Basic Organization of Nucleosomes

We know that general features of DNA sequences either favor or disfavor nucleosome formation. For example, sequences with high GC content or with AA or TT dinucleotides in periodic 10 base-pair intervals favor nucleosome formation, whereas sequences with tracts of deoxyadenosine nucleotides (poly(dA:dT)) disfavor nucleosome deposition (Hughes and Rando, 2009; Jiang and Pugh, 2009; Segal and Widom, 2009). However, beyond these general principles the details for how DNA sequence and cellular factors influence the positioning, occupancy, and other properties of nucleosomes are still unknown (Figure 2A).

Open in a separate window

Figure 2

Nucleosomal properties measured by high-resolution mapping studies

A. Numerous properties of individual nucleosomes can be extracted from histone mapping studies, including the spacing between nucleosomes, the presence of histone variants and posttranslational modifications, nucleosome fuzziness, occupancy, and position relative to a genomic feature. Fuzziness is the degree to which a nucleosome deviates from its consensus position in a population measurement. Occupancy is a measure of nucleosome density. B. Even a small change in a nucleosome’s location can alter a sequence’s accessibility to regulatory proteins. This schematic illustrates the rotational accessibility and inaccessibility of the DNA major groove on the surface of the core histone complex.

Many sequence-specific DNA-binding proteins determine their genomic locations by making precise molecular interactions with as few as 6–8 base pairs of a DNA. An equivalent level of specificity applied to nucleosomal DNA but dispersed over its ~147 base pairs would be difficult to discern. Moreover, unlike sequence-specific DNA binding proteins, nucleosomes do not adopt a single position. In a population of molecules, such positions can be quite variable or “fuzzy” (Figure 2A). As non-overlapping “beads-on-a-string,” the position of one nucleosome restricts the possible positions of adjacent nucleosomes. Consequently, the determinants of a nucleosome position may have distant origins, being propagated through adjacent nucleosomes.

Furthermore, chromatin remodeling complexes, such as SWR1, ISW2, and SWI/SNF, directly regulate nucleosome composition, positions, and occupancy levels, respectively. We know that these protein machines use the energy of ATP hydrolysis to drive nucleosomes to override intrinsic preferences for DNA sequence, but we do not know how they contribute to the organization that predominates in and around genes. Experiments aimed at understanding this question often delete or deplete remodeling complexes in vivo and then assess how the absence of these factors affects nucleosome organization.

Nucleosome Organization Influences Evolution

Nucleosome organization in vivo is not random, and it is now clear that DNA accessibility imparted by nucleosome positions alters DNA susceptibility to mutations, insertions, and deletions. This differential susceptibility shapes the genomic landscape, with mutations tending to be on nucleosomal DNA and insertions and deletions tending to be in linker regions (Sasaki et al., 2009). Nucleosome organization might impact human diversity, whereby single nucleotide polymorphisms tend to be enriched in nucleosome-free promoter regions near nucleosomal edges (Schuster et al., 2010). Further research in this area may be directed at understanding how nucleosome organization shapes the evolution of promoter and enhancer elements, including sequence conservation of cis-regulatory elements.

Nucleosome Organization Regulates Gene Expression

Repositioning a nucleosome by as little as a few base pairs may be sufficient to change the accessibility of a DNA regulatory element. If the element is located on linker DNA between nucleosomes, then it may be accessible (Figure 2B). If the element resides on nucleosomal DNA, then it may be inaccessible, particularly if the helical nature of DNA faces the site inward towards the histone core (Jiang and Pugh, 2009; Segal and Widom, 2009).

An alternative means for enhancing the accessibility of regulatory elements and coding sequences is through a complete or partial dismantling of nucleosomes. Such remodeling has been characterized by perturbing the cellular environment and mapping the resulting nucleosome reorganization (Schones et al., 2008). For example, genes that are induced by heat shock tend to lose nucleosomes in their promoter region (Shivaswamy et al., 2008). Experiments aimed at understanding the underlying mechanism of remodeling involve depleting or deleting remodeling factors and then examining how nucleosome positions and occupancy levels change. One general rule emerging from these experiments is that robust transcriptional activity involves nucleosome depletion whereas transcriptional regulation may involve nucleosome repositioning.

