Tissue-specific variation in DNA methylation levels along human chromosome 1
- Cecilia De Bustos†1, 8,
- Edward Ramos†2, 3, 9,
- Janet M Young†2,
- Robert K Tran4, 10,
- Uwe Menzel1,
- Cordelia F Langford5,
- Evan E Eichler3, 6,
- Li Hsu7,
- Steve Henikoff4, 6,
- Jan P Dumanski1 and
- Barbara J Trask2, 3Email author
© De Bustos et al; licensee BioMed Central Ltd. 2009
Received: 08 December 2008
Accepted: 08 June 2009
Published: 08 June 2009
DNA methylation is a major epigenetic modification important for regulating gene expression and suppressing spurious transcription. Most methods to scan the genome in different tissues for differentially methylated sites have focused on the methylation of CpGs in CpG islands, which are concentrations of CpGs often associated with gene promoters.
Here, we use a methylation profiling strategy that is predominantly responsive to methylation differences outside of CpG islands. The method compares the yield from two samples of size-selected fragments generated by a methylation-sensitive restriction enzyme. We then profile nine different normal tissues from two human donors relative to spleen using a custom array of genomic clones covering the euchromatic portion of human chromosome 1 and representing 8% of the human genome. We observe gross regional differences in methylation states across chromosome 1 between tissues from the same individual, with the most striking differences detected in the comparison of cerebellum and spleen. Profiles of the same tissue from different donors are strikingly similar, as are the profiles of different lobes of the brain. Comparing our results with published gene expression levels, we find that clones exhibiting extreme ratios reflecting low relative methylation are statistically enriched for genes with high expression ratios, and vice versa, in most pairs of tissues examined.
The varied patterns of methylation differences detected between tissues by our methylation profiling method reinforce the potential functional significance of regional differences in methylation levels outside of CpG islands.
DNA methylation is a major epigenetic modification that is vital to mammalian development . Methylation is carried out by DNA methyltransferases  and can suppress the initiation of transcription of a locus . Methylation plays a significant role in genomic imprinting, X-inactivation, and silencing of parasitic sequences [4–7]. Acquired DNA methylation differences might account for some phenotypic differences between monozygotic twins . Aberrant methylation can cause various syndromes [9–14] and contribute to tumorigenesis by decreasing activity of tumor suppressor genes , activating proto-oncogenes , or overall methylation imbalance  (reviewed in ).
In mammals, methylation occurs preferentially at cytosines that are followed by guanine (CpG). CpG dinucleotides are relatively infrequent in the human genome, except in CpG islands, which are small (0.2 to 2 kb) regions highly enriched in CpGs. Approximately 50 to 60% of gene promoters lie in CpG islands [19–21]. Methylation can inhibit binding of transcription factors to these sequences directly by altering the structure of the recognition site, or indirectly by recruiting repressive proteins with methyl-binding domains [22–25].
However, methylation of CpG island promoters does not always correlate with gene expression levels. Most inactive promoters are not methylated [3, 26]. The CpG islands of some tissue-specific genes remain unmethylated in tissues where the gene is not expressed (such as MyoD in non-muscle tissues) [27–29]. Intriguingly, while promoters are less methylated on the active X than on the inactive X, the opposite is true for CpGs in the bodies of genes on the X chromosome . Highly expressed autosomal genes were also found to have hypomethylated promoters and hypermethylated gene bodies in a recent genome-wide analysis . Moreover, clusters of hypomethylated regions not limited to CpG islands or promoters have been found interspersed in stretches of methylated CpGs . CpG methylation outside of CpG islands is thought to suppress transcription of transposable elements and other parasitic sequences [26, 33] and/or to suppress spurious initiation of transcription elsewhere, including within infrequently transcribed genes [34, 35]. Clearly more information on the state of methylation in and outside of CpG islands and across different tissues is needed to clarify the role methylation plays in transcriptional regulation.
Several approaches have been developed recently to scan genomes for sites of DNA methylation that might be important for tissue-specific gene regulation. These approaches include large-scale sampling of DNA sequences after bisulfite conversion of unmethylated cytosines to uracil [31, 33, 36–38], a variety of techniques measuring the differential sensitivity of methylated and unmethylated sequences to certain restriction enzymes (for example, RLGS, RDA, HELP, MCAM, and MSCC) [31, 32, 39–48], and quantitative analyses of promoter sequences immunoprecipitated with antibodies that enrich for methylated cytosines [26, 49, 50] (reviewed in ). Differences in methylation have been detected among somatic tissues at up to 15% of the loci analyzed in these studies [26, 31, 33, 36, 45, 48, 52].
Here, we employ a methylation profiling technique and a tiling microarray for human chromosome 1 to detect regional differences in the methylation state between tissues from the same individual. Our approach relies on a methylation-sensitive restriction enzyme to generate differently sized fragments from methylated versus non-methylated genomic DNA. Sample differences in fragment yield within a specified size range (80 to 2,500 bp) are detected by competitive hybridization of size-selected fragments to a tiling array of genomic clones. This approach has been used successfully to characterize methylation patterns in Arabidopsis methylation mutants [53, 54]. We have combined this technique with a custom chromosome 1 microarray with a resolution of approximately 100 kb .
Below, we first describe simulation experiments that show that log ratios measured by the array are a complex function of the methylation level of the hybridizing DNA and the GC-content of the array clones. We also predict from simulations that the array will be more responsive to methylation changes at dispersed CpG sites than to changes in CpG islands. We then use the array to profile chromosome 1 methylation in several tissues from two human donors, demonstrating that the technique yields highly reproducible results. The methylation profiles we obtain vary between different human tissues, but profiles of the same tissue from two different individuals are similar. We then describe our use of bisulfite sequencing to confirm an array finding of hypomethylation in heart DNA around the RYR2 gene and to confirm predictions from our simulations that observed log ratios are impacted by methylation differences outside of CpG islands. We analyze published data on the relative expression levels of genes and detect significantly different expression ratios for genes residing in regions exhibiting extremely high versus low methylation ratios in all six of the tissue comparisons analyzed.
Simulations predict impact of GC content and CpG distribution on fragment yield
These samples are allowed to competitively hybridize to 2,136 genomic clones (bacterial artificial chromosomes (BACs), P1-derived artificial chromosomes (PACs), fosmids, cosmids) on a microarray in the presence of Cot1 DNA to suppress repetitive sequences. The array contains 2,049 clones that cover most of the euchromatic portions of human chromosome 1, 17 clones that are distributed sparsely on chromosome X, and 70 other clones. Average log2 ratios of test-to-reference fluorescence intensities are generated from the duplicate spots for each clone whose measurements pass quality control tests (see Methods). Since the median size of the genomic inserts of the clones on the array is 159 kb, the signal ratio for a given clone is a complex function of the methylation states of Hpa II sites in the sample and reference (which might both be heterogeneous mixtures of cells with different methylation patterns) and the cumulative length (minus repeats) of the resulting Hpa II fragments within the size range of approximately 80 to 2,500 bp in the genomic interval covered by the clone.
