BCL6 promoter interacts with far upstream sequences with greatly enhanced activating histone modifications in germinal center B cells

Abstract

BCL6 encodes a transcriptional repressor that is essential for the germinal center (GC) reaction and important in lymphomagenesis. Although its promoter has been well studied, little is known concerning its possible regulation by more distal elements. To gain such information, we mapped critical histone modifications associated with active transcription within BCL6 as well as far upstream sequences at nucleosomal resolution in B-cell lines and in normal naive and GC B cells. Promoter-associated and intronic CpG islands (CGIs) in BCL6 showed a reciprocal pattern of histone modifications. Gene expression correlated with a paradoxical loss from the intronic CGI of histone H3 lysine-4 trimethylation, normally associated with transcription, suggesting that the intronic CGI may interfere with transcription. In an ∼110-kb region extending 150–260 kb upstream of BCL6, highly active histone modifications were present only in normal GC B cells and a GC B-cell line; this region overlaps with an alternative breakpoint region for chromosomal translocations and contains a GC-specific noncoding RNA gene. By chromosome conformation capture, we determined that the BCL6 promoter interacts with this distant upstream region. It is likely that transcriptional enhancers in this region activate BCL6 and overcome strong autorepression in GC B cells.

In lymphoma, aberrations affecting master regulatory transcription factors (TFs) block differentiation and contribute to tumor development (1, 2). An especially important stage of B-cell differentiation occurs in germinal centers (GCs), which are transient structures required for production of high-affinity antibodies. Cells enter the GC reaction as centroblasts and undergo somatic hypermutation of their Ig genes, which may increase or decrease affinity for antigen. After a burst of rapid proliferation, they exit the cell cycle as centrocytes, with four possible fates that depend in part upon affinity for antigen: death, further proliferation as centroblasts, differentiation into memory B cells, or differentiation into long-lived plasma cells that provide persistent humoral immunity (3).

BCL6, a transcriptional repressor, is centrally involved in these developmental decisions and is essential for the GC reaction (4). Naive B cells express moderate levels of BCL6 mRNA but little protein due to posttranscriptional mechanisms (2). Upon entry of antigen-activated B cells into the GC reaction, BCL6 mRNA and protein are rapidly induced, as transcription is increased and posttranscriptional inhibition is abrogated. Aberrant expression of BCL6 in GC B-cell lymphomas blocks differentiation at the centroblast stage. Aberrant expression may result from mutation of autorepressive sites in the promoter (5) or translocations at a major breakpoint region (MBR) in the first intron that remove autorepressive sites and substitute a heterologous promoter (6); however, breaks can occur at an “alternative breakpoint region” (ABR), located 245–285 kb upstream of the BCL6 transcription start site (TSS) (7). Because the selective advantage of these ABR translocations is almost certainly through dysregulated BCL6 transcription, the occurrence of ABR translocations strongly suggests that BCL6 transcription can be stimulated by sequences hundreds of kilobases from the gene and that distant regions might also exert such effects on the BCL6 promoter in normal cells.

To explore the possible existence of such distal regulatory elements, we used chromatin immunoprecipitation and hybridization to a custom DNA chip (ChIP-on-chip) to analyze the chromatin structure of large regions surrounding BCL6 in normal B lymphocytes and in cell lines representing various stages of B-cell development.

Results

ChIP-on-Chip Shows Histone Modifications That Correlate With Gene Expression.

ChIP of native core nucleosomes prepared by micrococcal nuclease digestion was followed by ligation-mediated PCR and hybridization to a high-density custom array (NimbleGen) interrogating >100 genes(Table S1) chosen for their likely roles in B-cell differentiation. In addition to normal centroblasts and naive follicular B cells, we analyzed three cell lines at different stages of B-cell development: SU-DHL-16 (hereafter DHL16), a DLBCL line, has a GC B-cell gene expression profile. U266, a myeloma cell line, represents the plasmablast/plasma cell stage. Finally, CL01, an atypical Burkitt lymphoma cell line, resembles founder centroblasts, which differ minimally from naive B cells and lack most of the distinctive markers of full centroblasts (8). We also compared our results to those of genome-wide high-throughput sequencing following ChIP (ChIP-Seq) on resting normal human CD4⁺ T cells (9). We analyzed four histone posttranslational modifications (PTMs)—namely, histone H3 lysine-4 trimethylation (H3K4me3), found at promoters and CpG islands; H3K4me1 (monomethylation), found at enhancers and other regulatory regions or adjacent to regions of H3K4me3; H3K27me3, associated with repression by polycomb complexes; and histone H3 doubly acetylated at lysines 9 and 14 (H3Ac), found at promoters and enhancers.