Histone modifications and variants have emerged as a fascinating yet still enigmatic means by which nucleosomes regulate gene expression. Histone variants, such as H2A.Z and H3.3, as well as certain acetylation sites, such as H3K9, 14 and H4K5, 8, 12, 16, create nucleosomes that may facilitate the eviction and/or repositioning of nucleosomes during transcription. Modifications, such as H3K4me3 and H3K36me3, bind proteins involved in the transcription cycle whereas other marks, such as H3K9me3 and H3K27me3, bind proteins that create inaccessible repressive chromatin (Kouzarides, 2007). Furthermore, other modifications mark the locations of transcriptional enhancers (Heintzman et al., 2009). With more than fifty different histone modifications available, the potential for combinatorial control is bewildering, and deciphering this code will certainly be a focus of future research for many years.

Nucleosomes at different positions in the genome serve unique functions. The first nucleosome downstream of a transcriptional start site may regulate the accessibility of the start site and/or the ability of RNA polymerase II to progress into a productive elongation state (although the evidence for such linkages has been correlative rather than causative). In contrast, the first nucleosome upstream of the transcriptional start site may regulate the accessibility of cis regulatory elements that bind sequence-specific transcription factors. Nucleosomes in the middle of genes may prevent spurious transcription initiation that might otherwise generate truncated gene products. Histone variants and modifications are selective to specific nucleosome positions (Figure 1), and thus are likely to endow nucleosomes with position-relevant functions. Because we do not currently know all the mechanistic steps of a transcription cycle (or the order of the steps), we have yet to learn how the combinatorial configuration of histone variants and modifications participate in transcriptional regulation.

The primary organization of nucleosomes across a genome may be largely invariant from cell type to cell type. However, highly targeted changes in positioning, occupancy, histone composition, and modifications probably help define cell types. Therefore, current emphasis is being placed on generating genome-wide maps of histone modification states as cells or organisms undergo developmental programs (Shi, 2007). In this regard, the ENCODE (ENCyclopedia Of DNA Elements) project has provided a major boost for the field, as it reports on a large number of modification states across model cell lines (Birney et al., 2007). From this and other publicly generated data, a multitude of questions can be addressed. What modification states are associated with tissue differentiation, cell identity, and epigenetic inheritance? How are these modifications “read” to elicit such programs? For example, one study reports that cells already committed to a lineage have more promoter regions with repressive histone modification marks than embryonic stem cells (Hawkins et al., 2010). Are these marks reducing options for pluripotency by locking down promoters?

Overview of Available Strategies

Data from MNase ChIP-Seq provide population averages from a large number of cells. To map nucleosome configurations on individual DNA molecules, ectopic expression of DNA methyltransferases may be used in vivo or added to nuclei. Nucleosomal DNA is identified because its sequence is eventually altered, whereas linker DNA is not. For example, the M.CviPI methyltransferase methylates cytosine in 5’-GC-3’ dinucleotides when it is present in linker DNA (Pardo et al., 2010). This methyl-cytosine, unlike cytosine, is protected against bisulfite conversion to uracil (and ultimately thymine) in vitro. After traditional Sanger sequencing, the configuration of a nucleosomal array on the original DNA molecule is inferred from the sequence. GC dinucleotides are inferred to be nucleosome-free whereas ~150 base-pair spans of GTs that are GCs in the reference (i.e., untreated) genome, are interpreted as nucleosomal.

The value of mapping an array of nucleosomes on a single DNA molecule is that adjacent nucleosome positions may appear to overlap in a population average, but are actually mutually exclusive when examined on a single molecule basis. Currently, this strategy has not been applied on a genomic scale, as it optimally requires high-throughput long-read (>1000 nucleotides) sequencing.

Regions depleted of nucleosomes are candidates for regulatory regions. Therefore, if the primary purpose of chromatin mapping is to screen for nucleosome-depleted regions in many cell types or under various conditions, then FAIRE (Formaldehyde-Assisted Isolation of Regulatory Elements) may be the appropriate method (Giresi and Lieb, 2009). FAIRE simply depends upon the differential partitioning of nucleosomal and nucleosome-free DNA in phenol-chloroform and aqueous phases. Thus, the major advantages of FAIRE are its simplicity and cost-efficiency. However, its resolution is low compared to mapping the location of individual nucleosomes.

