Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Microarray-based DNA methylation profiling: technology and applications

Microarray-based DNA methylation profiling: technology and applications Published online January 20, 2006 528–542 Nucleic Acids Research, 2006, Vol. 34, No. 2 doi:10.1093/nar/gkj461 Microarray-based DNA methylation profiling: technology and applications Axel Schumacher, Philipp Kapranov , Zachary Kaminsky, James Flanagan, 2 2 2 1 Abbas Assadzadeh, Patrick Yau , Carl Virtanen , Neil Winegarden , Jill Cheng , Thomas Gingeras and Arturas Petronis* The Krembil Family Epigenetics Laboratory, Centre for Addiction and Mental Health, 250 College Street, Toronto, ON, 1 2 Canada M5T 1R8, Affymetrix, Santa Clara, USA and The Microarray Centre, The University Health Network, 200 Elizabeth Street, Toronto, ON, Canada M5G 2C4 Received November 21, 2005; Revised December 20, 2005; Accepted January 5, 2006 INTRODUCTION ABSTRACT Over the last decade the field of DNA methylation has grown This work is dedicated to the development of a dramatically and become one of the most dynamic and rapidly technology for unbiased, high-throughput DNA developing branches of molecular biology. The methyl group methylation profiling of large genomic regions. In at the fifth position of the cytosine pyrimidine ring, that is this method, unmethylated and methylated DNA present in about 80% of CpG-dinucleotides in the human fractions are enriched using a series of treatments genome, can be of major functional significance and is with methylation sensitive restriction enzymes, and regarded as the ‘fifth base’ of the genome (1). DNA methyla- interrogated on microarrays. We have investigated tion, along with histone modifications (acetylation, methyla- various aspects of the technology including its rep- tion, phosphorylation and the like), are referred to as licability, informativeness, sensitivity and optimal epigenetic phenomena that control various genomic functions PCR conditions using microarrays containing oligo- without a change in nucleotide sequence (2). Such functions nucleotides representing 100 kb of genomic DNA include meiotic and mitotic recombination, replication, control of ‘parasitic’ DNA elements, establishing and maintenance of derived from the chromosome 22 COMT region in gene expression profiles, X chromosome inactivation as well as addition to 12 192 element CpG island microarrays. regulation of developmental programming and cell differenti- Several new aspects of methylation profiling are ation (3–6). Aberrations in epigenetic regulation, or ‘epimuta- provided, including the parallel identification of tions’, cause several paediatric syndromes (Prader–Willi confounding effects of DNA sequence variation, [OMIM #176270], Angelman [OMIM #105830], Beckwith– the description of the principles of microarray Wiedemann [OMIM #130650] and Rett [OMIM #312750]) design for epigenomic studies and the optimal (7) and may also predispose to cancer (8). choice of methylation sensitive restriction enzymes. Our understanding of the peculiarities of DNA methylation We also demonstrate the advantages of using the in the human genome is still very superficial. Based on the unmethylated DNA fraction versus the methylated review of available publications, our estimate is that <0.1% of the genome has been subjected to a detailed DNA modification one, which substantially improve the chances of analysis. The recently completed Human Genome sequencing detecting DNA methylation differences. We applied project did not attempt to differentiate between methylated and this methodology for fine-mapping of methylation unmethylated cytosines. To some extent our understanding of patterns of chromosomes 21 and 22 in eight indi- the dynamic state of genome-wide DNA methylation has been viduals using tiling microarrays consisting of over hampered by the lack of high-throughput technologies that 340 000 oligonucleotide probe pairs. The principles would interrogate DNA methylation profiles over large developed in this work will help to make epigenetic genomic regions. A gold standard technique in DNA methyla- profiling of the entire human genome a routine tion studies, the bisulfite modification-based fine mapping of met procedure. C (9), although precise, is very labour intensive and in *To whom correspondence should be addressed. The Krembil Family Epigenetics Laboratory, Room 28, Centre for Addiction and Mental Health, 250 College Street, Toronto, ON, Canada M4T 1R8. Tel: +1 416 5358501 4880; Fax: +1 416 979 4666; Email: [email protected] The Author 2006. Published by Oxford University Press. All rights reserved. The online version of this article has been published under an open access model. Users are entitled to use, reproduce, disseminate, or display the open access version of this article for non-commercial purposes provided that: the original authorship is properly and fully attributed; the Journal and Oxford University Press are attributed as the original place of publication with the correct citation details given; if an article is subsequently reproduced or disseminated not in its entirety but only in part or as a derivative work this must be clearly indicated. For commercial re-use, please contact [email protected] Nucleic Acids Research, 2006, Vol. 34, No. 2 529 most cases limited to short DNA fragments, often less than a DNA from the same brain region (prefrontal cortex) was used kilobase. as a control. Unmethylated sites were defined using a two-step The advent of microarray technologies that enabled the inter- analysis approach similar to the one used to determine tran- rogation of a large number of DNA/RNA fragments in a highly scription factor binding sites in the chromatin immunoprecip- parallel fashion has opened new opportunities for epigenetic itation (ChIP)-chip assay (27). First, a smoothing-window studies (10). A number of microarray-based technologies used Wilcoxon approach was applied to generate a P-value for epigenetic analyses are already available (11–23). However, graph for each individual where probe signal from the enriched all of these methods have some limitations, which renders them fraction was compared with the total genomic DNA in a one- unsuitable for some experimental setups. Additionally, many sided upper paired test. The window used in this report was technological parameters, such as the influence of DNA 501 bp. Second, three thresholds were applied to determine the sequence variations, amplification conditions and sensitivity boundaries of the unmethylated site: (i) an individual probe of the methods have not been investigated before. Here we threshold of P < 10 to determine if a probe is significantly present a detailed analysis of various parameters of epigenetic enriched in the unmethylated fraction compared with the profiling and provide a substantially improved microarray- control total genomic DNA; (ii) the maximum distance based high-throughput technology for DNA methylation between the two positive probes set to 250 bp and (iii) profiling of DNA regions that span from hundreds of kilobases the minimal size of a site set to 1 bp. The graphs can be to megabases. Eventually, this technology will be applied to downloaded from the internet (see Web resources). All the entire human genome, as exemplified by the methylation coordinates and annotation analysis were done on the April mapping of chromosomes 21 and 22 as reported here. 2003 version of the genome. Methylation -sensitive digestion of genomic MATERIALS AND METHODS DNA (gDNA) Microarray fabrication and data processing Prior to treatment with restriction enzymes, gDNA was sup- plemented with ‘spike’-DNAs (different concentrations of l COMT and CpG island microarrays were printed on Corning and Arabidopsis fragments), which were used as controls for CMT-GAPSII slides (Corning Life Sciences, Acton, MA) signal normalization. For enrichment of the unmethylated using a VersArray ChipWriter Pro System (Bio-Rad Labor- fraction, depending on the number of CpG dinucleotides to atories, Hercules, CA). For the COMT array, we designed 384 be interrogated, several combinations of methylation-sensitive oligonucleotides (Operon/Qiagen, US), each 50 bases long, enzymes, HpaII, Hin6I, AciI and HpyCH4IV, were used. representing every restriction fragment flanked by HpaII, gDNA was cleaved with a cocktail of these enzymes (10 U/ Hin6I and AciI restriction sites. In addition, control DNA ml in 2xY+/Tango buffer, Fermentas Life Sciences/Lithuania) fragments containing l phage, pBR322, FX174 and pUC57 for 8 h at 37 C. For enrichment of the methylated fraction, sequences were spotted on the slide. Each oligonucleotide was gDNA was cleaved by TasI or Csp6I (10 U/mlinG -buffer, diluted to a 25 mM solution and spotted four times to give a Fermentas) for 8 h at 65 C (TasI) or at 37 C (Csp6I). After the total of 1536 elements. In addition, 192 blank spots consisted restriction reaction, TasI was inactivated by 0.5 M EDTA. of SSC buffer and 48 spots contained Arabidopsis clones. The human CpG island array contains 12 192 sequenced CpG Adaptor–ligation island clones derived from a CpG island library that was ori- ginally created with MeCP2 DNA binding columns (24,25). For the ligation step, gDNA was supplemented with 8 GE Hybridized arrays were scanned on a GenePix 4000A scan- MspI-cleaved pBR322 plasmid (1 GE ¼ 1.45 pg/ 1 mg ner (Axon Instruments, Union City/CA) and analysed using gDNA), which was used as control for a potential ligation the GenePix 6.0 software. The GenePix PMT voltage for Cy3 bias. The ends of the cleaved DNA fragments were ligated and Cy5 channels were balanced with the histogram feature of to the unphosphorylated adaptors. Our adaptors contained a the scanner software to ensure a similar dynamic range for the sequence-specific protruding end, a non-target homologous two channels. Final scans were taken at 10 mm resolution, and core sequence, a specific antisense-overhang that prevents images for each channel were saved as separate 16-bit TIFF tandem repeat formation and blunt-end ligation, a ‘disruptor’ files. The emission signals for each channel were determined sequence that interrupts the original restriction sites after liga- by subtracting the local background from its corresponding tion, a new non-palindromic Alw26I (BsmAI) restriction site median average intensity. These raw data were either exported that enables the blunt-end cleavage of the adaptor from the into a custom Excel spreadsheet for subsequent data analysis target sequences (e.g. for library enrichment) and a non-5 - or directly imported into the Acuity 4.0 software (Axon complementary end. The CpG-overhang specific universal Instruments). The resulting datasets were normalized for the adaptor ‘U-CG1’ for the unmethylated DNA fraction ligates normalization features (spike-DNAs) and for signal intensity to DNA fragments generated by 11 CpG-methylation- (Lowess normalization). sensitive restriction enzymes HpaII, Hin6I (Hinp1I), Hpy- Profiling of unmethylated sites in the brain tissue of eight CH4IV, Bsu15I (ClaI, BspDI), AciI (SsiI), Psp1406I (AclI), adults was carried out using a tiling array spanning 12 Mb of Bsp119I (AsuII), Hin1I (AcyI, BsaHI), XmiI (AccI), NarI, non-repetitive sequence of chromosome 21 and 22 (q arms), BstBI (FspII) and also TaqI and MspI, which are not affected with probes spaced on average every 35 bp center-to-center by methylation of the internal cytosine. The adaptor represents (26). The genomic DNA from these individuals was cut with the annealing product of the two primers U-CG1a, 0 0 HpaII and Hin6I, amplified and hybridized to the microarray 5 -CGTGGAGACTGACTACCAGAT-3 , and U-CG1b, 0 0 as described previously (26,27). Unprocessed total genomic 5 -AGTTACATCTGGTAGTCAGTCTCCA-3 . 530 Nucleic Acids Research, 2006, Vol. 34, No. 2 The AATT-overhang specific adaptor ‘AATT-1’ for the The hybridization method used for the chromosome 21 and methylated DNA fraction fits to DNA ends produced by the 22 tiling arrays was described before (26,27). restriction enzyme TasI (TspEI), whereas the ‘TA-1’ adaptor Whole genome amplification fits to ends produced by Csp6I, BfaI or MseI, respectively: 0 0 AATT-1a, 5 -AATTGAGACTGACTACCAGAT-3 ; AAT- Genomic DNA was amplified using the GenomiPhi Kit 0 0 T-1b, 5 -AGTTACATCTGGTAGTCAGTCTC-3 ; TA-1a, (Amersham Biosciences) according to the manufacturer’s pro- 0 0 0 5 -TATGAGACTGACTACCAGAT-3 ; and TA-1b: 5 -AGT- tocol. Briefly, 10 ng of gDNA (1 ml) was mixed with 9 mlof TACATCTGGTAGTCAGTCTCA-3 . sample buffer, denatured at 95 C for 3 min, cooled on ice and All adapters were prepared by mixing equimolar amounts of then added to 9 ml of reaction buffer and 1 ml of Phi29 DNA the primer pairs, incubating the mixture at 80 C for 5 min, and polymerase. The reaction was incubated at 30 C for 16 h and then cooling it down to 4 C with 1 C/min. The double- then inactivated at 65 C for 10 min. stranded adaptors [200 pmol/ml] were added at 0.1 pmol per enzyme for each ng of the cleaved DNA (e.g. 0.3 pmol/ Bisulfite sequencing ng in a triple-digest HpaII/Hin6I/AciI). The ligation-mixture The methylation status of a number of CpG islands were with 400 ng template DNA was supplemented with 2 mlof 10· analysed by direct sequencing of sodium bisulphite modified ligation buffer (Fermentas), 1 ml ATP [10 mM] and water to gDNA (9). gDNA samples were subjected to bisulfite modi- 18 ml. The reaction was started in a thermal-cycler at 45 C for fication using a standard protocol (28). The primer sequences, 10 min, chilled on ice and 2 ml T4 ligase (Fermentas) was PCR conditions and cloning methods are provided in the added. The ligation reaction was carried out at 22 C for 18 h, Supplementary Data. followed by a heat-inactivation step at 65 C for 5 min. The mixture was then cooled down to room temperature with 1 C/ Genomic DNA min and stored at 4 C for subsequent procedures. Genomic DNA from all tissues was purified using standard laboratory methods (Phenol–Chloroform or Qiagen Blood and PCR Cell DNA Midi columns). To avoid cross reactivity of amine To control for a potential PCR bias, the DNA mixture was groups with the aminoallyl-labeling procedure, DNA samples supplemented with 2 GE FX174 plasmid (1 GE ¼ 1.8 pg of were stored in 0.5 M POPSO buffer (pH 8.0) instead of FX174 corresponding to 1 mg gDNA) that was cut with - Tris–EDTA. Male placental DNA was purchased from HpyCH4IV and ligated to the adaptor. PCR amplifications Sigma and the post mortem brain samples were provided were conducted for up to 25 cycles. A standard aminoallyl- by the Stanley Medical Research Institute. All parts of the PCR mixture included 400 ng of the ligate, 40 mlof10· study were approved by the CAMH review/ethics board. reaction-buffer (Sigma), 42 ml MgCl [25 mM], 3 ml aminoallyl-dNTP Mix [containing 15 mM aminoallyl- Web resources dUTP, 10 mM dTTP and 25 mM each dCTP, dGTP and All chromosome 21/22 tiling array data can be viewed in the dATP], 200 pmol primer (U-CG1a, AATT-1b or TA-1b, UCSC genome browser available via the methylation database respectively), 3 ml Taq enzyme (5 U/ml, NEB) and water to at www.epigenomics.ca. Additionally, the complete tiling a final volume of 400 ml. For PCR conditions and generation of array source data plus graphs that can be viewed in the Integ- dye-coupled adaptor products see Supplementary Data. rated Genome Browser (Affymetrix; www.affymetrix. com/support/developer/downloads/TilingArrayTools/index.affx) Array hybridizations and can be downloaded at http://transcriptome.affymetrix.com/ download/DataMethPaper (case sensitive). All coordinates and Each microarray slide was prehybridized with a mixture con- annotation analysis was done on the April 2003 version of the sisting of DIG Easy Hyb (Roche Diagnostics), 25 mg/ml tRNA genome. SNP data were derived from the SNP consortium, www. and 200 mg/ml BSA. The printed area was covered with the ncbi.nlm.nih.gov/SNP. prehybridization mixture under a coverslip for 1 h at 45 C. The OMIM numbers are derived from Online Mendelian Inheritance microarray slides were then washed in two changes of water in Man (OMIM), http://www.ncbi.nlm.nih.gov/entrez/query. for 2 min at 45 C, followed by two wash-steps at room tem- fcgi?db¼OMIM. Genome annotations were derived from the perature and a final wash-step in isopropanol for 1 min. The ReSeq database, http://www.ncbi.nlm.nih.gov/RefSeq/ and the slides were immediately blown dry with pressurized air and UCSC database, http://genome.ucsc.edu/cgi-bin/hgGateway. stored for hybridization. The hybridization mixtures were then pipetted onto the arrays and covered with Sigma Hybri-slips. The microarrays were placed in hybridization chambers (Corning Microarray Technologies, NY) and incubated on a RESULTS level surface for 16 h at 42 C for the COMT-arrays and Enrichment of the unmethylated fraction of gDNA 44–52 C for the CpG island microarrays in a covered water bath. The coverslips were removed by immersion of the arrays The strategy for enrichment of unmethylated portions of the in a wash solution containing 2· SSC and 0.5% SDS (washing genome is presented in Figure 1. gDNA is digested with buffer I). The array was washed twice for 15 min at 42–52 Cin methylation-sensitive restriction enzymes (Figure 1, middle washing buffer I (low stringency), followed by two wash-steps panel). Whereas methylated restriction sites remain in washing buffer II (0.5· SSC, 0.5% SDS), followed by 2 min unaltered, the sites containing unmethylated CpGs are of incubation in water. The