Many of the examined genes show marked differences in histone PTMs among the samples analyzed. Because of space limitations, only a few genes are described here. Although emphasis is on BCL6, two other genes, IRF4 (Fig. 1A and Fig. S1) and PRDM1 (Fig. 1B), with different patterns of expression are described to provide points of comparison. Other genes with stage-specific changes in histone PTMs include MYBL1 and cyclin D2 (CCND2) (Figs. S2 and S3), which are up-regulated and down-regulated in centroblasts, respectively.

Fig. 1.

Histone PTMs at the IRF4 and PRDM1 loci. (A and B) Shown are relative levels of pol II (cell lines only) and four histone PTMs. The gene structure with noncoding 5′- and 3′-UTR as narrower rectangles is shown above. Below are CGIs and mammalian conservation (UCSC Genome Browser). (A) IRF4. Normalized expression data are from the Affymetrix U133 Plus 2.0 platform (probe set 204562_at). T-cell data for pol II and H3K4me1 and -3 are from ref. 9. (B) PRDM1. Two alternative promoters are active in plasma cells. Expression data are from probe sets 202511_s_at (ATG5) and 228964_at (PRDM1).

PRDM1 and IRF4 have a pattern of expression reciprocal to that of BCL6, and both are essential for plasma cell development. The latter is activated by chromosomal translocations in some cases of multiple myeloma, a malignancy of plasma cells. The IRF4 promoter lies within a CpG island (CGI), and peaks of H3K4me3 and H3Ac (Fig. 1A and Fig. S1) are present within the CGI in all samples other than the centroblast cell line DHL16, in which the promoter has aberrant DNA methylation (Fig. S4). The peak height of H3Ac correlates with the degree of IRF4 expression. Conversely, DHL16 and centroblasts show very prominent peaks of the repressive PTM H3K27me3, which is much less in naive B cells and essentially absent in CL01 and U266. H3K4me3 in resting CD4⁺ T cells shows very similar peak locations to those of naive B cells. A similar pattern of large peaks of H3K27me3 present only in normal centroblasts and in DHL16 cells is seen in the cyclin D2 gene, CCND2, a target of repression by BCL6 (Fig. S3).

PRDM1 also shows histone PTMs that largely correlate with expression (Fig. 1B). Strikingly, several downstream evolutionarily conserved regions have the H3K4me1 PTM in most samples, but only U266, with very high PRDM1 expression, also shows high levels of H3Ac and of pol II in these regions. This result strongly suggests that these regions are enhancers, which presumably contribute to PRDM1 expression, as their activity does not correlate with expression of the other neighboring gene, ATG5.

BCL6, LPP, and the Intervening Regions Show Marked Differentiation Stage-Specific Histone Modifications Correlating with Expression.

BCL6 lies ∼400 kb from the closest upstream protein-encoding gene, lipoma preferred partner (LPP), identified by its involvement in a recurring translocation t(3;12) in lipomas (10). Within this 400-kb interval are numerous highly conserved regions (Fig. S5) that may represent regulatory sequences, such as enhancers. Gene expression profiling showed that not only BCL6 but also the neighboring gene LPP is part of the GC program (11). “Deep sequencing” of the Ramos Burkitt lymphoma cell line identified a number of ncRNAs (12) including one upstream of BCL6 that is highly expressed, at a level higher than that of transcripts for such essential proteins as the ribosomal components RPS13, RPL10A, and RPS21. Analysis of expressed sequence tags showed that this transcript is spliced.