DNase I hypersensitivity has been a classical means of mapping regions of accessible chromatin. Like FAIRE, it is a strategy used on a genomic scale in the ENCODE project (Hesselberth et al., 2009). However, due to frequent cleavages within nucleosomal DNA, nucleosome positions may be more difficult to discern when DNase I is used instead of MNase. Moreover, DNase I involves many sample handling steps and is complicated by technical variation in DNA digestion.

Initial Preparation of Mononucleosomes

From an operational perspective, there are two types of starting material for mapping nucleosomes: tissue excised from a multi-cellular eukaryotic organism, and minimally aggregated cells, including those drawn from blood, grown in tissue culture, or cultured as free-living microorganisms. Excised tissue may contain a heterogeneous mixture of cells, which may obscure chromatin patterns specific to a cell-type. Cellular heterogeneity may be minimized by highly selective and precise tissue excision, which may necessitate acquisition of less material. Although the minimum amount of excised material required to generate nucleosome maps is not known, a lower limit of ~10,000 cells drawn from blood may provide a guide (Adli et al., 2010). More commonly, large numbers of cells are easily collected from tissue culture or microorganisms, such as yeast for which10⁷–10⁸ cells are used. More sophisticated methods may be used to isolate tissue-specific nuclei (Deal and Henikoff, 2010).

The production of genome-wide nucleosome maps has variously used or avoided formaldehyde crosslinking (Figure 3, step 1). Formaldehyde essentially “freezes” existing protein-protein and protein-nucleic acid interactions in place, thereby preserving the in vivo status of interactions, without adverse effects on nucleosomes (Fragoso and Hager, 1997). Without crosslinking, nucleosomes may re-organize during cell harvesting, chromatin preparation, and chromatin fragmentation. However, for most genes in yeast, we and other laboratories have found that nucleosome organization is largely the same in the presence or absence of formaldehyde when MNase is used for chromatin fragmentation (Kaplan et al., 2009). Nevertheless, we have also found genomic regions where nucleosome organization varies in the absence of formaldehyde, and thus, we recommend a simple formaldehyde crosslinking step.

Open in a separate window

Figure 3

Flow chart for nucleosome preparation, mapping, and analysis

Images exemplify or illustrate the type of material or data at each stage. Steps 5 and 6 may be performed in either order.

Yeast and plants have cell walls, which require disruption through mechanical breakage (e.g., vigorous vortexing with glass beads) or enzymatic digestion (Albert et al., 2007; Rando, 2010) (Figure 3, step 2). Tissue or small whole animals, such as worms, may be disrupted by grinding of frozen material (Kolasinska-Zwierz et al., 2009). Tissue culture cells in sufficient quantities may be disrupted by douncing cells in a hypotonic buffer. If the amount of material is low, then it may be more practical to lyse with an ionic detergent, such as SDS, in combination with a freeze-thaw cycle (Adli et al., 2010). However, because SDS disruption is not compatible with subsequent MNase digestion, the chromatin must be fragmented by low resolution sonication.

Chromatin Fragmentation

The method of chromatin fragmentation is critical to producing nucleosome maps of a desired resolution (Figure 3, step 3). Sonication produces DNA fragments ranging from ~200 to ~700 base pairs. The heterogeneity of fragment size and cleavage sites makes sonication suitable for characterizing chromatin states over wide regions encompassing many nucleosomes, but it is not optimal for mapping individual nucleosomes. MNase digestion, on the other hand, produces DNA fragments with ends that correspond to the ends of nucleosomes and thus, produces maps with very high resolution.

One potential limitation of MNase digestion is its bias towards cleaving at A or T more frequently than at G or C. However, extensive MNase digestion that predominantly produces mononucleosomes largely, but not entirely, overcomes this bias because even unfavorable cleavage sites become cleaved. Furthermore, residual bias can be computationally compensated (Albert et al., 2007). A limitation of extensive MNase digestion is the production of subnucleosomal-sized DNA fragments, particularly at highly transcribed genes where the DNA on the surface of remodeled or partially disassembled nucleosomes may be more exposed (Weiner et al., 2010). The lack of nucleosome-sized DNA fragments in such regions may be interpreted as being entirely nucleosome free, as opposed to the presence of remodeled or partial nucleosomes that escape detection.