slides were then rinsed quickly cleaved by the enzymes, and DNA fragments with 5 -CpG in isopropanol and finally dried with pressurized air. protruding ends are generated. The proportion of interrogated Nucleic Acids Research, 2006, Vol. 34, No. 2 531 Figure 1. Schematic outline of the microarray-based method for identification of DNA methylation differences and DNA polymorphisms in genomic DNA. Left panel: analysis of DNA sequence variation. Middle panel: the main strategy of the method is based on enrichment of unmethylated DNA fragments. DNA samples are cleaved by methylation-sensitive restriction endonucleases, and the resulting DNA fragments are then selectively enriched by adaptor-specific aminoallyl-PCR’s, labelled and hybridized to microarrays. Right panel: alternative procedure to enrich the hypermethylated DNA fraction. CpG sites depends on the methylation-sensitive restriction sequences, double- or triple-digest combinations of AciI, enzymes used for the restriction of DNA. Based on our HpaII, HpyCH4IV and Hin6I are recommended. analysis of the CpG dinucleotides within the sites of After the digestion of gDNA, the double-stranded adaptor methylation-sensitive restriction enzymes across several U-CG1 is ligated to the CpG-overhangs. At this point, it is megabases of human gDNA, the combination of three expected that most of the relatively short (<1.5 kb) and amp- enzymes, HpaII, Hin6I and AciI, should interrogate 32% lifiable DNA fragments derive from the unmethylated DNA of all CpG dinucleotides in mammalian DNA (Table 1). The regions. To some extent, the length of the amplified fragments addition of two other relatively inexpensive methylation- depends on the primer annealing temperature of the PCR sensitive CpG-overhang generating enzymes, HpyCH4IV reaction (Figure 2A). Some ligation fragments, however, and Hin1I, would theoretically increase the proportion of may still contain methylated cytosines. A proportion of interrogated CpGs to 41%. Depending on the such fragments can be eliminated by treatment with met microarray-type, in our experiments we usually use either McrBC, which cleaves DNA containing C and will not a single enzyme or a ‘cocktail’ of up to three restriction act upon unmethylated DNA. McrBC restriction sites consist met enzymes. The application of a set of enzymes might be of two half-sites of the form (G/A) C, which can be separ- disadvantageous for the analysis of GC-rich regions as ated by up to 3 kb (29,30). Hence, as can be seen in Figure 2B, met such a strategy would produce restriction fragments too a proportion of DNA fragments with two or more (G/A) C short for an efficient hybridization. In the latter case, it is within the restriction fragment are cleaved and therefore advisable to use a smaller number of restriction enzymes. deleted from the subsequent enrichment steps. The remaining Based on our experimental results and computer-based pool of unmethylated DNA fragments is then enriched by analysis of 100 randomly selected CpG islands, the most aminoallyl-PCR amplification that uses primers complement- suitable restriction enzymes are Hin6I and HpaII, followed ary to the adaptor U-CG1. One important advantage of using by AciI and Hin1I (Table 1). In contrast, for regular DNA protruding ends in the adaptor–ligation step is that degraded 532 Nucleic Acids Research, 2006, Vol. 34, No. 2 Table 1. Enzymes that generate protruding ends in the restriction fragments, profiles, especially in the case of hypermethylated CpG islands which are complementary to the adaptors U-CG1, TA-1 and AATT-1 or when the overall level of methylation in the genome is low (e.g. in insects). Thus, we developed an additional, modified Enzymes Recognition Percentage Number of Number of method to previously published methods of enrichment of sequence coverage fragments fragments methylated sequences to complement our data from the of CpGs in (per kb) in (per kb) in human CpG islands* non-CpG unmethylated fraction (Figure 1, right panel). This enrichment gDNA (%) islands* method relies on cleavage with the 4 bp frequent cutters TasI (AATT#) and/or Csp6I (G#TAC). Alternatively, BfaI or MseI HpaII (BsiSI) CCGG 8.6 3.98 1.18 can be used in combination with the Csp6I-specific adaptor. Hin6I (HinP1I) GCGC 6.4 3.98 0.61 AciI (SsiI) CCGC 17.4 3.23 1.79 All four enzymes produce DNA fragments in mammalian Hin1I GPuCGPyC 2.0 1.92 0.11 genomes of an average length 400–750 bp. The recognition (AcyI, BsaHI) sequences of TasI and Csp6I are infrequent within GC-rich HpyCH4IV ACGT 6.6 1.31 1.08 regions, leaving most CpG-islands intact. The analysis of 50 Bsu15I ATCGAT 0.2 <0.01 0.02 randomly selected CpG islands and several megabases of dif- (ClaI, BspDI) NarI (MlyI) GGCGCC 0.6 1.08 <0.01 ferent chromosomes revealed that Csp6I would produce more Bsp119I TTCGAA 0.1 0.11 <0.01 informative fragments in CpG islands than a digest with MseI, (BstBI, AsuII) whereas TasI and MseI produce informative fragments pref- Psp1406I AACGTT 0.3 <0.01 0.05 erentially in DNA regions outside of CpG islands (AclI, PspI) XmiI (AccI) GTMKAC 0.1 0.19 0.34 (Table 1). After ligation to the AATT- and TA-overhang spe- TasI AATT na 0.80 2.88 cific adaptors ‘AATT-1’ and ‘TA-1’, the un- and hypo- Csp6I GTAC na 2.23 1.41 methylated ligation products are eliminated from the reaction MseI TTAA na 0.80 2.88 by cleavage with a cocktail of methylation-sensitive restriction BfaI CTAG na 1.56 1.55 enzymes such as HpaII, HhaI (Hin6I), HpyCH4IV, Hin1I and Asterisk (*) indicates the number of 50 bp to 1.5 kb long (‘informative’) frag- AciI. Compared with a single digestion with BstUI (17), a ments, derived from several Mb of randomly selected CpG island and non-CpG cocktail of restriction enzymes will delete a higher percentage island sequences on chromosomes 1, 2, 4, 5, 6, 9, 17, 19 and 20; bold numbers of unmethylated sequences from the DNA fraction. The represent the most informative enzymes; na ¼ not applicable; M ¼ Adenine or remaining pool of mostly hypermethylated DNA fragments Cytosine; K ¼ Guanine or Thymine. is subsequently enriched by the aminoallyl-PCR amplification as described for the unmethylated fraction, and then hybrid- gDNA fragments (which are common in human post mortem ized to a microarray (Figure 2C). tissues) will not be ligated and amplified, and therefore will not interfere with DNA methylation analysis. Microarray design Most previous microarray-based epigenetic studies target hypermethylated DNA sequences (15,17,31,32); however, Various aspects of the microarray-based DNA modification interrogation of the unmethylated fraction is significantly profiling were investigated on the oligonucleotide microarray more informative. For example, the 100 kb region of chromo- that interrogates 100 kb fragment on 22q11.2 (Figure 3A). In some 22 interrogated by our COMT oligonucleotide addition to the catechol-O-methyltransferase (COMT, [MIM array (TXNRD2-COMT-ARVCF region; Microarray Design), 116790]), this chromosomal region contains also the gene contains 2193 methylatable cytosines. Enrichment of the encoding the thioredoxin reductase 3 gene (TXNRD2, [MIM unmethylated fraction can generate up to 401 amplicons of 606448]) and the armadillo repeat gene deleted in velocardi- sufficient size (50–1.5 kb), each representing the methylation ofacial syndrome (ARVCF, [MIM 602269]). For maximal status of at least one cytosine. In contrast, the combination of informativeness, it is necessary to design oligonucleotides MseI (+BsuI, to remove unmethylated fragments), the most according to the restriction sites of the methylation sensitive frequently used enzymes for enrichment of the hypermethyl- endonucleases used for the treatment of gDNA (Figure 3B). ated fraction (15,17,31,32), would produce 227 amplicons. For the COMT array, 384 oligonucleotides were designed, Seventy-seven amplicons would either contain no CpG dinuc- each 50 nucleotides long, representing every restriction frag- leotides or would be too short to stringently hybridize to a ment flanked by HpaII, Hin6I and AciI restriction sites. In microarray. Of the remaining 150 fragments, 144 contain addition, control DNA fragments containing l phage, multiple CpGs; hence, they are not fully informative since a pBR322, FX174, pUC57 and Arabidopsis sequences were single unmethylated BsuI restriction site would eliminate the spotted on the array (Materials and Methods). Additionally, entire fragment from the eventual amplification. Overall, only we used 12 192 element containing CpG island- and 6 of the 2193 methylatable cytosines are truly informative, high-density chromosome 21/22-microarrays (Materials and and none of these CpG dinucleotides are targeted by BsuI. Methods). Computer-based analysis of 50 randomly selected CpG island sequences revealed that the unmethylated fraction derived Detection of confounding effects of DNA sequence from HpaII cleavage results in 22 times more fragments variation (19.9 fragments/kb) of the suitable size range (50 bp to 1.5 kb) than the hypermethylated fraction (0.9 fragments/kb) Since restriction enzymes are used in the enrichment of dif- using MseI. ferentially modified DNA fractions, DNA sequence variation Nevertheless, analysis of the hypermethylated DNA frac- may simulate epigenetic differences. However, until now, tion may also add some new information to the methylation microarray methods used in epigenetic studies have not been Nucleic Acids Research, 2006, Vol. 34, No. 2 533 Figure 2. Selective enrichment of restriction fragments with the universal adaptor U-CG1. (A) Scatter plot that shows a comparison of ligation products treated with McrBC versus the untreated sample on the COMT array. McrBC treated fragments that contained at least two methylated cytosines were cleaved and could not be amplified in the following adaptor-PCR, resulting in reduced signal intensities in the Cy5 channel. (B) Co-hybridization of enriched unmethylated (Figure 1, middle panel) and hypermethylated (Figure 1, right panel) fragments derived from the same DNA source to a CpG island microarray. A large portion of amplicons is present only in one of the enriched fractions (marked black for log >0.3 black, green for log <0.3). Although the hypermethylated fraction hybridized to 75% of the microarray spots, based on our DNA sequence analysis, only a small fraction of them provide epigenetic information in comparison with the unmethylated fraction. differentiating between real DNA methylation differences patterns, HpaII, Hin6I, AciI and HpyCH4IV. The majority and single nucleotide polymorphisms (SNPs) within the of these CpG-SNPs were located in AciI and HpaII restriction restriction sites of the applied restriction enzymes. This prob- sites, with Hin6I and HpyCh4IV sites containing fewer met lem applies to some extent also to the C antibody-based polymorphisms (data not shown). Another approach to test strategy (22), which does not differentiate unmethylated CpG for DNA polymorphisms is the use of restriction endonuclease and TpG dinucleotides. In order to exclude the impact of DNA isoschizomers with different sensitivity to CpG methylation. sequence variation, two approaches are suggested. One is to However, this approach is currently only possible for HpaII/ check the available SNP databases in order to identify the MspI as there are no isoschizomers for most other methylation DNA sequence variation within the restriction sites of the sensitive restriction enzymes. enzymes used. For example, our 100 kb COMT array contains The third approach to differentiate the DNA sequence a total of 273 SNPs (SNPper, http://snpper.chip.org/bio/ effects from the genuine epigenetic differences consists of snpper-enter), of which 101 (37%) reside within CpG dinuc- performing an identical microarray experiment on the leotides and 55 (20%) are located within the restriction site of same DNA sample that has been stripped of all methylated the four main enzymes used to interrogate methylation cytosines. Our protocol utilizes the Phi29 DNA polymerase 534 Nucleic Acids Research, 2006, Vol. 34, No. 2 Figure 3. (A) Structure and GC-content of the chromosomal region on human chromosome 22q11.2 that spans the catechol-o-methyltransferase gene (COMT), the thioredoxin reductase 2 gene (TXNRD2) and the armadillo repeat gene deleted in VCFS (ARVCF). Vertical black bars represent exons. (B) To determine the methylation profile of the 100 kb TXNRD2-COMT-ARVCF region, 384 oligonucleotides (50mers, black horizontal bars) were designed based on the restriction sites for the methylation-sensitive endonucleases, HpaII, Hin6I and AciI (additional alternative enzymes are HpyCH4IV or Hin1I). Depending on the methylation status of the CpG-dinucleotides several combinations of amplicons (grey horizontal bars) can potentially hybridize to the oligonucleotides. (C) Typical hybridization patterns of the hypomethylated fraction of human gDNA on the COMT oligonucleotide-microarray. As discussed in Results, the complexity and informativeness of the hybridization signals increases with increasing number of methylation-sensitive restriction enzymes. to amplify whole genomic DNA, which creates a copy of the genome with all methylated cytosines replaced by unmethylated cytosines. Amplified DNA samples are then subjectedtothe same stepsasdepictedinFigure1 and hybridized on the microarrays. In this experiment all of the outliers must be a result of DNA sequence variations within the restriction sites of the enzymes used. These data can then be plotted against the DNA methylation data, whichare assayedinparallel(Figure 4).Insixexperiments that used amplified genomic DNA, the number of SNP- based outliers (threshold log-ratio <0.3, >0.3) ranged from 272 to 741 (432 ± 165, mean ± SD), or 2.2–6.1% of 12 192 CpG islands. Out of these, 72–234 (120 ± 66, mean ± SD) were initially identified as DNA methylation differences in microarray experiments using the unmethyl- ated fraction derived from the triple-digest with HpaII, AciI and Hin6I. From the CpG island array studies, our estimate is that 10–30% of the outliers detected in DNA Figure 4. Combined methylation- and SNP-analysis on a CpG island micro- methylation experiment could be due to DNA sequence array. The data of two separate hybridizations of DNA samples derived from variation. post mortem brain of two individuals are plotted against each other. The Y-axis contains the data derived from a methylation analysis (triple-cleavage with HpaII, Hin6I and AciI), whereas the X-axis contains the SNP data derived from Reproducibility the hybridization of the same DNA samples, which were subjected to the entire genome amplification prior to cleavage by the methylation-sensitive restriction To test the reproducibility of the method, a genomic DNA enzymes (Materials and Methods). Scale: log (Cy5/Cy3); an increased log- sample was split and subjected to the procedure of enrichment value on the Y-axis is indicated by red versus a decreased log-value represented by green. Significant outliers (log-ratio <0.3, >0.3, 2-fold difference) can be of the unmethylated fraction. The resulting amplification pro- classified into four clusters (S ¼ SNPs, M ¼ DNA methylation differences), ducts were labelled with Cy5 and Cy3 and then co-hybridized enabling the differentiation of epigenetic differences and nucleotide on the COMT array, which contains probes that flank the polymorphisms between the test-samples. Amp ¼ Whole-genome amplified HpaII, Hin6I and AciI restriction fragments around the sample. COMT gene. The Cy3 and Cy5 hybridization intensities exhib- ited very similar values (R ¼ 0.997; Figure 5A). Analogous Another critical factor in the amplification of unmethylated experiments, including switch dye hybridizations, were or hypermethylated DNA fragments is to ensure that no repeated several times also with the CpG island arrays and sequence specific bias is introduced. The rate of amplification in all cases were highly reproducible (R > 0.97). of repetitive sequences generally declines faster than that of Nucleic Acids Research, 2006, Vol. 34, No. 2 535 less abundant fragments in the later cycles of PCR (33). With increasing amplification cycles, repetitive DNA strands reach relatively high concentration and begin re-annealing to each other during the steps below the DNA melting temperature. To avoid this, a two-temperature PCR that uses a combined high- temperature elongation–annealing step was applied. A series of experiments were performed investigating how the number of PCR cycles would affect the hybridization patterns. As can be seen in Figure 5B, the relative intensities of the hybridiza- tion signals of both single copy sequences and repetitive DNA fragments, were similar in the range of 20–30 amplification cycles (R ¼ 0.991). Only when increasing the cycle numbers beyond 40 cycles was a biased amplification of some DNA sequences observed (data not shown). Sensitivity To test if differentially represented DNA fragments in two different DNA samples can be detected by this method, prior to methylation-sensitive cleavage, human gDNA was ‘spiked’ with unmethylated heterologous DNA, l phage and pBR322 plasmid (Figure 5C). Each sample was supple- mented with a different amount of spike-DNA, therefore mimicking differentially methylated sequences. The exact amount of l and pBR322 corresponded to increasing numbers of human genomic equivalents (1 GE of ‘spike’ DNA equals 16.28 pg l/mg gDNA and 1.45 pg/mg gDNA of pBR322, respectively). Hence, each of the experiments compared the intensities generated by 1 GE of l plus 128 GE of pBR322 (Y-axis) versus 16 GE of l plus 8 GE of pBR322 (X-axis). While the plotted signal intensities of the human gDNA sequences are positioned on or close to the regression line (indicating no methylation difference), the l and pBR322 fragments were identified as outliers. The average signal Figure 5. Reproducibility and sensitivity of the method. (A)A COMT microarray scatter plot representing two sets of amplification products derived from the same DNA source but produced at different time points by different researchers. The high-correlation coefficient of signal intensities demonstrates a high reproduci- bility of the method. (B) Influence of the PCR cycle number. Scatter plot diagrams show hybridization signal intensities of the unmethylated fraction that was amplified using 20 PCR cycles (Cy3 channel) and 30 cycles (Cy5 channel). Amplification products of each PCR were co-hybridized to the COMT microarray that contained oligonucleotides representing single copy sequences (closed circles), partially repetitive sequences (grey squares; 15–99 copies/genome) and highly repetitive DNA fragments (open squares; >100 copies/genome), such as ALU and LINE repeats. (C) Scatter plot representing the unmethylated fraction of human gDNA ‘spiked’ with different amounts of control DNA. The test samples were hybridized to the COMT array and contained either a 16-fold excess of l DNA (16 genome equivalents [GE] versus 1 GE; 10 fragments) or a 16-fold excess of pBR322 (128 GE versus 8 GE; 2 fragments), respectively. The ampli- cons of the spiked DNA (representing unmethylated DNA) can be easily distinguished as outliers; whereas the signals representing gDNA are located close to the regression line. Median signal intensities of different length oligonucleo- tides (40–50 bases) that target a specific HpaII restriction fragment in l DNA reveal that the length of spotted sequences directly influences the spot intensity and therefore the sensitivity of the microarray. (D) Sensitivity of the CpG-island microarray hybridization. Control amplicon (2 mg) (post mortem brain, unmethy- lated fraction) was labelled with Cy5 and co-hybridized with 2mg (0% difference), 1.9mg (5% difference), 1.8mg(10% difference), 1.5mg (25% difference) or 1.0mg (50% difference) of Cy3-labelled amplicon. For each hybridization to a COMT array, the regression lines represent the overall intensity that mimics methylation differences over the entire sample. The decrease of amount of DNA is reflected in the angle of the regression lines, which deviated by 5–7% from the expected values. 