Semirandom Sampling to Detect Differentiation ... - Semantic Scholar

Journal of Gerontology: BIOLOGICAL SCIENCES 2004, Vol. 59A, No. 12, 1221–1233

Copyright 2004 by The Gerontological Society of America

Semirandom Sampling to Detect Differentiation-Related and Age-Related Epigenome Remodeling Valya R. Russanova, Tazuko H. Hirai, and Bruce H. Howard National Institute of Child Health and Human Development, National Institutes of Health, Bethesda, Maryland.

With completion of the human genome project, patterns of higher order chromatin structure can be easily related to other features of genome organization. A well-studied aspect of chromatin, histone H4 acetylation, is examined here on the basis of its role in setting competence for gene activation. Three applications of a new hybrid genome sampling–chromatin immunoprecipitation strategy are described. The first explores aspects of epigenome architecture in human fibroblasts. A second focuses on chromatin from HL-60 promyelocytic leukemia cells before and after differentiation into macrophage-like cells. A third application explores age-related epigenome change. In the latter, acetylation patterns are compared in human skin fibroblast chromatin from donors of various ages. Two sites are reported at which observed histone H4 acetylation differences suggest decreasing acetylation over time. The sites, located in chromosome 4p16.1 and 4q35.2 regions, appear to remodel during late fetal–early child development and from preadolescence through adult life, respectively.

H

IGHER order chromatin structure plays an integral role in the control of gene expression (1–3). Central to chromatin structure are histone modifications that serve to recruit regulatory proteins and remodeling complexes (4–9). The latter facilitate or retard further modifications, and ultimately regulate the recruitment of the transcription apparatus. Cumulatively, such mechanisms give rise to patterns of histone modification extending across the genome. Pattern features may be localized to tens of base pairs, but may also extend over hundreds of kilobases or even encompass entire chromosomes (10–13). Specialized chromatin domains are thought to form the basis for epigenetic memory in yeast and Drosophila systems (14–17). In vertebrates and plants, such domains act in conjunction with DNA methylation to store epigenetic information (18–20). Much of what we know concerning the role of chromatin in transcriptional regulation derives from well-studied gene loci (12,13,21–23). These are most often described in the context of development or differentiation pathways. Paradigms for developmentally linked chromatin remodeling and epigenetic memory are provided by studies on Drosophila homeodomain genes by Polycomb and Trithorax group gene products, as well as X chromosome dosage compensation (24). More recently, the genetically tractable Arabidopsis system is providing interesting examples of chromatin-mediated gene control (25,26). In vertebrates, epigenetic processes are notably evident in developmental regulation of the chicken beta-globin locus (12,13,21,24,27–29), X chromosome inactivation, and imprinting (19,30,31). T-cell differentiation also turns out to be a process associated with striking changes in chromatin and DNA methylation profiles (11,32). With completion of human and mouse genome assemblies, a more general picture of epigenetic processes should

become available. An emerging goal is to describe chromatin and DNA methylation patterns, which, taken together, form the epigenome for a given cell type. Experimental approaches that are or should soon become feasible in mammalian systems have been reported in recent work with Saccharomyces cerevisiae (33,34). One productive strategy in yeast relates gene expression profiles to chromatin regions through the use of microarrays. The relatively small size of the S. cerevisiae genome renders this practical. By contrast, the much greater complexity of mammalian genomes makes it difficult to obtain comparable array coverage. At the resolution of bacterial artificial chromosomes (BACs), a complete tiled representation of the human genome has been reported (35). To query smaller domains, reasonable alternatives are to construct microarrays restricted to small chromosomes, promoter-proximal regions, or CpG islands (36–43). The high cost and long lead time associated with microarrays have prompted several alternative strategies (44–46). In this article, we describe epigenome mapping using a complementary approach related to differential display (47,48). The approach allows general sampling of mammalian genomes and has been implemented in three ways. First, a screen was done to gain an overall profile of histone H4 acetylation patterns in chromatin from normal human fibroblasts. The results are consistent with large alternating regions of high- and low-acetylation distributed along the chromosome arms, as has been reported previously (24). Second, a screen was performed to compare histone H4 acetylation patterns in undifferentiated versus macrophage-like HL-60 cells. This revealed regions with substantial shifts to be relatively infrequent, being on the order of 0.5% of the epigenome. Third, the genome sampling approach was used to search for locus-specific acetylation differences in chromatin from human fibroblasts 1221

1222

RUSSANOVA ET AL.

taken at different passage levels or donor ages. This search was successful, revealing sites that appear to report agerelated epigenome change near the chromosome 4p and 4q termini. METHODS

Cell Culture Normal human skin fibroblasts from donors of different ages were obtained from the Coriell collection. [HSC172 human lung embryo fibroblasts (49) were a gift from Sam Goldstein.] Fetal cells used were: AG04392A, 16 weeks, population doubling level (PDL) 17; AG04449, 12 weeks, PDL 11; AG04431B, 15 weeks, PDL 18; AG04451, 16 weeks, PDL 13; AG04525, 17 weeks, PDL17. Cells from preadolescents were: GM03348D, 10 years, passage 9; GM2036A, 11 years, passage 14; GM00500, 10 years, passage 10; GM1582A, 11 years, passage 9. Cells from young adults were: AG07720B, 23.9 years, PDL 16; AG07719A, 27.9 years, PDL 16; AG04441B, 29.4 years, PDL 16; AG13153, 30.2 years, PDL 5. Cells from old adults were: GM08401, 75 years, passage 2; GM09918, 78 years, passage 10; GM00288B, 64 years, passage 10. Fibroblasts were grown in Dulbecco’s modified Eagle medium (MEM) (Gibco/BRL, Carlsbad, CA) supplemented with 10% fetal calf serum (FCS) or Eagle’s MEM with Earle’s salts with 15% FSC, 2 mM glutamine, and antibiotics (50 U of penicillin/ml and 50 mg of streptomycin/ml). Chinese hamster–human hybrid cell lines Y162-11C and X86T2 were grown as published (50). HL-60 cells were propagated in RPMI media, 10% fetal calf serum. Differentiation was induced by 12-O-tetradecanoylphorbol-13 acetate (TPA) treatment (51). Preparation of Nuclei and Chromatin HSC172 cells were grown to confluency, and, in most experiments, incubated for 18 hours with 20 mM sodium butyrate to inhibit histone deacetylase activity. HL-60 cells were not treated with butyrate prior to harvesting. HSC172 nuclei and histone-H1-depleted chromatin were prepared by a procedure described previously (52). Nuclei from HL-60 cells were isolated according to (51). Micrococcal nuclease concentrations were titrated in order to obtain chromatin fragments with a median length of 1500 bp. All solutions used during preparation of nuclei contained 10 mM sodium butyrate. Immunoprecipitation of Acetylated Chromatin Antibody against acetylated H4 was prepared by immunizing rabbits with chemically acetylated histone H4 (22,53). Affinity purification was accomplished with acetylated bovine serum albumin (BSA) coupled to BrCN Sepharose and characterized by enzyme-linked immunosorbent assay (ELISA) and Western blotting as described previously (53). Affinity-purified antibody (70 lg) was incubated with magnetic beads coated with antirabbit IgG (Dynal Biotech, Brown Deer, WI) as described by the manufacturer. Chromatin fragments (25 lg measured as DNA at 260 nm) were incubated with antibody-coated

