Transcription and Remodeling Produce Asymmetrically Unwrapped Nucleosomal Intermediates

Abstract

Nucleosomes are disrupted during transcription and other active processes, but the structural intermediates during nucleosome disruption in vivo are unknown. To identify intermediates, we mapped subnucleosomal protections in Drosophila cells using Micrococcal Nuclease followed by sequencing. At the first nucleosome position downstream of the transcription start site, we identified unwrapped intermediates, including hexasomes that lack either proximal or distal contacts. Inhibiting topoisomerases or depleting histone chaperones increased unwrapping, whereas inhibiting release of paused RNAPII or reducing RNAPII elongation decreased unwrapping. Our results indicate that positive torsion generated by elongating RNAPII causes transient loss of histone-DNA contacts. Using this mapping approach, we found that nucleosomes flanking human CTCF insulation sites are similarly disrupted. We also identified diagnostic subnucleosomal particle remnants in cell-free human DNA data as a relic of transcribed genes from apoptosing cells. Thus identification of subnucleosomal fragments from nuclease protection data represents a general strategy for structural epigenomics.

Keywords: cell-free DNA; hexasome; structural epigenomics; subnucleosome; torsional strain; transcription elongation.

Figure 1.. Asymmetric loss of histone-DNA contacts over the transcribed +1 nucleosome.

(A) Enrichment of different fragment lengths that are protected by MNase over the mean signal at the +1 nucleosome position ± 2000 bp plotted relative to the +1 nucleosome position. The distribution was generated from fragments that mapped to expressed genes with a defined +1 position (n=5273). Profiles are averages over a 50-bp sliding window. (B) Distribution of fragment lengths from a MNase-seq experiment (All Fragments) and after selection of <100 bp fragments from PAGE-purification (Gel-purified Fragments). (C) Distribution of 125 ± 5 bp fragment centers (open circles) plotted relative to the nucleosome centers of expressed genes (n=5273), over which the Gaussian fit for the distribution is plotted in red. (D) Same as (C) for highly positioned, non-transcribed nucleosome positions selected randomly (n=5273). (E) Same as (C) for 103 ± 5 bp fragments. The gray curves represent the individual Gaussian functions that make up the bimodal fit plotted in red. (F) Same as (D) for 103 ± 5 bp fragments. (E) Same as (C) for 90 ± 5 bp fragments. (G) Same as (E) for 90 ± 5 bp fragments. Profiles are averages over a 20-bp sliding window. The gray bars below the plots represent the protection of the fragment represented by the mean position of the Gaussian distributions. The map of DNA contacts to successive histone dimers in the nucleosome structure is represented below the gray bars on the same X-axis scale as the plots. See also Figure S1, S2, and S3; Movie S1.

Figure 2.. Subnucleosome protections in active chromatin.

A) Distribution of 103 ± 5 bp fragment centers (open circles) plotted relative to the −1 nucleosome centers of expressed Drosophila genes (n=3773), over which the Gaussian fit for the distribution is plotted in red. The gray curves represent the individual Gaussian functions that make up the bimodal fit plotted in red. B) Distribution of 103 ± 10 bp fragment centers (open circles) plotted relative to the −1 nucleosome centers relative to CTCF CUT&RUN sites in human K562 cells (n=15815). The distribution is obtained by subtracting the adjacent CTCF peaks. The Gaussian fit for the distribution is plotted in red. The gray curves represent the individual Gaussian functions that make up the bimodal fit plotted in red. C) Histone dimers that form contacts with different regions of nucleosomal DNA in the nucleosome structure are mapped at the top. For each length class, the mean of the Gaussian distribution that fits the fragment distribution from low-salt-soluble chromatin is used to center the representative fragment. The fragment length corresponds to the mean length shown on the left. For 112, 103, 90, 80, 68, and 58 bp length classes, the distribution of fragment centers was fit as the sum of two Gaussian functions and the thickness of the representative fragments shown here are scaled to reflect the relative areas under the curve of each Gaussian function. D) Same as A, but for total chromatin. E) Distribution of 68±5 bp fragment centers (open circles) plotted relative to the nucleosome centers of expressed genes (n=5273) for low-salt-soluble chromatin, over which the Gaussian fit for the distribution is plotted in red. The gray curves represent the individual Gaussian functions that make up the bimodal fit plotted in red. F) Same as (C) for total chromatin.

