Histone Replacement Marks the Boundaries of cis-Regulatory Domains

See allHide authors and affiliations

Science  09 Mar 2007:
Vol. 315, Issue 5817, pp. 1408-1411
DOI: 10.1126/science.1134004


Cellular memory is maintained at homeotic genes by cis-regulatory elements whose mechanism of action is unknown. We have examined chromatin at Drosophila homeotic gene clusters by measuring, at high resolution, levels of histone replacement and nucleosome occupancy. Homeotic gene clusters display conspicuous peaks of histone replacement at boundaries of cis-regulatory domains superimposed over broad regions of low replacement. Peaks of histone replacement closely correspond to nuclease-hypersensitive sites, binding sites for Polycomb and trithorax group proteins, and sites of nucleosome depletion. Our results suggest the existence of a continuous process that disrupts nucleosomes and maintains accessibility of cis-regulatory elements.

Chromatin can be differentiated by the replication-independent replacement of one histone variant with another (1). For example, histone H3.3 is deposited throughout the cell cycle, replacing H3 that is deposited during replication (24). Unlike replication-coupled assembly of H3, which occurs in gaps between old nucleosomes on daughter helices, the insertion of H3.3 is preceded by disruption of preexisting histones during transcription and other active processes (3, 5). We have previously shown that H3.3 replacement profiles resemble those for RNA polymerase II (2), which suggests that gradual replacement of H3.3 occurs in the wake of transiting polymerase to repair disrupted chromatin (1). Here, we ask whether histone replacement and nucleosome occupancy are also distinctive at cis-regulatory elements.

Log-phase Drosophila melanogaster S2 cells were induced to produce biotin-tagged H3.3 for two or three cell cycles (2). DNA was extracted from streptavidin pull-down assay and input material, labeled with Cy3 and Cy5 dyes, and cohybridized to microarrays. To provide a standard, we profiled biotin-tagged H3-containing chromatin in parallel. Analysis of H3.3/H3 levels over the entire 3R chromosome arm revealed that the ∼350-kb bithorax complex (BX-C) region displays the lowest H3.3/H3 ratio of any region of comparable size on 3R, and the Antennapedia homeotic gene complex (ANTP-C) also displays an unusually low H3.3/H3 ratio (Fig. 1A). Low H3.3/H3 ratios at the homeotic gene clusters are attributable to infrequent histone replacement, and not to low nucleosome occupancy, because H3.3 levels at the BX-C are far below the median (log2 = 0) for all of 3R, whereas H3 levels are slightly above the median overall (Fig. 1B). Even the heterochromatic chromosome 4 (6) includes only shorter (∼100-kb) stretches that are as depleted in H3.3 as the BX-C (Fig. 1C).

Fig. 1.

The BX-C is extensively depleted in H3.3. (A) H3.3/H3 log-ratio profile of chromosome arm 3R. Orientation is proximal (0 on base-pair scale) to distal (28 Mb). The location of the two homeotic gene clusters are indicated with horizontal bars. (B) H3.3 (red) and H3 (blue) profiles for the BX-C and flanking regions. (C) Same as (B) for chromosome 4. The locations of annotated transposons (brown boxes) and genes (black boxes) are shown beneath each panel, oriented 5′-to-3′ above the line and 3′-to-5′ below. Profiles are displayed as moving averages, with a 1-kb (A) or a 2.5-kb (B and C) window, in 100-bp intervals.

A close-up view of the BX-C iab region reveals the presence of several prominent H3.3 peaks (Fig. 2A). Notably, the seven highest peaks correspond to the functional boundaries of the seven proximal-to-distal cis-regulatory domains that regulate the abd-A (iab2 to iab4) and Abd-B (iab5 to iab8) homeotic genes successively from anterior to posterior in the abdomen (7). Conspicuous peaks of H3.3 also correspond to the bxd Polycomb response element (PRE) and to promoters within the Abd-B gene, which is known to be active in S2 cells (8, 9). Therefore, each of the most prominent H3.3 peaks in the region corresponds to a previously defined cis-regulatory element. Our findings are likely to be general, because in budding yeast, promoters and boundaries are also sites of intense histone replacement (10).

Fig. 2.