beads by rotating overnight at 48C in 50 mM NaCl, 10 mM Tris-HCl pH 8.0, 1 mM EDTA, 10 mM Na-butyrate, and proteinase inhibitor cocktail (22,53). Supernatant from the first magnetic separation combined with the first wash was taken as the unbound chromatin fraction. Beads were washed six times with 1 ml 150 mM NaCl, 10 mM Tris-HCl pH 8.0, 1 mM EDTA, 10 mM Na-butyrate, proteinase inhibitor cocktail (Roche Diagnostics, Indianapolis, IN). Samples were transferred to new tubes prior to a final wash. The bound acetylated chromatin was eluted by two incubations in 100 ll 1% SDS, 1% beta-mercaptoethanol for 15 minutes at 378C. DNA was purified by proteinase K digestion, phenol-chloroform extraction and ethanol precipitation in the presence of 40 lg/ml glycogen, or alternatively was purified using Qiagen PCR (polymerase chain reaction) purification kits (Valencia, CA). The concentration of DNA was determined fluorimetrically. The typical yield from 25 lg chromatin was 0.5–1 lg bound DNA.

PCR Amplification Phosphoglycerate kinase (PGK) sequence primers (54): 59GGTCTCGCACATTCTTCACGTC, 39GCAAGGAACCTTCCCGACTTAG Human a-satellite sequence primers (55): 59TCAGAAACTTCTTTGT, 39GAATGCTTCTGTCTAG First intron of the human ribosomal protein S-14 gene (56): 59CTGTTGCTGATTGGTTAGGC, 39TGAAGGAGAGAAGACTGGAG PCR amplification was performed in a 20 ll reaction mix containing 1–3 ng DNA, 13 buffer (PerkinElmer, Wellesley, MA), 1.5 or 2.5 mM MgCl, 2 lM deoxynucleotides, 0.2 lM primers, 1 unit Taq Polymerase (BoehringerMannheim/Roche), and 2 lCi 33P-dCTP (NEN, DuPont, Wilmington, DE). The cycling parameters were: 5 minutes at 948C, followed by 25 cycles of 15 seconds at 948C, 15 seconds at the annealing temperature, and 15 seconds at 728C. For longer PCR products, the nucleotide concentrations were increased to 10 lM and each cycling step was 30 seconds. For a-satellite PCR, the number of cycles was decreased to 15. Due to the high guanine plus cytosine content in the PGK promoter sequence, 10% glycerol and 3.5% formamide were included in the reaction buffer. As a check for linearity, reactions containing 1 or 3 ng DNA were performed in triplicate. PCR products were separated by 5% or 6% polyacrylamide gel electrophoresis (PAGE) containing Tris/borate/EDTA (TBE) and 7.5 M urea. Quantification was performed by phosphoimager scanning. Real-time PCR was performed using SYBR Green Faststart PCR kits (Qiagen) and a LightCycler instrument (Roche). One ng of each sample was analyzed in triplicate together with 0.1–3 ng input DNA used for standard curve calculation.

Genome Sampling—Chromatin Immunoprecipitation (gsChIP) DNA samples purified from bound (acetylated), unbound (under-acetylated), and input chromatin fractions were

AGE-RELATED EPIGENOME REMODELING

analyzed by PCR using combinations of random primer pairs similar to those utilized previously for RNA differential display (47,48). Each primer contains a 12nucleotide arbitrary sequence at the 39 end and 10 invariant nucleotides including a Hind III site at the 59 end. PCR reactions were performed essentially as described above, except that 25 lM dNTP and 1.25 lM of each primer were used. The cycling parameters were: initial denaturation step at 948C for 5 minutes, 4 cycles of 45 seconds at 948C, 60 seconds at 398C to 458C and 60 seconds at 728C, followed by 25 cycles of 45 seconds at 948C, 60 seconds at 608C, and 60 seconds at 728C (47) (Gene Hunter, Inc., Nashville, TN). PCR products were separated on 6% DNA sequencing gels. After autoradiography, the film and corresponding gel were aligned and fractions of interest excised. Eluted DNAs were reamplified and the resulting PCR products were purified on low-melting agarose gel for direct sequence analysis (Applied Biosystems, Foster City, CA). Primary sequences were aligned to the human genome using the National Center for Biotechnology Information (NCBI) blast utility to obtain candidate loci. Specific primers were then designed for these loci and used for quantitative or realtime PCR.

Statistical and Sequence Analysis Mean difference test (MDT) values were calculated using Mathematica version 5 software (Wolfram Research, Inc., Champaign, IL). Fisher’s Exact Test was used as implemented by O. Langsrud at http://www.matforsk.no/ola/ index.html. Distributions of interspersed repeat sequences were determined using the RepeatMasker utility (A. F. A. Smit and P. Green, ‘‘RepeatMasker’’ at http://repeatmasker.genome.washington.edu). This study utilized the highperformance computational capabilities of the Biowulf PC/ Linux cluster at the National Institutes of Health, Bethesda, MD (http://biowulf.nih.gov).