In order to simulate the relative fragment yield for an entire BAC at different methylation levels, we wrote a PERL script to randomly assign each Hpa II site in a BAC's sequence to be methylated or unmethylated, so as to achieve various user-specified average overall levels of methylation. The script performed in silico Hpa II digestion of the resulting partially methylated sequence and the size-fractionation step. It also masked the sequences using RepeatMasker , which mimics the Cot1 repeat-suppression step necessary to achieve specificity in array hybridization. The output is total unmasked base pairs in fragments passing the size filter, which simulates labeling by random priming .
Our simulations show that yield of total unmasked base pairs passing the size filter from the region corresponding to any given clone on the array decreases approximately linearly with increasing average per cent methylation. Figure 2B shows the results across a range of methylation levels for BAC RP11-47A4 (AL391809), an outlying clone in the heart-versus-spleen comparisons and the object of experimental validation below. This negative slope indicates that the array response is dominated by methylation status at Hpa II sites between larger fragments where increased methylation would reduce yield through the size filter (sites of type a in Figure 2A), and that smaller Hpa II fragments like those in CpG islands (sites of type c in Figure 2A) have a negligible effect on the yield of size-selected fragments corresponding to any array clone. Simulations for all clones on the array show a similar negative correlation between fragment yield through the size filter and methylation level. Figure 2C and 2D display results for two selected methylation levels, demonstrating that a sample uniformly methylated at 40% of sites (Figure 2C, black contours) yields more hybridizing material for all array BACs than a sample with 80% methylation (Figure 2C, gray contours) and positive predicted log ratios for all array BACs (Figure 2D). A caveat of our simulations is that we assumed that all CpG dinucleotides had an equal chance of being methylated. More sophisticated simulations might model local heterogeneity, such as that observed in our bisulfite sequencing analyses (see Results), or the differences between CpG islands and other regions, but would need a good model to describe how methylation is distributed across sites. Additional simulations show that the relationship between fragment yield and methylation level would have a positive slope only if CpG islands comprise more than 30% of a clone's sequence (data not shown), and that is not the case for any clone on the array. Thus, signal intensity differences between clones or samples are likely to be predominated by methylation differences at CpGs outside of CpG islands.
Simulations also demonstrate a strong relationship between fragment yield and the percentage of bases that are either G or C (GC%) in the probed sequence. Figure 2A helps explain how GC content can affect measured log2 ratios. In this schematic example, Hpa II sites are relatively frequent in BAC 1, so that many fragments pass the size filter in both samples 1 and 2. The increased methylation of sample 2 causes a relatively modest reduction in hybridizing material for BAC 1. In contrast, BAC 2 has lower GC content and thus larger Hpa II fragments, on average, and therefore the same increased methylation level in sample 2 has a much more dramatic effect, entirely preventing any material passing the 80 to 2,500 bp size filter. This more dramatic effect would result in a more extreme measured log ratio for BAC 2 than BAC 1 despite the fact that both BACs measure the same pair of samples with the same ratio of methylation levels. Figure 2C shows the positive relationship between total base pairs in Hpa II fragments predicted to pass our size filter and GC% for the regions represented by the clones on our array. This positive relationship is seen at all simulated methylation levels (40% and 80% are shown in Figure 2C), except at the extreme of 100% methylation, when the DNA would not be cut at all. Moreover, the log2 ratio of yield for two samples simulated to have different overall methylation levels (test low, reference high) is positive for all clones on the array, but the ratios are negatively correlated with clone GC content, even though simulated methylation levels were uniform (Figure 2D). The slope of this relationship between log2 ratio and GC% is influenced by choice of overall methylation levels simulated for the two samples, relative simulated levels of background signals on the array spots (for example, a fixed amount of non-specific signals or signals due to incompletely suppressed repetitive elements), and allowance for variation of actual methylation levels with GC%. In fact, some combinations of these variables can yield a generally positive (but curved) slope in the relationship between log2 ratio and GC% (Figure 2E). To demonstrate this effect, we simulated adding cross-hybridizing signal proportional to each BAC's repeat content to both test and reference channels. The predicted ratio for BACs of low GC content is greatly reduced, because the cross-hybridizing material (equal for test and reference) outweighs the small amount of specific hybridizing material. BACs of higher GC content receive much more specific hybridizing material, so the cross-hybridizing material has less effect and the ratios are much closer to those predicted without cross-hybridization (compare Figures 2D and 2E)
Experimental data confirm a dependency on GC content. Note that we do not know the overall average levels of methylation for any of the samples we used, but any overall difference in methylation levels between the two samples being compared would result in a correlation between GC content and measured log ratio. In the example provided in Figure 2F, a comparison of liver and spleen tissue from the same donor, both test (liver) and reference (spleen) signal intensities for clones on the array increase with increasing clone GC content, and the observed log2 ratios decrease with clone GC content (Figure 2G). This plot shows raw ratios, but ratios normalized by intensity, a common normalization method, show a similar relationship (not shown). Because much of the variance in log2 ratios across the genome is accounted for by variation in GC content, we normalized log2 ratios for each clone based on its GC% using a loess fit of the relationship between log2 ratio and GC% measured for each array experiment. All subsequent plots and analyses use GC-normalized values. Note that this normalization, while necessary to correct for clone-to-clone differences in measured ratios due to differences in GC content, excludes an ability to detect any true variation in methylation levels that might correlate with GC content. The GC-normalized log2 ratios allow us to identify potential differences in relative methylation levels for clones with similar GC content (Figure 2H).
Tissue differences in methylation
We next used the profiling method to detect methylation differences between tissues from the same individual. We evaluated lung, heart, liver, four regions of the brain (cerebellum, medulla oblongata, occipital lobe and pons), each from two individuals, as well as testis or ovary from single individuals. Each tissue was directly compared with spleen from the same individual to control for any inter-individual variation in genomic content (for example, segmental copy number variants and/or restriction fragment length polymorphisms). We also conducted conventional array-comparative genomic hybridization (CGH) analyses on the tissue samples to verify that no significant somatic differences in copy number existed within the same individual for sequences represented by our clone array (data not shown). The female lung sample was excluded from further analyses, because it had an unusually noisy genomic copy-number profile, suggestive of random genome instability, with relative copy number correlating strongly with methylation profile ratios (data not shown).
Pearson correlation coefficients, R2.
Same tissue, Same individual
Same tissue, Different individual
All, except lung
All, except T & O
testis vs ovary
Different tissue, Same individual
All*, except lung
*only cerebellum represented from brain
heart vs. ovary
heart vs. testis
heart vs. cerebellum
heart vs. liver
liver vs. ovary
liver vs. testis
liver vs. cerebellum
cerebellum vs. ovary
cerebellum vs. testis
All four brain parts
The methylation profiles of the various lobes of the brain map are highly correlated, with many of the peaks and valleys in the profile landscapes mapping to the same chromosomal domains (Figure 5). This similarity is best appreciated in the XY-plots provided in Additional files 2, 3, 4, their correlation coefficients summarized in Table 1, and the hierarchical cluster plots in Additional file 5. These pronounced common differences might be the cumulative result of many genes in these regions that are regulated in a tissue-specific manner, in this case brain- or spleen-specific. The median Pearson correlation coefficient of ratios measured for the four different brain portions (each compared with spleen) from the same donor is 0.45 (range 0.27 to 0.93, n = 108). This level of correlation approaches what we observed when comparing the same tissue taken from different individuals (see above).