Figure 3.. RNAPII drives subnucleosome production at the +1 nucleosome.

(A) Heatmap of 3’NT Z-scores relative to the +1 nucleosome position (n=5269). The Z-scores are calculated for each gene over a window of dyad±100 bp after smoothing the 3’NT profile using a 20 bp running average. The genes are divided into five clusters obtained using k-means clustering. (B) 3’NT normalized counts averaged over the genes in each cluster relative to the +1 nucleosome position. Profiles are averages over a 20-bp sliding window. (C) Distribution of 112±5 bp fragment centers for each of the five 3’NT clusters shown in (A) plotted relative to the +1 nucleosome center. Profiles are averages over a 20-bp sliding window. (D) Same as (C) for 103±5 bp fragment centers. (E) Same as (C) for 90±5 bp fragment centers. (F) Mean ± S.E.M. of the cross-correlation asymmetry of 112±5 bp fragments versus 147±5 fragments around the +1 nucleosome position for genes of each 3’NT cluster. (G) Same as (F) for 103±5 bp fragments. (H) Same as (F) for 90±5 bp fragments. (I) Genes were divided into six groups based on total 3’NT signal at the +1 nucleosome position ±90 bp. Mean ± S.E.M. of the cross-correlation asymmetry of 112±5 bp fragments versus 147±5 fragments around the +1 nucleosome position for genes in each group is plotted. (J) Same as (I) for 103±5 bp fragments. (K) Same as (I) for 90±5 bp fragments. (L) Mean ± S.E.M. of the cross-correlation asymmetry of subnucleosomal fragments versus 147±5 fragments around each nucleosome position for all expressed genes are plotted.

Figure 4.. RNAPII elongation preferentially drives loss of contacts to the distal dimer.

(A) Ratio of loss of contacts to the distal dimer over the proximal dimer of the 90±5, 103±5, and 112±5 bp fragment center distributions calculated after resampling same number of fragments in each length class from the Control, DRB, and heat-shock datasets. (B) The 112±5 bp (top) and 103±5 bp (bottom) fragment center distribution of the control sample subtracted from the heat-shock sample, showing lowest and highest quartiles of loss of contacts to the distal dimer upon heat-shock. (C) Average RNAPII ChIP-seq enrichment of the control sample subtracted from the heat-shock sample plotted over the +1 nucleosome position of genes in the highest and lowest quartiles of loss of contacts to the distal dimer upon heat shock, as defined by 112±5 bp fragments (top) and 103±5 bp fragments (bottom). (D) Cross-correlation asymmetry of genes responding to heat-shock in the control sample and the heat-shock sample for 112±5 bp fragments (top, n=28) and 103±5 bp fragments (bottom, n=29). P-values are from paired Wilcoxon tests. (H) Same as (A) for topoisomerase inhibition datasets. The p-values for comparisons of treatments vs. control are listed in Table S1. See also Figure S4.

Figure 5.. Chaperones protect nucleosomes during transcription.

