Background

Covalent modification of different bases of DNA occurs throughout genomes [1, 2]. In mammals, the typical DNA modification is methylation of cytosine in the CG dinucleotide with over 70% of CG dinucleotides being methylated [3–5]. Several single nucleotide resolution methylation maps show that unmethylated cytosines occur in clusters representing 2% to 3% of the genome, mainly in the CG rich regions termed CG islands (CGIs), which encompass 1% of the genome. Approximately 70% of promoters contain a CGI, and these promoters tend to regulate housekeeping genes [6], and the hypomethylation of these regions is critical for cellular function [4, 7, 8]. However, a number of hypomethylated regions (HMRs) exist outside of CGIs (non-CGI HMRs) and promoters, and these tend to vary across cell types and tissues [4, 7–15], suggestive of a possible regulatory role for non-CGIs as enhancer elements. Indeed, non-CGI HMRs have been shown to be enriched for transcription factor binding sites (TFBSs) [14], and are associated with expression of nearby tissue-specific genes [4, 8, 16, 17], indicating that hypomethylation of non-CGI regions is a hallmark of regulatory function.

In addition to the methylation status of regulatory elements, the primary organizational unit of chromatin, the nucleosome, is an important factor impacting the regulatory capacity of the genome. Nucleosomes restrict access to factors requiring the DNA template and nucleosome loss, depletion, or rearrangement is indicative of transcriptional activity in regulatory regions including promoters and enhancers across species [18, 19]. Moreover, the nucleosome itself is an important factor signaling regulatory activity. For example, nucleosomes harboring specific post-translational modifications have been shown to associate with active chromatin (for example, monomethylation of lysine 4 of histone H3 is enriched in nucleosomes flanking active or poised enhancers [20]). Nucleosomes have also been implicated in the targeting of DNA methylation, as they have also been shown to serve as tethering sites for DNA demethylases that, in turn, contribute to the formation of repressive chromatin states [21, 22].

Several studies have begun to investigate the relationship between DNA methylation and nucleosome organization. Many studies focus on the effects of DNA methylation on nucleosome stability. It has been demonstrated that CG methylation is reduced by the presence of nucleosomes [23–27], however, methylation is enriched within nucleosome core DNA in vivo[28]. Thus, the role of nucleosome positioning on DNA methylation as well as the effects of methylation on nucleosome positioning are not resolved, and it has been suggested that both can influence each other [29]. Recent studies focusing on specific regulatory regions (promoters and distal enhancers) indicate that nucleosome reorganization and depletion generally accompanies demethylation of regulatory elements leading to activation of these regions in vivo[29–31]. It has recently been shown that nucleosomes tend to be highly organized over regions that are differentially methylated between tissues even when these regions are not active [30].

Here, we have investigated nucleosome organization at the boundaries of HMRs where the DNA methylation transition occurs. To this end, we have generated genome-wide nucleosome maps produced by MNase digestion of nucleosomal DNA obtained from primary tissues from mouse fibroblasts and keratinocytes, followed by high-throughput sequencing. We next compared the nucleosome organization surrounding HMRs that we have recently identified in both cell types [5, 32] and found that nucleosomes are enriched at the HMR boundaries. We find that nucleosome organization at non-CGI HMR boundaries, which tend to be tissue-specific, is independent of methylation status. In addition, in contrast to HMRs in CGIs, nucleosomes at the boundaries of non-CGI HMRs are predicted by a model of intrinsic nucleosome occupancy [33], suggesting that nucleosomes are localized at their preferred sites, implicating a role for DNA sequence in demarcating these putative regulatory regions genome-wide.

Results

Genome-wide nucleosome maps of mouse dermal fibroblasts and keratinocytes

Micrococcal nuclease (MNase)-digested nucleosomal DNA from primary cultures of new-born female mouse fibroblasts and keratinocytes was sequenced using an Illumina HiSeq sequencer [34–38]. The resulting reads, mapped to the mouse genome, allowed the generation of map of nucleosome occupancy at high (27-35X) coverage for each cell type (Additional file 1: Table S1, Additional file 2: Figure S1).