536 Nucleic Acids Research, 2006, Vol. 34, No. 2 intensity ratio of l oligonucleotides was 15.4, which is very silencing in mammalian genomes. The scatter plot shows two close to the ratio of spiked-DNA (16:1). The intensity values distinct spot areas, which represent predominantly unmethyl- for pBR322 were not as linear and exhibited a 6.5- to 10-fold ated fragments in placenta (yellow spots) and brain (orange difference (expected the same ratio of 1:16), most likely due to spots), respectively (Figure 7A). About 11% of the CpG island saturation effects of the hybridization. fragments exhibited 2-fold or more signal intensity difference In order to determine the sensitivity of the hybridization per between the two tissues. Some of the strongest brain-specific se, a control amplicon DNA was compared with itself but by signals could be identified for CpG islands associated decreasing the amounts of DNA by 5, 10, 25 and 50%. On the with neuronal genes such as DPYSL5, FABP7, DIRAS2, global level, the regression lines [y ¼ f(x)] reflected reprodu- GRIN3A, SLC24A3 and DSCAML1, whereas strong placenta- cible differences of the amount of amplicon DNA used in the specific outliers were associated with genes expressed in hybridization and varied by 5–7% from the expected values placenta, such as PCM1, CCND1, HA-1 and ADAMTSL1. (Figure 5D). Individual sites exhibited a lower accuracy, Overall, analysis revealed that brain DNA harboured notably which depended on the signal intensity, i.e. the stronger the more unmethylated CpG islands than placenta DNA. signal, the closer the observed spot intensity was to the expec- ted one. The rate of false outliers (log-ratio <0.3; >0.3; 2-fold Verification of detected methylation differences difference) was on average 3%. Usually, replication of Several loci that displayed methylation differences in our microarray experiments reduced the degree of aberration experiments were selected for verification by the sodium bisul- (log-ratio <0.3; >0.3) below 2% for all types of microarrays. fite modification mapping of methylated cytosines (Materials and Methods). The technique is based on the reaction of gDNA Examples of DNA methylation profiles with sodium bisulfite under conditions such that cytosine is Identification of DNA modification differences is provided in a deaminated to uracil but 5-methylcytosine remains unaltered. series of examples below. The COMT oligonucleotide array In the sequencing of amplified products, all uracil and thymine met was used to identify DNA methylation changes in a brain residues are detected as thymine and only C residues remain tumour (Figure 6A). In contrast to the pair of control brain as cytosine. The sites for the methylation-sensitive restriction DNA samples, where hybridization signals are close to the enzymes used in our experiments showed the expected regression line (indicating similar DNA methylation patterns), methylation difference across the DNA samples, as exempli- a visible proportion of the hybridization signals originating fied for CpG island clones located in the promoter region from the unmethylated DNA fraction of the brain tumour of galectin-1 and in the promoter region of a brain-specific deviates from the regression line. More subtle changes in transcript CR606704 (Figure 7B and C). DNA methylation patterns have been identified when post mortem brain tissues of healthy individuals were compared Chromosome-wide mapping of DNA methylation with the same tissues from schizophrenia patients (A. Schu- differences macher, A. Petronis, manuscript in preparation; representative example is shown in Figure 6B). The differences of the cancer Analysis of the unmethylated fraction from brain specific and psychosis studies show that diseases other than cancer DNA of eight adults using a chromosome 21/22 tiling array may reveal more subtle epigenetic differences, and therefore, detected 488–747 unmethylated sites per sample (Table 2). the informativeness and sensitivity of the epigenetic profiling This number increased to 977 in a merged map, showing that method is of critical importance. many sites were common between different individuals. The Another application of the technology includes epigenetic vast majority of the sites (90%) were positioned outside of 0 0 profiling of different tissues. One example of tissue specific the 5 ends and 5 flanking regions of the genes consistent with effects is shown using the CpG island microarrays that contain abundant transcriptional activity and a significant fraction of 12 192 CpG island clones of whom 8025 represent unique transcription factor binding sites found outside of known sequences. CpG islands tend to be found in many promoter annotations (26,27,34). The unmethylated sites outside of sequences and their methylation has profound effects on gene the 5 ends of known genes were about equally distributed Figure 6. Applications of the epigenetic profiling technology. (A) Changes of methylation profiles at TXNRD2-COMT-ARVCF in a brain tumour. The data from two different microarrays experiments are superimposed over each other. The analysis of two post mortem brain samples (closed dots) reveals no major difference in methylation levels, whereas the signal intensities vary significantly in the brain tumour (grey dots) when compared with the normal brain. (B) The comparison of DNA methylation profiles using the COMT microarray in brain tissue of a healthy control and a schizophrenia patient displays subtle epigenetic differences. Nucleic Acids Research, 2006, Vol. 34, No. 2 537 Figure 7. Examples of applications using a CpG island microarray. (A) Hybridization of the unmethylated fraction of placenta DNA and post mortem brain DNA to a CpG island array. Two pools of CpG island elements could be identified, which display extensively different methylation levels between these tissues(Note: some of the identified differences could be due to DNA sequence variation). (B) To validate the identified methylation differences, several CpG islands were subjected to bisulfite modification based mapping of methylated cytosines as exemplified for CpG island clones 22_B_12 (promoter region of Galectin-1) and 52_C_03 (promoter region of a brain-specific transcript, CR606704). The top sequence shows the reverse strand () of the original restriction sites, the bottom sequence displays the bisulfite-modified DNA. For each bisulfite-modified CpG-island, 8–10 clones were sequenced per tissue. Sequence 52_C_03 revealed several fully methylated CpG’s in placenta, which were unmethylated in brain. In contrast, clone 22_B_12 showed subtler methylation differences (15–100%), depending on the position of CpG-dinucleotide. (C) Methylation patterns of clones 22B_12 and 52_C_03 derived from bisulfite sequencing of 10–12 clones per tissue. The yellow boxes indicate CpG dinucleotides that are shown in the sequenced graph (Figure 7B). Table 2. Interindividual differences and distribution of the detected unmethylated sites with respect to the known genes as defined by the combined set of RefSeq and UCSC known genes for each brain DNA sample (M17–M25) and the merged map 0 0 0 0 0 0 Individual 3 -flanking 3 ter 5 -flanking 5 flanking–3 flanking 5 ter Distal Internal Total Site coverage (bp) #M17 chr21/22 13/12 2/16 8/20 2/4 10/20 64/122 98/97 488 64943/134730 %Total 5.1 3.7 5.7 1.2 6.1 38.1 40.0 #M18 chr21/22 17/22 9/15 13/29 3/3 16/28 95/191 134/152 727 98456/236797 %Total 5.4 3.3 5.8 0.8 6.1 39.3 39.3 #M19 chr21/22 15/24 11/14 12/27 2/5 14/21 86/173 119/130 653 88290/221721 %Total 6.0 3.8 6.0 1.1 5.4 39.7 38.1 #M21 chr21/22 20/24 12/18 15/29 2/5 14/22 102/184 143/157 747 109595/252347 %Total 5.9 4.0 5.9 0.9 4.8 38.3 40.2 #M22 chr21/22 18/20 8/17 9/29 3/6 15/24 86/169 127/143 674 87604/213453 %Total 5.6 3.7 5.6 1.3 5.8 37.8 40.1 #M23 chr21/22 12/15 4/13 10/25 2/3 10/21 68/150 101/111 545 70912/163322 %Total 5.0 3.1 6.4 0.9 5.7 40.0 38.9 #M24 chr21/22 14/18 5/12 7/20 4/3 10/20 61/158 88/107 527 65639/187229 %Total 6.1 3.2 5.1 1.3 5.7 41.6 37.0 #M25 chr21/22 17/15 7/13 10/18 3/3 9/22 65/171 102/97 552 69937/171073 %Total 5.8 3.6 5.1 1.1 5.6 42.8 36.1 Merged chr21/22 26/28 13/22 19/36 4/9 19/34 142/237 187/201 977 152148/314374 %Total 5.5 3.6 5.6 1.3 5.4 38.8 39.7 0 0 0 0 0 0 ‘5 ter’ or ‘3 ter’ refers to a 5 or 3 terminal site internal and within 1 kb of a gene boundary ‘5 flanking’ or ‘3 flanking’ refers to a site outside and within 5 kb of a gene 0 0 boundary; ‘internal’ refers to an intronic site and ‘distal’ refers to an intergenic site outside of the 5 kb/+1 kb boundaries. A site can also be both 5 and 3 flanking in 0 0 a gene rich region and referred as ‘5 flanking–3 flanking’. between sites residing within introns of known genes and unmethylated sites detected in this study cover 0.47 Mb outside of the gene boundaries. Interestingly, while some or 4% of the 12 Mb of non-repetitive sequences of chromo- genes, like BCR, showed a large number of sites inside somes 21 and 22 interrogated in the combined map of all eight the gene boundaries, some loci, like C21ORF55 spanning individuals with an average of 0.28 Mb (2.3%) in any given 150 kb, were essentially devoid of internal unmethylated individual. Maps of the methylation patterns (average value sites and in some cases, such as the SIM2 locus, the unmethyl- of the eight tested individuals) of the q-arms of chromosome ated sites were limited to the first intron (Figure 8A–C). Such 21 and 22 are shown in Figure 9A–B. Detailed maps of all intragenic methylation may inhibit inappropriate transcrip- individuals for chromosome 21 and 22, linked to the UCSC tional initiation at cryptic sites (35) or may serve as regulators Genome Browser (http://genome.ucsc.edu) are also available of alternate transcripts as can be seen for SIM2. Overall, on our web-based methylation database (Web Recourses). 538 Nucleic Acids Research, 2006, Vol. 34, No. 2 Figure 8. Profiles of unmethylated sites in three loci on human chromosomes 21 and 22 (501 bp window, Materials and Methods): BCR (A), C21ORF55 (B) and SIM2 (C) for human brain DNA (average of eight individuals, M17-M25). The graphs are based on P-values for each individual interrogation that show the significance of the enrichment in the unmethylated fraction versus total gDNA. The P-values were converted to the (10 log ) scale, such that, for example, P-value of 10 4 12 4 becomes 40. The vertical axes are adjusted to represent probes in the 40–120 range (P-values of 10 –10 ), thus only probes that pass P < 10 threshold are shown. Enlarged is a part of the chr 22q11.21 region (181 bp window), spanning breakpoints found in the generation of the two alternative forms of the Philadelphia chromosome translocation. C ¼ gDNA control. A comparison of the hypomethylation tracks with data from fragments representing DNA methylation profiles over large the Affymetrix transcriptome project (26,36) indicates that segments of gDNA. Building on the principles described in many of the unmethylated chromosomal regions overlap earlier publications (11–23) our method addresses a series of with mapped transcriptional active regions (Figure 9A–C, bot- critical issues and exhibits several advantages. An earlier tom tracks). These DNA methylation data complement exist- method (18) used a sucrose gradient to enrich the unmethyl- ing studies on transcriptional activity and histone ated DNA fraction. This method, however, requires a large modifications on human chromosomes 21 and 22 (37). We amount of DNA template and is rather imprecise in terms found that in the majority of cases, specific histone modifica- of the upper limit of the fragments that are subjected to tion patterns reported by Bernstein et al. (37) for the human hybridization. Other microarray methods for DNA methyla- tion analysis can be categorized into three main classes which hepatoma cell line HepG2 overlapped notably with the observed DNA methylation patterns. An example is shown are based on: (i) identification of bisulfite induced C!T trans- in Figure 9C for the PEX26 gene that is ubiquitously tran- itions (11–13,38,39), (ii) cleavage of gDNA by methylation- scribed in most tissues. The gene harbours an extensively sensitive restriction enzymes and (iii) immunocapturing with unmethylated CpG rich region in its promoter. The compar- antibodies against methylated cytosines. In the bisulfite arrays, ison of the different epigenetic profiles of both studies shows each tested CpG is represented by a pair of either C(G) or T(A) that the same genomic region was also highly acetylated at nucleotides. The arrays contain oligonucleotides that measure Lysine 9 and 14 of histone 3 (H3), accompanied with H3 the C(G)/T(A) ratio in the bisulfite treated DNA (correspond- met di- and trimethylation of Lysine 4. A comparison of histone ing to C/C in the native DNA). Although informative and modification tracks and our hypomethylation patterns for the precise, these microarrays can contain only a limited number q-arms of chromosome 21 and 22 revealed that H3 acetylation of oligonucleotides because treatment with bisulfite degener- and Lys4 methylation usually correlated with unmethylated ates the 4 nt code, resulting in a loss of specificity for a large CpGs. portion of the genome. For example, after bisulfite treatment all of the possible 16 permutations of a four base sequence containing unmethylated C and T (CCCC, CTCT, CCCT, CCTT, TCTC, TTTC, TTTT and so on) will become identical DISCUSSION TTTT. The bisulfite method is also laborious and cannot be Microarray based technology for DNA modification analysis easily applied to profile a large set of samples. Furthermore, it enables the highly parallel screening of numerous restriction is difficult to design suitable oligonucleotides that would Nucleic Acids Research, 2006, Vol. 34, No. 2 539 Figure 9. Genomic views showing unmethylated regions on chromosomes 21 and 22. (A and B): The top tracks (dark red) in the two chromosomal graphs shows the average amount of hypomethylation in the brain cortex of eight adult individuals. Also displayed are known genes (dark blue) and CpG islands (green). The bottom tracks display transcriptome data derived from 11 different tissues from the Affymetrix transcriptome phase 2 study (36). The track is coloured blue in areas that are thought to be transcribed at a statistically significant level. Regions that have a significant homology to other chromosomal regions or that overlap putative pseudogenes are coloured in lighter shades of blue. All other regions of the track are colored brown. (C) Enlarged is a part of chromosome 22q11.21, containing the peroxisome biogenesis factor 26 (PEX26, MIM 608666) that shows correlation between histone modifications and unmethylated DNA in its promoter region. The top three tracks represent histone modification data for H3 Lys4 dimethylation (orange bar), H3 Lys4 trimethylation (blue bar) and H3 Lys9/14 acetylation (yellow bar) (37). Underneath are