning sites (Fig. 3A), which underscores the importance of flanking repelling elements for positioning in vivo. We term such elements with strong G/C cores and A/T flanks “container sites” to emphasize the proposed positioning mechanism (Fig. 3D). This positioning signal is different from a 10 bp dinucleotide periodicity observed in populations of nucleosome core segments isolated from a variety of species[19,20] and proposed to contribute to precise positioning and/or rotational setting of DNA on nucleosomes[20] on a fine scale(Supplementary Fig. 7). G/C rich signals are known to promote nucleosome occupancy[18,21], while AA-rich sequences repel nucleosomes[4], and our data demonstrate that precise arrangement of a core-length attractive segment flanked by repelling sequences can produce a strongly positioned nucleosome (Fig. 3D).

Figure 3

Sequence signals that drive nucleosome positioning

a, Sequence signals within sites containing moderately positioned in vitro nucleosomes (stringency > 0.5). Distance from the positioned dyad to a given dinucleotide are plotted along the X axis; Y axis represents frequency of a given k-mer divided by its genome-wide expectation. The 147 bp footprint of a nucleosome is indicated by an orange band. b, Changes in AA dinucleotide usage with increasing positioning stringency. X and Y axes same as in (a). Shown are curves of AA usage within the sites of increasingly positioned dyads (stringency cutoffs of 0.4, 0.5, 0.6, 0.7). c, Sequence signals within sites containing in vitro positioned nucleosomes (stringency > 0.5) that also have high in vivo stringency (stringency > 0.4). X and Y axes same as in (a). d, Schematic depiction of the container site positioning mechanism. The C/G-rich core area (green) favors occupancy, but does not precisely position the nucleosome (top). Adding flanking A/T-rich repelling elements (purple, bottom) restricts the position of the nucleosome. e, Nucleosome organization around container sites in vivo and in vitro. X axis represents distances from the dyads to container sites(based on 300,000 container sites). Frequencies of nucleosome dyads around those sites are plotted along the Y axis. The upper plot shows distribution of in vivo dyads across CD4+ cells, CD8+ cells, and granulocytes. The ovals depict hypothetical nucleosome positions across the site with color intensities reflecting their positioning strength. The lower plot shows distribution of dyads in vitro and in MNase control.

Dyad frequencies around container sites (Fig 3E) show a strong peak of enrichment in vivo, confirming that DNA positions nucleosomes in vivo over these sites. Additionally, wave-like patterns emanate from these sites in vivo (but not in vitro), reflecting the nucleation of phased arrays by positioned cellular nucleosomes. Viewing these results in light of the nucleosome barrier model[22], which proposes that nucleosomes are packed into positioned and phased arrays against a chromatin barrier, we conclude that sequence-positioned nucleosome can initiate propagation of adjacent stereotypically positioned nucleosomes. Importantly, wave periods around container sites are shorter in granulocytes than in T-cells, allowing tissue-specific variation in linker length (Fig. 1D) to alter placement of nucleosomes over distances of as much as 1 kb from an initial container site. Functional consequences of such rearrangements might include global shifts in regulatory properties that could contribute to distinct transcription factor accessibility profiles in different cell types.

The cellular environment can drive nucleosomes to sequences not intrinsically favorable to being occupied, as evident in a genome-wide comparison of observed nucleosome coverage of all possible tetranucleotides between the granulocyte and the in vitro data (Fig. 4A). In vitro, nucleosome occupancy is strongly associated with AT/GC content, but this preference is abolished in vivo; the exception are C/G rich tetramers that contain CpG dinucleotides, which show a 30% reduction in apparent nucleosome occupancy despite having high core coverage in vitro. Consistent with this, CpG islands are five-fold depleted for observed nucleosome coverage in vivo (Fig. 4B). No such decrease is observed in the in vitro dataset.

Figure 4

Influence of gene regulatory function on nucleosome positioning

a, Comparison of sequence preferences of nucleosomes in vivo and in vitro. Normalized nucleosome core coverage in vivo (granulocytes) for a given sequence 4-mer are plotted along the X axis. In vitro core coverage is plotted along the Y axis. Each data point on the plot represents one of the 256 possible 4-mers (colored according to their G/C content). The diagonal line depicts the positions in the plot for which sequence-based preferences of nucleosomes would be the same in vivo and in vitro. b, Nucleosome core coverage over CpG islands in vivo and in vitro. X axis represents coordinates within CpG islands (0–100%) and flanking upstream of the transcriptional start sites (TSS) (left) and downstream of the TSS (right). Normalized frequencies of nucleosome cores in vivo (upper plot) and in vitro (lower plot)are plotted along the Y axis. c, In vivo CD4+ T-cell nucleosome organization around promoters. X axis represents distance from the TSS (blue arrow). Normalized frequencies of nucleosome dyads are plotted along the Y axis. Shown are nucleosome arrangements within four gene groups (not expressed 0–0.1 RPKM, low expressed 0.1–1 RPKM, moderately expressed 1–8 RPKM, highly expressed >8 RPKM). Pie chart depicts distribution of RPKM values across gene groups. d, RNA Pol II binding signal within highly expressed genes (orange curve) and H3K4me3-marked nucleosome dyad frequency (green curve) within highly expressed genes(>8 RPKM). Nucleosomes show consistent positions, indicated by grey lines pointing to nucleosome centers. e, Schematic depiction of nucleosome organization around promoters of repressed and active genes. Promoters of repressed genes do not have a well-defined nucleosome organization, while promoters of active genes have a nucleosome free region (NFR, blue), RNA Pol II (orange) localized at the NFR boundary, and positioned nucleosomes (red) radiating from the NFR. Height of the ovals represents nucleosome frequency(inferred from c). f, Nucleosome distribution around the top 1000 NRSF sites in vivo and in vitro. Distances from the NRSF binding sites are plotted along the X axis. Y axis represents the normalized frequency of nucleosome dyads. Blue ovals depict hypothetical nucleosome positions. NRSF binding site is shown by the green rectangle.