In sharp contrast to the similarity of the brain tissue profiles, we observe gross differences between the methylation profiles of other tissues, perhaps reflecting regions that are differentially regulated between tissues (Figure 4, Additional file 4). The correlation coefficients are notably low when ratios measured for different organs (each relative to spleen) of the same donor are compared (median R2 = 0.1, range 0 to 0.44 excluding lung and including a only single brain part (cerebellum) in these comparisons, n = 108) (Table 1). Thus, methylation profiles of the same tissue in different individuals correlate more strongly than do those of different tissues from the same individual.
The heart-versus-spleen methylation profiles identify an intermediate number of clones that deviate reproducibly and significantly from the median ratio in independent comparisons (Figure 4, Additional files 2 and 7). Many of these clones were not identified as outliers in other tissue comparisons, suggesting a real difference in the methylation status between heart and spleen in these genomic regions. Below, we focus our bisulfite sequencing assays of methylation state on two of these clones with outlying values.
Bisulfite confirmation of methylation differences outside of CpG islands
We conducted bisulfite sequence analyses to confirm methylation differences between heart and spleen for sequence represented by RP11-47A4 (AL391809), which showed very high heart-versus-spleen log2 ratios in all replicates of both donors (Additional files 8 and 9). These very high relative heart:spleen ratios are predicted to reflect significantly less methylation of CpGs, and thus more frequent cutting with Hpa II, in heart than spleen in this locus compared with other regions of chromosome 1 after accounting for fluctuations in GC content. This clone contains part of a heart-specific gene, RYR2, making it an interesting candidate for confirmation of this prediction by bisulfite sequencing. These analyses also confirm our simulations, which predict that the method is most responsive to methylation differences outside of CpG islands.
Although our simulations predict that the differences between heart and spleen detected by the array are dominated by methylation differences such as those outside of CpG islands, we nevertheless examined the methylation state of RYR2's CpG island, which contains regulatory elements, GC boxes essential for transcription of this gene, and spans a total of 181 CpGs. We successfully amplified two portions of the CpG island, containing a total of 53 CpGs. Overall, CpGs in the island were less methylated than CpGs distributed outside of the island in both heart and spleen. CpGs in the first 251-bp CpG island amplicon (P1) just upstream of the GC-boxes are approximately 3.5-fold less often methylated on average in heart than spleen, but the second 303-bp amplicon (P2) containing 35 CpGs, including five Hpa II sites just downstream of the GC boxes, was almost entirely unmethylated in both heart and spleen (Figure 6, bottom; Additional file 9). Only 0% and 0.5% of assayed CpGs in P2 were methylated in heart and spleen, respectively. Thus, if these combined results apply to the entire CpG island, methylation differences outside of the CpG island must account for most of the high signal ratio reported by this clone.
We also find no significant methylation differences between heart and spleen in the CpG island of ATP2B4, which is partially contained in RP11-90O23, another clone exhibiting a high log2 ratio in the heart-versus-spleen array comparisons (Additional file 7). ATP2B4 (PMCA4b) is involved in calcium homeostasis and has a role in controlling cardiac hypertrophy in response to increased load on the heart . We compared the methylation status of 27 CpGs representing approximately 80% of the CpG island of ATP2B4 in heart and spleen using bisulfite sequence analysis. The 27 CpGs were similarly unmethylated in heart and spleen (data not shown). We found average methylation levels per site of only 0.41% and 1.0% in heart and spleen, respectively. Thus, methylation differences in the region in and around ATP2B4 must account for RP11-90O23's observed high log2 ratio in the heart-versus-spleen comparison, as we find no methylation differences in ATP2B4's CpG island.
Regions with outlying methylation ratios contain genes with outlying expression ratios
In order to examine the possible biological significance of the methylation ratios measured by our arrays, we compared relative methylation levels with relative expression levels of corresponding genes. In all six of the tissue comparisons examined, we found that BACs with apparent high relative levels of methylation (that is, low array ratios) tended to contain genes that were expressed at lower relative levels than BACs with lower methylation levels. This relationship is expected if regional methylation assayed by our array, like promoter methylation, is associated with suppression of transcriptional initiation . In detail, we compared our results (averaged across arrays representing the same tissue comparison) with expression data obtained by Ge et al. , who hybridized RNA from each of 36 normal human tissues singly to Affymetrix oligonucleotide microarrays to determine expression levels of around 20,000 human genes (see Methods). For each gene, we calculated a ratio of expression levels for six of the tissue pairs evaluated in our methylation studies, with spleen as the reference tissue in each case (Ge et al. did not generate expression data for members of the remaining tissue pairs). We compared the methylation array ratios of BAC-sized genomic regions with the expression ratios of genes whose 5' ends mapped within those genomic regions, recognizing that such an analysis likely ignores facets of the complex cause-and-effect relationships between the methylation levels in different parts of a gene (promoter, body of gene, and so on) with that gene's expression levels.
DNA methylation is widely recognized as an epigenetic modification that plays a key role in specifying which portions of the genome are utilized at any given time or place in eukaryotic organisms, from plants to humans. Therefore, determining where those marks are made in the genome and how these marks are interpreted by the proteins that transcribe genes, prevent transcription of other sequences, package chromatin in the nucleus, or perform other functions on DNA, has become an important goal. We have generated chromosome-wide methylation profiles of different tissues using an approach that relies on the action of a methylation-sensitive restriction endonuclease and measurement of differences in the relative yield of size-selected fragments between two samples. Hybridization of these fragments to a tiling array of large-insert clones reports coarse regional relative differences in methylation state. Hybridization to high-resolution oligomer arrays can reveal much finer-scale fluctuations in methylation levels along the chromosome (our unpublished data).
We obtained very different methylation profiles from different organs, suggesting the existence of tissue-specific epigenetic modification patterns across chromosome 1. These patterns are conserved across individuals, as the values measured for the same tissue obtained from different donors were strongly correlated. As one might expect given the high inter-individual similarity in methylation profiles for a given tissue, replicate analyses of the same tissue from the same donor were even more strongly correlated, demonstrating that our technique gives highly reproducible results. We find that methylation profiles of the same tissue correlate better across individuals than do profiles of different tissues from the same individual. This result corroborates findings of other studies that used alternative techniques to measure relative methylation levels [31, 33, 63].