the tracks for the average methylation patterns (unmethylated sites) observed in brain and the individual methylation patterns of all tested individuals (dark red). It is noteworthy that methylation patterns exhibit some interindividual differences (indicated by arrows). exhibit similar melting temperatures since the specificity of Methods relying on the enrichment and detection of hyper- base discrimination varies considerably (12). Using our methylated DNA have predominantly been used to identify approach, arrays can contain an almost unlimited number of abnormally methylated CpG islands in malignant cells (15– oligonucleotides: coverage can range from individual genes to 17,31). Although this strategy seems to be useful for detecting entire chromosomes represented by millions of oligonuc- major epigenetic changes in some regions of the genome, the leotides on glass chips. Whole genome tiling arrays are already overall proportion of interrogated CpG sites is substantially available for Arabidopsis thaliana and Escherichia coli, and lower compared with that achieved using approaches based will soon be available for the entire human genome. on the analysis of the unmethylated fraction. As shown in Restriction enzyme based methods are used to enrich either Results, we have estimated that interrogation of the unmethyl- the hypermethylated or unmethylated fraction of gDNA. ated fraction of gDNA could be up to several hundred 540 Nucleic Acids Research, 2006, Vol. 34, No. 2 folds more efficient than analysing the hypermethylated frac- the methylation patterns generated from a substantially larger tion. Furthermore, since unmethylated cytosines are less number of cells from the same tissue. abundant in the genome than methylated cytosines (depending There are, however, also some of limitations to the techno- on the tissue, 70–90% of cytosines are methylated), analysis of logy described in this article. The methylation sensitive the smaller unmethylated fraction of gDNA is more sensitive restriction enzymes do not interrogate every cytosine, and to detect subtle changes. For example, an increase of 10% with our current design, more than half of CpG sites remain met from the normal density of C would result in a 100% uninterrogated. This may be critical when the phenotypic out- (from 20 to 10%) difference in the unmethylated fraction, comes are determined by a methylation change at an isolated but only a 12% (from 80 to 90%) difference in the hyper- cytosine that is not within the restriction site of a methylation methylated fraction of gDNA. The unmethylated fraction has sensitive restriction enzyme. This problem may be partially been used in some approaches employing class II microarray overcome by the application of the same arrays to the CpG methods, for instance by using the methylation-specific specific immunoprecipitation technique (MeDIP) (22) in addi- McrBC enzyme (23) to deplete the hypermethylated fraction. tion to histone modification analysis through ChIP technology, However, the remaining unmethylated DNA fragments which identifies DNA sequences associated with modified (>1 kb) have to be gel-purified, requiring large amounts of histones (10). DNA and histone modifications seem to be starting material. Additionally, the McrBC method may not be inter-dependent, and consequently the possibility of a com- able to differentiate between dense and sparse methylation bined approach that interrogates both DNA methylation and within relatively short DNA fragments. For example, the chromatin modification in parallel might be a productive 2 kb human COMT promoter region, which contains 27 approach to the fine mapping of epigenetic changes. Also, McrBC target sites, can be cut to shorter than 1 kb fragments asymmetrical C sites (CpNpN) that are found in plants in cases where there are 2 (7%) or 27 (100%) methylated and some fungi such as Neurospora crassa are difficult to McrBC sites. Furthermore, the McrBC method cannot differ- detect, although some methylation-sensitive type IIs restric- entiate between unmethylated and polymorphic cytosines. tion enzymes are available (e.g. Esp3I or BveI). However, Another method to enrich the unmethylated fraction uses methylation of asymmetrical sites in animal organisms is 0 0 the rare cutter NotI (5 -GCGGCCGC-3 ) (19–21). However, not common. Additionally, this array method can also be NotI sites are not well represented in the genome and will only modified for analysis of methylated adenines in plants and provide a very superficial overview of genomic methylation bacteria. patterns. An alternative to these methods is the use of anti- In summary, the use of microarrays targeted at unmethyl- bodies specific for methylated cytosines [MeDIP (22)]. In this ated cytosines is a high-throughput approach to profile DNA method, antibodies are used to immunocapture methylated methylation patterns across the genome. The ability to analyse genomic fragments. However, this approach requires large minute amounts of DNA may enable the epigenetic screening amounts of gDNA (>8 mg) and also relies on the enrichment of DNA in plasma, serum or other body fluids, as well as in of the less informative hypermethylated fraction of the prenatal diagnostics. Although all the examples provided in genome. this work investigated human DNA, the same strategies can be In our analyses, we have addressed another important used for the epigenetic analyses of numerous other species. It issue: the interference of DNA polymorphisms that may is evident that epigenetic profiling should be performed in a simulate DNA modification differences across individuals. systematic, unbiased fashion and not limited to the tradition- Data from the SNP consortium indicate that roughly every ally preferable regions such as CpG islands. Outside of CpG 360th nucleotide in the human genome represents an SNP. In islands, numerous other genomic loci exist that may be sites humans, 2.16 million SNPs are detected in CpG dinuc- for important epigenetic modification, including enhancers, leotides, and such CpG SNPs are 6.7-fold more abundant imprinting control elements (41) or the regions that encode than expected (40). Depending on the restriction enzyme regulatory RNA elements. combination, our CpG island array-based studies demon- The above described technology, in combination with strated that 10–30% of all outliers initially detected as existing epigenetic profiling methods, may help to identify methylation differences contained SNPs (Figure 4). Informa- inter-individual variation in genome-wide methylation pat- tion on the SNPs and other polymorphisms such as deletions, terns as well as epigenetic changes that arise during tissue inversions or duplications within the restriction sites of the differentiation and the understanding of the epigenetic effects enzymes used for the enrichment of the unmethylated or of various environmental factors. Of particular interest is the hypermethylated fractions is helpful in differentiating the application of high-throughput DNA methylation analyses to epigenetic variations from the DNA sequence ones. To min- address the molecular basis of various non-Mendelian irregu- imize the effects of DNA polymorphisms, it may be also larities of complex diseases, such as discordance of mono- beneficial to compare affected tissue and healthy cells zygotic twins, remissions and relapses of a disease, parent of from the same individual. origin- and sex-effects, and tissue- and site-specificity (42). Another advantage of PCR-based methylation profiling Further technological developments may include building methods is the ability to work with limited DNA resources. high-resolution oligonucleotide-based microarrays spanning Although our basic protocol requires about 500 ng of gDNA, the entire human genome, improving the enrichment strate- the amount of template DNA can be significantly lower. In gies through the application of more specialized methylation our recent experiments, methylation patterns at the COMT sensitive restriction enzymes, and substantial reduction in region generated from a relatively small number of Jurkat the amount of initial template DNA down to the amount tissue culture cells (up to 500 cells, or 3 ng of DNA) did of a haploid or diploid genome. All these developments not reveal any significant differences when compared with will provide the basis for identifying the methylation profile Nucleic Acids Research, 2006, Vol. 34, No. 2 541 of the entire genome in a single cell, one of the ‘quantum 17. Huang,T.H., Perry,M.R. and Laux,D.E. (1999) Methylation profiling of CpG islands in human breast cancer cells. Hum. Mol. Genet., 8, leaps’ in post-genomic biology (43). 459–470. 18. Tompa,R., McCallum,C.M., Delrow,J., Henikoff,J.G., van Steensel,B. and Henikoff,S. (2002) Genome-wide profiling of DNA methylation reveals transposon targets of CHROMOMETHYLASE3. Curr. Biol., SUPPLEMENTARY DATA 12, 65–68. 19. Li,J., Protopopov,A., Wang,F., Senchenko,V., Petushkov,V., Supplementary Data are available at NAR Online. Vorontsova,O., Petrenko,L., Zabarovska,V., Muravenko,O., Braga,E. et al. (2002) NotI subtraction and NotI-specific microarrays to detect copy number and methylation changes in whole genomes. Proc. Natl Acad. Sci. USA, 99, 10724–10729. ACKNOWLEDGEMENTS 20. Yamamoto,F. and Yamamoto,M. (2004) A DNA microarray-based methylation-sensitive (MS)-AFLP hybridization method for genetic and This research has been supported by the Special Initiative grant epigenetic analyses. Mol. Genet. Genomics, 271, 678–686. from the Ontario Mental Health Foundation, and also by 21. Ching,T.T., Maunakea,A.K., Jun,P., Hong,C., Zardo,G., Pinkel,D., NARSAD, CIHR, NIH, the Stanley Foundation, and the Albertson,D.G., Fridlyand,J., Mao,J.H., Shchors,K. et al. (2005) Crohn’s and Colitis Foundation of Canada. A.S. holds a Epigenome analyses using BAC microarrays identify evolutionary CIHR Michael Smith Fellowship. Funding to pay the Open conservation of tissue-specific methylation of SHANK3. Nature Genet., 37, 645–651. Access publication charges for this article was provided by 22. Weber,M., Davies,J.J., Wittig,D., Oakeley,E.J., Haase,M., Lam,W.L. and the Ontario Mental Health Foundation. Schubeler,D. (2005) Chromosome-wide and promoter-specific analyses identify sites of differential DNA methylation in normal and transformed Conflict of interest statement. None declared. human cells. Nature Genet., 37, 853–862. 23. Lippman,Z., Gendrel,A.-V., Colot,V. and Martienssen,R. (2005) Profiling DNA methylation patterns using genomic tiling microarrays. Nature Methods, 2, 219–224. REFERENCES 24. Cross,S.H., Charlton,J.A., Nan,X. and Bird,A.P. (1994) Purification of 1. Bird,A.P. (1986) CpG-rich islands and the function of DNA methylation. CpG islands using a methylated DNA binding column. Nature Genet., Nature, 321, 209–213. 6, 236–244. 2. Henikoff,S. and Matzke,M.A. (1997) Exploring and explaining 25. Heisler,L.E., Torti,D., Boutros,P.C., Watson,J., Chan,C., Winegarden,N., epigenetic effects. Trends Genet., 13, 293–295. Takahashi,M., Yau,P., Huang,T.H., Farnham,P.J. et al. (2005) CpG Island microarray probe sequences derived from a physical library are 3. Wolffe,A.P. and Matzke,M.A. (1999) Epigenetics: regulation through representative of CpG Islands annotated on the human genome. repression. Science, 286, 481–486. Nucleic Acids Res., 33, 2952–2961. 4. Reik,W., Dean,W. and Walter,J. (2001) Epigenetic reprogramming in 26. Kapranov,P., Cawley,S.E., Drenkow,J., Bekiranov,S., Strausberg,R.L., mammalian development. Science, 293, 1089–1093. Fodor,S.P. and Gingeras,T.R. (2002) Large-scale transcriptional 5. Grewal,S.I. and Moazed,D. (2003) Heterochromatin and epigenetic activity in chromosomes 21 and 22. Science, 296, 916–919. control of gene expression. Science, 301, 798–802. 27. Cawley,S., Bekiranov,S., Ng,H.H., Kapranov,P., Sekinger,E.A., 6. Suter,C.M., Martin,D.I. and Ward,R.L. (2004) Germline epimutation of Kampa,D., Piccolboni,A., Sementchenko,V., Cheng,J., Williams,A.J. MLH1 in individuals with multiple cancers. Nature Genet., 36, 497–501. et al. (2004) Unbiased mapping of transcription factor binding sites along 7. Walter,J. and Paulsen,M. (2003) Imprinting and disease. Semin. Cell Dev. human chromosomes 21 and 22 points to widespread regulation of Biol., 14, 101–110. noncoding RNAs. Cell, 116, 499–509. 8. Laird,P.W. (2003) The power and the promise of DNA methylation 28. Hajkova,P., el-Maarri,O., Engemann,S., Oswald,J., Olek,A. and Walter,J. markers. Nature Rev. Cancer, 3, 253–266. (2002) DNA-methylation analysis by the bisulfite-assisted genomic 9. Frommer,M., McDonald,L.E., Millar,D.S., Collis,C.M., Watt,F., sequencing method. Methods Mol. Biol., 200, 143–154. Grigg,G.W., Molloy,P.L. and Paul,C.L. (1992) A genomic sequencing 29. Sutherland,E., Coe,L. and Raleigh,E.A. (1992) McrBC: a multisubunit protocol that yields a positive display of 5-methylcytosine residues in GTP-dependent restriction endonuclease. J. Mol. Biol., 225, 327–348. individual DNA strands. Proc. Natl Acad. Sci. USA, 89, 1827–1831. 10. van Steensel,B. and Henikoff,S. (2003) Epigenomic profiling using 30. Kruger,T., Wild,C. and Noyer-Weidner,M. (1995) McrB: a prokaryotic microarrays. Biotechniques, 35, 346–350, 352–344, 356–347. protein specifically recognizing DNA containing modified cytosine 11. Adorjan,P., Distler,J., Lipscher,E., Model,F., Muller,J., Pelet,C., residues. EMBO J., 14, 2661–2669. Braun,A., Florl,A.R., Gutig,D., Grabs,G. et al. (2002) Tumour class 31. Yan,P.S., Chen,C.M., Shi,H., Rahmatpanah,F., Wei,S.H. and Huang,T.H. prediction and discovery by microarray-based DNA methylation analysis. (2002) Applications of CpG island microarrays for high-throughput Nucleic Acids Res., 30, e21. analysis of DNA methylation. J. Nutr., 132, 2430S–2434S. 12. Balog,R.P., de Souza,Y.E., Tang,H.M., DeMasellis,G.M., Gao,B., 32. Chen,C.M., Chen,H.L., Hsiau,T.H., Hsiau,A.H., Shi,H., Brock,G.J., Avila,A., Gaban,D.J., Mittelman,D., Minna,J.D., Luebke,K.J. et al. Wei,S.H., Caldwell,C.W., Yan,P.S. and Huang,T.H. (2003) Methylation (2002) Parallel assessment of CpG methylation by two-color target array for rapid analysis of CpG island hypermethylation in hybridization with oligonucleotide arrays. Anal Biochem., 309, multiple tissue genomes. Am. J. Pathol., 163, 37–45. 301–310. 33. Mathieu-Daude,F., Welsh,J., Vogt,T. and McClelland,M. (1996) DNA 13. Gitan,R.S., Shi,H., Chen,C.M., Yan,P.S. and Huang,T.H. (2002) rehybridization during PCR: the ‘Cot effect’ and its consequences. Methylation-specific oligonucleotide microarray: a new potential for Nucleic Acids Res., 24, 2080–2086. 34. Kampa,D., Cheng,J., Kapranov,P., Yamanaka,M., Brubaker,S., high-throughput methylation analysis. Genome Res., 12, 158–164. Cawley,S., Drenkow,J., Piccolboni,A., Bekiranov,S., Helt,G. et al. (2004) 14. Hatada,I., Kato,A., Morita,S., Obata,Y., Nagaoka,K., Sakurada,A., Sato,M., Novel RNAs identified from an in-depth analysis of the transcriptome of Horii,A., Tsujimoto,A. and Matsubara,K. (2002) A microarray-based human chromosomes 21 and 22. Genome Res., 14, 331–342. method for detecting methylated loci. J. Hum. Genet., 47, 448–451. 35. Bird,A.P. (1995) Gene number, noise reduction and biological 15. Shi,H., Wei,S.H., Leu,Y.W., Rahmatpanah,F., Liu,J.C., Yan,P.S., complexity. Trends Genet., 11, 94–100. Nephew,K.P. and Huang,T.H. (2003) Triple analysis of the cancer 36. Cheng,J., Kapranov,P., Drenkow,J., Dike,S., Brubaker,S., Patel,S., Long,J., epigenome: an integrated microarray system for assessing gene Stern,D., Tammana,H., Helt,G. et al. (2005) Transcriptional maps of 10 expression, DNA methylation, and histone acetylation. Cancer human chromosomes at 5-nucleotide resolution. Science, 308, 1149–1154. Res., 63, 2164–2171. 37. Bernstein,B.E., Kamal,M., Lindblad-Toh,K., Bekiranov,S., Bailey,D.K., 16. Yan,P.S., Efferth,T., Chen,H.L., Lin,J., Rodel,F., Fuzesi,L. and Huebert,D.J., McMahon,S., Karlsson,E.K., Kulbokas,E.J.,III, Huang,T.H. (2002) Use of CpG island microarrays to identify colorectal Gingeras,T.R. et al. (2005) Genomic maps and comparative analysis of tumors with a high degree of concurrent methylation. Methods, 27, histone modifications in human and mouse. Cell, 120, 169–181. 162–169. 542 Nucleic Acids Research, 2006, Vol. 34, No. 2 38. Shi,H., Maier,S., Nimmrich,I., Yan,P.S., Caldwell,C.W., Olek,A. and polymorphic sites and suppression of variation in CpG islands. J. Mol. Huang,T.H. (2003) Oligonucleotide-based microarray for DNA Biol., 327, 303–308. methylation analysis: principles and applications. J. Cell Biochem., 41. Schumacher,A., Buiting,K., Zeschnigk,M., Doerfler,W. and 88, 138–143. Horsthemke,B. (1998) Methylation analysis of the PWS/AS region does 39. Hou,P., Ji,M., Liu,Z., Shen,J., Cheng,L., He,N. and Lu,Z. (2003) A not support an enhancer-competition model. Nature Genet., 19, 324–325. microarray to analyze methylation patterns of p16(Ink4a) gene 5 -CpG 42. Petronis,A. (2001) Human morbid genetics revisited: relevance of islands. Clin. Biochem., 36, 197–202. epigenetics. Trends Genet., 17, 142–146. 40. Tomso,D.J. and Bell,D.A. (2003) Sequence context at human single 43. Collins,F.S., Green,E.D., Guttmacher,A.E. and Guyer,M.S. (2003) A nucleotide polymorphisms: overrepresentation of CpG dinucleotide at vision for the future of genomics research. Nature, 422, 835–847. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Nucleic Acids Research Oxford University Press

Microarray-based DNA methylation profiling: technology and applications

Download PDF
15 pages

Loading next page...