As validation of the quality of these data, we assessed the nucleosome occupancy surrounding the transcription start site (TSS). We subdivided mouse promoters based on the presence of CGIs (+/-CGI) and their methylation status [5, 32]. We find that unmethylated promoters (+/-CGI) in both tissues are characterized by a nucleosome-depleted region upstream of the TSS, with phased nucleosomes (165 base pairs (bps) periodicity) occurring towards the body of the gene (Additional file 2: Figure S2). The degree of nucleosome depletion upstream of the TSS as well as the strength of the phasing is positively correlated with transcription, as previously observed in yeast [38, 39] and vertebrates [40–42]. The next largest group, are promoters that are methylated but do not contain a CGI (-HMR/-CGI: 8,803/28,283). In contrast to the previous groups, promoters of the most highly expressed genes in this class do not have a nucleosome depleted region or the phased nucleosome pattern similar to Saga-containing promoters in yeast [43] (Additional file 2: Figure S2).

Nucleosomes localize at boundaries of hypomethylated regions of keratinocytes and fibroblasts

We next investigated the in vivo nucleosome organization surrounding 49,233 fibroblast hypomethylated regions (HMRs) [5] and 71,495 keratinocyte HMRs [32]. One difficulty in such an analysis is the assignment of the boundaries of HMRs. These boundaries could occur at the first unmethylated CG dinucleotide of the HMR, the methylated CG directly upstream and downstream of the HMR, or somewhere between the two. To gain insight into nucleosome distributions across HMRs, we aligned HMRs either by the first/second unmethylated CG or by the adjacent methylated CG (Additional file 2: Figure S3 and S4) located on average 150 bps from the first unmethylated CG (Additional file 2: Figure S5a and b). We observe a weak averaged nucleosome profile when HMR boundaries are assigned using the adjacent methylated CGs (Additional file 2: Figures S3a, S4a-d). However, when HMR boundaries are set using the first unmethylated CG, we find a striking pattern in both cell types with nucleosomes enriched at the boundaries with a periodic array of nucleosomes propagating into the methylated region (Figure 1a, d, g and Additional file 2: Figure S3b). Similar results were obtained when defining the HMR boundaries using the second unmethylated CG (Additional file 2: Figures S3c, S4e-h), although this approach yielded a less periodic nucleosome signal outside of the HMR boundaries (Additional file 2: Figure S5c).

Figure 1
figure 1

Patterns of nucleosome occupancy at hypomethylated regions (HMRs) of mouse fibroblasts and keratinocytes. (a-c) Heatmap of nucleosome densities at HMRs sorted by length in fibroblasts (Fb.) for (a) all HMRs, (b) HMRs overlapping with CGIs and (c) HMRs not overlapping with CGIs. (d-f) Average nucleosome occupancy measured in fibroblasts (blue) and keratinocytes (red) and intrinsic nucleosome occupancy scores (INOS, black) for (d) all fibroblast HMRs, (e) fibroblast HMRs overlapping CGIs (+CGIs), and (f) non-CGI containing fibroblast HMRs (-CGIs). HMRs are aligned by the first (5′) unmethylated CG dinucleotides. (g-i) same as in (a-c) but for mouse keratinocytes (Ker).

Full size image

Since a substantial fraction of HMR regions found in our methylome maps do not overlap a CGI (72.5% of all fibroblast HMRs and 81.2% of all identified keratinocyte HMRs), we next divided the HMRs into two groups: those that overlap with CGIs (Figure 1b, e, h) and those that do not (Figure 1c, f, i). For fibroblast HMRs overlapping CGIs, nucleosomes localize to the unmethylated CG boundary with phased nucleosomes extending into the methylated region, with a nucleosome depleted region in the middle of the HMR (Figure 1e). This nucleosome depleted region is at the TSS of expressed genes in CGIs (Additional file 2: Figure S2). Indeed, 78% (10,529/13,520) (Additional file 2: Figure S6a) of fibroblast HMRs within CGIs contain TSSs and the observed nucleosome depletion is presumably caused by RNA Pol II and associated factors that preferentially associate with CGI promoters (Additional file 2: Figure S2) [42, 44–46]. The 13,460 keratinocyte HMRs in CGIs are primarily promoters with nucleosomes at the boundary (Additional file 2: Figure S6c), however, the interior of these HMRs are more occupied by nucleosomes than the corresponding group in fibroblasts (Figure 1e, h), potentially reflecting the changes observed in the terminal stages of epidermal differentiation [47, 48].