Conspicuous histone replacement at cis-regulatory domain boundaries and DNaseI-hypersensitive sites within the BX-C. (A) Map and histone profiles for the abdominal and flanking regions of the BX-C, including four well-mapped PRE-boundaries [magenta boxes (11, 13, 25)]. The scale at top indicates genomic location on 3R, with genes and PRE-boundaries indicated on the line below. The five Abd-B promoters are marked with horizonal arrows. H3.3 (red) and H3 (blue) log2-ratio profiles are displayed as 1-kb moving averages. Boundaries between adjacent cis-regulatory domains are indicated with brackets above the log-ratio plots. (B) H3.3/H3 log ratios (brown) are shown for the PREs (magenta) and boundaries (blue) for which DNaseI-hypersensitive sites have been mapped (indicated with vertical arrows).

A characteristic feature of both boundaries and PREs in the BX-C is that they span deoxyribonuclease I (DNaseI)–hypersensitive sites in a variety of cell types, including S2 cells (11). To better delineate histone replacement patterns in the vicinity of hypersensitive sites, we tiled the entire BX-C at 20-bp resolution (fig. S1). The bxd, Mcp, Fab-7, and Fab-8 PRE-boundaries each encompass conspicuous peaks of H3.3 abundance (Fig. 2B) that closely correspond to all the known nuclease-hypersensitive sites within the region (1113). Nuclease hypersensitivity identifies sites of relatively accessible DNA, so that their correspondences to peaks of histone replacement suggest that continuous disruption of nucleosomes exposes cis-regulatory DNA relative to surrounding regions.

PRE-boundary elements in the BX-C and other regions are binding sites for multiple Polycomb group (PcG) proteins, which have been mapped in an S2 cell line at high resolution (8). If the process that disrupts nucleosomes also facilitates PcG binding, then we would expect a correspondence between peaks of PcG binding and peaks of H3.3. Indeed, when we compared H3.3 profiles with those for Enhancer-of-zeste (EZ) and Posterior-sex-combs (PSC) PcG proteins, we found that all 10 peaks of PcG binding in the abdominal region are also local peaks of H3.3 (Fig. 3A and table S1). Likewise at the ANTP-C, all 13 peaks of PcG binding in the Scr-Antp region correspond to high levels of H3.3 (Fig. 3B and table S1). H3.3 enrichment at PcG-binding sites is not attributable to higher nucleosome occupancy, because essentially identical results were obtained for H3.3/H3 profiles (figs. S2 and S3).

Fig. 3.

Sites of PcG binding correspond to local peaks of histone replacement and reduced nucleosome occupancy. Comparison of the H3.3 log-ratio profile to EZ and PSC profiles (8) at (A) the BX-C (from Fig. 2A) and (B) the ANTP-C. Locations of prominent EZ and PSC peaks are marked with vertical dotted lines. Arbitrary scaling was used to facilitate visual comparison between H3.3/H3 log ratios and linear EZ and PSC profiles. Actual scales are shown in fig. S5. (C) H3.3 log-ratio profiles aligned around EZ+PSC peaks for the BX-C and ANTP-C regions (dotted line) and for the remainder of the genome (solid line), showing moving averages using a 500-bp window. (D) Same as (C), except showing the moving averages for nucleosomal/genomic DNA log-ratio profiles.

Not all PREs in the BX-C are found to be sites of PcG binding; for example, neither Fab-7 nor Fab-8 is detectably bound by EZ or PSC (8). The fact that all PcG sites are peaks of histone replacement, but not vice versa, suggests that histone replacement at PREs and boundaries is constitutive and independent of the expression of the homeotic genes that they regulate. For example, Abd-B is expressed at high levels in S2 cells and displays the typical H3.3 5′ peak for an active gene (Fig. 2A), whereas Ubx and abd-A are nearly inactive (8, 9), yet the PREs and boundaries regulating all three genes are sites of conspicuous histone replacement over a low background.

We also examined histone replacement averaged over the 175 genomewide EZ+PSC peaks outside of the BX-C and ANTP-C (table S1) and observed an H3.3 peak centered over the PcG maximum (Fig. 3C and fig. S4). Therefore, the strong association between PcG protein binding and histone replacement is not limited to homeotic gene clusters. The genomewide H3.3 peak is higher than that for the BX-C and ANTP-C, presumably because other PcG-binding sites are not superimposed over such deep H3.3 valleys (fig. S5).

The colocalization of PcG-binding sites and local peaks of H3.3 suggests that the process that disrupts nucleosomes locally maintains the accessibility of cis-regulatory DNA to PcG proteins. If so, then there should be a lower average occupancy of nucleosomes over sites of PcG protein binding than over their surrounding regions (8, 14). To test this possibility, we hybridized nucleosomal DNA and fragmented genomic DNA on the same microarrays and measured nucleosomal/genomic DNA log ratios. Around peak regions of EZ+PSC binding, nucleosomal DNA was clearly depleted on average (Fig. 3D and table S1), similar to the depletion seen for active gene promoters (2) (fig. S1), and essentially the same results were obtained with different methods for genomic DNA fragmentation (fig. S6). We conclude that the correspondence between histone replacement and nucleosome depletion is a genomewide feature of PcG-binding sites.