We hypothesize that the decreased nucleosome occupancy of promoters could be due to promoter-related functions of mammalian CpG islands, similar to promoter-associated nucleosome-free regions observed in flies[23] and yeast[5], which do not have CpG islands. We therefore analyzed transcription-dependent nucleosome packaging around promoters. As in other organisms[23-27], promoters of active genes have a nucleosome-free region (NFR) of about 150 bp overlapping the transcriptional start site and arrays of well-positioned and phased nucleosomes that radiate from the NFR (Fig. 4C). A notable reduction in apparent nucleosome occupancy extends up to 1 kb into the gene body. We also observed consistent nucleosome coordinates in an independent data set of H3K4me3-bearing nucleosomes[16] (Fig. 4D). Comparison of the nucleosome data (Fig 4D) with binding patterns of RNA Polymerase II[16] (Fig 4D) around active promoters indicates that phasing of positioned nucleosomes can be explained by packing of nucleosomes against Pol II stalled at the promoter, with Pol II potentially acting as the “barrier”. The set of inactive promoters, by contrast, exhibits neither a pronounced depletion of nucleosomes, nor a positioning and phasing signal (Fig. 4C). The transition of an inactive promoter to an active one is therefore likely to involve eviction of nucleosomes, coupled with positioning and phasing of nucleosomes neighboring RNA Pol II (Fig. 4E). These results suggest that CpG-rich segments in mammalian promoters override intrinsic signals of high nucleosome affinity (Supplementary Fig. 8) to become active; this would be in contrast to fly and yeast, where AT-rich promoters may comprise intrinsic sequence signals that are particularly prone to nucleosome eviction[28].

To explore how regulatory factors interact with sequence signals to influence nucleosome organization outside of promoters, we focused on binding sites of the NRSF repressor protein[15] and the insulator protein CTCF. NRSF and CTCF sites are flanked by arrays of positioned nucleosomes (Fig. 4G, Supplementary Fig. 9), consistent with barrier-driven packing previously reported for CTCF[29,30]. Both proteins occupy additional linker space, with NRSF taking up an extra 37 bp and CTCF 74 bp. In agreement with sequence-based predictions[21], both CTCF and NRSF sites intrinsically encode high nucleosome occupancy as can be seen from the in vitro data (Fig. 4F, Supplementary Fig. 9), but this signal is overridden in vivo by occlusion of these sites from associating with nucleosomes. Additionally, phasing of nucleosomes around these regulatory sites is more compact in granulocytes compared to T-cells (Supplementary Fig. 9), again exemplifying the importance of cellular parameters for placement of nucleosomes.

Our genome-wide, deep sequence data of nucleosome positions facilitated an initial characterization of the determinants of nucleosome organization in primary human cells. Spacing of nucleosomes differs between cell types and between distinct epigenetic domains in the same cell type, and is influenced by transcriptional activity. We confirm positioning preferences in regulatory elements such as promoters and chromatin regulator binding sites, but find that the majority of the human genome exhibits little if any detectable positioning. The influence of sequence on positioning of nucleosomes in vivo is modest but detectable. Despite DNA sequence being a potent driver of nucleosome organization at certain sites, the cellular environment often overrides sequence signals and can drive nucleosomes to occupy intrinsically unfavorable DNA elements or evict nucleosomes from intrinsically favorable sites. We find evidence for the barrier model for nucleosome organization, and that barriers can be nucleosomes (positioned by container sites), RNA polymerase II (stalled at the promoter), or sequence-specific regulatory factors. Our nucleosome maps should be useful for investigating how nucleosome organization affects gene regulation and vice versa, as well as for pinpointing the mechanisms driving regional heterogeneity of nucleosome spacing.

Methods summary

Neutrophil granulocytes, CD4+ and CD8+ T-cells were isolated from donor blood using Histopaque density gradients and Ig-coupled beads against blood cell surface makers (pan T and CD4+ microbeads, Miltenyi Biotec). Nucleosome cores were prepared as previously described[7]; cells were snap-frozen and crushed to release chromatin, followed by micrococcal nuclease treatment. In vitro nucleosomes were prepared by combining human genomic DNA with recombinantly-derived histone octamers at an average ratio of 1 octamer per 850 bps. Unbound DNA was then digested using micrococcal nuclease. After digestion, reactions were stopped with EDTA, samples were treated with proteinase K, and nucleosome-bound DNA was extracted with phenol-chloroform and precipitated with ethanol (Supplementary methods). Purified DNA was size-selected (120–180 bp) on agarose to obtain mononucleosome cores, followed by sequencing library construction. RNA was isolated by homogenizing purified cells in Trizol, poly-A RNA was purified using Qiagen Oligotex kit and RNA-seq libraries were constructed using SOLiD Whole Transcriptome Analysis kit. All sequence data was obtained using SOLiD 35 bp protocol and aligned using the SOLiD pipeline against the human hg18 reference genome. Downstream analyses were all conducted using custom scripts (supplementary methods).