Different chromatin samples and preparations of MNase (i.e., commercial lots) may yield different degrees of MNase digestion. Therefore, it is prudent to titrate the MNase to achieve ~80% of the DNA as mononucleosomal, which is detected by electrophoresis as a band at ~150 base pairs (Figure 3, step 3) (Rando, 2010). In addition, pooling chromatin that has been fragmented to various extents from an MNase titration may help avoid biased isolation of mononucleosome subpopulations that differ in accessibility.

Fragmentation by sonication releases insoluble chromatin fragments from the pellet to the supernatant. MNase treatment solubilizes mononucleosomes in yeast but is often less efficient in fly and mammalian systems. Therefore, a brief sonication in these latter two systems improves solubilization, without creating additional fragmentation. Alternatively, salt extractions of increasing strength can be used to selectively solubilize “active” chromatin (Henikoff et al., 2009). Gel analysis of histones and DNA released to the supernatant versus that retained in the pellet can be conducted to confirm full extraction.

Chromatin Immunoprecipitation (ChIP)

Perhaps the most frequent use of nucleosome mapping is to characterize the distribution of histone modification states or histone variants. In these cases, immobilized antibodies against the particular modification or variant are necessary to immunoprecipitate (or “ChIP”) chromatin fragments possessing the specific modification or variant (Figure 3, step 4)(Liu et al., 2005). Because only a small percentage of DNA becomes crosslinked to histones by formaldehyde, immunoprecipitation should be conducted in the presence of detergent (e.g., 0.05% SDS) to eliminate uncrosslinked DNA. Many antibodies, such as those against H3K4me3 and H2A.Z, are commercially available, providing a level of standardization and quality-control of antibody specificity. However, one limitation of any antibody targeted against a modification is its potential to cross-react with the same or a similar modification located at other sites. Alternatively, an antibody may not recognize its epitope if a nearby amino acid is also modified, and such interfering modification might be present in only a subpopulation of the nucleosomes. Synthetic peptides harboring the modification or potentially confounding secondary modifications can be used to verify antibody specificity.

Detection

Historically, genome-wide detection of chromatin began with the use of low-resolution DNA microarrays in yeast. PCR probes of each intergenic and genic region were arrayed onto glass slides upon which fluorescently-labeled ChIP material was hybridized (reviewed in Jiang and Pugh, 2009). Higher resolution was achieved with microarrays containing overlapping 50-nucleotide probes tiled every 20 base pairs across a small region of the yeast genome. Next high-density microarrays that spanned entire genomes were developed. These arrays, which remain in use today probably for a limited time, can generate maps of individual nucleosomes but with lower resolution compared to deep sequencing. Deep sequencing has the additional advantages of less background, better coverage, and a larger dynamic range compared to microarrays. That said, the fuzziness of nucleosome positions over a population (Figure 2A) precludes full realization of deep sequencing’s intrinsic high resolution.

Regardless of the fragmentation method or whether ChIP is use, the resulting DNA should be gel purified in the 120–170 base-pair range to remove nonspecific, subnucleosomal, and polynucleosomal DNA fragments (Figure 3, step 5). Currently, deep sequencing of nucleosomal DNA requires library preparation, which essentially involves ligating DNA adapters to the ends of gel-purified mononucleosomal DNA (Figure 3, step 6). This allows for PCR amplification of the sample and creates a template by which sequencing initiates. By this stage, users typically have given their samples to a sequencing facility, which will construct the libraries for sequencing using kits provided by the manufactures of the sequencing instrument (Figure 3, step 7). Research laboratories that produce large numbers of libraries may develop their own library preparation protocols, which enhance cost efficiency. Adaptor sequences are available at company web sites, and their ligation involves standard molecular biology manipulations. In this case, greater DNA yields may be obtained by gel purifying after library preparation and PCR amplification.

Currently, the Illumina Genome Analyzer and the Applied Biosystems SOLiD sequencers are the most widely used deep sequencers for this type of work. Although a variety of deep sequencers will likely be available in the near future, the key instrument parameter for nucleosome mapping (and for ChIP-Seq in general) is not the read length but rather the tag count, which is the number of different DNA molecules that can be sequenced and mapped to the reference genome. In general, a technology platform should meet these minimum specifications: minimal steps for library construction, a sequencing read or tag length of ~35 nucleotides, read accuracy of >99%, turnaround time of less than a few days, and a cost of under US$10,000 per run.