Large numbers of peaks were found for each of the analyzed histone PTMs in a 0.8-Mb region including BCL6, two downstream genes, somatostatin (SST) and RTP2, and part of LPP (Fig. 2 and Fig. S6). The peaks for H3K4me3 and H3Ac are highly similar in location. A striking finding is the presence in normal centroblasts of numerous peaks of H3K4me3, H3K4me1, and H3Ac over broad regions lying between BCL6 and LPP, which are markedly reduced or absent in naive B cells. DHL16 shows a pattern similar to normal centroblasts, but with a broadening of the activating PTMs H3Ac and H3K4me3; that is, the histone modifications extend a greater distance from the peak centers than in normal centroblasts. H3K4me1 shows an even more marked pattern of broadening in DHL16 with almost continuous H3K4me1 over a ∼110-kb region upstream of the BCL6 promoter (Fig. 2 and Fig. S7), overlapping the ABR. Thus, the GC program is associated with activation of large chromatin regions far upstream of BCL6.

Fig. 2.

Histone PTMs in 0.8 Mb surrounding and upstream of BCL6. Gene structures of SST, RTP2, BCL6, and part of LPP are shown, as is an ncRNA described in the text. Also shown is the approximate extent of the “alternative breakpoint region” (ABR) at which chromosomal translocations occur, as well as the specific position of a cloned and sequenced breakpoint junction. Levels of BCL6, LPP, and RPL13A mRNAs and the ncRNA were analyzed by real-time quantitative RT-PCR. Normalized miR-28-5p and -3p values (2^−ΔCt) assume perfect PCR efficiency.

Peaks of H3K27me3, associated with polycomb-mediated repression, are found downstream of BCL6 in normal cells and in all of the cell lines studied; in the two normal cell samples, the highest peaks are found at the promoter region of SST, which is silenced in B cells. Elsewhere, however, there is very little H3K27me3. In marked contrast to the cells that express BCL6, CL01 has lost H3Ac and H3K4me3 throughout the entire region, with retention only of several peaks of H3K4me1. H3K27me3 is found throughout the region. This finding suggests that aberrant epigenetic changes have affected the entire region, abrogating BCL6 transcription.

Because BCL6 is repressed in plasma cells and plasmablasts, we anticipated a loss of activating, and a gain of repressive, histone PTMs in U266 cells. Surprisingly, H3K4me3 and H3Ac in U266 cells are found adjacent to its promoter and in a large region of the first intron, superficially similar to the pattern in BCL6-expressing cells; this is described in more detail below. Except for the BCL6 and LPP promoters, little H3Ac is found in U266 cells. Several peaks of H3K4me1 and -3 are retained, and a few peaks (e.g., at 189.0 Mb) are either present only in U266 or are much higher in this sample. H3K27me3 peaks in U266 are seen both within BCL6 and upstream (Figs. 2 and 3).

Fig. 3.

Histone PTMs at the BCL6 promoter region. The major and a minor TSS (exon 1A) and of an alternative promoter and first exon (1B) are shown for BCL6. Histone PTMs are shown; see Fig. 1 legend for details. Data for methylcytosine in human ES cells are from ref. 15. Seven conserved BCL6 binding sites predicted by MatInspector (39) (http://www.genomatix.de) show the match to one of two sequence matrices, with “1” being a perfect match. To the right is shown the distribution of H3K4me3 within the BCL6 promoter region and first intron as determined by summing the values at all of the points within each region. The intronic CGI is subdivided into four regions as demarcated by the CTCF sites.

As expected, BCL6 and LPP expression was higher in normal centroblasts than in normal naive B cells and virtually absent from the myeloma cell line, U266 (Fig. 2). In addition to encoding a multifunctional protein, LPP is the host gene for microRNAs miR-28-5p and -3p (which lie outside the analyzed region). As with LPP itself, expression of miR-28 is elevated in normal centroblasts, but in the GC B-cell lines analyzed, expression is relatively low. Although the data shown here do not include normal plasma cells, results from normal mouse plasma cells show that LPP, along with BCL6, is markedly down-regulated (13). Using primers spanning its first intron, we also found that the ncRNA described above was present at much higher levels in centroblasts and DHL16 than in normal naive B cells or cell lines at different stages of B-cell differentiation (Fig. 2).