While most organ profiles were dissimilar, there were notable exceptions. Interestingly, the profiles for the various lobes of the brain were strongly correlated, suggesting that the shared methylation pattern of these tissues was established in a shared developmental precursor cell type and/or that the functions of these cells are sufficiently similar to be reflected in a similar pattern of epigenetic modifications. We also observe that some sites stand out as deviating from the median in multiple tissues. For example, some clones show similar relative log2 values in brain profiles and either testis or ovary profiles (Additional file 4). These similarities might represent spleen-specific methylation changes (spleen was used as a common reference sample) or changes common to these other organs. Other deviant sites are not shared by these different organs, such that, in fact, testis and ovary profiles are poorly correlated with each other.
Our simulations indicate that significant deviations of signal ratios from the median are influenced predominantly by methylation differences in CpG dinucleotides in Hpa II sites outside of CpG islands. Thus, this method complements approaches that focus on the methylation states of CpG dinucleotides in CpG islands. Our bisulfite sequencing results support the predictions of these simulations. For example, many more of the significant methylation differences that might explain the high heart:spleen signal ratio observed for the RYR2 locus, a gene with heart-specific function, were found outside of CpG islands than in them. An important conclusion from these results is that CpG islands and flanking DNA can have different methylation patterns, and that tissue-specific genes can be differentially methylated in regions other than their CpG islands. This conclusion is similar to that drawn from other studies which found that both the bodies and promoters of highly expressed genes are differentially methylated compared with inactive genes [30, 31, 38, 50].
Together, our observations point to strikingly different patterns of methylation at Hpa II sites outside of CpG islands in various tissues. Methylation differences in the non-CpG-island compartment of the genome are largely unappreciated and merit further study to understand their possible functional consequence(s). Of all the tissues analyzed, cerebellum showed the greatest fluctuations in relative fragment yield along the chromosome, perhaps reflecting gross variation in the density of brain-expressed genes across the chromosome. In contrast, the profiles for liver, heart and lung showed much less regional variation with respect to spleen, suggesting a more even distribution (or lack) of genes expressed specifically in these tissues and associated tissue-specific epigenetic modifications.
Our findings are consistent with recent studies that have found a comparatively low frequency of tissue-specific differentially methylated regions associated with CpG-island promoters , while differences have been detected in CpG-poor promoters , outside CpG-dense promoters , and in CpG islands far from any known transcription start site . Our data also reinforce the conclusions of Khulan et al. , who used a different approach (HELP) on mouse tissue samples to find that tissue-specific differentially methylated regions are frequent and not confined to gene promoter regions. Applying yet another methylation profiling method (RLGS), two groups found a highly significant enrichment of differentially methylated Not I sites away from CpG islands in mouse tissues [48, 65]. Interestingly, a study by Oakes et al. found 8-fold more hypomethylated Not I sites (among 2,600 analyzed) in testis than somatic tissues, and found relatively few brain-specific differentially methylated sites . In contrast, we detect many more apparently differentially methylated regions in cerebellum and other brain parts than testes when both are compared with spleen.
These observations raise the possibility that hypomethylation of CpGs outside of CpG islands might correlate with tissue-specific transcriptional activity of genes or the great many non-canonical transcripts increasingly being recognized . Indeed, when we compare our measures of relative methylation of genomic clones on the chromosome 1 array and relative expression levels of constituent genes measured for the same tissue types in other individuals by Ge et al. , we find that the subsets of array clones with log2 values above or below our MAD-defined thresholds (implying relatively low or high methylation, respectively) tend to include genes with higher or lower relative expression levels, respectively. It is worth noting that the direction of this trend is similar to that found for promoters, not gene bodies, according to recent large-scale studies that found gene bodies to be more highly methylated in highly expressed genes than in inactive genes [31, 38, 50]. The regional methylation differences we observe here might correlate with the density of other marks of active/inactive chromatin as proposed by Eckhardt et al.  and/or reflect (or influence) regional compartmentalization within the nucleus . Alternatively, methylation of dispersed CpGs might serve to suppress spurious transcription from cryptic promoters (as proposed for Arabidopsis [54, 68]), raising the interesting question as to why the relative level of protection by methylation might vary among tissues.
In conclusion, the methylation landscapes that we have generated for various tissues suggest a complex pattern of, and possible function for, methylation differences outside of CpG islands. Future studies of regional methylation, regional transcriptional activity, and large-scale organization of the nucleus will help further our understanding of these regional epigenetic differences.
Materials and methods
Two sets of nine tissues derived from two phenotypically normal adult individuals, one female and one male, were provided by the NCI-funded Cooperative Human Tissue Network http://chtn.nci.nih.gov (CHTN-32364 and CHTN-32505, respectively).
Digestion, size-fractionation, and labeling of DNA samples
We used spleen as the reference in all tissue comparisons to keep the denominator for each clone's ratio roughly constant across our experiments. Spleen DNA was used as reference for the intra-individual tissue comparisons because it was the tissue for which DNA was most abundant for both donors. Phenol- and chloroform-extracted and then isopropanol-precipitated DNA derived from test and reference tissue samples were divided into tubes (20 to 60 μg per tube) and independently processed and analyzed. Each DNA sample was digested to completion with the methylation-sensitive restriction enzyme, Hpa II (New England Biolabs). We monitored completion of digestion by agarose gel electrophoresis (data not shown). The digested DNA was fractionated by size on 5% to 30% sucrose gradients by using a Beckman SW-40TI swing-out rotor as previously described . Fractions containing fragments smaller than 2.5 kb and greater than 80 bp as judged by gel electrophoresis were selected, pooled, and precipitated with isopropanol (Figure 1). The resulting DNA was analyzed by agarose gel electrophoresis to confirm that this process had effectively selected fragments 80 to 2,500 bp in length (data not shown). One microgram each of digested, size-selected test and reference DNA was differentially labeled with Cy3-dCTP (green) or Cy5-dCTP (red) (both GE Healthcare), respectively, by random priming using the Bioprime DNA Labeling System (Invitrogen).
Chromosome 1 tiling array hybridization
Preparation and validation of the human chromosome 1 genomic-clone microarray was previously described [55, 70]. Briefly, a total of 2,136 large-insert genomic clones were spotted in duplicate onto a glass slide. For simplicity, we will refer to this array as a 'BAC array', although the spotted clones include BACs, PACs, fosmids, and cosmids. Of these clones, 2,049 represent a tiling path covering 213 Mb, approximately 96% of the euchromatic regions of chromosome 1. Seventeen clones represent sparsely spaced segments of the X chromosome, with the first at 35 Mb and the last at 145 Mb with median spacing of approximately 6 Mb. These clones had been included on the array as controls for sex-mismatched conventional array CGH assays for other studies. The remaining 70 clones were included on the array as they were previously thought to map to chromosomes 1 or X, but are excluded from the results we report as their chromosomal mapping is now uncertain. Clone coordinates in the May 2004 assembly of the human genome sequence (Build 35) were provided by the Sanger Center and are available upon request. These coordinates typically imply clones longer than the spans shown on the NCBI or UCSC genome browsers due to trimming of sequences during genome assembly to eliminate overlapping sequence.