Most HMRs are not in CGIs and they have a similar nucleosome pattern at the boundaries (Figure 1c, f, i). The 35,713 fibroblast and 58,035 keratinocyte HMRs without CGIs have nucleosomes at the boundary with phased nucleosomes spreading into the adjacent methylated region. However, in contrast to the HMRs within CGIs, the nucleosomes at the boundary of non-CGI HMRs are predictable by a model of intrinsic nucleosome sequence preference (intrinsic nucleosome occupancy scores (INOS)) [33] suggesting that the localization of nucleosomes at these boundaries might be driven by DNA sequence (Figure 1f).

Nucleosome localization at non-CGI HMR boundaries does not depend on methylation status

We next asked whether the observed nucleosome pattern in HMRs outside of CGIs is due to their methylation status. We subdivided HMRs into three groups: those that are common in both cell types and those that are differentially methylated between the two tissues. For the purposes of clarity, we will refer to these differentially methylated regions as either keratinocyte or fibroblast tissue-specific HMRs (TS-HMRs) to mean those HMRs that are found in only one of the two tissues compared in this study. In contrast to common HMRs, in which a substantial portion (36.3%) occurs within CGIs, TS-HMRs show only 0.4% overlap with CGIs. Common HMRs outside of CGIs have nucleosomes at their boundaries as predicted by INOS with one to three strongly phased nucleosomes extending into the methylated region (Figure 2a, c, e). A similar pattern is seen in the nucleosome occupancy profiles of TS-HMRs, albeit with weaker phasing surrounding the HMR (Figure 2b, d). The stronger phasing may be due to active transcription [49] as approximately 14% common non-CGI HMR contain a TSS (Additional file 2: Figure S6e). In contrast, 4% of TS-HMRs contain a TSS within 2 kbp of the HMR (Additional file 2: Figure S6b, d).Cross-comparison of nucleosome occupancy profiles in differentially methylated regions between keratinocytes and fibroblasts shows a peak of nucleosome occupancy at the TS-HMR boundary in both cell types. This peak at the boundaries corresponds to regions of high INOS (Figure 2f, g), suggesting that DNA sequence encoded nucleosome occupancy and not methylation status is an important contributor to the observed nucleosome localization at non-CGI HMR boundaries.

Figure 2
figure 2

Nucleosome organization in non-CGI HMRs is not dependent on methylation status. (a, b) Heatmap of nucleosome density in fibroblasts (Fb.) surrounding non-CGI containing HMRs for (a) common (found at both keratinocytes and fibroblasts), and (b) fibroblast-specific HMRs. (c, d) Heatmap of nucleosome density surrounding non-CGI containing HMRs in keratinocyte (Ker.) for (c) common and (d) keratinocyte-specific HMRs. For all heatmaps, HMRs in each class are sorted by length. (e-g) Average nucleosome occupancy measured in fibroblasts (blue), and keratinocytes (red), and intrinsic nucleosome occupancy scores (INOS, black) surrounding the 5′ unmethylated CG of the HMR boundary for (e) common non-CGI HMRs, (f) fibroblast specific non-CGI HMRs, and (g) keratinocyte-specific non-CGI HMRs.