RESULTS

Antibody Specificity and Control Studies An affinity-purified polyclonal antiacetyl-lysine antibody, acLHA, was employed throughout for immunofractionation (12,53). This antibody primarily recognizes multiacetylated forms of histone H4 (Figure 1A). For chromatin immunoprecipitation (ChIP), oligonucleosomes were separated into bound and nonbound fractions, extracted to recover the corresponding DNA fragment preparations, and then adjusted to equal concentrations for PCR amplification and analysis. With this procedure, loci packaged in more highly acetylated domains than average are represented preferentially in the bound fraction, and the converse is true for loci packaged in under-acetylated regions. Initial experiments focused on the establishment of positive controls. Oligonucleosomes were prepared from two hamster lines, Y162-11C and X86T2 cells; these lines carry a single active human X chromosome or a single inactive X chromosome, respectively (50,54). Immunofractionation was performed and chromatin structure was

1223

probed at the X-chromosome-linked phosphoglycerate kinase 1 (PGK1) locus, known to be associated with euchromatin in the Y162-11C cell line, and with heterochromatin in the X86T2 cell line. As shown in Figure 1B, oligonucleosomes derived from the active X chromosome appeared, as expected, predominantly in the bound fraction, whereas identically prepared material from the inactive X chromosome distributed with a bias to the unbound fraction. In additional controls prepared from human normal fibroblast cells, a-satellite repeat sequences (55) were represented preferentially in the unbound fraction; conversely, the ribosomal protein S-14 first intron, localized within a housekeeping gene and near an early-firing origin of DNA replication (47,56) (and thus likely to reside in a euchromatin region), was efficiently precipitated by the anti-acetylated histone H4 antibody.

Linking Genome Sampling With ChIP Having established differential immunoprecipitation at several known loci, we next modified this approach to obtain a more general picture of H4 acetylation states across the human genome. Oligonucleosomes were prepared from normal, that is, nonimmortalized, human fetal lung (HSC172) fibroblasts (57). As before, pools of anti-acetyl H4 bound and nonbound oligonucleosomes were extracted and the DNA content normalized prior to PCR reactions. Arbitrary pairs of DNA primers were used with four lowstringency cycles followed by 25 standard cycles, that is, conditions similar to those employed for RNA differential display (47,48). An example is shown in Figure 2A of the products obtained. In general, triplicate PCR reactions for each primer pair–DNA combination proved very useful as a means to distinguish reproducible from nonreproducible products. A series of reactions was performed with 16 arbitrary primer pairs, yielding more than 1000 reproducible products. Bound to nonbound ratios of bands from the first eight primer pair sets, plotted according to the presumed histone H4 acetylation level, exhibit a roughly normal distribution (Figure 2B). From the gels used to generate these initial data, 91 bands were excised, reamplified, and subjected to DNA sequence analysis, yielding informative sequence information in 65 cases. Of these, 45 were selected for further study, including 17 loci potentially representing highly acetylated regions and 14 candidates for very underacetylated regions. An additional 14 loci were taken as controls. The positions of the selected loci are plotted onto build 34 of the human genome in Figure 2C. The candidates for both the highly acetylated and under-acetylated regions distribute along the chromosome arms, consistent with semirandom sampling of the epigenome. Confirmation by Real-Time PCR and Local Mapping Deviations from a 1:1 distribution in the differential display protocol might reflect one or more unexpected artifacts. To address this issue, we designed independent specific-primer sets for 26 single-copy loci. Real-time PCR was then performed to compare bound and nonbound material, yielding high-quality data in 20 cases. Several additional loci were confirmed by standard PCR verified

1224

RUSSANOVA ET AL.

Figure 1. Control experiments to verify experimental approach. A: Specificity of the antibody against acetylated histone H4. Total histones were isolated from HeLa cells treated with butyrate, separated by Triton X-100-urea PAGE (polyacrylamide gel electrophoresis), and analyzed by Western blotting. Left panel: incubation with anti-acetyl-H4; right panel: gel stained with Coomassie blue. Histone types as well as acetylated forms of H4 are indicated. B: Controls for chromatin immunoprecipitation (ChIP) procedure. Chromatin fragments were separated into bound and unbound fractions using an antibody against acetylated histone H4. DNA was purified and polymerase chain reaction performed using primers specific for the ribosomal protein S-14 gene [1], alpha-satellite [2], or the human PGK1 gene [3 and 4]. Templates for samples 3 and 4 derived from inactive and active human X chromosomes, respectively. In each comparison, total DNA concentrations were equalized in the samples from bound and unbound fractions. Gels were analyzed by PhosphorImager scanning (Amersham Biosciences/GE Healthcare, Milwaukee, WI) and data were quantified from 3–6 independent experiments. Results are expressed as ratios between bound (B) and nonbound (NB) levels set to a log scale.


1225

Figure 2. Highly acetylated and under-acetylated sites identified in human fibroblasts. A: genome sampling/chromatin immunoprecipitation (gsChIP) gel. DNA samples from bound (B) and nonbound (NB) chromatin fractions were subjected to polymerase chain reaction (PCR) using pairs of arbitrary primers. Products were separated by 6% sequencing PAAG for direct comparison. Triplicate PCR reactions are represented for each DNA template. H ¼ product from highly acetylated locus (mostly bound); L ¼ products from under-acetylated loci (mostly nonbound). B: Distribution of bands from 8 primer pairs. Approximately 850 reproducible and clearly separated bands were counted from autoradiography films. Bands were grouped according to B/NB ratios and plotted to show presumed acetylation distributions. C: Locations of identified loci along chromosomes. Chromosome lengths and positions depicted are based on build 34 of the human genome. Centromeres are drawn as gray circles. Loci shown as open triangles ¼ hyperacetylated; black triangles ¼ under-acetylated; gray triangles ¼ near equal or variable distribution between bound and unbound fractions.