Labeled test and reference samples were mixed with 85 μg of Cot-1 DNA (Invitrogen) and hybridized to arrays under conditions described elsewhere . Each experiment was done in triplicate using aliquots of the DNA isolated from each tissue sample, but processed independently. The replicates were independently digested, size selected, labeled and hybridized, and paired with similarly processed spleen DNA from the same donor. Replicate arrays of a given tissue usually employed different independently processed replicates of the spleen reference DNA as well (see Additional file 1).
Array data analysis
Image acquisition was performed using an Axon 4000 B scanner and hybridization intensities were analyzed with the GenePixPro image analysis software (both Axon Instruments). For each spot, GenePixPro gave raw intensity values with the surrounding background subtracted for each wavelength scanned (635 nm and 532 nm for red and green channels, respectively). A custom R script, Normalization.r, was used to log2-transform these values, calculate the ratio of red to green fluorescence intensities, and apply a loess normalization procedure  based on GC content of the BACs (see below). A script then processed the output file to identify and eliminate clones whose duplicate spots reported significantly different normalized log2 ratios (that is, with standard deviation (SD) > 0.28) or had been flagged by GenePixPro as bad spots (that is, spots having less than 80% of pixels with intensities more than one SD above the background pixel intensity in either wavelength channel). We also eliminated results for subtelomeric clone RP5-857K21, as it appeared as an outlier on almost every array. The log2 ratios for the two spots for each remaining clone were averaged, and the median of all the clone averages was calculated and set to zero.
Our computer simulations of the method (see below and Results) indicated that relative fragment yield at a specified methylation level, as well as predicted intensity ratios for a pair of samples having different set methylation levels, vary with the GC content of the clones on the array. Therefore, we adjusted for GC content as follows: for each block of spots on each array, a loess curve was fitted to the relationship between the BACs' GC contents (in per cent) and their raw log2 ratios. The loess predicted value for each BAC's log2 ratio based on its GC content was subtracted from its real raw log ratio to give a normalized log ratio.
Using the GC-normalized data for each array, we then calculated the MAD for a contiguous set of 97 clones in 1q31–1q32 that contains relatively few Hpa II sites and showed little clone-to-clone variation in self-to-self comparisons and little deviation from the median in any of the sample comparisons. This group of clones was used in each array experiment as an internal measure of experimental noise to help distinguish potentially biologically meaningful deviations from experimental noise, which could vary from one array to another. This region of clones, whose midpoints are between positions 184,314,284 and 196,448,439 bp in chromosome 1's sequence in Build 35, is marked in each methylation profile with a bar marked 'MAD'. We used the MAD value as it is relatively robust to outliers and does not assume that values are normally distributed. The MAD value was calculated by subtracting the median log2 ratio for these selected clones from each clone's log2 ratio to give a deviation value; the MAD value is the median of the absolute values of these deviations. The SD of a normal distribution can be estimated to be 1.48 times the MAD value. We identified clones anywhere on chromosome 1 whose average log2 ratio values deviated below or above the overall median by >4.88 MAD units (that is, >3.29 times the SD), which would correspond to deviations with statistical significance at the predictive value of 0.001 based on a normal distribution with two-sided P value. In the self-to-self comparisons, we also used all clones to calculate the MAD and identify outliers beyond the P < 0.001 cutoff.
These outlying values, marked in each methylation profile, should include loci having true methylation differences between samples, since only 0.1% of the ratios drawn from a normal distribution (that is, two false-positive clones per array) are expected to deviate from the median by >4.88 MAD units. However, this threshold does not exclude all false positives due to experimental variation. The 97-clone region of 1q31–32 shows less experimental variation than the rest of the chromosome even in direct comparisons of the same tissue sample (for example, female medulla in Additional file 11). Only 0.5% of the ratios in this direct comparison deviate from the median by >4.88 MAD units when MAD is calculated over the entire array, whereas 2.2% do when MAD is calculated over the 97-clone region.
To control for potential genomic copy number differences between test and reference genomes, we also competitively hybridized 1 μg each of sonicated test and reference total DNA, mixed with 100 μg of Cot1 DNA (Invitrogen), to replicas of the same microarrays using the same conditions as described above . No significant deviations in copy number were observed across chromosome 1 in any tissue sample relative to the spleen sample from the same donor in these conventional array-based comparative genomic hybridization (array CGH) assays, except the female lung sample (data not shown). Because log2 ratios for the methylation-profiling array correlated very strongly with the wildly varying log2 ratios from the conventional CGH array for this sample (in isolation, each profile appeared to have large random noise), we eliminated this sample from further analysis. Genomic content abnormalities were noted for this lung sample in another study .
Note that it is not possible using this comparative method alone to determine what ratio represents an equivalent methylation state in test and reference DNA, as is possible with some sequencing-based methods (for example, [36, 49]). Thus, while a normalized log2 ratio of 0 (the median) might represent equivalent actual levels of methylation in a comparison of two normal tissue samples, it probably would not in a comparison of cells deficient in DNA methyltransferase activity and control cells.
In order to understand the response of our array to differently methylated genomes, we developed a PERL script, METHBATCHSIM, to simulate methylation states for a sequence of interest, such as a particular clone on the array, and the corresponding yield of fragments through our methylation-profiling procedure. First, we generated a list of Hpa II restriction sites on the sequence and used RepeatMasker  to generate a repeat-masked sequence file. METHBATCHSIM reads in the Hpa II and masked-sequence files and randomly designates which restriction sites in each clone contain methylated CpGs and which are unmethylated, for a specified overall methylation proportion. Specifically, we used the rbinom function of the R package  to randomly assign methylation status of each site, using the desired overall methylation fraction (between 0% and 100%, in steps of 1%) as the 'prob' parameter, the number of restriction sites in the BAC as the 'n' parameter, and the 'size' parameter set at 1. The sequence is then virtually digested with Hpa II based on the assigned methylation states, and we determine which of the resulting fragments are within the specified size range (80 to 2,500 bp). We simulated labeling by random priming by summing the total unmasked base pairs in all fragments meeting the size-range criterion. The final output summarizes these totals, averaged over 1,000 simulation runs. We repeated this simulation using the sequence of each clone on the array.
Other bioinformatic analyses
CpG island coordinates were taken from the UCSC Genome Browser http://genome.ucsc.edu CpG Island track, where CpG islands are defined as regions of >200 bp with GC content ≥ 50% and a ratio >0.6 of observed number of CG dinucleotides to expected number on the basis of the numbers of Gs and Cs in the segment. Gene coordinates were obtained from the UCSC's RefSeq track.
In order to test whether the relative methylation levels we measured relate to transcription levels, we compared our results with expression data obtained by Ge et al. , who hybridized RNA from each of 36 normal human tissues singly to Affymetrix oligonucleotide microarrays to determine expression levels of approximately 20,000 human genes. No RNA was available to study expression levels in samples from the same donors used for our methylation studies, but because methylation array ratios were highly correlated between the two donors we studied, it seemed reasonable to assume that methylation states would be similar in the tissues of donors used by Ge et al. and, therefore, reasonable to think that if any consistent correlation between regional methylation and expression levels exists, it might be observed even if different tissue donors are used for the methylation and expression arrays. We averaged methylation array ratios for the same tissue comparisons (that is, across sets of six arrays for each of cerebellum, heart and liver, and across sets of three arrays for each of lung, testis and ovary, compared with spleen in all cases). We used these cross-array averages to recalculate the MAD-based thresholds used to classify BACs as having outlying methylation array values.