In Drosophila, many cis-regulatory elements, including PREs and boundaries, are bound by the trxG proteins, Zeste and GAGA factor (GAF) (15). To test the possibility that histone replacement is enhanced and nucleosome occupancy is reduced where Zeste protein preferentially binds, we aligned 390 Zeste-binding sites identified by high-resolution chromatin immunoprecipitation (ChIP) combined with tiling microarrays (ChIP-chip profiling) (16) and averaged log ratios of H3.3/H3 and nucleosome occupancy. We observed a prominent maximum of histone replacement and a sharp minimum of nucleosome occupancy centered over the point of alignment (Fig. 4A and table S2). Similar results were obtained for predicted GAF sites (figs. S7 and S8), which suggests that nucleosome disruption is a general feature of trxG protein DNA-binding sites. H3.3 enrichment at PcG- and trxG protein–binding sites results from a replication-independent replacement process, because essentially identical profiles were obtained for H3.3core, which lacks the N-terminal tail and does not assemble during replication (fig. S9).

Fig. 4.

Binding sites for trxG proteins and poised promoters are associated with conspicuous histone replacement and reduced nucleosome occupancy. (A) Average H3.3/H3 and nucleosome occupancy log-ratio profiles aligned at 390 Zeste-binding sites. (B) Average H3.3 (red) and H3 (blue) log ratios at uninduced heat shock genes. Dotted gray lines are histone profiles for all annotated genes on 3R, shown in decreasing intensity from the most active (top 20%) to the least active (bottom 20%) gene sets. Very similar H3.3 profiles were obtained in three different experiments (fig. S10).

Like Fab-7 and Fab-8, heat shock gene promoters are prominent sites of GAF binding, nuclease hypersensitivity, and reduced nucleosome occupancy (17). Heat shock protein Hsp70 genes are constitutively “poised” for rapid induction, but do not produce detectable mRNAs in the uninduced state. We aligned Hsp70 genes at their 5′ ends and averaged H3.3 and H3 profiles. For comparison, we averaged similarly aligned H3.3 and H3 profiles for all 2165 genes on 3R with known 5′ and 3′ ends, divided into quintiles based on expression levels. H3.3 patterns were similar to those of highly active genes (Fig. 4B and fig. S10), with histone replacement levels peaking on either side of heat shock promoters. As do transcriptionally active gene promoters (2), heat shock genes display prominent H3.3 and H3 dips in abundance that are attributable to partial nucleosome depletion (17). Constitutive histone replacement also appears to be a feature of poised promoters in vertebrates, because H3.3 is strongly enriched in the upstream region of the chicken folate receptor gene, regardless of whether the gene is active or inactive (18).

What process maintains the chromatin of cis-regulatory elements in a state of flux? Many DNA-binding and chromatin-binding proteins involved in gene regulation display short residence times on DNA (19), and some mouse transcription factors show dynamic behavior at their functional binding sites (20, 21). A model for this process has been proposed, involving alternating cycles of nucleosome disruption by a Brahma-related SWI/SNF chromatin-remodeler and transcription factor binding (21). The binding of PcG and trxG proteins is also dynamic (22, 23), and we propose that a similar cycle of nucleosome disruption and factor binding takes place at boundaries and PREs. Nucleosome disruption by SWI/SNF remodeling complexes would occasionally evict nucleosomes (24) and transiently expose DNA, which would become available to other diffusible factors, including PcG proteins. The continued local presence of nucleosome remodelers would result in another cycle of remodeling, nucleosome depletion, nuclease hypersensitivity, and histone replacement at the site. This model could account for the diversity of trxG proteins (15), which include DNA-binding proteins (Zeste and GAF), nucleosome remodelers (Brahma and Kismet), and histone methyltransferases (Trithorax and Ash1) that are specific for H3K4, a modification that is highly enriched on H3.3. The resulting dynamic process would allow for proteins that promote opposite epigenetic outcomes to act at common cis-regulatory sites.

Supporting Online Material

Materials and Methods

Figs. S1 to S10

Tables S1 to S3


References and Notes

Stay Connected to Science

Navigate This Article