1226

RUSSANOVA ET AL.

through calibration for linearity with respect to DNA input. In all but one instance, the results and the original differential display results were in agreement. Quantification of the real-time PCR data for the confirmed loci is shown in Figure 3A. From these and additional experiments (see below), we conclude that candidates generated by genomic differential display are correctly identified in almost all instances. An important question is whether genome sampling as described here predominantly detects very local chromatin states, for example, limited to one or two nucleosomes, or provides marker loci for larger regions of atypical chromatin. Evidence in support of the latter possibility was provided by walking across »20 kb (kilobase) regions flanking two loci, h40 and h90, characterized respectively by under-acetylation or high-acetylation. The under-acetylated locus, h40, is flanked by loci having similar acetylation levels (Figure 3C). The highly acetylated locus h90 (Figure 3B) lies just upstream from the transcription start point of the trichorhinophalangeal syndrome I (TRPS1) gene; it is also adjacent to a CpG island. Each position checked, extending from h90 across the CpG island and the transcription start site for TRPS1, exhibits elevated acetylation. This is consistent with previous reports that CpG islands and their flanking regions are frequently overacetylated (12,58,59).

Regional Transcripts and CpG Islands The human genome is believed to be composed of generich and gene-poor regions, the latter often referred to as gene deserts (60). We asked whether the loci identified in this study might show a nonrandom distribution with respect to local gene density or expression levels. Taking the gsChIP-identified subsets characterized by low and high acetylation, respectively, overlapping or nearby genes are observed more frequently in the latter (4/14 vs 14/16 closer than 50 kb; p ¼ .0015, Fisher’s exact test; see also Table 1). Related to gene distribution are additional parameters, including both the presence or absence of nearby CpG islands and the number of regional expressed sequence tags (ESTs) documented by the human genome project. To analyze these parameters, flanking regions were extended to include a total of 125 kb (the human genome assembly build 34 provided finished sequences for all but one locus). Strong statistical relations were observed, with under-acetylated loci lying uniformly within EST-poor regions relative to highly acetylated loci (p ¼ .007, MDT; see Figure 4A). Genome regions flanking under-acetylated loci are devoid of CpG islands, whereas such islands are common in the regions flanking highly acetylated loci (p ¼ .002, MDT; Figure 4B). A question that can be asked is whether the CpG and EST distributions observed are merely those that would be seen by landing at random points across the human genome. To address this issue, an additional set of 44 loci was included, representing randomly chosen chromosome positions centered within 125 kb regions. ESTs associated with such regions show wide variation in abundance, similar to highly acetylated loci. By contrast, comparisons between underacetylated and random subsets reveal highly significant

differences in both EST abundance ( p ¼ .0001, MDT) and CpG island frequency (p ¼ .002, MDT; not shown).

Differentiation-Associated Chromatin Remodeling Of considerable value would be a general means to identify regions of local chromatin remodeling in association with changes in differentiation state. In vertebrates, such remodeling has been described in detail through studies on the chicken beta-globin and human immunoglobulin gene loci. We asked whether gsChIP applied at an intermediate scale would be sufficiently sensitive to detect chromatin state changes associated with the human HL-60 promyelocyte-to-macrophage differentiation program. PCR reactions were performed with 27 sets of arbitrary primers taking oligonucleosomes from promyelocytic cells and in parallel from phorbol ester-treated macrophage-like cells. The display patterns obtained were greater than 99% identical, perhaps surprising given the morphological differences between these respective cell types. Still, three loci were identified at which changes in acetylation level accompany differentiation. These changes were confirmed using independent sets of specific primers and real-time PCR (Figure 5A). Additional confirmation was obtained by mapping alterations in chromatin acetylation across a region spanning » 30 kb (Figure 5B). Further characterization remains to be done concerning the size of the regions apparently remodeled in association with differentiation, as well as concerning potential impacts on the expression levels of nearby genes. However, these experiments confirm that chromatin changes occurring in the context of a simple differentiation program can be discovered by semirandom genome sampling as described here. Age-Related Chromatin Remodeling A subject of major interest has been to what extent postnatal development and adult life span in vertebrates are accompanied by age-related epigenome remodeling. This has been extensively investigated with respect to DNA methylation (61), but very little is known concerning locusspecific histone modifications. We realized that gsChIP could be of special value in the latter context, as the requisite first step to find such changes is a relatively unbiased search approach. Human HSC lung embryo fibroblasts were compared at early-passage and late-passage levels (19–27 PDL and 55– 60 PDL, respectively). PCR reactions were performed with 16 sets of arbitrary primers, yielding several candidate loci at which histone acetylation appeared to vary. One of these, h9, exhibited decreasing histone H4 acetylation as cells progressed from early passage to midpassage, that is, substantially before the onset of cellular senescence (Figure 6A and unpublished results). The h9 band was recovered and sequenced, revealing it to be located in the chromosome 4p16.1 region, approximately 11 Mb from the 4p terminus (position 10,967,271 in human genome build 34). We postulated that altered acetylation at h9 might also be evident on examination of fetal and postnatal fibroblasts harvested at comparable early-passage levels. Comparisons were done using the appropriate arbitrary primer pair to


1227

Figure 3. Verification and mapping of genome sampling/chromatin immunoprecipitation (gsChIP) sites. A: Real-time polymerase chain reaction (PCR) values for loci estimated by gsChIP to be under-acetylated (L) or highly acetylated (H). DNA was eluted from gel bands, sequenced, and loci mapped on the human genome. Real-time PCR was performed with locus-specific primers using SYBR Green Fast-start PCR kits on a LightCycler instrument (Roche). One ng of each sample was analyzed in triplicate with 0.11 to 3 ng input DNA used for standard curve preparation. Shown are B/NB ratios set to a log scale. B and C: Mapping across representative under-acetylated (h40) and highly acetylated (h90) regions. Primer pairs distributed across »20 Kb regions flanking h40 (C) and h90 (B) were used in real-time PCR assays performed in triplicate. ARs (amplified regions) are shown to scale. LINEs, SINEs and LTRs are shaded black for conserved elements and gray for less conserved repeats. TRPS1 ¼ trichorhinophalangeal syndrome I gene; LINE ¼ long interspersed nuclear element; SINE ¼ short interspersed nuclear element; LTR ¼ long-terminal repeat.