In principle, biases during ligation, post-construction PCR amplification, gel purification, and sequencing can result in biased tag production, which may influence the apparent occupancy level and position of nucleosomes (Stein et al., 2010). Nevertheless, such biases may be compensated computationally because they manifest as anomalously high tag counts at specific genomic coordinates. For example, setting an upper limit on the normalized tag counts at a particular coordinate may correct such statistical outliers (Kaplan et al., 2009). In practice, sequencing bias may be rather innocuous because data is often aggregated in a way that eliminates outliers.

From Mapped Tags to Nucleosomes

The relevant part of a mapped tag is the genomic coordinate of its 5’ end (lowest coordinate on the forward strand, highest coordinate on the reverse strand). It corresponds to a single measured nucleosome border, if the sequenced library originates from mononucleosomal DNA. The sequence specificity bias that is inherent to MNase digestion and incomplete protection of borders by histones collude to create imprecision in a mapped border, which in practice may be largely moot, because nucleosome positions are intrinsically imprecise.

Nonetheless, the resulting tag can be used to represent a nucleosome in multiple ways (Figure 5A): 1) As a strand-specific nucleosome border (unshifted tag) (Barski et al., 2007); 2) As an entire nucleosome by extending the tag in the 3’ direction to a length of 147 base pairs (extended tag) (Kaplan et al., 2009); or, 3) As a single coordinate representing a presumed nucleosome midpoint by shifting the 5’ end of the tag 73 nucleotides towards the 3’ direction (shifted tag) (Albert et al., 2007). For paired-end sequencing, the midpoint between the highest and lowest coordinates of the two tags defines the nucleosome midpoint (Figure 5B). The midpoint can also be extended 73 nucleotides in both directions to define the presumed nucleosome length. When sonication is used to fragment chromatin, the resulting libraries largely lack single nucleosome precision. Nevertheless, their sequenced 5’ ends can be shifted in the 3’ direction by the average fragmentation size to estimate the midpoint of the nucleosome.

Open in a separate window

Figure 5

Turning tags into nucleosomes

Solid red and blue lines represent nucleosomal DNA ready for sequencing. The nucleosomal DNA is produced as MNase-resistant DNA fragments. In a population of molecules, the DNA fragments will have heterogeneous ends due to biases in digestion efficiency at different sequences, as well as the nucleosome not residing at a single position (i.e. “fuzzy” positioning). The asterisk, representing a unique sequence, provides a frame of reference. A. In “fragment” or “single-read sequencing,” the DNA library is sequenced from only one of the adapters (green) (except in reading the barcode) and in the direction indicated by the blue or red arrows. Consequently, each nucleosome border is measured independently as a population. Tags can then be extended to 147 nucleotides or their 5’ ends shifted by 73 nucleotides, as indicated to the right. Either way, the resulting frequency distributions, while looking different, have exactly the same uncertainty. B. Paired-end sequencing allows both ends of the same DNA molecule to be sequenced. The midpoint of the pair defines the consensus nucleosome midpoint, which can be extended 73 nucleotides in both directions (right side).

Clusters of tags can be aggregated to define a consensus nucleosome position, which then represents a population average (Figure 3, step 8). For this, our laboratory uses GeneTrack software, which was the first peak calling software developed for mapping nucleosomes or any other data by ChIP-Seq (Albert et al., 2007). GeneTrack converts tag counts at each coordinate into a smoothed Gaussian distribution across multiple coordinates, implementing a user-defined standard deviation. GeneTrack then sums all instances of the distribution to create a smoothed continuous landscape across the genome. Local peaks are then identified, starting with the highest peak. A user-defined exclusion zone (e.g., 147 nucleotides) is centered over the peak to represent the steric exclusion of a nucleosome and prevent the calling of secondary peaks within the exclusion zone.

Peak calling can be performed with unshifted tags on each DNA strand separately. The consensus midpoint for the nucleosome is then the midpoint distance between a peak on one strand and the next downstream (3’) peak but located on the opposite strand. Alternatively, shifted tags from each strand can be first combined, and then applied to GeneTrack. The former may be more accurate as it involves position-specific correction factors. The latter involves only a single correction for an entire dataset and may be more appropriate for raw data display in a browser.