 
/lp/oxford-university-press/microarray-based-dna-methylation-profiling-technology-and-applications-y5PSQu08ne

References (43)

Publisher
Oxford University Press
Copyright
© The Author 2006. Published by Oxford University Press. All rights reserved
 The online version of this article has been published under an open access model. Users are entitled to use, reproduce, disseminate, or display the open access version of this article for non-commercial purposes provided that: the original authorship is properly and fully attributed; the Journal and Oxford University Press are attributed as the original place of publication with the correct citation details given; if an article is subsequently reproduced or disseminated not in its entirety but only in part or as a derivative work this must be clearly indicated. For commercial re-use, please contact [email protected]
ISSN
0305-1048
eISSN
1362-4962
DOI
10.1093/nar/gkj461
pmid
16428248
Publisher site
See Article on Publisher Site

Abstract

Published online January 20, 2006 528–542 Nucleic Acids Research, 2006, Vol. 34, No. 2 doi:10.1093/nar/gkj461 Microarray-based DNA methylation profiling: technology and applications Axel Schumacher, Philipp Kapranov , Zachary Kaminsky, James Flanagan, 2 2 2 1 Abbas Assadzadeh, Patrick Yau , Carl Virtanen , Neil Winegarden , Jill Cheng , Thomas Gingeras and Arturas Petronis* The Krembil Family Epigenetics Laboratory, Centre for Addiction and Mental Health, 250 College Street, Toronto, ON, 1 2 Canada M5T 1R8, Affymetrix, Santa Clara, USA and The Microarray Centre, The University Health Network, 200 Elizabeth Street, Toronto, ON, Canada M5G 2C4 Received November 21, 2005; Revised December 20, 2005; Accepted January 5, 2006 INTRODUCTION ABSTRACT Over the last decade the field of DNA methylation has grown This work is dedicated to the development of a dramatically and become one of the most dynamic and rapidly technology for unbiased, high-throughput DNA developing branches of molecular biology. The methyl group methylation profiling of large genomic regions. In at the fifth position of the cytosine pyrimidine ring, that is this method, unmethylated and methylated DNA present in about 80% of CpG-dinucleotides in the human fractions are enriched using a series of treatments genome, can be of major functional significance and is with methylation sensitive restriction enzymes, and regarded as the ‘fifth base’ of the genome (1). DNA methyla- interrogated on microarrays. We have investigated tion, along with histone modifications (acetylation, methyla- various aspects of the technology including its rep- tion, phosphorylation and the like), are referred to as licability, informativeness, sensitivity and optimal epigenetic phenomena that control various genomic functions PCR conditions using microarrays containing oligo- without a change in nucleotide sequence (2). Such functions nucleotides representing 100 kb of genomic DNA include meiotic and mitotic recombination, replication, control of ‘parasitic’ DNA elements, establishing and maintenance of derived from the chromosome 22 COMT region in gene expression profiles, X chromosome inactivation as well as addition to 12 192 element CpG island microarrays. regulation of developmental programming and cell differenti- Several new aspects of methylation profiling are ation (3–6). Aberrations in epigenetic regulation, or ‘epimuta- provided, including the parallel identification of tions’, cause several paediatric syndromes (Prader–Willi confounding effects of DNA sequence variation, [OMIM #176270], Angelman [OMIM #105830], Beckwith– the description of the principles of microarray Wiedemann [OMIM #130650] and Rett [OMIM #312750]) design for epigenomic studies and the optimal (7) and may also predispose to cancer (8). choice of methylation sensitive restriction enzymes. Our understanding of the peculiarities of DNA methylation We also demonstrate the advantages of using the in the human genome is still very superficial. Based on the unmethylated DNA fraction versus the methylated review of available publications, our estimate is that <0.1% of the genome has been subjected to a detailed DNA modification one, which substantially improve the chances of analysis. The recently completed Human Genome sequencing detecting DNA methylation differences. We applied project did not attempt to differentiate between methylated and this methodology for fine-mapping of methylation unmethylated cytosines. To some extent our understanding of patterns of chromosomes 21 and 22 in eight indi- the dynamic state of genome-wide DNA methylation has been viduals using tiling microarrays consisting of over hampered by the lack of high-throughput technologies that 340 000 oligonucleotide probe pairs. The principles would interrogate DNA methylation profiles over large developed in this work will help to make epigenetic genomic regions. A gold standard technique in DNA methyla- profiling of the entire human genome a routine tion studies, the bisulfite modification-based fine mapping of met procedure. C (9), although precise, is very labour intensive and in *To whom correspondence should be addressed. The Krembil Family Epigenetics Laboratory, Room 28, Centre for Addiction and Mental Health, 250 College Street, Toronto, ON, Canada M4T 1R8. Tel: +1 416 5358501 4880; Fax: +1 416 979 4666; Email: [email protected] The Author 2006. Published by Oxford University Press. All rights reserved. The online version of this article has been published under an open access model. Users are entitled to use, reproduce, disseminate, or display the open access version of this article for non-commercial purposes provided that: the original authorship is properly and fully attributed; the Journal and Oxford University Press are attributed as the original place of publication with the correct citation details given; if an article is subsequently reproduced or disseminated not in its entirety but only in part or as a derivative work this must be clearly indicated. For commercial re-use, please contact [email protected] Nucleic Acids Research, 2006, Vol. 34, No. 2 529 most cases limited to short DNA fragments, often less than a DNA from the same brain region (prefrontal cortex) was used kilobase. as a control. Unmethylated sites were defined using a two-step The advent of microarray technologies that enabled the inter- analysis approach similar to the one used to determine tran- rogation of a large number of DNA/RNA fragments in a highly scription factor binding sites in the chromatin immunoprecip- parallel fashion has opened new opportunities for epigenetic itation (ChIP)-chip assay (27). First, a smoothing-window studies (10). A number of microarray-based technologies used Wilcoxon approach was applied to generate a P-value for epigenetic analyses are already available (11–23). However, graph for each individual where probe signal from the enriched all of these methods have some limitations, which renders them fraction was compared with the total genomic DNA in a one- unsuitable for some experimental setups. Additionally, many sided upper paired test. The window used in this report was technological parameters, such as the influence of DNA 501 bp. Second, three thresholds were applied to determine the sequence variations, amplification conditions and sensitivity boundaries of the unmethylated site: (i) an individual probe of the methods have not been investigated before. Here we threshold of P < 10 to determine if a probe is significantly present a detailed analysis of various parameters of epigenetic enriched in the unmethylated fraction compared with the profiling and provide a substantially improved microarray- control total genomic DNA; (ii) the maximum distance based high-throughput technology for DNA methylation between the two positive probes set to 250 bp and (iii) profiling of DNA regions that span from hundreds of kilobases the minimal size of a site set to 1 bp. The graphs can be to megabases. Eventually, this technology will be applied to downloaded from the internet (see Web resources). All the entire human genome, as exemplified by the methylation coordinates and annotation analysis were done on the April mapping of chromosomes 21 and 22 as reported here. 2003 version of the genome. Methylation -sensitive digestion of genomic MATERIALS AND METHODS DNA (gDNA) Microarray fabrication and data processing Prior to treatment with restriction enzymes, gDNA was sup- plemented with ‘spike’-DNAs (different concentrations of l COMT and CpG island microarrays were printed on Corning and Arabidopsis fragments), which were used as controls for CMT-GAPSII slides (Corning Life Sciences, Acton, MA) signal normalization. For enrichment of the unmethylated using a VersArray ChipWriter Pro System (Bio-Rad Labor- fraction, depending on the number of CpG dinucleotides to atories, Hercules, CA). For the COMT array, we designed 384 be interrogated, several combinations of methylation-sensitive oligonucleotides (Operon/Qiagen, US), each 50 bases long, enzymes, HpaII, Hin6I, AciI and HpyCH4IV, were used. representing every restriction fragment flanked by HpaII, gDNA was cleaved with a cocktail of these enzymes (10 U/ Hin6I and AciI restriction sites. In addition, control DNA ml in 2xY+/Tango buffer, Fermentas Life Sciences/Lithuania) fragments containing l phage, pBR322, FX174 and pUC57 for 8 h at 37 C. For enrichment of the methylated fraction, sequences were spotted on the slide. Each oligonucleotide was gDNA was cleaved by TasI or Csp6I (10 U/mlinG -buffer, diluted to a 25 mM solution and spotted four times to give a Fermentas) for 8 h at 65 C (TasI) or at 37 C (Csp6I). After the total of 1536 elements. In addition, 192 blank spots consisted restriction reaction, TasI was inactivated by 0.5 M EDTA. of SSC buffer and 48 spots contained Arabidopsis clones. The human CpG island array contains 12 192 sequenced CpG Adaptor–ligation island clones derived from a CpG island library that was ori- ginally created with MeCP2 DNA binding columns (24,25). For the ligation step, gDNA was supplemented with 8 GE Hybridized arrays were scanned on a GenePix 4000A scan- MspI-cleaved pBR322 plasmid (1 GE ¼ 1.45 pg/ 1 mg ner (Axon Instruments, Union City/CA) and analysed using gDNA), which was used as control for a potential ligation the GenePix 6.0 software. The GenePix PMT voltage for Cy3 bias. The ends of the cleaved DNA fragments were ligated and Cy5 channels were balanced with the histogram feature of to the unphosphorylated adaptors. Our adaptors contained a the scanner software to ensure a similar dynamic range for the sequence-specific protruding end, a non-target homologous two channels. Final scans were taken at 10 mm resolution, and core sequence, a specific antisense-overhang that prevents images for each channel were saved as separate 16-bit TIFF tandem repeat formation and blunt-end ligation, a ‘disruptor’ files. The emission signals for each channel were determined sequence that interrupts the original restriction sites after liga- by subtracting the local background from its corresponding tion, a new non-palindromic Alw26I (BsmAI) restriction site median average intensity. These raw data were either exported that enables the blunt-end cleavage of the adaptor from the into a custom Excel spreadsheet for subsequent data analysis target sequences (e.g. for library enrichment) and a non-5 - or directly imported into the Acuity 4.0 software (Axon complementary end. The CpG-overhang specific universal Instruments). The resulting datasets were normalized for the adaptor ‘U-CG1’ for the unmethylated DNA fraction ligates normalization features (spike-DNAs) and for signal intensity to DNA fragments generated by 11 CpG-methylation- (Lowess normalization). sensitive restriction enzymes HpaII, Hin6I (Hinp1I), Hpy- Profiling of unmethylated sites in the brain tissue of eight CH4IV, Bsu15I (ClaI, BspDI), AciI (SsiI), Psp1406I (AclI), adults was carried out using a tiling array spanning 12 Mb of Bsp119I (AsuII), Hin1I (AcyI, BsaHI), XmiI (AccI), NarI, non-repetitive sequence of chromosome 21 and 22 (q arms), BstBI (FspII) and also TaqI and MspI, which are not affected with probes spaced on average every 35 bp center-to-center by methylation of the internal cytosine. The adaptor represents (26). The genomic DNA from these individuals was cut with the annealing product of the two primers U-CG1a, 0 0 HpaII and Hin6I, amplified and hybridized to the microarray 5 -CGTGGAGACTGACTACCAGAT-3 , and U-CG1b, 0 0 as described previously (26,27). Unprocessed total genomic 5 -AGTTACATCTGGTAGTCAGTCTCCA-3 . 530 Nucleic Acids Research, 2006, Vol. 34, No. 2 The AATT-overhang specific adaptor ‘AATT-1’ for the The hybridization method used for the chromosome 21 and methylated DNA fraction fits to DNA ends produced by the 22 tiling arrays was described before (26,27). restriction enzyme TasI (TspEI), whereas the ‘TA-1’ adaptor Whole genome amplification fits to ends produced by Csp6I, BfaI or MseI, respectively: 0 0 AATT-1a, 5 -AATTGAGACTGACTACCAGAT-3 ; AAT- Genomic DNA was amplified using the GenomiPhi Kit 0 0 T-1b, 5 -AGTTACATCTGGTAGTCAGTCTC-3 ; TA-1a, (Amersham Biosciences) according to the manufacturer’s pro- 0 0 0 5 -TATGAGACTGACTACCAGAT-3 ; and TA-1b: 5 -AGT- tocol. Briefly, 10 ng of gDNA (1 ml) was mixed with 9 mlof TACATCTGGTAGTCAGTCTCA-3 . sample buffer, denatured at 95 C for 3 min, cooled on ice and All adapters were prepared by mixing equimolar amounts of then added to 9 ml of reaction buffer and 1 ml of Phi29 DNA the primer pairs, incubating the mixture at 80 C for 5 min, and polymerase. The reaction was incubated at 30 C for 16 h and then cooling it down to 4 C with 1 C/min. The double- then inactivated at 65 C for 10 min. stranded adaptors [200 pmol/ml] were added at 0.1 pmol per enzyme for each ng of the cleaved DNA (e.g. 0.3 pmol/ Bisulfite sequencing ng in a triple-digest HpaII/Hin6I/AciI). The ligation-mixture The methylation status of a number of CpG islands were with 400 ng template DNA was supplemented with 2 mlof 10· analysed by direct sequencing of sodium bisulphite modified ligation buffer (Fermentas), 1 ml ATP [10 mM] and water to gDNA (9). gDNA samples were subjected to bisulfite modi- 18 ml. The reaction was started in a thermal-cycler at 45 C for fication using a standard protocol (28). The primer sequences, 10 min, chilled on ice and 2 ml T4 ligase (Fermentas) was PCR conditions and cloning methods are provided in the added. The ligation reaction was carried out at 22 C for 18 h, Supplementary Data. followed by a heat-inactivation step at 65 C for 5 min. The mixture was then cooled down to room temperature with 1 C/ Genomic DNA min and stored at 4 C for subsequent procedures. Genomic DNA from all tissues was purified using standard laboratory methods (Phenol–Chloroform or Qiagen Blood and PCR Cell DNA Midi columns). To avoid cross reactivity of amine To control for a potential PCR bias, the DNA mixture was groups with the aminoallyl-labeling procedure, DNA samples supplemented with 2 GE FX174 plasmid (1 GE ¼ 1.8 pg of were stored in 0.5 M POPSO buffer (pH 8.0) instead of FX174 corresponding to 1 mg gDNA) that was cut with - Tris–EDTA. Male placental DNA was purchased from HpyCH4IV and ligated to the adaptor. PCR amplifications Sigma and the post mortem brain samples were provided were conducted for up to 25 cycles. A standard aminoallyl- by the Stanley Medical Research Institute. All parts of the PCR mixture included 400 ng of the ligate, 40 mlof10· study were approved by the CAMH review/ethics board. reaction-buffer (Sigma), 42 ml MgCl [25 mM], 3 ml aminoallyl-dNTP Mix [containing 15 mM aminoallyl- Web resources dUTP, 10 mM dTTP and 25 mM each dCTP, dGTP and All chromosome 21/22 tiling array data can be viewed in the dATP], 200 pmol primer (U-CG1a, AATT-1b or TA-1b, UCSC genome browser available via the methylation database respectively), 3 ml Taq enzyme (5 U/ml, NEB) and water to at www.epigenomics.ca. Additionally, the complete tiling a final volume of 400 ml. For PCR conditions and generation of array source data plus graphs that can be viewed in the Integ- dye-coupled adaptor products see Supplementary Data. rated Genome Browser (Affymetrix; www.affymetrix. com/support/developer/downloads/TilingArrayTools/index.affx) Array hybridizations and can be downloaded at http://transcriptome.affymetrix.com/ download/DataMethPaper (case sensitive). All coordinates and Each microarray slide was prehybridized with a mixture con- annotation analysis was done on the April 2003 version of the sisting of DIG Easy Hyb (Roche Diagnostics), 25 mg/ml tRNA genome. SNP data were derived from the SNP consortium, www. and 200 mg/ml BSA. The printed area was covered with the ncbi.nlm.nih.gov/SNP. prehybridization mixture under a coverslip for 1 h at 45 C. The OMIM numbers are derived from Online Mendelian Inheritance microarray slides were then washed in two changes of water in Man (OMIM), http://www.ncbi.nlm.nih.gov/entrez/query. for 2 min at 45 C, followed by two wash-steps at room tem- fcgi?db¼OMIM. Genome annotations were derived from the perature and a final wash-step in isopropanol for 1 min. The ReSeq database, http://www.ncbi.nlm.nih.gov/RefSeq/ and the slides were immediately blown dry with pressurized air and UCSC database, http://genome.ucsc.edu/cgi-bin/hgGateway. stored for hybridization. The hybridization mixtures were then pipetted onto the arrays and covered with Sigma Hybri-slips. The microarrays were placed in hybridization chambers (Corning Microarray Technologies, NY) and incubated on a RESULTS level surface for 16 h at 42 C for the COMT-arrays and Enrichment of the unmethylated fraction of gDNA 44–52 C for the CpG island microarrays in a covered