Full size image

Nucleosome organization at extended HMRs

A common feature of HMRs shared between keratinocytes and fibroblasts is that the boundaries are not always identical. There exist 15,747 HMRs in which the one boundary is identical; however the other boundary is extended in one cell type [32]. The extended unmethylated region is on average 280 bps in length with a median of 170 bps (Figure 3a and b and Additional file 2: Figure S7a and b). The nucleosome localization at the common boundary of these extended HMRs is similar to the boundaries of those observed for HMRs with identical boundaries (Figure 3c and Additional file 2: Figures S7c, S8a and b). However for the extended HMRs, nucleosomes are displaced from their optimal positions at the identical boundary and phased nucleosomes extend into the methylated region, reflecting the influence of CGI-containing HMRs in the averaged profiles (2,453/7,497 or 32.7% of such HMRs overlap a CGI). The tissue-specific boundary of the shorter HMR (boundary 2) also has a positioned nucleosome and some modest phasing (Figure 3c and Additional file 2: Figure S7c). The extended boundary of the longer HMR (boundary 3) is reminiscent of boundaries in TS-HMRs with nucleosomes at the boundary, with little to no phasing of nucleosomes (Figure 3c and Additional file 2: Figure S7c). Nucleosomes centered on the boundary of both short and extended HMRs (boundaries 2 and 3) are in a region of high INOS, suggesting a prominent role for DNA sequence in specifying the boundaries of extended HMRs.

Figure 3
figure 3

Nucleosome organization at extended HMRs in fibroblasts. (a) Heatmap of CG methylation in fibroblasts (Fb) and keratinocytes (Ker) for the common HMRs in which one boundary is identical, but the other boundary is extended in fibroblasts. The boundaries of these HMRs are grouped into three types: identical in two cells (boundary 1), not identical with short end (boundary 2), and with long end (boundary 3). These HMRs are aligned by boundary 2. (b) Heatmap of nucleosome positioning in fibroblasts and keratinocytes for the overlapping HMRs extended in fibroblasts aligned by boundary 2. (c) Average nucleosome density in fibroblasts (blue), keratinocytes (red), and intrinsic nucleosome occupancy scores (INOS, black) for each boundary type of the extended HMRs.

Full size image

Sequence features enriched at HMR boundaries

Our findings identify nucleosomes at boundaries of HMRs. However, the area under the ROC curve (AUC) score assessing the ability of a single CG dinucleotide to predict a nucleosome at the boundary of a HMR is modest (AUC =0.52, Additional file 2: Figure S9a). In contrast, the INOS calculation is more robust at predicting if a nucleosome will localize to these regions (AUC =0.79, Additional file 2: Figure S9a). Moreover, the predictive power of INOS is higher for the nucleosomes that are most highly enriched at HMR boundaries (that is, the top 20% of nucleosome peaks at boundaries, with an AUC =0.83 for INOS (AUC =0.81 for all the fibroblast-specific HMR boundaries), Additional file 2: Figure S9b), providing evidence that the broader sequence contexts in which these unmethylated CGs occur is important for the enrichment of nucleosomes at HMR boundaries.

While INOS shows a general positive correspondence to the observed nucleosome occupancy at TS-HMR boundaries, this correlation is not perfect (Additional file 2: Figure S9). We therefore sought to identify additional DNA sequence features besides INOS that contribute to the observed nucleosome arrangements at HMR boundaries in vivo. We computed the enrichment of all 6-mers centered on the unmethylated CG at both HMR boundaries. Motifs enriched in the top 20% of all in vivo nucleosome peaks include GC rich 6-mers (CACGTG, CACGGG, CTCGTG) and motifs enriched at the HMR boundaries not bound by nucleosomes (bottom 20%) tend to be AT-rich (AACGTT, AACGTG) (Figures 4a and b, and Additional file 2: Figure S10a and b), consistent with base composition being a dominant contributor to INOS [33]. Interestingly, the E-box motif (CACGTG) is the most frequently occurring 6-mer at the unmethylated CG at the boundary at TS-HMRs that have high observed nucleosome occupancy in both cell types (Figure 4a and b and Additional file 2: Figure S10a and b) [7], being enriched approximately 2.0-fold at boundaries (Figure 4c and d and Additional file 2: Figure S10c and d). We also find that the E-Box motif is predicted to be well bound by nucleosomes and potentially its function is not as a TFBS but as a nucleosome binding site [50], as 150-bps sequences containing this motif in the mouse genome tend to be in sequences with higher than average INOS (INOS with E-box motif =0.59 vs. genome average =0.23). Taken together, this suggests that overlapping genomic signals, TFBS, and intrinsic nucleosome sequence preferences, contribute to the potential regulatory functions of TS-HMRs.

