Structured nucleosome fingerprints enable high-resolution mapping of chromatin architecture within regulatory regions

Abstract

Transcription factors canonically bind nucleosome-free DNA, making the positioning of nucleosomes within regulatory regions crucial to the regulation of gene expression. Using the assay of transposase accessible chromatin (ATAC-seq), we observe a highly structured pattern of DNA fragment lengths and positions around nucleosomes in Saccharomyces cerevisiae, and use this distinctive two-dimensional nucleosomal "fingerprint" as the basis for a new nucleosome-positioning algorithm called NucleoATAC. We show that NucleoATAC can identify the rotational and translational positions of nucleosomes with up to base-pair resolution and provide quantitative measures of nucleosome occupancy in S. cerevisiae, Schizosaccharomyces pombe, and human cells. We demonstrate the application of NucleoATAC to a number of outstanding problems in chromatin biology, including analysis of sequence features underlying nucleosome positioning, promoter chromatin architecture across species, identification of transient changes in nucleosome occupancy and positioning during a dynamic cellular response, and integrated analysis of nucleosome occupancy and transcription factor binding.

Figure 1.

ATAC-seq signal is highly structured around nucleosomes. (A) ATAC-seq (green) insertion track for S. cerevisiae shows enrichment of insertions at accessible chromatin regions, similar to DNase-seq cut density (orange). Both tracks were smoothed by 150 bp and scaled so that the maximum density in the region is 1. (B) Fragment-size distribution for S. cerevisiae ATAC-seq samples. (C) Insertion probabilities for ATAC-seq (teal), genomic DNA (purple), and predicted by sequence bias (orange) (see Methods) around nucleosomes defined by chemical mapping. (D) Schematic illustration of expected V-plot pattern around a well-positioned nucleosome. (E) V-plot (fragment size versus fragment center position) of ATAC-seq fragments around well-positioned nucleosomes called by chemical mapping, with inset showing region with nucleosome-spanning fragments.

Figure 2.

NucleoATAC enables high-resolution nucleosome positioning. (A) Schematic of NucleoATAC workflow. First, the V-plot nucleosome signature is cross-correlated against a 2D fragment size versus fragment midpoint representation of ATAC-seq data at a locus. The signal is then normalized by a background model (based on sequence bias and read depth) to obtain a normalized signal. Nucleosome occupancy is calculated using the local fraction of nucleosomal fragments. The normalized cross-correlation signal and nucleosome occupancy tracks are used to assign nucleosome and nucleosome-free (NFR) positions. (B) Distance of dyad calls from different assays (ATAC on top panel; MNase on bottom) using either NucleoATAC (green) or DANPOS2 (orange). (C) AA/TT dinucleotide pattern around nucleosome dyad calls determined by chemical mapping (top panel), or from ATAC-seq (middle panel), or MNase-seq (bottom panel) using either NucleoATAC (green) or DANPOS2 (orange).

Figure 3.

V-plot derived from S. cerevisiae can be used as a template to apply NucleoATAC to other species. (A) Fragment-size distributions for S. cerevisiae (purple), S. pombe (orange), and human GM12878 cell line (teal). (B) S. pombe V-plot based on chemical map calls for S. pombe. (C) S. cerevisiae V-plot normalized to match S. pombe fragment-size distribution. (D) Comparison of NucleoATAC concordance with chemical mapping for S. pombe when using V-plots in B or C. (E) S. cerevisiae V-plot normalized to match human GM12878 fragment sizes. (F) 147-bp MNase fragment density around calls for GM12878 made by either NucleoATAC (black curve in upper panel) or DANPOS (black and gray curves in lower panel; gray curve is restricted to top calls to match the number of calls made by NucleoATAC).

Figure 4.

NucleoATAC reveals differences in nucleosome architecture between species. (A) Power spectrum density at 1/10.5 bp for each dinucleotide from 19 to 60 bp from NucleoATAC-called dyads for S. cerevisiae, S. pombe, and human (left to right). (B) Pair-wise correlation between dinucleotide frequencies for each species. (C) Distances between adjacent nucleosomes in three species. (D) Nucleosome-free region lengths for three species. (E) Positive NucleoATAC cross-correlation signal aggregated at TSS in three species. Cartoons show canonical nucleosome positioning at TSS for each species, with more transparent nucleosome ovals representing nucleosomes that are less consistently positioned among different TSSs.

Figure 5.

NucleoATAC reveals dynamics of nucleosome positioning and occupancy during osmotic stress response. (A) Promoter accessibility (top) and expression (bottom) changes over the osmotic stress time-course for genes showing an increase in accessibility from 0 to 15 min (green), a decrease in accessibility from 0 to 15 min (orange), or no significant change in accessibility between 0 and 15 min (purple). (B) Distribution of −1 and +1 nucleosome shifts for promoters with increasing accessibility and promoters with steady accessibility. (C) Distribution of −1 and +1 nucleosome occupancy changes for promoters with increasing accessibility and promoters with steady accessibility. (D) Individual occupancy traces for genes with significantly increased accessibility and characterized by either (1) downstream shifts in nucleosome positioning, (2) depletion of the −1 nucleosome, or (3) depletion of the +1 nucleosome during the first 15 min of the osmotic stress response. These categories do overlap.

Figure 6.

Changes in nucleosome positioning and occupancy during osmotic stress are linked to expression changes and mediated by TF binding. (A) Distribution of expression changes for promoters showing increased accessibility as well as different types of changes in nucleosome positioning or occupancy. (B) GO Term enrichment graph for genes with increased accessibility and depletion of the −1 nucleosome during the first 15 min of osmotic stress. (C) Distribution of changes in −1 nucleosome occupancy (top) and correlation between −1 nucleosome depletion and expression increases (bottom) for promoters bound by different TFs. (D) Distribution of changes in −1 nucleosome occupancy (top) and correlation between −1 nucleosome depletion and expression increases (bottom) for promoters with different Sko1 binding patterns.

Figure 7.

NucleoATAC defines stereotyped TF-nucleosome relationships. (A) Nucleosome dyad density relative to CTCF binding site for nucleosomes called previously with DANPOS (top), MNase (middle), or NucleoATAC (bottom). (B) Nucleosome occupancy distributions for sequence-specific TFs. (C) NucleoATAC nucleosome signal (left), ATAC-seq insertion profile (middle), and NFKB subunit ChIP-seq signal for NFKB at sites with different nucleosome occupancies (right). Insertion frequency normalized by sequence bias model. ChIP-seq intensities for each subunit were normalized such that the maximum intensity for the sites with 0 to 0.1 nucleosome occupancy was 1.