H3K4me3 in the BCL6 Intronic CGI Inversely Correlates with Expression.

Surprisingly, both BCL6-expressing and nonexpressing cells (except CL01) have a broad pattern of H3K4me3 in ∼6 kb of the BCL6 promoter region and first intron (Fig. 3). Genome-wide ChIP-Seq in human resting CD4⁺ T cells and embryonic stem cells (ESCs) also showed H3K4me3 in this region (9, 14). Using a low-stringency definition, two large CGIs are predicted in BCL6: one at the promoter and one within the first intron. Genome-wide analysis of cytosine methylation was recently reported for human ESCs and a fibroblast line, IMR90 (15). Analysis of these data confirmed that the two CGIs were largely unmethylated in both cell types, whereas the neighboring regions were methylated (Fig. 3). In a separate study, much of these two CGIs was examined in numerous tissues and found to be unmethylated in all (16).

Both our data and the T-cell ChIP-Seq data show a pattern of H3K4me3 peaks in the first intron at ∼200 bp separation, consistent with positioned nucleosomes. Positions of wider spacings are found to be due to binding sites for CCCTC-binding factor (CTCF) in T cells, which shows four peaks of CTCF within the first intron (sites 2–5, Fig. 3), as well as peaks upstream and downstream of the gene (sites 1 and 6). We found CTCF binding by ChIP at the same sites in DHL16, CL01, and U266 cells (Fig. S8), but with varying degrees of enrichment. Because cohesin is known to interact with CTCF, we also tested for binding of its subunit RAD21 (Fig. S8) and found enrichment at each of the sites in U266 cells and, with the possible exception of sites 3 and 4, in CL01 cells. In DHL16, RAD21 binding was detected at the flanking sites 1 and 6, but binding to the intronic sites 2–5 was reduced or absent.

The relative quantity of H3K4me3 in the promoter and intronic CGIs differs considerably among the normal samples and cell lines: strikingly, the fraction of H3K4me3 found in the intronic CGI is much higher in cells either not expressing BCL6 or expressing it at very low levels (U266, ESCs, and T cells) compared with the expressing cells (Fig. 3). In addition, among the three samples expressing BCL6, the expression level varies inversely with the degree of intronic H3K4me3. Among the genes we studied, this phenomenon is unique to BCL6, as H3K4me3 showed either no change or an increase in H3K4me3 in samples with higher levels of expression of the other genes examined (Fig. 1 and Figs. S2 and S3).

BCL6 Promoter Interacts with Far Upstream Regions.

To determine whether the promoter region interacts with potential enhancers far upstream of BCL6 that are marked with H3K4me1 and other modifications, we used “chromosome conformation capture” (3C) to detect chromatin looping. Formaldehyde-fixed nuclei from several cell lines were digested with EcoRI; fragments were ligated to create novel junctions between fragments brought into proximity by looping; and a primer and probe at the promoter region (the “anchor fragment”) and reverse primers at numerous restriction fragments at differing distances from the promoter were used to test for formation of these novel junctions (Fig. 4A).

Fig. 4.

Chromosome conformation capture (3C) analysis of the BCL6 locus. (A) Analysis of interactions of the EcoRI fragment containing the BCL6 promoter (anchor fragment). (B) Interactions of an upstream anchor fragment. Data were obtained from at least three biological replicates for each datapoint; error bars are SEM.