Figure 4
figure 4

E-box motifs are enriched at the boundaries of non-CGI HMRs. (a, b) Comparison of 6-mer occurrences surrounding the (a) 5′ unmethylated CG and (b) 3′ unmethylated CG at the boundary of fibroblast-specific HMRs in the top 20% and bottom 20% of in vivo nucleosome peaks. (c) Ratio of observed/expected frequencies of the CACGTG pattern at -5 to +5 CGs for fibroblast HMRs (5′ CG): all HMRs (All, black circle), fibroblast-specific HMRs (black square) and fibroblast-specific HMRs with the top 20% in vivo nucleosome density at the boundary CG (Top 20%, black triangle). (d) Same as in (c) but at the 3′ boundary of fibroblast HMRs (3′ CG).

Full size image

Discussion

We have compared genome-wide nucleosome maps in mouse fibroblasts and keratinocytes with methylome data and have found that nucleosomes are enriched at the boundaries of HMRs. Furthermore, the methylation status of the HMR does not affect nucleosome localization at the boundary. The localization of nucleosomes at the boundaries of HMRs in CGIs is not predicted by INOS, indicating that additional biochemical forces are organizing HMR boundaries. Promoters in CGIs have a similar nucleosome organization to yeast, with a nucleosome-depleted region at the TSS and phased nucleosomes extending into the gene body [42]. However, the biochemical mechanisms that produce these positioned nucleosomes are different. In yeast, the nucleosome depleted promoter is AT rich and nucleosomes do not bind these sequences well. In contrast, mammalian promoters tend to be GC-rich and nucleosome depleted even though nucleosomes preferentially bind these sequences [45, 46]. Potentially, the DNA sequences that occur in CGIs [51] recruit remodelers [52] that are critical for the displacement of nucleosomes from favored binding positions. More impressively, in the case of HMRs outside of CGIs, which tend to be tissue-specific, we show that INOS explains a substantial fraction (AUC =0.81) of the nucleosome localization at boundaries, providing evidence that DNA sequence demarcates the boundaries of TS-HMRs in the genome. Sequence preferences of nucleosomes also play a role in the definition of the boundaries of extended HMRs, however, the exact mechanisms that regulate HMR length between tissue types warrant further investigation.

E-box motifs are enriched at boundaries of TS-HMRs in both cell types, particularly for HMRs with a nucleosome at the boundary. While some transcription factors, such as nuclear receptor proteins can bind nucleosomal DNA, B-HLH proteins cannot [50]. Nucleosome enrichment within HMR boundaries confirms and extends previous observations regarding the encoding of nucleosome occupancy over regulatory regions [45, 46]. Since many enhancers tend to be cell-type specific, high nucleosome occupancy of TFBS can restrict their utilization to a certain cell type. In addition, nucleosomes can reinforce and promote cooperative interactions between TFs in displacing nucleosomes from their preferred sites, providing higher specificity in gene regulation [53, 54]. The E-box motif is a sequence that nucleosome preferentially binds suggesting its presence at the HMR boundary is not to be a TFBS but a nucleosome localization site. Alternatively, the E-Box may function as both, sometimes being bound by a nucleosome and other times being bound by a B-HLH protein. Whether or not HLH proteins do in fact bind these E-Box sequences, suggesting equilibrium between a nucleosome and HLH protein binding and potential enhancer function of TS-HMRs, however, remains to be determined.

Conclusions

We have mapped genome-wide nucleosome organization in two mouse primary tissue types, providing an important resource for the understanding of chromatin structure and gene regulation. Comparing these maps with HMRs obtained from single-base pair resolution methylomes has allowed the identification of the nucleosome arrangements in these regions. In particular, we find that nucleosomes are enriched at the boundaries of HMRs. Nucleosome organization at HMR boundaries is independent of methylation status. For HMRs not in CGI, boundaries are calculated to be well bound by nucleosome as occurs in vivo. Since hypomethylation of non-CGI regions is a hallmark of regulatory activity, our findings have important implications for the specification of chromatin architecture at regulatory regions in the genome.