Starting with Ge et al.'s raw probe-level expression-array data (GEO series GSE2361), we applied standard background adjustment, normalization and log-transformation methods using the robust multi-array average algorithm from Bioconductor's affy package  to obtain expression levels for each probe set (a probe set is a set of around 20 oligonucleotide probes that together represent a portion of a single gene). For each of the six tissue comparisons made (Figure 7, Additional file 10), we combined normalized expression results for the relevant tissue types to obtain log2 expression ratios. A table of Affymetrix probe set names, their corresponding gene symbols, and genomic coordinates of those genes in Build 35 of the human genome assembly was obtained from the SCGAP Hematopoietic Stem Cells Project . We then plotted the expression ratios of probe sets corresponding to genes whose 5' ends mapped in BACs of each three implied categories of implied relative methylation levels: 'High', 'Mid' or 'Low'. These conservative classifications correspond to BACs with ratios on the methylation-sensitive arrays that exceed MAD-based thresholds discussed above (that is, for Figure 7, 4.88 MAD units from the median, a threshold chosen such that we would expect approximately 0.05% of BACs on the array to fall in each of the Low or High categories by chance). In each boxplot, the median is indicated by the thick line, the box spans the middle two quartiles of the distribution, the small circles are outliers, and the whiskers extend to the most extreme data point which is no more than 1.5 times the length of the box away from the box.
In order to test the statistical significance of the observed differences in relative expression ratios between genes in BACs of different methylation ratio categories, we needed to account for the many-to-many relationships that exist between BACs, genes and probe sets. A BAC can contain more than one gene, a gene's 5' end can be present in more than one overlapping BAC, and some genes are represented on the expression array by more than one probe set (located in different parts of the gene). We therefore fitted generalized estimating equations to the data (using R's geeglm function from the geepack package ), treating datapoints from the same BAC as groups. We also gave each BAC-probe set datapoint a weight equal to the reciprocal of the number of times its gene symbol appeared in the table of all BAC-probe set pairs (that is, a gene represented by three probe sets and whose 5' end is in two overlapping BACs would receive a weight of 1/6 for each of its six datapoints). For each tissue comparison, we performed two statistical tests: (a) a test of whether expression ratios differ between BACs in the 'High' and 'Low' methylation categories (P values on square brackets, Figure 7 and Additional file 10); (b) a trend test for whether expression ratios are linearly related to methylation category, when methylation levels are coded as -1 (for low array ratio and 'High' implied relative methylation level), 0 ('Mid'), or 1 (for high array ratio and 'Low' implied relative methylation level) (P values on arrows, Figure 7 and Additional file 10). In each test, we used the 'Gaussian' family to model variation in expression ratios and 'independence' as the correlation structure.
PCR, cloning and sequencing of bisulfite-modified DNA
Total genomic DNA was extracted from male heart and spleen samples, and unmethylated cytosines were converted to uracil by bisulfite treatment according to published protocols . Briefly, each DNA sample was first digested with Bam H1, the enzyme was removed by phenol and chloroform extraction, and the DNA was subsequently ethanol precipitated and resuspended in H2O. NaOH was added to each DNA sample to a final concentration of 0.3 M, and the DNA was denatured for 2 min at 97°C followed by incubation for 30 min at 39°C. A solution of sodium bisulfite and hydroquinone was immediately added to 3.3 M and 0.67 mM final concentrations, respectively, and samples were incubated for 16 h at 55°C with 95°C spikes for 5 min every 3 h. Samples were desalted using QIAquick PCR Purification columns (Qiagen) according to the manufacturer's instructions. NaOH was added to each converted sample to a final concentration of 0.3 M, and the samples were incubated at 37°C for 15 min to remove excess sodium bisulfite. DNA samples were then ethanol precipitated and resuspended in 100 μl H2O.
Primers for selected regions were designed using the BiSearch Web Server  and are provided upon request. Each amplification was done with touchdown PCR using 2 μl Ex-Taq enzyme (Takara Mirus Bio) and 0.5 μl of 50 μM primers in a total volume of 25 μl with the following conditions: 95°C for 2 min, 30 cycles of 95°C for 30s, 66°C for 30s (stepping down 2°C every 2 cycles until 56°C), and 72°C for 1 min, followed by 72°C for 10 min.
Amplified products from converted DNA were gel-purified with QIAquick Gel Extraction columns (Qiagen) according to the manufacturer's instructions and subsequently cloned into the pCR2.1 plasmid vector using the TOPO TA Cloning Kit (Invitrogen). Clones with inserts were identified by PCR amplification using M13 reverse and T7 forward primers using TOPO cloning instructions. Amplified M13-T7 products were purified with Sephacryl S-300 (Amersham BioSciences). Seven to thirty-two clones were then sequenced successfully from each PCR product using the same primers on an ABI 3730 (Applied Biosystems). Sequences were aligned and analyzed using PhredPhrap and Consed [80–82].
All array data relevant to this manuscript have been submitted to the Gene Expression Omnibus (GEO) under accession number GSE12925.
- ATP2B4 :
Ca-transporting plasma membrane ATPase isoform 4
bacterial artificial chromosome
comparative genomic hybridization
median absolute deviation
P1-derived artificial chromosome
- RYR2 :
ryanodine cardiac receptor 2
We thank Sean McGrath and Arkadiusz Piotrowski for technical assistance with tissue and cell samples, Jeff Delrow and Cassie Neal in the FHCRC Genomics facility for technical assistance, and Peter Ellis for printing the BAC arrays. This work was supported in part by a graduate fellowship from the Department of Education, Universities and Research of the Basque Government (to CDB), a Ford Foundation Predoctoral Diversity Fellowship (to ER), an NIH training grant fellowship T32-CA09657 (to RKT), the Wellcome Trust (to CFL), the Howard Hughes Medical Institute (to EEE and SH), NIH RO1 HD0435679 (to EEE), the Swedish Cancer Society, Swedish Children's Cancer Foundation, and the US Army Medical Research and Materiel Command award No. W81XWH-04-1-0269 (to JPD), and the Fred Hutchinson Cancer Research Center. This manuscript is based on work completed while ER was employed at the University of Washington. The opinions expressed here are the author's own and do not reflect the views of the Department of Health and Human Services.