RUSSANOVA ET AL.

1228

Table 1. Summary Analysis of Under-Acetylated and Highly Acetylated Regions Overlapping Gene or Gene ,50 kb Locus chrState* Chr Pos(bld34) Distant (Comment)y mRNA/ESTsz cpgi§ h04 h06 h07 h12 h14 h22 h32 h37 h38

L L L L L L L L L

X 88290116 20 41570598 7 69872430 3 68524375 16 73870316 4 35884816 14 70474677 12 53458013 1 187764975

h39 h40 h44 h46 h80 h11

L L L L L H

4 94550476 14 24220310 4 164503214 10 112798270 2 156242152 17 39441746

h49 h53 h55 h56

H H H H

6 1431101 3 130157067 5 57455101 4 99823816

h57 h58

H H

2 5

h59 h64

H H

18 21144397 2 207029045

h68 h69 h70

H H H

4 15 8

h74 h77 h78

H H H

2 120755398 22 24918468 2 191721964

h82

H

5

h90

H

8 116633475

99803409 96166432

2665818 34680715 73063517

95793101

No Yes (intron) No No No No Yes (intron) No Yes (30 kb from 39 end) Yes (intron) No No No No Yes (10 kb from 39 end; 27 kb from 59 end) No Yes (intron) No Yes (2 kb from 59end; 28 kb from 39end) Yes (exon) Yes (intron; 4 kb from 39 end) Yes (intron) Yes (43 kb from 39 end) Yes (intron) Yes (intron) Yes (25 kb from 59 end) Yes (intron) Yes (intron) Yes (4.5 kb from 59 end) Yes (7 kb from 39 end) Yes (intron)

1 33 3 36 6 61 68 0 30

0 0 0 0 0 0 0 0 0

10 2 3 2 1 286

0 0 0 0 0 0

6 1157 8 499

3 2 0 0

176 853

1 3

87 150

1 0

NA 80 67

NA 0 1

89 115 202

0 1 1

101

0

110

2

Notes: *chrState. This field and fields to right refer to 125 Kb regions flanking indicated loci. z Human UniGene cluster map. § Number of CpG islands. L ¼ under-acetylated; H ¼ highly acetylated. y

generate the h9 band, but substituting chromatin preparations from several fetal skin fibroblast lines and skin fibroblasts derived from young adults. The results obtained confirmed this idea (Figure 6B and unpublished results). Either very small or no differences at h9 were observed in experiments comparing chromatin from young and old adults (not shown). To search for epigenome remodeling during subsequent life span, chromatin was prepared and pooled from preadolescents versus old adults (ages 10–11 years vs 64–78 years). PCR reactions were done with 40 sets of arbitrary

primer pairs. As in previous comparisons, the gsChIP patterns obtained were almost identical. However, one striking difference was observed at a site termed here p14. Sequencing of the p14 band identified it as being derived from the chromosome 4q35.2 region (position 188,927,735), about 3 Mb from the 4q telomere. While the p14 sequence overlaps a Tigger 1 element belonging to the MER2 repetitive sequence family, there is sufficient divergence from other such sequences in human genome build 34 for the identification to be unambiguous. However, it cannot be excluded that another Tigger 1 repeat element of equal or greater sequence similarity remains to be integrated into a subsequent human genome assembly. DISCUSSION Much remains to be learned at the genome level concerning histone modification patterns and their remodeling in association with differentiation pathways, development, and aging. Here we describe the application of a genome sampling strategy coupled with chromatin immunoprecipitation to explore such aspects of epigenome structure. Several points of a technical nature apply in interpreting the results. The antibody used throughout for chromatin immunoprecipitation recognizes multiacetylated forms of histone H4 (22,53). These forms are abundant; however, it cannot be excluded that in sampling many genome loci, the antibody might recognize other multiacetylated proteins at a small number of sites. The sampling of genome loci as implemented here should be considered as semirandom. While a very large number of chromosome positions should be accessible, extended blocks of low-complexity sequence, simple tandem repeats, and other repeat arrays are necessarily under-represented. A major component of the human and other mammalian genomes is composed of interspersed LINE (long interspersed nuclear elements), SINE (short interspersed nuclear elements), and retrovirus-like elements. This being so, we expected a substantial percentage of sampled loci to overlap partially or completely with repeat sequences. Fortunately, it turns out that unambiguous genome assignments can be made in the great majority of cases. Reamplified fragments containing repetitive elements often contain junctions with unique sequence or adjacent repeats that facilitate identification. At other loci, repetitive elements are sufficiently diverged from other copies for the assignment to be straightforward. We routinely searched flanking genome regions for partial matches to the arbitrary primers used in the original PCR reaction, and these were confirmed for all loci listed in the study. For future work, the data compiled should be useful in setting search parameters. Within the 12 bp variable regions, minimum primer annealing corresponded to matches at the two 39 terminal bases, with no more than six mismatches at the remaining positions; in addition, the sum of mismatches for the two primer variable regions did not exceed nine (unpublished observations). Finally, for candidate loci residing within moderately conserved repetitive elements, the entire genome was searched to exclude the occurrence of other elements having equal or better matches to the primers (unpublished data).


1229

Figure 4. Distributions of ESTs and CpG islands in 125 kb regions flanking underacetylated and highly acetylated loci. A: EST values were taken from Unigen maps for build 34 of the human genome sequence. B: CpG islands were scored as defined by (69). EST ¼ expressed sequence tag.