Even without looping, generation of novel junctions with neighboring restriction fragments is expected, but at a frequency rapidly declining with distance. Such a drop was found with probes immediately either upstream (5′) or downstream of BCL6, but broad peaks were seen at a greater distance, one centered on ∼60 kb upstream and one extending from ∼150 kb to at least 250 kb upstream of BCL6, the latter corresponding to the broad region of H3K4me1 seen in DHL16 cells near the ABR. Surprisingly, looping was detected not only in cells with high BCL6 expression (DHL16) but also in those with virtually no expression (U266). Looping was found even in CL01, in which the repressive modification H3K27me3 was present throughout the BCL6 and LPP region. Although all three cell lines showed interactions between the promoter and upstream regions, the patterns were different. In particular, fragments close to or containing the ncRNA gene showed greater looping interactions with the BCL6 promoter fragment in DHL16, which expresses the ncRNA, than in the other two cell lines analyzed, in which both genes are silenced. We also designed a primer and probe to detect the interactions of an anchor fragment within the more distant upstream region that showed interaction with the promoter. Using these sequences, we also detected a sharp peak of interaction with the promoter fragment in both DHL16 and U266 cells (Fig. 4B), and a precipitous drop in interaction was found for fragments 3′ of the promoter fragment.

Discussion

Of the >100 genes we analyzed, those coexpressed in B cells and T cells typically showed broadly similar patterns of histone modifications when comparing our B-cell ChIP-chip data with available T-cell ChIP-Seq data (e.g., Fig. 1A). For many genes, the height of peaks of H3K4me3 and, especially, H3Ac at the promoter region correlated with gene expression, whereas the presence of K3K27me3 near the promoter region negatively correlated with expression (Fig. 1 and Figs. S2 and S3). For several of the genes on our custom array, conserved sequences distant from the promoter were bracketed by nucleosomes with histone PTMs that varied in a pattern that correlated with gene expression. For example, several regions downstream of PRDM1 were associated with H3K4me1 in most samples, but H3Ac was considerably increased at these sites in U266 cells, which highly express the gene; additional downstream sites of H3K4me1 and H3Ac were seen only in expressing cells (Fig. 1B). These downstream sites bound pol II only in the two cell lines expressing PRDM1, but not in DHL16. Some such peaks may neighbor enhancers that interact with, and regulate, promoter-bound pol II.

The most dramatic example of marked differences in the pattern of histone PTMs among the samples was in the region of the BCL6 and LPP loci and the intervening sequences. The correlated changes in expression and histone PTMs suggest that BCL6, LPP, and the ncRNA are coregulated. LPP, a component of the actin cytoskeleton, superficially seems unlikely to play a role in regulating the GC program. The protein, however, has been shown to shuttle between the nucleus and the cytoplasm and to act as a coactivator for some ETS family proteins (17). Targets of miRNAs miR-28-5p and -3p, embedded within LPP, have not been analyzed in detail, but recently miR-28-5p was shown to target the cell cycle inhibitor CDKN1A and thus may enhance proliferation of GC B cells (18).

Transcription of a ncRNA in the region between BCL6 and LPP is also intriguing. Recent analysis identified >1,000 long ncRNAs (19); their expression frequently correlates with that of the closest neighboring gene, which had an increased likelihood of encoding a transcription factor, as is the case here. A recent study showed that at least 38% of 469 long ncRNAs expressed in the cell types examined bound to either PRC2 or CoREST repressive complexes (20). A human ncRNA, HOTAIR, transcribed from the HOXC cluster, has been shown to bind PRC2 and to repress transcription across 40 kb of the HOXD cluster (21), which lies on a different chromosome. Thus, it will be valuable to determine whether the GC B-cell-specific ncRNA upstream of BCL6 also can repress other loci.

The general loss of activating histone PTMs and gain of H3K27me3 in CL01 cells throughout the BCL6/LPP region were striking. In contrast, regions between BCL6 and LPP showed marked enhancements in H3K4me1, -me3, and, especially H3Ac in centroblasts compared with naive B-cells, corresponding to the expression pattern of BCL6, LPP, and the ncRNA; these changes were even more dramatic in the centroblast cell line DHL16. We hypothesize that these regions contain multiple powerful enhancers activated in centroblasts that stimulate the the BCL6 promoter and overcome the strong autorepression by BCL6 protein.