- Li E, Bestor TH, Jaenisch R: Targeted mutation of the DNA methyltransferase gene results in embryonic lethality. Cell. 1992, 69: 915-926. 10.1016/0092-8674(92)90611-F.View ArticlePubMedGoogle Scholar
- Goll MG, Bestor TH: Eukaryotic cytosine methyltransferases. Annu Rev Biochem. 2005, 74: 481-514. 10.1146/annurev.biochem.74.010904.153721.View ArticlePubMedGoogle Scholar
- Bird A: DNA methylation patterns and epigenetic memory. Genes Dev. 2002, 16: 6-21. 10.1101/gad.947102.View ArticlePubMedGoogle Scholar
- Pfeifer GP, Tanguay RL, Steigerwald SD, Riggs AD: In vivo footprint and methylation analysis by PCR-aided genomic sequencing: comparison of active and inactive X chromosomal DNA at the CpG island and promoter of human PGK-1. Genes Dev. 1990, 4: 1277-1287. 10.1101/gad.4.8.1277.View ArticlePubMedGoogle Scholar
- Li E, Beard C, Jaenisch R: Role for DNA methylation in genomic imprinting. Nature. 1993, 366: 362-365. 10.1038/366362a0.View ArticlePubMedGoogle Scholar
- Yoder JA, Walsh CP, Bestor TH: Cytosine methylation and the ecology of intragenomic parasites. Trends Genet. 1997, 13: 335-340. 10.1016/S0168-9525(97)01181-5.View ArticlePubMedGoogle Scholar
- Martienssen RA, Colot V: DNA methylation and epigenetic inheritance in plants and filamentous fungi. Science. 2001, 293: 1070-1074. 10.1126/science.293.5532.1070.View ArticlePubMedGoogle Scholar
- Fraga MF, Ballestar E, Paz MF, Ropero S, Setien F, Ballestar ML, Heine-Suner D, Cigudosa JC, Urioste M, Benitez J, Boix-Chornet M, Sanchez-Aguilera A, Ling C, Carlsson E, Poulsen P, Vaag A, Stephan Z, Spector TD, Wu YZ, Plass C, Esteller M: Epigenetic differences arise during the lifetime of monozygotic twins. Proc Natl Acad Sci USA. 2005, 102: 10604-10609. 10.1073/pnas.0500398102.PubMed CentralView ArticlePubMedGoogle Scholar
- Amir RE, Veyver Van den IB, Wan M, Tran CQ, Francke U, Zoghbi HY: Rett syndrome is caused by mutations in X-linked MECP2, encoding methyl-CpG-binding protein 2. Nat Genet. 1999, 23: 185-188. 10.1038/13810.View ArticlePubMedGoogle Scholar
- Brannan CI, Bartolomei MS: Mechanisms of genomic imprinting. Curr Opin Genet Dev. 1999, 9: 164-170. 10.1016/S0959-437X(99)80025-2.View ArticlePubMedGoogle Scholar
- Hansen RS, Wijmenga C, Luo P, Stanek AM, Canfield TK, Weemaes CM, Gartler SM: The DNMT3B DNA methyltransferase gene is mutated in the ICF immunodeficiency syndrome. Proc Natl Acad Sci USA. 1999, 96: 14412-14417. 10.1073/pnas.96.25.14412.PubMed CentralView ArticlePubMedGoogle Scholar
- Okano M, Bell DW, Haber DA, Li E: DNA methyltransferases Dnmt3a and Dnmt3b are essential for de novo methylation and mammalian development. Cell. 1999, 99: 247-257. 10.1016/S0092-8674(00)81656-6.View ArticlePubMedGoogle Scholar
- Nicholls RD, Knepper JL: Genome organization, function, and imprinting in Prader-Willi and Angelman syndromes. Annu Rev Genomics Hum Genet. 2001, 2: 153-175. 10.1146/annurev.genom.2.1.153.View ArticlePubMedGoogle Scholar
- Sparago A, Cerrato F, Vernucci M, Ferrero GB, Silengo MC, Riccio A: Microdeletions in the human H19 DMR result in loss of IGF2 imprinting and Beckwith-Wiedemann syndrome. Nat Genet. 2004, 36: 958-960. 10.1038/ng1410.View ArticlePubMedGoogle Scholar
- Paz MF, Wei S, Cigudosa JC, Rodriguez-Perales S, Peinado MA, Huang TH, Esteller M: Genetic unmasking of epigenetically silenced tumor suppressor genes in colon cancer cells deficient in DNA methyltransferases. Hum Mol Genet. 2003, 12: 2209-2219. 10.1093/hmg/ddg226.View ArticlePubMedGoogle Scholar
- Rideout WM, Eversole-Cire P, Spruck CH, Hustad CM, Coetzee GA, Gonzales FA, Jones PA: Progressive increases in the methylation status and heterochromatinization of the myoD CpG island during oncogenic transformation. Mol Cell Biol. 1994, 14: 6143-6152.PubMed CentralView ArticlePubMedGoogle Scholar
- Laird PW, Jaenisch R: The role of DNA methylation in cancer genetics and epigenetics. Annu Rev Genet. 1996, 30: 441-464. 10.1146/annurev.genet.30.1.441.View ArticlePubMedGoogle Scholar
- Esteller M: Cancer epigenomics: DNA methylomes and histone-modification maps. Nat Rev Genet. 2007, 8: 286-298. 10.1038/nrg2005.View ArticlePubMedGoogle Scholar
- Larsen F, Gundersen G, Lopez R, Prydz H: CpG islands as gene markers in the human genome. Genomics. 1992, 13: 1095-1107. 10.1016/0888-7543(92)90024-M.View ArticlePubMedGoogle Scholar
- Wang Y, Leung FC: An evaluation of new criteria for CpG islands in the human genome as gene markers. Bioinformatics. 2004, 20: 1170-1177. 10.1093/bioinformatics/bth059.View ArticlePubMedGoogle Scholar
- Ioshikhes IP, Zhang MQ: Large-scale human promoter mapping using CpG islands. Nat Genet. 2000, 26: 61-63. 10.1038/79189.View ArticlePubMedGoogle Scholar
- Prendergast GC, Ziff EB: Methylation-sensitive sequence-specific DNA binding by the c-Myc basic region. Science. 1991, 251: 186-189. 10.1126/science.1987636.View ArticlePubMedGoogle Scholar
- Ohlsson R, Renkawitz R, Lobanenkov V: CTCF is a uniquely versatile transcription regulator linked to epigenetics and disease. Trends Genet. 2001, 17: 520-527. 10.1016/S0168-9525(01)02366-6.View ArticlePubMedGoogle Scholar
- Chen C, Yang MC, Yang TP: Evidence that silencing of the HPRT promoter by DNA methylation is mediated by critical CpG sites. J Biol Chem. 2001, 276: 320-328. 10.1074/jbc.M007096200.View ArticlePubMedGoogle Scholar
- Jaenisch R, Bird A: Epigenetic regulation of gene expression: how the genome integrates intrinsic and environmental signals. Nat Genet. 2003, 33 (Suppl): 245-254. 10.1038/ng1089.View ArticlePubMedGoogle Scholar