water bath. The coverslips were removed by immersion of the arrays The strategy for enrichment of unmethylated portions of the in a wash solution containing 2· SSC and 0.5% SDS (washing genome is presented in Figure 1. gDNA is digested with buffer I). The array was washed twice for 15 min at 42–52 Cin methylation-sensitive restriction enzymes (Figure 1, middle washing buffer I (low stringency), followed by two wash-steps panel). Whereas methylated restriction sites remain in washing buffer II (0.5· SSC, 0.5% SDS), followed by 2 min unaltered, the sites containing unmethylated CpGs are of incubation in water. The slides were then rinsed quickly cleaved by the enzymes, and DNA fragments with 5 -CpG in isopropanol and finally dried with pressurized air. protruding ends are generated. The proportion of interrogated Nucleic Acids Research, 2006, Vol. 34, No. 2 531 Figure 1. Schematic outline of the microarray-based method for identification of DNA methylation differences and DNA polymorphisms in genomic DNA. Left panel: analysis of DNA sequence variation. Middle panel: the main strategy of the method is based on enrichment of unmethylated DNA fragments. DNA samples are cleaved by methylation-sensitive restriction endonucleases, and the resulting DNA fragments are then selectively enriched by adaptor-specific aminoallyl-PCR’s, labelled and hybridized to microarrays. Right panel: alternative procedure to enrich the hypermethylated DNA fraction. CpG sites depends on the methylation-sensitive restriction sequences, double- or triple-digest combinations of AciI, enzymes used for the restriction of DNA. Based on our HpaII, HpyCH4IV and Hin6I are recommended. analysis of the CpG dinucleotides within the sites of After the digestion of gDNA, the double-stranded adaptor methylation-sensitive restriction enzymes across several U-CG1 is ligated to the CpG-overhangs. At this point, it is megabases of human gDNA, the combination of three expected that most of the relatively short (<1.5 kb) and amp- enzymes, HpaII, Hin6I and AciI, should interrogate 32% lifiable DNA fragments derive from the unmethylated DNA of all CpG dinucleotides in mammalian DNA (Table 1). The regions. To some extent, the length of the amplified fragments addition of two other relatively inexpensive methylation- depends on the primer annealing temperature of the PCR sensitive CpG-overhang generating enzymes, HpyCH4IV reaction (Figure 2A). Some ligation fragments, however, and Hin1I, would theoretically increase the proportion of may still contain methylated cytosines. A proportion of interrogated CpGs to 41%. Depending on the such fragments can be eliminated by treatment with met microarray-type, in our experiments we usually use either McrBC, which cleaves DNA containing C and will not a single enzyme or a ‘cocktail’ of up to three restriction act upon unmethylated DNA. McrBC restriction sites consist met enzymes. The application of a set of enzymes might be of two half-sites of the form (G/A) C, which can be separ- disadvantageous for the analysis of GC-rich regions as ated by up to 3 kb (29,30). Hence, as can be seen in Figure 2B, met such a strategy would produce restriction fragments too a proportion of DNA fragments with two or more (G/A) C short for an efficient hybridization. In the latter case, it is within the restriction fragment are cleaved and therefore advisable to use a smaller number of restriction enzymes. deleted from the subsequent enrichment steps. The remaining Based on our experimental results and computer-based pool of unmethylated DNA fragments is then enriched by analysis of 100 randomly selected CpG islands, the most aminoallyl-PCR amplification that uses primers complement- suitable restriction enzymes are Hin6I and HpaII, followed ary to the adaptor U-CG1. One important advantage of using by AciI and Hin1I (Table 1). In contrast, for regular DNA protruding ends in the adaptor–ligation step is that degraded 532 Nucleic Acids Research, 2006, Vol. 34, No. 2 Table 1. Enzymes that generate protruding ends in the restriction fragments, profiles, especially in the case of hypermethylated CpG islands which are complementary to the adaptors U-CG1, TA-1 and AATT-1 or when the overall level of methylation in the genome is low (e.g. in insects). Thus, we developed an additional, modified Enzymes Recognition Percentage Number of Number of method to previously published methods of enrichment of sequence coverage fragments fragments methylated sequences to complement our data from the of CpGs in (per kb) in (per kb) in human CpG islands* non-CpG unmethylated fraction (Figure 1, right panel). This enrichment gDNA (%) islands* method relies on cleavage with the 4 bp frequent cutters TasI (AATT#) and/or Csp6I (G#TAC). Alternatively, BfaI or MseI HpaII (BsiSI) CCGG 8.6 3.98 1.18 can be used in combination with the Csp6I-specific adaptor. Hin6I (HinP1I) GCGC 6.4 3.98 0.61 AciI (SsiI) CCGC 17.4 3.23 1.79 All four enzymes produce DNA fragments in mammalian Hin1I GPuCGPyC 2.0 1.92 0.11 genomes of an average length 400–750 bp. The recognition (AcyI, BsaHI) sequences of TasI and Csp6I are infrequent within GC-rich HpyCH4IV ACGT 6.6 1.31 1.08 regions, leaving most CpG-islands intact. The analysis of 50 Bsu15I ATCGAT 0.2 <0.01 0.02 randomly selected CpG islands and several megabases of dif- (ClaI, BspDI) NarI (MlyI) GGCGCC 0.6 1.08 <0.01 ferent chromosomes revealed that Csp6I would produce more Bsp119I TTCGAA 0.1 0.11 <0.01 informative fragments in CpG islands than a digest with MseI, (BstBI, AsuII) whereas TasI and MseI produce informative fragments pref- Psp1406I AACGTT 0.3 <0.01 0.05 erentially in DNA regions outside of CpG islands (AclI, PspI) XmiI (AccI) GTMKAC 0.1 0.19 0.34 (Table 1). After ligation to the AATT- and TA-overhang spe- TasI AATT na 0.80 2.88 cific adaptors ‘AATT-1’ and ‘TA-1’, the un- and hypo- Csp6I GTAC na 2.23 1.41 methylated ligation products are eliminated from the reaction MseI TTAA na 0.80 2.88 by cleavage with a cocktail of methylation-sensitive restriction BfaI CTAG na 1.56 1.55 enzymes such as HpaII, HhaI (Hin6I), HpyCH4IV, Hin1I and Asterisk (*) indicates the number of 50 bp to 1.5 kb long (‘informative’) frag- AciI. Compared with a single digestion with BstUI (17), a ments, derived from several Mb of randomly selected CpG island and non-CpG cocktail of restriction enzymes will delete a higher percentage island sequences on chromosomes 1, 2, 4, 5, 6, 9, 17, 19 and 20; bold numbers of unmethylated sequences from the DNA fraction. The represent the most informative enzymes; na ¼ not applicable; M ¼ Adenine or remaining pool of mostly hypermethylated DNA fragments Cytosine; K ¼ Guanine or Thymine. is subsequently enriched by the aminoallyl-PCR amplification as described for the unmethylated fraction, and then hybrid- gDNA fragments (which are common in human post mortem ized to a microarray (Figure 2C). tissues) will not be ligated and amplified, and therefore will not interfere with DNA methylation analysis. Microarray design Most previous microarray-based epigenetic studies target hypermethylated DNA sequences (15,17,31,32); however, Various aspects of the microarray-based DNA modification interrogation of the unmethylated fraction is significantly profiling were investigated on the oligonucleotide microarray more informative. For example, the 100 kb region of chromo- that interrogates 100 kb fragment on 22q11.2 (Figure 3A). In some 22 interrogated by our COMT oligonucleotide addition to the catechol-O-methyltransferase (COMT, [MIM array (TXNRD2-COMT-ARVCF region; Microarray Design), 116790]), this chromosomal region contains also the gene contains 2193 methylatable cytosines. Enrichment of the encoding the thioredoxin reductase 3 gene (TXNRD2, [MIM unmethylated fraction can generate up to 401 amplicons of 606448]) and the armadillo repeat gene deleted in velocardi- sufficient size (50–1.5 kb), each representing the methylation ofacial syndrome (ARVCF, [MIM 602269]). For maximal status of at least one cytosine. In contrast, the combination of informativeness, it is necessary to design oligonucleotides MseI (+BsuI, to remove unmethylated fragments), the most according to the restriction sites of the methylation sensitive frequently used enzymes for enrichment of the hypermethyl- endonucleases used for the treatment of gDNA (Figure 3B). ated fraction (15,17,31,32), would produce 227 amplicons. For the COMT array, 384 oligonucleotides were designed, Seventy-seven amplicons would either contain no CpG dinuc- each 50 nucleotides long, representing every restriction frag- leotides or would be too short to stringently hybridize to a ment flanked by HpaII, Hin6I and AciI restriction sites. In microarray. Of the remaining 150 fragments, 144 contain addition, control DNA fragments containing l phage, multiple CpGs; hence, they are not fully informative since a pBR322, FX174, pUC57 and Arabidopsis sequences were single unmethylated BsuI restriction site would eliminate the spotted on the array (Materials and Methods). Additionally, entire fragment from the eventual amplification. Overall, only we used 12 192 element containing CpG island- and 6 of the 2193 methylatable cytosines are truly informative, high-density chromosome 21/22-microarrays (Materials and and none of these CpG dinucleotides are targeted by BsuI. Methods). Computer-based analysis of 50 randomly selected CpG island sequences revealed that the unmethylated fraction derived Detection of confounding effects of DNA sequence from HpaII cleavage results in 22 times more fragments variation (19.9 fragments/kb) of the suitable size range (50 bp to 1.5 kb) than the hypermethylated fraction (0.9 fragments/kb) Since restriction enzymes are used in the enrichment of dif- using MseI. ferentially modified DNA fractions, DNA sequence variation Nevertheless, analysis of the hypermethylated DNA frac- may simulate epigenetic differences. However, until now, tion may also add some new information to the methylation microarray methods used in epigenetic studies have not been Nucleic Acids Research, 2006, Vol. 34, No. 2 533 Figure 2. Selective enrichment of restriction fragments with the universal adaptor U-CG1. (A) Scatter plot that shows a comparison of ligation products treated with McrBC versus the untreated sample on the COMT array. McrBC treated fragments that contained at least two methylated cytosines were cleaved and could not be amplified in the following adaptor-PCR, resulting in reduced signal intensities in the Cy5 channel. (B) Co-hybridization of enriched unmethylated (Figure 1, middle panel) and hypermethylated (Figure 1, right panel) fragments derived from the same DNA source to a CpG island microarray. A large portion of amplicons is present only in one of the enriched fractions (marked black for log >0.3 black, green for log <0.3). Although the hypermethylated fraction hybridized to 75% of the microarray spots, based on our DNA sequence analysis, only a small fraction of them provide epigenetic information in comparison with the unmethylated fraction. differentiating between real DNA methylation differences patterns, HpaII, Hin6I, AciI and HpyCH4IV. The majority and single nucleotide polymorphisms (SNPs) within the of these CpG-SNPs were located in AciI and HpaII restriction restriction sites of the applied restriction enzymes. This prob- sites, with Hin6I and HpyCh4IV sites containing fewer met lem applies to some extent also to the C antibody-based polymorphisms (data not shown). Another approach to test strategy (22), which does not differentiate unmethylated CpG for DNA polymorphisms is the use of restriction endonuclease and TpG dinucleotides. In order to exclude the impact of DNA isoschizomers with different sensitivity to CpG methylation. sequence variation, two approaches are suggested. One is to However, this approach is currently only possible for HpaII/ check the available SNP databases in order to identify the MspI as there are no isoschizomers for most other methylation DNA sequence variation within the restriction sites of the sensitive restriction enzymes. enzymes used. For example, our 100 kb COMT array contains The third approach to differentiate the DNA sequence a total of 273 SNPs (SNPper, http://snpper.chip.org/bio/ effects from the genuine epigenetic differences consists of snpper-enter), of which 101 (37%) reside within CpG dinuc- performing an identical microarray experiment on the leotides and 55 (20%) are located within the restriction site of same DNA sample that has been stripped of all methylated the four main enzymes used to interrogate methylation cytosines. Our protocol utilizes the Phi29 DNA polymerase 534 Nucleic Acids Research, 2006, Vol. 34, No. 2 Figure 3. (A) Structure and GC-content of the chromosomal region on human chromosome 22q11.2 that spans the catechol-o-methyltransferase gene (COMT), the thioredoxin reductase 2 gene (TXNRD2) and the armadillo repeat gene deleted in VCFS (ARVCF). Vertical black bars represent exons. (B) To determine the methylation profile of the 100 kb TXNRD2-COMT-ARVCF region, 384 oligonucleotides (50mers, black horizontal bars) were designed based on the restriction sites for the methylation-sensitive endonucleases, HpaII, Hin6I and AciI (additional alternative enzymes are HpyCH4IV or Hin1I). Depending on the methylation status of the CpG-dinucleotides several combinations of amplicons (grey horizontal bars) can potentially hybridize to the oligonucleotides. (C) Typical hybridization patterns of the hypomethylated fraction of human gDNA on the COMT oligonucleotide-microarray. As discussed in Results, the complexity and informativeness of the hybridization signals increases with increasing number of methylation-sensitive restriction enzymes. to amplify whole genomic DNA, which creates a copy of the genome with all methylated cytosines replaced by unmethylated cytosines. Amplified DNA samples are then subjectedtothe same stepsasdepictedinFigure1 and hybridized on the microarrays. In this experiment all of the outliers must be a result of DNA sequence variations within the restriction sites of the enzymes used. These data can then be plotted against the DNA methylation data, whichare assayedinparallel(Figure 4).Insixexperiments that used amplified genomic DNA, the number of SNP- based outliers (threshold log-ratio <0.3, >0.3) ranged from 272 to 741 (432 ± 165, mean ± SD), or 2.2–6.1% of 12 192 CpG islands. Out of these, 72–234 (120 ± 66, mean ± SD) were initially identified as DNA methylation differences in microarray experiments using the unmethyl- ated fraction derived from the triple-digest with HpaII, AciI and Hin6I. From the CpG island array studies, our estimate is that 10–30% of the outliers detected in DNA Figure 4. Combined methylation- and SNP-analysis on a CpG island micro- methylation experiment could be due to DNA sequence array. The data of two separate hybridizations of DNA samples derived from variation. post mortem brain of two individuals are plotted against each other. The Y-axis contains the data derived from a methylation analysis (triple-cleavage with HpaII, Hin6I and AciI), whereas the X-axis contains the SNP data derived from Reproducibility the hybridization of the same DNA samples, which were subjected to the entire genome amplification prior to cleavage by the methylation-sensitive restriction To test the reproducibility of the method, a genomic DNA enzymes (Materials and Methods). Scale: log (Cy5/Cy3); an increased log- sample was split and subjected to the procedure of enrichment value on the Y-axis is indicated by red versus a decreased log-value represented by green. Significant outliers (log-ratio <0.3, >0.3, 2-fold difference) can be of the unmethylated fraction. The resulting amplification pro- classified into four clusters (S ¼ SNPs, M ¼ DNA methylation differences), ducts were labelled with Cy5 and Cy3 and then co-hybridized enabling the differentiation of epigenetic differences and nucleotide on the COMT array, which contains probes that flank the polymorphisms between the test-samples. Amp ¼ Whole-genome amplified HpaII, Hin6I and AciI restriction fragments around the sample. COMT gene. The Cy3 and Cy5 hybridization intensities exhib- ited very similar values (R ¼ 0.997; Figure 5A). Analogous Another critical factor in the amplification of unmethylated experiments, including switch dye hybridizations, were or hypermethylated DNA fragments is to ensure that no repeated several times also with the CpG island arrays and sequence specific bias is introduced. The rate of amplification in all cases were highly reproducible (R > 0.97). of repetitive sequences generally declines faster than that of Nucleic Acids Research, 2006, Vol. 34, No. 2 535 less abundant fragments in the later cycles of PCR (33). With increasing amplification cycles, repetitive DNA strands reach relatively high concentration and begin re-annealing to each other during the steps below the DNA melting temperature. To avoid this, a two-temperature PCR that uses a combined high- temperature elongation–annealing step was applied. A series of experiments were performed investigating how the number of PCR cycles would affect the hybridization patterns. As can be seen in Figure 5B, the relative intensities of the hybridiza- tion signals of both single copy sequences and repetitive DNA fragments, were similar in the range of 20–30 amplification cycles (R ¼ 0.991). Only when increasing the cycle numbers beyond 40 cycles was a biased amplification of some DNA sequences observed (data not shown). Sensitivity To test if differentially represented DNA fragments in two different DNA samples can be detected by this method, prior to methylation-sensitive cleavage, human gDNA was ‘spiked’ with unmethylated