(A) Enrichment of 90±5 bp fragments over the mean signal at the +1 nucleosome position ± 5000 bp plotted relative to the +1 nucleosome position. The distribution was generated for a list of expressed genes with a defined +1 position and 90 bp fragment coverage in all four datasets (n=2713). Profiles are averages over a 50-bp sliding window. (B) Ratio of distal over proximal dimer loss of the 90±5 bp fragment center distributions calculated after resampling same number of fragments in the length class from the Control (dsGFP), Spt16, and SSRP RNAi. The p-values for comparisons of treatments vs. control are listed in Table S2. (C) Enrichment of centers of 147±5 bp fragments from H3K27ac CUT&RUN experiments relative to the +1 nucleosome position. The distribution was generated from fragments that mapped to expressed genes (n=5273) and non-expressed genes (n=3091). Profiles are averages over a 20-bp sliding window. (D) Distribution of 125±5 bp fragment centers (open circles) plotted relative to the nucleosome centers of expressed genes (n=5273), over which the Gaussian fit for the distribution is plotted in red. (E) Same as (D) for 112±5 bp fragments. The gray curves represent the individual Gaussian functions that make up the bimodal fit plotted in red. (F) Same as (E) for 103±5 bp fragments. (G) Same as (E) for 90±5 bp fragments. (H) Distribution of 112±5 bp fragment centers relative to the +1 nucleosome position for genes belonging to two clusters as defined in Figure S3 (I) Same as (F) for 103±5 bp fragments (J) Same as (F) for 90±5 bp fragments. See also Table S2.

Figure 6.. Subnucleosomal particles are hexasomes and unwrapped octamers.

(A) Experimental schematic of the sequential ChIP protocol that can distinguish hexasomes from unwrapped octamers. (B) Distribution of the centers of 103±5 bp fragments from the first ChIP (top) and the second ChIP (middle); tabulation of the areas under the curve for the two Gaussian functions (left peak and right peak) normalized to the area under the curve of the Gaussian fit of centers of 147±5 bp fragments (bottom). (C) Same as (B) for 90±5 bp fragments. Fragment center profiles were plotted for expressed genes with a defined +1 nucleosome position (n=5273) averaged over a 20 bp sliding window. (D) Model for nucleosomal intermediates formed during transcription. RNAPII elongation leads to distal dimer loss and/or loss of contacts to the distal dimer, whereas RNAPII stalling at the nucleosome entry site leads to proximal dimer loss and/or loss of contacts to the proximal dimer. See also Figures S5, S6, and S7.

Figure 7.. Cell free-DNA features subnucleosomal protections that correlate with transcription.

A) Length distribution of fragments whose centers are within the +1 dyad ± 50 bp. The top 1500 genes expressed in lymphoid/myeloid cell lines/tissues have a lower frequency of nucleosomal fragments and a higher frequency of shorter, subnucleosomal fragments compared to non-expressed genes. B) Distribution of centers of 92±10 bp fragments relative to the +1 dyad. A bimodal Gaussian curve fits the fragment center distribution, implying formation of hexasomes with either proximal or distal H2A/H2B dimer loss. The top 1500 genes expressed in lymphoid/myeloid cell lines/tissues have a higher frequency of 92±10 bp fragments, especially corresponding to loss of contacts with the H2A/H2B dimer distal to the promoter, compared to non-expressed genes. C) Gene expression (Fragments per kilobase of transcript per million mapped reads (FPKM) values) of MOLT-4, a lymphoid-derived cell line, plotted against the subnucleosome enrichment as a 2D histogram with hexagonal binning. The blue points represent the average values of successive 200 gene bins ranked by subnucleosome enrichment. The black line is the line of best fit for the blue points. D) Same as (C) for expression levels of skin tissue. E) Receiver operator characteristic (ROC) curves are plotted for cfDNA from a healthy individual (IH02) and from individuals diagnosed with cancer (IC15, IC17, IC20, IC35, and IC37). The legend indicates the AUC for each sample. F) To ensure that comparisons have equal depth of sequencing, a fraction of reads was randomly chosen from all samples 100 times and the AUC calculated with the chosen reads. Box plots represent the distribution of AUC from 100 shuffles. P-values of comparison of cancer states with IH02 are listed in Table S3. G) Same as (E) using correlation coefficients of FFT intensity and expression levels obtained from the Supplement of (Snyder et al., 2016). H) Mean ± standard error of the mean plotted for the percentage difference between AUC of disease states and AUC of the healthy state. The p-value for the comparison of FFT and +1 position was calculated using paired-one-tailed student’s t-test (n=5). The p-values for comparisons and read counts of different samples are listed in Table S3 and S4.