Characterizing the nuclease accessibility of DNA in human cells to map higher order structures of chromatin

Abstract

Packaging of DNA into chromatin regulates DNA accessibility and consequently all DNA-dependent processes. The nucleosome is the basic packaging unit of DNA forming arrays that are suggested, by biochemical studies, to fold hierarchically into ordered higher-order structures of chromatin. This organization has been recently questioned using microscopy techniques, proposing an irregular structure. To address the principles of chromatin organization, we applied an in situ differential MNase-seq strategy and analyzed in silico the results of complete and partial digestions of human chromatin. We investigated whether different levels of chromatin packaging exist in the cell. We assessed the accessibility of chromatin within distinct domains of kb to Mb genomic regions, performed statistical analyses and computer modelling. We found no difference in MNase accessibility, suggesting no difference in fiber folding between domains of euchromatin and heterochromatin or between other sequence and epigenomic features of chromatin. Thus, our data suggests the absence of differentially organized domains of higher-order structures of chromatin. Moreover, we identified only local structural changes, with individual hyper-accessible nucleosomes surrounding regulatory elements, such as enhancers and transcription start sites. The regulatory sites per se are occupied with structurally altered nucleosomes, exhibiting increased MNase sensitivity. Our findings provide biochemical evidence that supports an irregular model of large-scale chromatin organization.

Figure 1.

Isolation of nucleosomal DNA after differential MNase hydrolysis. (A) Nucleosomal DNA ladder of HeLa cells incubated with increasing concentrations of MNase. The agarose gel is showing the analysis of the purified DNA after MNase hydrolysis. The molecular weight marker is shown in lane 9 and the positions of subnucleosomal, mono-, di- and tri-nucleosomal DNA are indicated. The scheme below illustrates the hypothetical results of a partial and full hydrolysis of differentially accessible chromatin domains. (B) Simulation of low- and high-MNase digestions (see Materials and Methods for details) of an equally accessible linker DNA with p_cut(linker) = constant (left panel) and of a nucleosomal array containing a central compacted chromatin domain with reduced linker accessibility p_cut(linker, compact) = 0.8·p_cut(linker), (right panel). The relative sensitivity of intra-nucleosomal (intra) and inter-nucleosomal (inter) cleavage is indicated. (C) Large scale low- and high-MNase hydrolysis of chromatin (lanes 2 and 3) and subsequent isolation of nucleosomal DNA corresponding to the mono-, di-, and tri-nucleosomes (lanes 5 to 10). (D, E) 2D and 3D FISH experiments with isolated tri-nucleosomal DNA. Metaphase FISH (D) and 3D FISH (E) were performed with Cy3/FITC labelled tri-nucleosomal DNA derived from high-/low-MNase digestions, as indicated. (F) Low- ad high-digested mono- and di-nucleosomal DNA were subjected to library preparation and high throughput sequencing. The fragment size distributions of the mapped paired-end reads for mono- (left panel) and di-nucleosomal DNA (right panel) are shown. The shaded boxes highlight the fraction of isolated mono-nucleosomal (140–200 bp) and di-nucleosomal (250–500 bp) fragments.

Figure 2.

MNase sequence preferences dominate differential chromatin fragmentation. (A) Genome browser plot showing the nucleosome occupancy distribution along chromosome 6. Differences of high- and low-MNase digestions were calculated as log₂ fold changes (log FC) of the average profile intensity in 1 Mb non-overlapping windows. GC content variation is displayed as the deviation from the genome wide average (40.5%). R is the Pearson correlation coefficient determined for the log FC (high/low) and the GC content deviation. The mappability track at the bottom illustrates how uniquely 100mer sequences align to a region of the genome (41). (B) Average GC content of isolated mono- and di-nucleosomal fragments. The genome wide GC average is indicated by the dashed line. (C–E) Genome wide correlation of GC content und nucleosome occupancy profiles. Dots represent 1 Mb non-overlapping windows; the coefficient of determination (R²) refers to the result of a simple linear regression between GC content and the average signal of (C) log FC of high-MNase versus low-MNase, (D) high-MNase, (E) low-MNase.

Figure 3.