Supporting this model, 3C identified interactions between the BCL6 promoter region and these far upstream regions. Several genes have been shown to be regulated by enhancers at great distances (22–24). In many cases, these long-range interactions are cell specific, but in others they are found even in cells not expressing the genes. For example, the promoters of the coregulated T_H2 cytokine genes IL4, IL5, and IL13 interact with each other even in fibroblasts, but the locus control region for these genes participates in these interactions in CD4⁺ T cells and natural killer cells, but not in fibroblasts or B cells (25). The fact that the ncRNA gene lies within a region that interacts with the BCL6 promoter helps to explain the possible coregulation of the two genes. Our 3C data demonstrate that at least some of the looping within this large region is present even in cells that do not express BCL6 or LPP. The preexistence of looping in cells before activation of upstream enhancers would be expected to allow BCL6 to respond rapidly to developmental signals.

In many cases, the protein CTCF participates in long-distance interactions (26) by recruiting the cohesin complex (27), which forms a ring that entraps pairs of DNA molecules (28). BCL6 is unusual in having four clustered CTCF sites within its first intron, in addition to upstream and downstream sites. All of the BCL6 CTCF sites are conserved (Fig. S9) and match CTCF consensus sequences. CTCF binds these sites not only in B cells, but also in T cells and fibroblasts (9, 29). The association of the cohesin component RAD21 with the CTCF sites suggests that they are involved in some form of looping, which is likely to influence looping between the BCL6 promoter and the upstream interacting regions that we have identified. Because the pattern of RAD21 binding varies among the three cell lines tested (Fig. S8), it is likely that the pattern of looping varies as well.

BCL6 contains two large CGIs, one overlapping the promoter and one within the first intron (Fig. 3). A feature of most CGIs is that their CpGs lack cytosine methylation in all tissues irrespective of the activity of the associated gene. Recent genome-wide “methylome” analysis in a human ESC line and a fibroblast line (15) showed that both BCL6 CGIs were largely unmethylated (Fig. 3); a recent analysis of DNA methylation that included much of both of these regions also showed methylation to be absent in all tissues tested (16). There is a strong but partial association between unmethylated CGIs and other undermethylated regions (UMRs) both with TSSs and with histone H3K4me3 (16, 30). Thus, the pattern of H3K4me3 found at BCL6 in ES cells is likely to be the default condition based in part upon the high frequency of CpGs in the region. Given the association between H3K4me3 and gene transcription, it was surprising that H3K4me3 within the intronic CGI is highest in nonexpressing cells. This raises the possibility that H3K4me3 at the intronic CGI actually interferes with promoter activity. This pattern of H3K4me3 is reminiscent of the recent identification of at least 50 genes in which an internal UMR is methylated exclusively in the cell type in which the gene is expressed (16). Evidence was provided that in many cases the UMR contains the promoter for an antisense transcript, which was proposed to interfere with activity of the promoter of the gene in which it is embedded. In an analogous model, antisense transcription initiated from the intronic CGI would interfere with the true BCL6 promoter. As evidence for this possibility, analysis of nascent RNA in IMR90 fibroblasts (31) identified 33 antisense reads within the BCL6 first intron, but none elsewhere within the gene. Alternatively, the intronic CGI might compete with the promoter for interaction with enhancers, either elsewhere in the first intron, upstream of the promoter, or in the far distal regions we have found by 3C to interact with the promoter.

A number of possibilities may explain the relative loss of H3K4me3 in the BCL6 intronic CGI in centroblasts. For example, the intronic CGI DNA may be methylated only in centroblasts; however, immunoprecipitation with antimethyldeoxycytosine antibodies failed to show DNA methylation in DHL16, despite the ready detection of DNA methylation in CL01 throughout both BCL6 CGIs, including the promoter region (Fig. S4). An intriguing alternative possibility is that BCL6 itself may bind this region and lead to reduction in H3K4me3. In this regard, it is interesting that BCOR, a BCL6 corepressor, associates with FBXL10, a H3K4me3 demethylase (32, 33). The BCL6 promoter region and intron 1 contain seven conserved predicted BCL6 binding sites (Fig. 3). ChIP-on-chip analysis of the BCL6 gene in a GC B-cell lymphoma using antibodies to BCL6 (34) showed not only a peak at the promoter, but also an additional, strong peak centered ∼6 kb from the TSS within the first intron, at which a high-affinity BCL6 binding site is predicted (Fig. 3). Thus, BCL6 protein may have both negative and positive effects on its own transcription, the latter through inactivating a region that interferes with promoter activity.