heterologous DNA, l phage and pBR322 plasmid (Figure 5C). Each sample was supple- mented with a different amount of spike-DNA, therefore mimicking differentially methylated sequences. The exact amount of l and pBR322 corresponded to increasing numbers of human genomic equivalents (1 GE of ‘spike’ DNA equals 16.28 pg l/mg gDNA and 1.45 pg/mg gDNA of pBR322, respectively). Hence, each of the experiments compared the intensities generated by 1 GE of l plus 128 GE of pBR322 (Y-axis) versus 16 GE of l plus 8 GE of pBR322 (X-axis). While the plotted signal intensities of the human gDNA sequences are positioned on or close to the regression line (indicating no methylation difference), the l and pBR322 fragments were identified as outliers. The average signal Figure 5. Reproducibility and sensitivity of the method. (A)A COMT microarray scatter plot representing two sets of amplification products derived from the same DNA source but produced at different time points by different researchers. The high-correlation coefficient of signal intensities demonstrates a high reproduci- bility of the method. (B) Influence of the PCR cycle number. Scatter plot diagrams show hybridization signal intensities of the unmethylated fraction that was amplified using 20 PCR cycles (Cy3 channel) and 30 cycles (Cy5 channel). Amplification products of each PCR were co-hybridized to the COMT microarray that contained oligonucleotides representing single copy sequences (closed circles), partially repetitive sequences (grey squares; 15–99 copies/genome) and highly repetitive DNA fragments (open squares; >100 copies/genome), such as ALU and LINE repeats. (C) Scatter plot representing the unmethylated fraction of human gDNA ‘spiked’ with different amounts of control DNA. The test samples were hybridized to the COMT array and contained either a 16-fold excess of l DNA (16 genome equivalents [GE] versus 1 GE; 10 fragments) or a 16-fold excess of pBR322 (128 GE versus 8 GE; 2 fragments), respectively. The ampli- cons of the spiked DNA (representing unmethylated DNA) can be easily distinguished as outliers; whereas the signals representing gDNA are located close to the regression line. Median signal intensities of different length oligonucleo- tides (40–50 bases) that target a specific HpaII restriction fragment in l DNA reveal that the length of spotted sequences directly influences the spot intensity and therefore the sensitivity of the microarray. (D) Sensitivity of the CpG-island microarray hybridization. Control amplicon (2 mg) (post mortem brain, unmethy- lated fraction) was labelled with Cy5 and co-hybridized with 2mg (0% difference), 1.9mg (5% difference), 1.8mg(10% difference), 1.5mg (25% difference) or 1.0mg (50% difference) of Cy3-labelled amplicon. For each hybridization to a COMT array, the regression lines represent the overall intensity that mimics methylation differences over the entire sample. The decrease of amount of DNA is reflected in the angle of the regression lines, which deviated by 5–7% from the expected values. 536 Nucleic Acids Research, 2006, Vol. 34, No. 2 intensity ratio of l oligonucleotides was 15.4, which is very silencing in mammalian genomes. The scatter plot shows two close to the ratio of spiked-DNA (16:1). The intensity values distinct spot areas, which represent predominantly unmethyl- for pBR322 were not as linear and exhibited a 6.5- to 10-fold ated fragments in placenta (yellow spots) and brain (orange difference (expected the same ratio of 1:16), most likely due to spots), respectively (Figure 7A). About 11% of the CpG island saturation effects of the hybridization. fragments exhibited 2-fold or more signal intensity difference In order to determine the sensitivity of the hybridization per between the two tissues. Some of the strongest brain-specific se, a control amplicon DNA was compared with itself but by signals could be identified for CpG islands associated decreasing the amounts of DNA by 5, 10, 25 and 50%. On the with neuronal genes such as DPYSL5, FABP7, DIRAS2, global level, the regression lines [y ¼ f(x)] reflected reprodu- GRIN3A, SLC24A3 and DSCAML1, whereas strong placenta- cible differences of the amount of amplicon DNA used in the specific outliers were associated with genes expressed in hybridization and varied by 5–7% from the expected values placenta, such as PCM1, CCND1, HA-1 and ADAMTSL1. (Figure 5D). Individual sites exhibited a lower accuracy, Overall, analysis revealed that brain DNA harboured notably which depended on the signal intensity, i.e. the stronger the more unmethylated CpG islands than placenta DNA. signal, the closer the observed spot intensity was to the expec- ted one. The rate of false outliers (log-ratio <0.3; >0.3; 2-fold Verification of detected methylation differences difference) was on average 3%. Usually, replication of Several loci that displayed methylation differences in our microarray experiments reduced the degree of aberration experiments were selected for verification by the sodium bisul- (log-ratio <0.3; >0.3) below 2% for all types of microarrays. fite modification mapping of methylated cytosines (Materials and Methods). The technique is based on the reaction of gDNA Examples of DNA methylation profiles with sodium bisulfite under conditions such that cytosine is Identification of DNA modification differences is provided in a deaminated to uracil but 5-methylcytosine remains unaltered. series of examples below. The COMT oligonucleotide array In the sequencing of amplified products, all uracil and thymine met was used to identify DNA methylation changes in a brain residues are detected as thymine and only C residues remain tumour (Figure 6A). In contrast to the pair of control brain as cytosine. The sites for the methylation-sensitive restriction DNA samples, where hybridization signals are close to the enzymes used in our experiments showed the expected regression line (indicating similar DNA methylation patterns), methylation difference across the DNA samples, as exempli- a visible proportion of the hybridization signals originating fied for CpG island clones located in the promoter region from the unmethylated DNA fraction of the brain tumour of galectin-1 and in the promoter region of a brain-specific deviates from the regression line. More subtle changes in transcript CR606704 (Figure 7B and C). DNA methylation patterns have been identified when post mortem brain tissues of healthy individuals were compared Chromosome-wide mapping of DNA methylation with the same tissues from schizophrenia patients (A. Schu- differences macher, A. Petronis, manuscript in preparation; representative example is shown in Figure 6B). The differences of the cancer Analysis of the unmethylated fraction from brain specific and psychosis studies show that diseases other than cancer DNA of eight adults using a chromosome 21/22 tiling array may reveal more subtle epigenetic differences, and therefore, detected 488–747 unmethylated sites per sample (Table 2). the informativeness and sensitivity of the epigenetic profiling This number increased to 977 in a merged map, showing that method is of critical importance. many sites were common between different individuals. The Another application of the technology includes epigenetic vast majority of the sites (90%) were positioned outside of 0 0 profiling of different tissues. One example of tissue specific the 5 ends and 5 flanking regions of the genes consistent with effects is shown using the CpG island microarrays that contain abundant transcriptional activity and a significant fraction of 12 192 CpG island clones of whom 8025 represent unique transcription factor binding sites found outside of known sequences. CpG islands tend to be found in many promoter annotations (26,27,34). The unmethylated sites outside of sequences and their methylation has profound effects on gene the 5 ends of known genes were about equally distributed Figure 6. Applications of the epigenetic profiling technology. (A) Changes of methylation profiles at TXNRD2-COMT-ARVCF in a brain tumour. The data from two different microarrays experiments are superimposed over each other. The analysis of two post mortem brain samples (closed dots) reveals no major difference in methylation levels, whereas the signal intensities vary significantly in the brain tumour (grey dots) when compared with the normal brain. (B) The comparison of DNA methylation profiles using the COMT microarray in brain tissue of a healthy control and a schizophrenia patient displays subtle epigenetic differences. Nucleic Acids Research, 2006, Vol. 34, No. 2 537 Figure 7. Examples of applications using a CpG island microarray. (A) Hybridization of the unmethylated fraction of placenta DNA and post mortem brain DNA to a CpG island array. Two pools of CpG island elements could be identified, which display extensively different methylation levels between these tissues(Note: some of the identified differences could be due to DNA sequence variation). (B) To validate the identified methylation differences, several CpG islands were subjected to bisulfite modification based mapping of methylated cytosines as exemplified for CpG island clones 22_B_12 (promoter region of Galectin-1) and 52_C_03 (promoter region of a brain-specific transcript, CR606704). The top sequence shows the reverse strand () of the original restriction sites, the bottom sequence displays the bisulfite-modified DNA. For each bisulfite-modified CpG-island, 8–10 clones were sequenced per tissue. Sequence 52_C_03 revealed several fully methylated CpG’s in placenta, which were unmethylated in brain. In contrast, clone 22_B_12 showed subtler methylation differences (15–100%), depending on the position of CpG-dinucleotide. (C) Methylation patterns of clones 22B_12 and 52_C_03 derived from bisulfite sequencing of 10–12 clones per tissue. The yellow boxes indicate CpG dinucleotides that are shown in the sequenced graph (Figure 7B). Table 2. Interindividual differences and distribution of the detected unmethylated sites with respect to the known genes as defined by the combined set of RefSeq and UCSC known genes for each brain DNA sample (M17–M25) and the merged map 0 0 0 0 0 0 Individual 3 -flanking 3 ter 5 -flanking 5 flanking–3 flanking 5 ter Distal Internal Total Site coverage (bp) #M17 chr21/22 13/12 2/16 8/20 2/4 10/20 64/122 98/97 488 64943/134730 %Total 5.1 3.7 5.7 1.2 6.1 38.1 40.0 #M18 chr21/22 17/22 9/15 13/29 3/3 16/28 95/191 134/152 727 98456/236797 %Total 5.4 3.3 5.8 0.8 6.1 39.3 39.3 #M19 chr21/22 15/24 11/14 12/27 2/5 14/21 86/173 119/130 653 88290/221721 %Total 6.0 3.8 6.0 1.1 5.4 39.7 38.1 #M21 chr21/22 20/24 12/18 15/29 2/5 14/22 102/184 143/157 747 109595/252347 %Total 5.9 4.0 5.9 0.9 4.8 38.3 40.2 #M22 chr21/22 18/20 8/17 9/29 3/6 15/24 86/169 127/143 674 87604/213453 %Total 5.6 3.7 5.6 1.3 5.8 37.8 40.1 #M23 chr21/22 12/15 4/13 10/25 2/3 10/21 68/150 101/111 545 70912/163322 %Total 5.0 3.1 6.4 0.9 5.7 40.0 38.9 #M24 chr21/22 14/18 5/12 7/20 4/3 10/20 61/158 88/107 527 65639/187229 %Total 6.1 3.2 5.1 1.3 5.7 41.6 37.0 #M25 chr21/22 17/15 7/13 10/18 3/3 9/22 65/171 102/97 552 69937/171073 %Total 5.8 3.6 5.1 1.1 5.6 42.8 36.1 Merged chr21/22 26/28 13/22 19/36 4/9 19/34 142/237 187/201 977 152148/314374 %Total 5.5 3.6 5.6 1.3 5.4 38.8 39.7 0 0 0 0 0 0 ‘5 ter’ or ‘3 ter’ refers to a 5 or 3 terminal site internal and within 1 kb of a gene boundary ‘5 flanking’ or ‘3 flanking’ refers to a site outside and within 5 kb of a gene 0 0 boundary; ‘internal’ refers to an intronic site and ‘distal’ refers to an intergenic site outside of the 5 kb/+1 kb boundaries. A site can also be both 5 and 3 flanking in 0 0 a gene rich region and referred as ‘5 flanking–3 flanking’. between sites residing within introns of known genes and unmethylated sites detected in this study cover 0.47 Mb outside of the gene boundaries. Interestingly, while some or 4% of the 12 Mb of non-repetitive sequences of chromo- genes, like BCR, showed a large number of sites inside somes 21 and 22 interrogated in the combined map of all eight the gene boundaries, some loci, like C21ORF55 spanning individuals with an average of 0.28 Mb (2.3%) in any given 150 kb, were essentially devoid of internal unmethylated individual. Maps of the methylation patterns (average value sites and in some cases, such as the SIM2 locus, the unmethyl- of the eight tested individuals) of the q-arms of chromosome ated sites were limited to the first intron (Figure 8A–C). Such 21 and 22 are shown in Figure 9A–B. Detailed maps of all intragenic methylation may inhibit inappropriate transcrip- individuals for chromosome 21 and 22, linked to the UCSC tional initiation at cryptic sites (35) or may serve as regulators Genome Browser (http://genome.ucsc.edu) are also available of alternate transcripts as can be seen for SIM2. Overall, on our web-based methylation database (Web Recourses). 538 Nucleic Acids Research, 2006, Vol. 34, No. 2 Figure 8. Profiles of unmethylated sites in three loci on human chromosomes 21 and 22 (501 bp window, Materials and Methods): BCR (A), C21ORF55 (B) and SIM2 (C) for human brain DNA (average of eight individuals, M17-M25). The graphs are based on P-values for each individual interrogation that show the significance of the enrichment in the unmethylated fraction versus total gDNA. The P-values were converted to the (10 log ) scale, such that, for example, P-value of 10 4 12 4 becomes 40. The vertical axes are adjusted to represent probes in the 40–120 range (P-values of 10 –10 ), thus only probes that pass P < 10 threshold are shown. Enlarged is a part of the chr 22q11.21 region (181 bp window), spanning breakpoints found in the generation of the two alternative forms of the Philadelphia chromosome translocation. C ¼ gDNA control. A comparison of the hypomethylation tracks with data from fragments representing DNA methylation profiles over large the Affymetrix transcriptome project (26,36) indicates that segments of gDNA. Building on the principles described in many of the unmethylated chromosomal regions overlap earlier publications (11–23) our method addresses a series of with mapped transcriptional active regions (Figure 9A–C, bot- critical issues and exhibits several advantages. An earlier tom tracks). These DNA methylation data complement exist- method (18) used a sucrose gradient to enrich the unmethyl- ing studies on transcriptional activity and histone ated DNA fraction. This method, however, requires a large modifications on human chromosomes 21 and 22 (37). We amount of DNA template and is rather imprecise in terms found that in the majority of cases, specific histone modifica- of the upper limit of the fragments that are subjected to tion patterns reported by Bernstein et al. (37) for the human hybridization. Other microarray methods for DNA methyla- tion analysis can be categorized into three main classes which hepatoma cell line HepG2 overlapped notably with the observed DNA methylation patterns. An example is shown are based on: (i) identification of bisulfite induced C!T trans- in Figure 9C for the PEX26 gene that is ubiquitously tran- itions (11–13,38,39), (ii) cleavage of gDNA by methylation- scribed in most tissues. The gene harbours an extensively sensitive restriction enzymes and (iii) immunocapturing with unmethylated CpG rich region in its promoter. The compar- antibodies against methylated cytosines. In the bisulfite arrays, ison of the different epigenetic profiles of both studies shows each tested CpG is represented by a pair of either C(G) or T(A) that the same genomic region was also highly acetylated at nucleotides. The arrays contain oligonucleotides that measure Lysine 9 and 14 of histone 3 (H3), accompanied with H3 the C(G)/T(A) ratio in the bisulfite treated DNA (correspond- met di- and trimethylation of Lysine 4. A comparison of histone ing to C/C in the native DNA). Although informative and modification tracks and our hypomethylation patterns for the precise, these microarrays can contain only a limited number q-arms of chromosome 21 and 22 revealed that H3 acetylation of oligonucleotides because treatment with bisulfite degener- and Lys4 methylation usually correlated with unmethylated ates the 4 nt code, resulting in a loss of specificity for a large CpGs. portion of the genome. For example, after bisulfite treatment all of the possible 16 permutations of a four base sequence containing unmethylated C and T (CCCC, CTCT, CCCT, CCTT, TCTC, TTTC, TTTT and so on) will become identical DISCUSSION TTTT. The bisulfite method is also laborious and cannot be Microarray based technology for DNA modification analysis easily applied to profile a large set of samples. Furthermore, it enables the highly parallel screening of numerous restriction is difficult to design suitable oligonucleotides that would Nucleic Acids Research, 2006, Vol. 34, No. 2 539 Figure 9. Genomic views showing unmethylated regions on chromosomes 21 and 22. (A and B): The top tracks (dark red) in the two chromosomal graphs shows the average amount of hypomethylation in the brain cortex of eight adult individuals. Also displayed are known genes (dark blue) and CpG islands (green). The bottom tracks display transcriptome data derived from 11 different tissues from the Affymetrix transcriptome phase 2 study (36). The track is coloured blue in areas that are thought to be transcribed at a statistically significant level. Regions that have a significant homology to other chromosomal regions or that overlap putative pseudogenes are coloured in lighter shades of blue. All other regions of the track are colored brown. (C) Enlarged is a part of chromosome 22q11.21, containing the peroxisome biogenesis factor 26 (PEX26, MIM 608666) that shows correlation