As mentioned in the introduction, other approaches to map the epigenome are being pursued (34,36–46,62). How efficient is semirandom sampling in comparison to these alternative approaches? For small scale experiments, it is easy to implement gsChIP and to obtain results rapidly. For more extensive sampling, automated liquid handling instruments are certainly desirable. High-end sampling should be comparable to microarrays constructed from PCR products with a layout of approximately 32,000 positions. A simple calculation indicates that 40 arbitrary primers, taken as all combinations, suffice to carry out 780 distinct PCR sampling reactions. Based on our experience, each reaction yields reproducible information for 50–100 loci, thus providing sampling information for » 40,000–80,000 chromosome positions. This is attractive in terms of the investment required, at least relative to PCR-based microarrays, where the cost would include 80,000–160,000 primers and fabrication of the arrays themselves. In terms of flexibility, the same or similar arbitrary primer sets should be usable for any higher eukaryotic genome or combination of genomes of interest. The fraction of a given genome that is accessible by semirandom sampling is a somewhat difficult issue to evaluate, since it depends in part on the average size of the chromatin regions in which the characterized loci reside. Based on results here and in the research literature, a conservative estimate for distinguishable histone H4 acetylation domains is at least 20 kb. Taking this value, a medium scale sampling with arbitrary primers—for example 100 primer pairs—should cover roughly 5% of the human genome. This remains a relatively small fraction; however, a common goal will be to obtain seed examples of differentiation-linked chromatin changes. Taking the above numbers and a differentiating cell type in which »100 complex sequence regions undergo remodeling, the proba-

bility of discovering at least a few such changes is in fact quite high (.99% in the absence of unexpected sources of bias). Such seed examples, in addition to their intrinsic interest, should frequently provide the basis for bioinformatics-based queries to search for additional instances. Beyond a description and comments on the potential of gsChIP, this article provides three practical examples of its utility. With respect to static epigenome architecture, a finding of some interest is the comparative homogeneity of regions flanking under-acetylated loci, both with respect to CpG islands and to ESTs. The contrast is valid for comparisons to regions near highly acetylated loci as well as to regions flanking randomly selected chromosome positions. CpG islands near genes are highly acetylated, presumably due to associated transcription factors that recruit histone acetyltransferase complexes (12,58,59). Likewise, polymerase complexes typically require multiple acetyltransferases for efficient assembly. Thus, the under-acetylated loci identified here seem most likely due to a lack of nearby entry sites for histone acetyltransferase complexes. An important application of gsChIP is to identify regions of chromatin remodeling associated with differentiation programs. Along these lines, there is scant information at the genome level, and much valuable information to be gained. We chose HL-60 as a model system in which homogeneous populations of prepromyelocytes growing in suspension can be readily separated from adherent macrophage-like derivatives. Nuclei of the latter have somewhat more compact nuclei than their parent cells at the level of microscopy; nevertheless, it was uncertain whether regions exhibiting a change in acetylation would be common or, at the other extreme, too rare for detection. It is notable that remodeling events turn out to be infrequent but detectable, and that a moderate-sized screen consisting of 27 primer pairs was adequate to reveal these events.

1230

RUSSANOVA ET AL.

Figure 5. Local chromatin remodeling associated with HL-60 differentiation. A. Control and TPA-treated HL-60 cells were compared by genome sampling/ chromatin immunoprecipitation (gsChIP). Real-time polymerase chain reaction (PCR) data are shown for three loci exhibiting differential amplification (Unstable) and for control loci (Stable). B: Acetylation levels across »25 kb region flanking locus L28. Black triangles show position of original locus. Fold change values are indicated below.

Our initial motivation to develop gsChIP was in fact based on a desire to search for age-related epigenome remodeling. X-chromosome reactivation is well described in mouse models (63–66), but to date has been detected in human cells only at relatively low levels (67,68). Likewise, most CpG island silencing with age occurs in too small a fraction of cells to have been detectable by the approach described here (61). The identification of two candidate chromatin loci exhibiting

age-related differences in histone H4 acetylation suggests that this form of remodeling may occur in a larger fraction of cells than the previously observed epigenetic changes. How many comparable loci might be found by genome sampling remains to be seen; however, as noted above, it seems likely that a relatively small fraction of the epigenome has been effectively queried so far. An important question is whether mapping of the regions flanking the h9 and p14 sites will


1231

Figure 6. Local chromatin differences between fibroblasts related to passage level or donor age. A: Comparison of chromatin fractions derived from early and late passage HSC172 human lung embryo fibroblasts. Segment of genome sampling/chromatin immunoprecipitation (gsChIP) gel pattern is shown. Arrow indicates h9 product. Early passage corresponds to population doubling level (PDL) 19; Late, PDL 60. B ¼ bound; NB ¼ nonbound material. Total DNA concentrations in B and NB fractions were equalized prior to polymerase chain reaction (PCR) amplification. B: Comparison of materials derived from fetal and young adult skin fibroblasts. Segment of gsChIP gel with h9 product indicated by arrow. Fetal skin fibroblasts were derived from a 16-week-old embryo (AG04392A, PDL 17). Young adult fibroblasts were from individuals of age range 24 to 30 yr (see Methods). C: Comparison of chromatin preparations derived from preadolescent and old donors (see Methods). Segment of gsChIP gel with p14 product indicated by arrow.

reveal extended areas of change, as is found at the L28 locus in association with HL-60 differentiation. Before addressing this question, we chose to develop a much more efficient set of techniques for building regional epigenome maps. These techniques, together with results relating to the h9 and p14 loci, are described elsewhere (70). ACKNOWLEDGMENTS We thank Arthur Riggs for providing the Y162-11C and X86T2 cell lines. We thank David Landsman and Jonathan Epstein for help with bioinformatics. HSC172 human lung embryo fibroblasts (49) were a gift from Sam Goldstein, formerly of the University of Arkansas. Address correspondence to Bruce Howard, MD, NICHD, NIH, Bldg. 31, Rm. 2A25, 6 Center Dr., Bethesda, MD 20892. E-mail: howard@helix. nih.gov