Nucleosomes in AT-rich regions are partially depleted in high-, but not in low-MNase conditions. (A) Average G/C-di-nucleotide frequency, deduced from GG, GC, CG, CC occurrences, centred on mapped mono-nucleosomal fragment midpoints. A random distribution (grey) was simulated on 1 million randomly generated fragments of 147 bp lengths. Vertical lines indicate nucleosome boundaries (±73 bp). (B) Genome browser snapshot showing the GC bias of nucleosome annotation after high MNase digestions. Nucleosome positions, which have been called with the DANPOS2 toolkit (80), are represented by blue boxes. Arrows indicate over-digested nucleosomes in high MNase condition. GC content was calculated in 50 bp windows with 10 bp sliding steps. The genome wide GC average is given by the dashed line. (C) Genome wide correlation of GC content and quantity of nucleosome prediction. Dots represent 1 Mb non-overlapping windows; (D) Kernel density plot of the nucleosome count in 1 Mb sized windows. (E) Simulation of low- and high-MNase digestions of a nucleosome strand exhibiting nucleosomes with variable MNase sensitivity. A higher intra-nucleosome cleavage probability p_cut(nuc, AT+) was assigned to a stretch simulating an AT-rich region in the middle of the chromatin strand. Used cleavage probabilities: p_cut(nuc, AT+) = 2·p_cut(nuc) and p_cut(linker) = 10·p_cut(nuc).

Figure 4.

Comparison of chromatin accessibility within annotated chromatin regions. (A) Genome wide correlation of GC content and differential MNase signal. Dots represent non-overlapping windows of varying genome sizes (size given at the top). (B) Genome browser snapshot showing H3K27me3 (ENCODE, ENCFF958BAN) marked domains and local H3K27ac sites (ENCODE, ENCFF311EWS) highlighted by black rectangle compared to differential MNase digestions. log₂ FC was calculated in 10 kb non-overlapping windows. (C) Correlation of H3K27me3 (blue) / H3K27ac (orange) enrichment over input with differential MNase digestion enrichment. The average signal was calculated in 10 kb non-overlapping windows spanning the whole genome. Windows were grouped on the basis of the ChIP enrichment signal into windows showing no ChIP enrichment (ChIP-signal < input signal) and windows enriched for the histone modification were further subdivided into quartiles of signal enrichment. (D) Differential MNase signal (upper panel) in annotated chromatin states (ROADMAP) compared to the size of the chromatin states (lower panel). (B–D) The MNase signal was normalized for the underlying GC-bias (see Supplementary Figure S11 and Materials and Methods for details).

Figure 5.

Differences of low- and high-MNase analysis at local sites. (A) Genome browser snapshot around an accessible enhancer marked by DNase I HS peak (ENCODE, ENCFF567PRQ). (B–E) Averaged MNase-seq profiles around: (B) DNase I HS sites (ENCODE, ENCFF692NCU); average G/C—di-nucleotide content is indicated by the grey line. (C) TSS of highly expressed genes (left panel) and not expressed genes (right panel). FPKM values determined by RNA-Seq analysis (ENCODE, ENCFF000DNW) were used to estimate the level of gene expression. Genes not detected in RNA-Seq analysis (FPKM = 0) were considered as not expressed and those genes showing the 25% highest FPKM values were considered as highly expressed. (D) Active enhancer sites in HeLa cells as defined in (49). (E) CTCF sites (ENCODE, ENCFF002DCW). (F) Heatmaps of high-MNase (left panel) and low-MNase (right panel) are centred on CTCF binding sites and sorted in descending order of ChIP-Seq density at detected CTCF sites. (G) Simulation of low- and high-MNase digestions of a nucleosome array exhibiting nucleosomes with variable MNase sensitivity and hyper-accessible DNA sites. A higher intra-nucleosome cleavage probability p_cut(nuc, AT+) was assigned to a domain, simulating an AT-rich region in the middle of the chromatin strand. In the centre of the nucleosomal array an hyper-accessible site was introduced, representing a fragile nucleosome with markedly increased intra-nucleosome cleavage probability p_cut(nuc, fragile), and the two adjacent DNA linker sites were assigned with enhanced inter-nucleosomal cleavage probability p_cut(linker, hyperaccessible). Chosen cleavage probabilities: p_cut(nuc, AT+) = 2·p_cut(nuc), p_cut(nuc, fragile) = 4·p_cut(nuc), p_cut(linker) = 10·p_cut(nuc), p_cut(linker, hyper-accessible) = 12·p_cut(nuc). (H) Schematic model summarising the results of this study.