between histone modifications and unmethylated DNA in its promoter region. The top three tracks represent histone modification data for H3 Lys4 dimethylation (orange bar), H3 Lys4 trimethylation (blue bar) and H3 Lys9/14 acetylation (yellow bar) (37). Underneath are the tracks for the average methylation patterns (unmethylated sites) observed in brain and the individual methylation patterns of all tested individuals (dark red). It is noteworthy that methylation patterns exhibit some interindividual differences (indicated by arrows). exhibit similar melting temperatures since the specificity of Methods relying on the enrichment and detection of hyper- base discrimination varies considerably (12). Using our methylated DNA have predominantly been used to identify approach, arrays can contain an almost unlimited number of abnormally methylated CpG islands in malignant cells (15– oligonucleotides: coverage can range from individual genes to 17,31). Although this strategy seems to be useful for detecting entire chromosomes represented by millions of oligonuc- major epigenetic changes in some regions of the genome, the leotides on glass chips. Whole genome tiling arrays are already overall proportion of interrogated CpG sites is substantially available for Arabidopsis thaliana and Escherichia coli, and lower compared with that achieved using approaches based will soon be available for the entire human genome. on the analysis of the unmethylated fraction. As shown in Restriction enzyme based methods are used to enrich either Results, we have estimated that interrogation of the unmethyl- the hypermethylated or unmethylated fraction of gDNA. ated fraction of gDNA could be up to several hundred 540 Nucleic Acids Research, 2006, Vol. 34, No. 2 folds more efficient than analysing the hypermethylated frac- the methylation patterns generated from a substantially larger tion. Furthermore, since unmethylated cytosines are less number of cells from the same tissue. abundant in the genome than methylated cytosines (depending There are, however, also some of limitations to the techno- on the tissue, 70–90% of cytosines are methylated), analysis of logy described in this article. The methylation sensitive the smaller unmethylated fraction of gDNA is more sensitive restriction enzymes do not interrogate every cytosine, and to detect subtle changes. For example, an increase of 10% with our current design, more than half of CpG sites remain met from the normal density of C would result in a 100% uninterrogated. This may be critical when the phenotypic out- (from 20 to 10%) difference in the unmethylated fraction, comes are determined by a methylation change at an isolated but only a 12% (from 80 to 90%) difference in the hyper- cytosine that is not within the restriction site of a methylation methylated fraction of gDNA. The unmethylated fraction has sensitive restriction enzyme. This problem may be partially been used in some approaches employing class II microarray overcome by the application of the same arrays to the CpG methods, for instance by using the methylation-specific specific immunoprecipitation technique (MeDIP) (22) in addi- McrBC enzyme (23) to deplete the hypermethylated fraction. tion to histone modification analysis through ChIP technology, However, the remaining unmethylated DNA fragments which identifies DNA sequences associated with modified (>1 kb) have to be gel-purified, requiring large amounts of histones (10). DNA and histone modifications seem to be starting material. Additionally, the McrBC method may not be inter-dependent, and consequently the possibility of a com- able to differentiate between dense and sparse methylation bined approach that interrogates both DNA methylation and within relatively short DNA fragments. For example, the chromatin modification in parallel might be a productive 2 kb human COMT promoter region, which contains 27 approach to the fine mapping of epigenetic changes. Also, McrBC target sites, can be cut to shorter than 1 kb fragments asymmetrical C sites (CpNpN) that are found in plants in cases where there are 2 (7%) or 27 (100%) methylated and some fungi such as Neurospora crassa are difficult to McrBC sites. Furthermore, the McrBC method cannot differ- detect, although some methylation-sensitive type IIs restric- entiate between unmethylated and polymorphic cytosines. tion enzymes are available (e.g. Esp3I or BveI). However, Another method to enrich the unmethylated fraction uses methylation of asymmetrical sites in animal organisms is 0 0 the rare cutter NotI (5 -GCGGCCGC-3 ) (19–21). However, not common. Additionally, this array method can also be NotI sites are not well represented in the genome and will only modified for analysis of methylated adenines in plants and provide a very superficial overview of genomic methylation bacteria. patterns. An alternative to these methods is the use of anti- In summary, the use of microarrays targeted at unmethyl- bodies specific for methylated cytosines [MeDIP (22)]. In this ated cytosines is a high-throughput approach to profile DNA method, antibodies are used to immunocapture methylated methylation patterns across the genome. The ability to analyse genomic fragments. However, this approach requires large minute amounts of DNA may enable the epigenetic screening amounts of gDNA (>8 mg) and also relies on the enrichment of DNA in plasma, serum or other body fluids, as well as in of the less informative hypermethylated fraction of the prenatal diagnostics. Although all the examples provided in genome. this work investigated human DNA, the same strategies can be In our analyses, we have addressed another important used for the epigenetic analyses of numerous other species. It issue: the interference of DNA polymorphisms that may is evident that epigenetic profiling should be performed in a simulate DNA modification differences across individuals. systematic, unbiased fashion and not limited to the tradition- Data from the SNP consortium indicate that roughly every ally preferable regions such as CpG islands. Outside of CpG 360th nucleotide in the human genome represents an SNP. In islands, numerous other genomic loci exist that may be sites humans, 2.16 million SNPs are detected in CpG dinuc- for important epigenetic modification, including enhancers, leotides, and such CpG SNPs are 6.7-fold more abundant imprinting control elements (41) or the regions that encode than expected (40). Depending on the restriction enzyme regulatory RNA elements. combination, our CpG island array-based studies demon- The above described technology, in combination with strated that 10–30% of all outliers initially detected as existing epigenetic profiling methods, may help to identify methylation differences contained SNPs (Figure 4). Informa- inter-individual variation in genome-wide methylation pat- tion on the SNPs and other polymorphisms such as deletions, terns as well as epigenetic changes that arise during tissue inversions or duplications within the restriction sites of the differentiation and the understanding of the epigenetic effects enzymes used for the enrichment of the unmethylated or of various environmental factors. Of particular interest is the hypermethylated fractions is helpful in differentiating the application of high-throughput DNA methylation analyses to epigenetic variations from the DNA sequence ones. To min- address the molecular basis of various non-Mendelian irregu- imize the effects of DNA polymorphisms, it may be also larities of complex diseases, such as discordance of mono- beneficial to compare affected tissue and healthy cells zygotic twins, remissions and relapses of a disease, parent of from the same individual. origin- and sex-effects, and tissue- and site-specificity (42). Another advantage of PCR-based methylation profiling Further technological developments may include building methods is the ability to work with limited DNA resources. high-resolution oligonucleotide-based microarrays spanning Although our basic protocol requires about 500 ng of gDNA, the entire human genome, improving the enrichment strate- the amount of template DNA can be significantly lower. In gies through the application of more specialized methylation our recent experiments, methylation patterns at the COMT sensitive restriction enzymes, and substantial reduction in region generated from a relatively small number of Jurkat the amount of initial template DNA down to the amount tissue culture cells (up to 500 cells, or 3 ng of DNA) did of a haploid or diploid genome. All these developments not reveal any significant differences when compared with will provide the basis for identifying the methylation profile Nucleic Acids Research, 2006, Vol. 34, No. 2 541 of the entire genome in a single cell, one of the ‘quantum 17. Huang,T.H., Perry,M.R. and Laux,D.E. (1999) Methylation profiling of CpG islands in human breast cancer cells. Hum. Mol. Genet., 8, leaps’ in post-genomic biology (43). 459–470. 18. Tompa,R., McCallum,C.M., Delrow,J., Henikoff,J.G., van Steensel,B. and Henikoff,S. (2002) Genome-wide profiling of DNA methylation reveals transposon targets of CHROMOMETHYLASE3. Curr. Biol., SUPPLEMENTARY DATA 12, 65–68. 19. Li,J., Protopopov,A., Wang,F., Senchenko,V., Petushkov,V., Supplementary Data are available at NAR Online. Vorontsova,O., Petrenko,L., Zabarovska,V., Muravenko,O., Braga,E. et al. (2002) NotI subtraction and NotI-specific microarrays to detect copy number and methylation changes in whole genomes. Proc. Natl Acad. Sci. USA, 99, 10724–10729. ACKNOWLEDGEMENTS 20. Yamamoto,F. and Yamamoto,M. (2004) A DNA microarray-based methylation-sensitive (MS)-AFLP hybridization method for genetic and This research has been supported by the Special Initiative grant epigenetic analyses. Mol. Genet. Genomics, 271, 678–686. from the Ontario Mental Health Foundation, and also by 21. Ching,T.T., Maunakea,A.K., Jun,P., Hong,C., Zardo,G., Pinkel,D., NARSAD, CIHR, NIH, the Stanley Foundation, and the Albertson,D.G., Fridlyand,J., Mao,J.H., Shchors,K. et al. (2005) Crohn’s and Colitis Foundation of Canada. A.S. holds a Epigenome analyses using BAC microarrays identify evolutionary CIHR Michael Smith Fellowship. Funding to pay the Open conservation of tissue-specific methylation of SHANK3. Nature Genet., 37, 645–651. Access publication charges for this article was provided by 22. Weber,M., Davies,J.J., Wittig,D., Oakeley,E.J., Haase,M., Lam,W.L. and the Ontario Mental Health Foundation. Schubeler,D. (2005) Chromosome-wide and promoter-specific analyses identify sites of differential DNA methylation in normal and transformed Conflict of interest statement. None declared. human cells. Nature Genet., 37, 853–862. 23. Lippman,Z., Gendrel,A.-V., Colot,V. and Martienssen,R. (2005) Profiling DNA methylation patterns using genomic tiling microarrays. Nature Methods, 2, 219–224. REFERENCES 24. Cross,S.H., Charlton,J.A., Nan,X. and Bird,A.P. (1994) Purification of 1. Bird,A.P. (1986) CpG-rich islands and the function of DNA methylation. CpG islands using a methylated DNA binding column. Nature Genet., Nature, 321, 209–213. 6, 236–244. 2. Henikoff,S. and Matzke,M.A. (1997) Exploring and explaining 25. Heisler,L.E., Torti,D., Boutros,P.C., Watson,J., Chan,C., Winegarden,N., epigenetic effects. Trends Genet., 13, 293–295. Takahashi,M., Yau,P., Huang,T.H., Farnham,P.J. et al. (2005) CpG Island microarray probe sequences derived from a physical library are 3. Wolffe,A.P. and Matzke,M.A. (1999) Epigenetics: regulation through representative of CpG Islands annotated on the human genome. repression. Science, 286, 481–486. Nucleic Acids Res., 33, 2952–2961. 4. Reik,W., Dean,W. and Walter,J. (2001) Epigenetic reprogramming in 26. Kapranov,P., Cawley,S.E., Drenkow,J., Bekiranov,S., Strausberg,R.L., mammalian development. Science, 293, 1089–1093. Fodor,S.P. and Gingeras,T.R. (2002) Large-scale transcriptional 5. Grewal,S.I. and Moazed,D. (2003) Heterochromatin and epigenetic activity in chromosomes 21 and 22. Science, 296, 916–919. control of gene expression. Science, 301, 798–802. 27. Cawley,S., Bekiranov,S., Ng,H.H., Kapranov,P., Sekinger,E.A., 6. Suter,C.M., Martin,D.I. and Ward,R.L. (2004) Germline epimutation of Kampa,D., Piccolboni,A., Sementchenko,V., Cheng,J., Williams,A.J. MLH1 in individuals with multiple cancers. Nature Genet., 36, 497–501. et al. (2004) Unbiased mapping of transcription factor binding sites along 7. Walter,J. and Paulsen,M. (2003) Imprinting and disease. Semin. Cell Dev. human chromosomes 21 and 22 points to widespread regulation of Biol., 14, 101–110. noncoding RNAs. Cell, 116, 499–509. 8. Laird,P.W. (2003) The power and the promise of DNA methylation 28. Hajkova,P., el-Maarri,O., Engemann,S., Oswald,J., Olek,A. and Walter,J. markers. Nature Rev. Cancer, 3, 253–266. (2002) DNA-methylation analysis by the bisulfite-assisted genomic 9. Frommer,M., McDonald,L.E., Millar,D.S., Collis,C.M., Watt,F., sequencing method. Methods Mol. Biol., 200, 143–154. Grigg,G.W., Molloy,P.L. and Paul,C.L. (1992) A genomic sequencing 29. Sutherland,E., Coe,L. and Raleigh,E.A. (1992) McrBC: a multisubunit protocol that yields a positive display of 5-methylcytosine residues in GTP-dependent restriction endonuclease. J. Mol. Biol., 225, 327–348. individual DNA strands. Proc. Natl Acad. Sci. USA, 89, 1827–1831. 10. van Steensel,B. and Henikoff,S. (2003) Epigenomic profiling using 30. Kruger,T., Wild,C. and Noyer-Weidner,M. (1995) McrB: a prokaryotic microarrays. Biotechniques, 35, 346–350, 352–344, 356–347. protein specifically recognizing DNA containing modified cytosine 11. Adorjan,P., Distler,J., Lipscher,E., Model,F., Muller,J., Pelet,C., residues. EMBO J., 14, 2661–2669. Braun,A., Florl,A.R., Gutig,D., Grabs,G. et al. (2002) Tumour class 31. Yan,P.S., Chen,C.M., Shi,H., Rahmatpanah,F., Wei,S.H. and Huang,T.H. prediction and discovery by microarray-based DNA methylation analysis. (2002) Applications of CpG island microarrays for high-throughput Nucleic Acids Res., 30, e21. analysis of DNA methylation. J. Nutr., 132, 2430S–2434S. 12. Balog,R.P., de Souza,Y.E., Tang,H.M., DeMasellis,G.M., Gao,B., 32. Chen,C.M., Chen,H.L., Hsiau,T.H., Hsiau,A.H., Shi,H., Brock,G.J., Avila,A., Gaban,D.J., Mittelman,D., Minna,J.D., Luebke,K.J. et al. Wei,S.H., Caldwell,C.W., Yan,P.S. and Huang,T.H. (2003) Methylation (2002) Parallel assessment of CpG methylation by two-color target array for rapid analysis of CpG island hypermethylation in hybridization with oligonucleotide arrays. Anal Biochem., 309, multiple tissue genomes. Am. J. Pathol., 163, 37–45. 301–310. 33. Mathieu-Daude,F., Welsh,J., Vogt,T. and McClelland,M. (1996) DNA 13. Gitan,R.S., Shi,H., Chen,C.M., Yan,P.S. and Huang,T.H. (2002) rehybridization during PCR: the ‘Cot effect’ and its consequences. Methylation-specific oligonucleotide microarray: a new potential for Nucleic Acids Res., 24, 2080–2086. 34. Kampa,D., Cheng,J., Kapranov,P., Yamanaka,M., Brubaker,S., high-throughput methylation analysis. Genome Res., 12, 158–164. Cawley,S., Drenkow,J., Piccolboni,A., Bekiranov,S., Helt,G. et al. (2004) 14. Hatada,I., Kato,A., Morita,S., Obata,Y., Nagaoka,K., Sakurada,A., Sato,M., Novel RNAs identified from an in-depth analysis of the transcriptome of Horii,A., Tsujimoto,A. and Matsubara,K. (2002) A microarray-based human chromosomes 21 and 22. Genome Res., 14, 331–342. method for detecting methylated loci. J. Hum. Genet., 47, 448–451. 35. Bird,A.P. (1995) Gene number, noise reduction and biological 15. Shi,H., Wei,S.H., Leu,Y.W., Rahmatpanah,F., Liu,J.C., Yan,P.S., complexity. Trends Genet., 11, 94–100. Nephew,K.P. and Huang,T.H. (2003) Triple analysis of the cancer 36. Cheng,J., Kapranov,P., Drenkow,J., Dike,S., Brubaker,S., Patel,S., Long,J., epigenome: an integrated microarray system for assessing gene Stern,D., Tammana,H., Helt,G. et al. (2005) Transcriptional maps of 10 expression, DNA methylation, and histone acetylation. Cancer human chromosomes at 5-nucleotide resolution. Science, 308, 1149–1154. Res., 63, 2164–2171. 37. Bernstein,B.E., Kamal,M., Lindblad-Toh,K., Bekiranov,S., Bailey,D.K., 16. Yan,P.S., Efferth,T., Chen,H.L., Lin,J., Rodel,F., Fuzesi,L. and Huebert,D.J., McMahon,S., Karlsson,E.K., Kulbokas,E.J.,III, Huang,T.H. (2002) Use of CpG island microarrays to identify colorectal Gingeras,T.R. et al. (2005) Genomic maps and comparative analysis of tumors with a high degree of concurrent methylation. Methods, 27, histone modifications in human and mouse. Cell, 120, 169–181. 162–169. 542 Nucleic Acids Research, 2006, Vol. 34, No. 2 38. Shi,H., Maier,S., Nimmrich,I., Yan,P.S., Caldwell,C.W., Olek,A. and polymorphic sites and suppression of variation in CpG islands. J. Mol. Huang,T.H. (2003) Oligonucleotide-based microarray for DNA Biol., 327, 303–308. methylation analysis: principles and applications. J. Cell Biochem., 41. Schumacher,A., Buiting,K., Zeschnigk,M., Doerfler,W. and 88, 138–143. Horsthemke,B. (1998) Methylation analysis of the PWS/AS region does 39. Hou,P., Ji,M., Liu,Z., Shen,J., Cheng,L., He,N. and Lu,Z. (2003) A not support an enhancer-competition model. Nature Genet., 19, 324–325. microarray to analyze methylation patterns of p16(Ink4a) gene 5 -CpG 42. Petronis,A. (2001) Human morbid genetics revisited: relevance of islands. Clin. Biochem., 36, 197–202. epigenetics. Trends Genet., 17, 142–146. 40. Tomso,D.J. and Bell,D.A. (2003) Sequence context at human single 43. Collins,F.S., Green,E.D., Guttmacher,A.E. and Guyer,M.S. (2003) A nucleotide polymorphisms: overrepresentation of CpG dinucleotide at vision for the future of genomics research. Nature, 422, 835–847.

Journal

Nucleic Acids Research – Oxford University Press

Published: Jan 1, 2006

There are no references for this article.