Because some activating (H3K4me3) and repressive (e.g., H3K27me3) histone PTMs are heritable, alterations in these PTMs following DNA damage or chance errors may persist in subsequent cell generations as “epimutations,” which potentially could affect genes at a considerable distance. We suspect that both the tightly repressed pattern in CL01 cells and the abnormally active pattern in DHL16 cells represent epimutations that prevent differentiation into centroblasts or block exit from the centroblast stage. It will be of interest to test primary tumor samples to determine whether epimutations commonly affect master regulators of differentation in lymphoma cells, blocking differentiation.

Materials and Methods

Primary Cells and Cell Lines.

Naive B cells and centroblasts were purified as previously described (35) from human tonsils with approval of the Institutional Review Board. By flow cytometry, isolated naive B cells (IgD⁺CD38^loCD27⁻) were 92–94% pure, and centroblasts (CD77⁺CD38^hi) were 96–98% pure. SU-DHL16, CL01, and U266 cell lines were grown in RPMI medium complemented with 10% FBS, 1 mM L-glutamine, and penicillin/streptomycin.

ChIP.

For analysis of histone modifications, native ChIP was performed following micrococcal nuclease digestion of chromatin from 20 × 10⁶ cells to obtain mononucleosomes as described previously (9), with a few modifications. For other ChIP experiments, cells were cross-linked with formaldehyde (1% final concentration). Chromatin was solubilized by sonication, resulting in DNA of 200–400 bp. For each ChIP, chromatin was incubated with protein A Dynabeads (Dynal Biotech) bearing 4 μg prebound specific antibody. ChIP-validated antibodies were as follows: anti-H3-K9,K14-diacetyl and anti-H3K27me3 from Upstate (06-599 and 07-449); anti-H3K4me1, anti-H3K4me3, anti-polII, and anti-RAD21 from Abcam (Ab8895, Ab8580, Ab5408, and Ab992); and anti-CTCF from Millipore (07-729). ChIP DNA enrichment was validated by real-time PCR. For ChIP-on-chip, input and ChIP DNA samples were amplified by ligation-mediated PCR as described previously (36) with a few modifications. Amplified DNA labeled with Cy5 (ChIP products) and Cy3 (input) was cohybridized by the NimbleGen service laboratory to a custom 387,981-feature oligonucleotide chip (NimbleGen Systems), tiling portions of >100 genes (Table S1). Details on ChIP, ChIP-on-chip, and data manipulation are reported in SI Materials and Methods. The data discussed in this publication have been deposited in the Gene Expression Omnibus (GEO; National Center for Biotechnology Information) and are accessible through GEO Series accession no. GSE19910 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=fjqpjwmuiiekezm&acc=GSE19910).

Gene Expression Profiling.

RNA was isolated using the All Prep DNA/RNA mini kit (Qiagen) and processed with the Genechip 3′IVT express kit (Affymetrix). Gene expression profiling experiments were performed on GeneChip HG U133 plus 2 arrays (Affymetrix) according to the manufacturer's instructions. miRNA expression profiling was conducted using the TaqMan human microRNA array set v2.0 with 300 ng total RNA, which was reversed transcripted with Megaplex RT Primers and preamplified using Megaplex PreAmp Primers. Real-time PCR was run on an Applied Biosystems 7900HT Fast Real-Time PCR System in 384-well plates. Data were normalized to U6 small nuclear RNA.

3C.

The 3C experiments were performed as described (37), with a few modifications; additional controls for the analysis and quantification of the 3C were included as described (38). A total of 10 × 10⁶ cells were formaldehyde crosslinked and digested with EcoRI (New England Biolabs). After enzyme inactivation, novel junctions resulting from chromatin looping were generated by DNA ligation and quantified by real-time PCR using Taqman probes and primers (Table S2). Details on 3C are reported in SI Materials and Methods.