Methods

Mouse primary keratinocytes and dermal fibroblasts

NIH research guidelines and IACUC approved animal study protocols were followed in this study. Keratinocytes and dermal fibroblasts were cultured from newborn wild type according to the protocol described previously [55]. Primary keratinocytes were seeded at a density of one mouse epidermis per 10 cm dish or equivalent in calcium and magnesium free SMEM (GIBCO Laboratories, Grand Island, NY, USA), supplemented with 8% Chelex (Bio-Rad, Richmond, CA, USA) treated FBS (Atlanta Biologicals, Inc.) and 0.2 mM calcium (CaCl2). Dermal fibroblasts were also seeded at a density of one mouse dermis per 10-cm-dish or equivalent in DMEM/F12: GlutaMAX medium (Invitrogen) with 10% FBS.

Micrococcal nuclease (MNase) digestion and mapping of MNase-seq data

Primary cultures of female primary mouse dermal fibroblasts or keratinocytes were harvested and washed once with ice cold PBS. Resuspended cell pellet (approximately 108 cells) were resuspended in 5 mL of ice cold NP-40 lysis buffer (10 mM Tris–HCl (pH 7.4), 10 mM NaCl, 3 mM MgCl2, 0.5% NP-40) and incubated for 5 min in ice. Nuclei were washed with MNase digestion buffer (10 mM Tris–HCl (pH 7.4), 15 mM NaCl, 60 mM KCl) and resuspended in MNase digestion buffer containing 1 mM CaCl2. Nuclei were digested with MNase for 10 and 15 min at 37 C. The MNase digestion was stopped by putting the samples on ice and adding 100 mM EDTA and 10 mM EGTA (pH 7.5). MNase digested DNA were purified with the Qiagen PCR purification kit after digestion with protease K (Qiagen). Isolated DNA was eluted in elution buffer (Qiagen). Mnase digested DNA fragments were separated on a 2% agarose gel. DNA corresponding to the mononucleosomal bands was gel extracted and pulled from all digestion. The libraries for sequencing were prepared according to the standard protocol for the Illumina HiSeq2000 sequencing platform with 102 bp paired end reads. Paired end reads of MNase-seq data were aligned using Novoalign software (http://www.novocraft.com/) with default parameters for both primary dermal fibroblasts and keratinocytes to the mouse genome (UCSC build mm9). Reads mapping to more than one location were discarded, and data were filtered to include only those sequences that had a mate pair match within 100 to 160 bp. Counts were recorded at the midpoint of the mate pair alignment, and Gaussian smoothing was applied to yield a continuous measure of nucleosome signal across the entire genome [56].

Annotations

TSS information was extracted from RefGene annotations downloaded from the UCSC genome browser (https://genome.ucsc.edu/) for version of mm9. For genes with identical transcript start and stop sites, only one was retained. We downloaded the CG island annotations from the UCSC genome browser for version mm9 of the mouse genome.

Female mouse primary dermal fibroblast and keratinocyte methylomes

The fibroblast methylome was produced and released by our group [5]. Generation of the genome-wide primary female mouse keratinocyte methylation data is described in the accompanied paper [32]. Regions of hypomethylation (hypomethylated regions (HMRs)) were defined using a two-state Hidden Markov Model (HMM) described in [7].

Intrinsic nucleosome occupancy scores

We predicted intrinsic nucleosome occupancy scores (INOSs) across HMRs using the Lasso linear model described in [33]. For each HMR, we calculated the INOSs at every base pair using a sliding window of 147 bp.

Data submission

Two biological replicates of MNase-seq for each primary dermal fibroblasts and keratinocytes are in the process of GEO submission. Keratinocyte methylome data have been submitted to the GEO database with accession number (GSE44918) [32]. Two biological replicates of keratinocyte mRNA-seq data are in the process of submission to the GEO database [32]. Fibroblast methylome and mRNA-seq data have been obtained from the GEO accession number (GSE44942) [5]. The second biological replicate RNA-seq data of dermal fibroblasts are in the process of submission to the GEO database [32]. All the sequencing data for fibroblasts and keratinocytes will be submitted to the GEO database with accession numbers (GSE44942 and GSE44918 respectively). The data can also be obtained from the authors upon request.