REFERENCES 1. Horn PJ, Peterson CL. Molecular biology. Chromatin higher order folding—wrapping up transcription. Science. 2002;297:1824–1827. 2. Richards EJ, Elgin SC. Epigenetic codes for heterochromatin formation and silencing: rounding up the usual suspects. Cell. 2002;108:489–500. 3. Rice JC, Allis CD. Code of silence. Nature. 2001;414:258–261. 4. Cheung P, Allis CD, Sassone-Corsi P. Signaling to chromatin through histone modifications. Cell. 2000;103:263–271. 5. Hake SB, Xiao A, Allis CD. Linking the epigenetic ‘language’ of covalent histone modifications to cancer. Br J Cancer. 2004;90:761–769. 6. Lachner M, Jenuwein T. The many faces of histone lysine methylation. Curr Opin Cell Biol. 2002;14:286–298. 7. Sun JM, Chen HY, Davie JR. Effect of estradiol on histone acetylation dynamics in human breast cancer cells. J Biol Chem. 2001;276:49435– 49442. 8. Iizuka M, Smith MM. Functional consequences of histone modifications. Curr Opin Genet Dev. 2003;13:154–160. 9. Fischle W, Wang Y, Allis CD. Histone and chromatin cross-talk. Curr Opin Cell Biol. 2003;15:172–183.

1232

RUSSANOVA ET AL.

10. Turner BM, Birley AJ, Lavender J. Histone H4 isoforms acetylated at specific lysine residues define individual chromosomes and chromatin domains in Drosophila polytene nuclei. Cell. 1992;69:375–384. 11. Zhou W, Chang S, Aune TM. Long-range histone acetylation of the Ifng gene is an essential feature of T cell differentiation. Proc Natl Acad Sci U S A. 2004;101:2440–2445. 12. Hebbes TR, Clayton AL, Thorne AW, Crane-Robinson C. Core histone hyperacetylation co-maps with generalized DNase I sensitivity in the chicken beta-globin chromosomal domain. EMBO J. 1994;13: 1823–1830. 13. Litt MD, Simpson M, Recillas-Targa F, Prioleau MN, Felsenfeld G. Transitions in histone acetylation reveal boundaries of three separately regulated neighboring loci. EMBO J. 2001;20:2224–2235. 14. Hall IM, Shankaranarayana GD, Noma K, et al. Establishment and maintenance of a heterochromatin domain. Science. 2002;297:2232–2237. 15. Wallrath LL, Elgin SC. Position effect variegation in Drosophila is associated with an altered chromatin structure. Genes Dev. 1995;9:1263–1277. 16. Aparicio OM, Gottschling DE. Overcoming telomeric silencing: a trans-activator competes to establish gene expression in a cell cycle-dependent way. Genes Dev. 1994;8:1133–1146. 17. Henikoff S. A reconsideration of the mechanism of position effect. Genetics. 1994;138:1–5. 18. Bird A. DNA methylation patterns and epigenetic memory. Genes Dev. 2002;16:6–21. 19. Reik W, Walter J. Genomic imprinting: parental influence on the genome. Nat Rev Genet. 2001;2:21–32. 20. Li E, Beard C, Jaenisch R. Role for DNA methylation in genomic imprinting. Nature. 1993;366:362–365. 21. Litt MD, Simpson M, Gaszner M, David Alis C, Felsenfeld G. Correlation Between histone lysine methylation and developmental changes at the chicken beta-globin locus. Science. 2001;293:2453– 2455. 22. Hebbes TR, Thorne AW, Crane-Robinson C. A direct link between core histone acetylation and transcriptionally active chromatin. EMBO J. 1988;7:1395–1402. 23. Issa JP. Age-related epigenetic changes and the immune system. Clin Immunol. 2003;109:103–108. 24. Jeppesen P, Turner BM. The inactive X chromosome in female mammals is distinguished by a lack of histone H4 acetylation, a cytogenetic marker for gene expression. Cell. 1993;74:281–289. 25. Luo S, Preuss D. Strand-biased DNA methylation associated with centromeric regions in Arabidopsis. Proc Natl Acad Sci U S A. 2003;100:11133–11138. 26. He Y, Michaels SD, Amasino RM. Regulation of flowering time by histone acetylation in Arabidopsis. Science. 2003;302:1751–1754. 27. Bulger M, Sawado T, Schubeler D, Groudine M. ChIPs of the betaglobin locus: unraveling gene regulation within an active domain. Curr Opin Genet Dev. 2002;12:170–177. 28. Horak CE, Mahajan MC, Luscombe NM et al. GATA-1 binding sites mapped in the beta-globin locus by using mammalian chIp-chip analysis. Proc Natl Acad Sci U S A. 2002;99:2924–2929. 29. Belyaev N, Keohane AM, Turner BM. Differential underacetylation of histones H2A, H3 and H4 on the inactive X chromosome in human female cells. Hum Genet. 1996;97:573–578. 30. Sleutels F, Barlow DP. The origins of genomic imprinting in mammals. Adv Genet. 2002;46:119–163. 31. Saitoh S, Wada T. Parent-of-origin specific histone acetylation and reactivation of a key imprinted gene locus in Prader-Willi syndrome. Am J Hum Genet. 2000;66:1958–1962. 32. Chowdhury D, Sen R. Stepwise activation of the immunoglobulin mu heavy chain gene locus. EMBO J. 2001;20:6394–6403. 33. Robyr D, Suka Y, Xenarios I, et al. Microarray deacetylation maps determine genome-wide functions for yeast histone deacetylases. Cell. 2002;109:437–446. 34. Robyr D, Grunstein M. Genomewide histone acetylation microarrays. Methods. 2003;31:83–89. 35. Till BJ, Colbert T, Tompa R, et al. High-throughput TILLING for functional genomics. Meth Mol Biol. 2003;236:205–220. 36. van Steensel B, Henikoff S. Epigenomic profiling using microarrays. Biotechniques. 2003;35:346–350, 352–354, 356–357. 37. Lo AW, Magliano DJ, Sibson MC, et al. A novel chromatin immunoprecipitation and array (CIA) analysis identifies a 460-kb

38.

39. 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50.

51.

52.

53.

54.

55. 56. 57. 58. 59.

60.