A number of other “peak-calling” algorithms have been used to define consensus nucleosome positions. Hidden Markov modeling has been applied to microarray data (Yuan et al., 2005) and can infer which DNA segments are occupied by nucleosomes after the algorithm trains on a data set. On the other hand, template filtering aims to classify peak patterns (Weiner et al., 2010), then shifts peak pairs on opposite strands individually by whatever distances maximize area overlap between the two peaks. In contrast, Model-Based Analysis of ChIP-Seq (MACS) uses empirical modeling of the length of protein-DNA interaction sites in combination with local biases in the genome based on a Poisson distribution (Zhang et al., 2008). Other peak-calling software, which may be applicable to nucleosome mapping, includes PeakFinder, FindPeaks, SISSRs, QuEST, CisGenome, PeakSeq, and Hpeak. Indeed, Laajala et al. (2009) compare these programs for use with ChIP-Seq data.

Nucleosome Occupancy

Nucleosome occupancy can be assessed for a consensus nucleosome location or on a per base pair basis (Figure 5). The former measures the number of shifted tags residing within ±73 base pairs of a consensus nucleosome midpoint (Mavrich et al., 2008). By this measure, a single occupancy value is attributed to a consensus nucleosome position defined by a single coordinate. Preferential MNase digestion sites and fuzziness of the nucleosome may influence the consensus position but will have little effect on its occupancy level.

In contrast, occupancy measured on a per base-pair basis counts the number of extended tags that cover each genomic coordinate without defining any consensus position (Kaplan et al., 2009). By this measure, occupancy levels at coordinates will be influenced by preferential MNase digestion sites and nucleosome fuzziness.

Comparing occupancy levels across datasets requires normalization because the number of tags delivered by a sequencer is, to a first approximation, defined by the user rather than reflective of any biological property. Normalization simply involves setting the total number of uniquely mappable tags between samples equal. This assumes that the number of nucleosomes between the compared samples is indeed equal in the biological setting. Because different extents of MNase digestion may influence occupancy levels measured on a per base-pair basis (Weiner et al. 2010), digestion uniformity (i.e. ~80% mononucleosomal) is a critical parameter.

Although the amount of histones present across samples may be approximately constant, a particular histone modification may vary across samples. In such situations, if the total level of a histone modification across the whole genome can be measured independently, for example by immunoblotting, then total sample tag counts can be scaled to reflect this measured level (Figure 6). Measured levels of a particular histone modification (i.e. tag counts) at specific genomic locations have two contributing biological factors: nucleosome occupancy level and the amount of modification per nucleosome. The latter is the desired metric and can be derived by dividing the measured modification level for a consensus nucleosome or genomic interval by the corresponding level of nucleosome occupancy (Figure 6).

Open in a separate window

Figure 6

Data normalization

This schematic depicts an immunoblot measuring the bulk levels of chromatin-associated histone H3 and a particular histone modification. The corresponding immunoblot signals in a reference sample are set to 100. Assuming total levels of H3 are constant between samples, the relative level of each modification can be assessed. Nucleosomal tags generated from MNase ChIP-Seq can then be proportionally adjusted to reflect the relative modification state. This does not preclude further normalization on a locus-by-locus basis in which modification densities are calculated for a given amount of core histone (typically, H3) present in the sample.

Normalized levels of nucleosome occupancy or normalized modification densities can be compared across datasets on a per nucleosome basis or over defined intervals, such as every 500 base pairs. Datasets in which a large proportion of the occupancy levels are distributed over a rather narrow range of values will yield poor correlations between datasets. By analogy, any portion of a calm ocean will look like any other portion of a calm ocean. Data can be filtered in order to compare only nucleosome or intervals that have particularly high or low occupancy levels (Kaplan et al., 2009). However, one caveat to this filtering is that any resulting correlation is applicable only to those regions.

We thank Mike Kladde, Bryan J. Venters, Shinichiro Wachi, Kuangyu Yen, Liye Zhang, Sujana Ghosh, Megha Wal, Kiran Batta, Jing Hu, and Christine H. Walsh for helpful discussions. This work was supported by NIH grant HG004160.