CENP-A-binding neocentromere DNA. Genome Res. 2001;11: 448–457. Lian Z, Kluger Y, Greenbaum DS, et al. Genomic and proteomic analysis of the myeloid differentiation program: global analysis of gene expression during induced differentiation in the MPRO cell line. Blood. 2002;100:3209–3220. Kurdistani SK, Grunstein M. In vivo protein-protein and proteinDNA crosslinking for genomewide binding microarray. Methods. 2003;31:90–95. Albertson DG, Pinkel D. Genomic microarrays in human genetic disease and cancer. Hum Mol Genet. 2003;12(Spec No 2):R145–R152. Lee MP. Genome-wide analysis of epigenetics in cancer. Ann N Y Acad Sci. 2003;983:101–109. Shi H, Wei SH, Leu YW, et al. Triple analysis of the cancer epigenome: an integrated microarray system for assessing gene expression, DNA methylation, and histone acetylation. Cancer Res. 2003;63:2164–2171. Smith RJ, Dean W, Konfortova G, Kelsey G. Identification of novel imprinted genes in a genome-wide screen for maternal methylation. Genome Res. 2003;13:558–569. Barski A, Frenkel B. ChIP Display: novel method for identification of genomic targets of transcription factors. Nucleic Acids Res. 2004;32:e104. Liang G, Lin JC, Wei V, et al. Distinct localization of histone H3 acetylation and H3-K4 methylation to the transcription start sites in the human genome. Proc Natl Acad Sci U S A. 2004;101:7357–7362. Roh TY, Ngau WC, Cui K, Landsman D, Zhao K. High-resolution genome-wide mapping of histone modifications. Nat Biotechnol. 2004;22:1013–1016. Linskens MH, Feng J, Andrews WH, et al. Cataloging altered gene expression in young and senescent cells using enhanced differential display. Nucleic Acids Res. 1995;23:3244–3251. Liang P, Pardee AB. Differential display of eukaryotic messenger RNA by means of the polymerase chain reaction. Science. 1992;257:967–971. Gupta RS, Goldstein S. Diphtheria toxin resistance in human fibroblast cell strains from normal and cancer-prone individuals. Mutat Res. 1980;73:331–338. Hansen RS, Ellis NA, Gartler SM. Demethylation of specific sites in the 59 region of the inactive X-linked human phosphoglycerate kinase gene correlates with the appearance of nuclease sensitivity and gene expression. Mol Cell Biol. 1988;8:4692–4699. O’Neill LP, Turner BM. Histone H4 acetylation distinguishes coding regions of the human genome from heterochromatin in a differentiationdependent but transcription-independent manner. EMBO J. 1995; 14:3946–3957. Russanova VR, Driscoll CT, Howard BH. Adenovirus type 2 preferentially stimulates polymerase III transcription of Alu elements by relieving repression: a potential role for chromatin. Mol Cell Biol. 1995;15:4282–4290. Mutskov VJ, Russanova VR, Dimitrov SI, Pashev IG. Histones associated with non-nucleosomal rat ribosomal genes are acetylated while those bound to nucleosome-organized gene copies are not. J Biol Chem. 1996;271:11852–11857. Pfeifer GP, Riggs AD. Chromatin differences between active and inactive X chromosomes revealed by genomic footprinting of permeabilized cells using DNase I and ligation-mediated PCR. Genes Dev. 1991;5:1102–1113. Koch JE, Kolvraa S, Petersen KB, Gregersen N, Bolund L. Oligonucleotide-priming methods for the chromosome-specific labelling of alpha satellite DNA in situ. Chromosoma. 1989;98:259–265. Tasheva ES, Roufa DJ. Densely methylated DNA islands in mammalian chromosomal replication origins. Mol Cell Biol. 1994;14:5636–5644. Gupta RS, Siminovitch L. Diphtheria toxin resistance in Chinese hamster cells: genetic and biochemical characteristics of the mutants affected in protein synthesis. Somatic Cell Genet. 1980;6:361–379. Tazi J, Bird A. Alternative chromatin structure at CpG islands. Cell. 1990;60:909–920. Myers FA, Evans DR, Clayton AL, Thorne AW, Crane-Robinson C. Targeted and extended acetylation of histones H4 and H3 at active and inactive genes in chicken embryo erythrocytes. J Biol Chem. 2001;276:20197–20205. Nobrega MA, Ovcharenko I, Afzal V, Rubin EM. Scanning human gene deserts for long-range enhancers. Science. 2003;302:413.


61. Issa JP. Living longer: the aging epigenome. In: Beck S, Olek A, eds. The Epigenome. Weinheim: Wiley-VCH; 2003:141–149. 62. Pelling AL, Thorne AW, Crane-Robinson C. A human genomic library enriched in transcriptionally active sequences (aDNA library). Genome Res. 2000;10:874–886. 63. Cattanach BM. Position effect variegation in the mouse. Genet Res. 1974;23:291–306. 64. Wareham KA, Lyon MF, Glenister PH, Williams ED. Age related reactivation of an X-linked gene. Nature. 1987;327:725–727. 65. Gartler SM, Goldman MA. Reactivation of inactive X-linked genes. Dev. Genet. 1994;15:504–514. 66. Bennett-Baker PE, Wilkowski J, Burke DT. Age-associated activation of epigenetically repressed genes in the mouse. Genetics. 2003;165: 2055–2062.

1233

67. Migeon BR, Axelman J, Beggs AH. Effect of ageing on reactivation of the human X-linked HPRT locus. Nature. 1988;335:93–96. 68. Holliday R. Understanding Ageing. Cambridge: Cambridge University Press; 1995. 69. Gardiner-Garden M, Frommer M. CpG islands in vertebrate genomes. J Mol Biol. 1987;196:261–282. 70. Russanova VR, Hirai TH, Tchernov AV, Howard BH. Mapping development-related and age-related chromatin remodeling by a high throughput ChIP-HPLC approach. J Gerontol Biol Sci. 2004;59A: 1234–1243. Received August 3, 2004 Accepted October 1, 2004 Decision Editor: James R. Smith, Jr.