- Weber M, Hellmann I, Stadler MB, Ramos L, Pääbo S, Rebhan M, Schübeler D: Distribution, silencing potential and evolutionary impact of promoter DNA methylation in the human genome. Nat Genet. 2007, 39: 457-466. 10.1038/ng1990.View ArticlePubMedGoogle Scholar
- Bird AP, Taggart MH, Nicholls RD, Higgs DR: Non-methylated CpG-rich islands at the human alpha-globin locus: implications for evolution of the alpha-globin pseudogene. EMBO J. 1987, 6: 999-1004.PubMed CentralPubMedGoogle Scholar
- Jones PA, Wolkowicz MJ, Rideout WM, Gonzales FA, Marziasz CM, Coetzee GA, Tapscott SJ: De novo methylation of the MyoD1 CpG island during the establishment of immortal cell lines. Proc Natl Acad Sci USA. 1990, 87: 6117-6121. 10.1073/pnas.87.16.6117.PubMed CentralView ArticlePubMedGoogle Scholar
- McKeon C, Ohkubo H, Pastan I, de Crombrugghe B: Unusual methylation pattern of the alpha 2 (l) collagen gene. Cell. 1982, 29: 203-210. 10.1016/0092-8674(82)90104-0.View ArticlePubMedGoogle Scholar
- Hellman A, Chess A: Gene body-specific methylation on the active X chromosome. Science. 2007, 315: 1141-1143. 10.1126/science.1136352.View ArticlePubMedGoogle Scholar
- Ball MP, Li JB, Gao Y, Lee JH, LeProust EM, Park IH, Xie B, Daley GQ, Church GM: Targeted and genome-scale strategies reveal gene-body methylation signatures in human cells. Nat Biotechnol. 2009, 27: 361-368. 10.1038/nbt.1533.PubMed CentralView ArticlePubMedGoogle Scholar
- Khulan B, Thompson RF, Ye K, Fazzari MJ, Suzuki M, Stasiek E, Figueroa ME, Glass JL, Chen Q, Montagna C, Hatchwell E, Selzer RR, Richmond TA, Green RD, Melnick A, Greally JM: Comparative isoschizomer profiling of cytosine methylation: the HELP assay. Genome Res. 2006, 16: 1046-1055. 10.1101/gr.5273806.PubMed CentralView ArticlePubMedGoogle Scholar
- Eckhardt F, Lewin J, Cortese R, Rakyan VK, Attwood J, Burger M, Burton J, Cox TV, Davies R, Down TA, Haefliger C, Horton R, Howe K, Jackson DK, Kunde J, Koenig C, Liddle J, Niblett D, Otto T, Pettett R, Seemann S, Thompson C, West T, Rogers J, Olek A, Berlin K, Beck S: DNA methylation profiling of human chromosomes 6, 20 and 22. Nat Genet. 2006, 38: 1378-1385. 10.1038/ng1909.PubMed CentralView ArticlePubMedGoogle Scholar
- Suzuki MM, Kerr AR, De Sousa D, Bird A: CpG methylation is targeted to transcription units in an invertebrate genome. Genome Res. 2007, 17: 625-631. 10.1101/gr.6163007.PubMed CentralView ArticlePubMedGoogle Scholar
- Weber M, Schübeler D: Genomic patterns of DNA methylation: targets and function of an epigenetic mark. Curr Opin Cell Biol. 2007, 19: 273-280. 10.1016/j.ceb.2007.04.011.View ArticlePubMedGoogle Scholar
- Meissner A, Mikkelsen TS, Gu H, Wernig M, Hanna J, Sivachenko A, Zhang X, Bernstein BE, Nusbaum C, Jaffe DB, Gnirke A, Jaenisch R, Lander ES: Genome-scale DNA methylation maps of pluripotent and differentiated cells. Nature. 2008, 454: 766-770.PubMed CentralPubMedGoogle Scholar
- Lister R, O'Malley RC, Tonti-Filippini J, Gregory BD, Berry CC, Millar AH, Ecker JR: Highly integrated single-base resolution maps of the epigenome in Arabidopsis. Cell. 2008, 133: 523-536. 10.1016/j.cell.2008.03.029.PubMed CentralView ArticlePubMedGoogle Scholar
- Deng J, Shoemaker R, Xie B, Gore A, LeProust EM, Antosiewicz-Bourget J, Egli D, Maherali N, Park IH, Yu J, Daley GQ, Eggan K, Hochedlinger K, Thomson J, Wang W, Gao Y, Zhang K: Targeted bisulfite sequencing reveals changes in DNA methylation associated with nuclear reprogramming. Nat Biotechnol. 2009, 27: 353-360. 10.1038/nbt.1530.PubMed CentralView ArticlePubMedGoogle Scholar
- Kawai J, Hirotsune S, Hirose K, Fushiki S, Watanabe S, Hayashizaki Y: Methylation profiles of genomic DNA of mouse developmental brain detected by restriction landmark genomic scanning (RLGS) method. Nucleic Acids Res. 1993, 21: 5604-5608. 10.1093/nar/21.24.5604.PubMed CentralView ArticlePubMedGoogle Scholar
- Ushijima T, Morimura K, Hosoya Y, Okonogi H, Tatematsu M, Sugimura T, Nagao M: Establishment of methylation-sensitive-representational difference analysis and isolation of hypo- and hypermethylated genomic fragments in mouse liver tumors. Proc Natl Acad Sci USA. 1997, 94: 2284-2289. 10.1073/pnas.94.6.2284.PubMed CentralView ArticlePubMedGoogle Scholar
- Gonzalgo ML, Liang G, Spruck CH, Zingg JM, Rideout WM, Jones PA: Identification and characterization of differentially methylated regions of genomic DNA by methylation-sensitive arbitrarily primed PCR. Cancer Res. 1997, 57: 594-599.PubMedGoogle Scholar
- Toyota M, Ho C, Ahuja N, Jair KW, Li Q, Ohe-Toyota M, Baylin SB, Issa JP: Identification of differentially methylated sequences in colorectal cancer by methylated CpG island amplification. Cancer Res. 1999, 59: 2307-2312.PubMedGoogle Scholar
- Yan PS, Chen CM, Shi H, Rahmatpanah F, Wei SH, Huang TH: Applications of CpG island microarrays for high-throughput analysis of DNA methylation. J Nutr. 2002, 132: 2430S-2434S.PubMedGoogle Scholar
- Chen CM, Chen HL, Hsiau TH, Hsiau AH, Shi H, Brock GJ, Wei SH, Caldwell CW, Yan PS, Huang TH: Methylation target array for rapid analysis of CpG island hypermethylation in multiple tissue genomes. Am J Pathol. 2003, 163: 37-45.PubMed CentralView ArticlePubMedGoogle Scholar
- Song F, Smith JF, Kimura MT, Morrow AD, Matsuyama T, Nagase H, Held WA: Association of tissue-specific differentially methylated regions (TDMs) with differential gene expression. Proc Natl Acad Sci USA. 2005, 102: 3336-3341. 10.1073/pnas.0408436102.PubMed CentralView ArticlePubMedGoogle Scholar
- Frigola J, Song J, Stirzaker C, Hinshelwood RA, Peinado MA, Clark SJ: Epigenetic remodeling in colorectal cancer results in coordinate gene suppression across an entire chromosome band. Nat Genet. 2006, 38: 540-549. 10.1038/ng1781.View ArticlePubMedGoogle Scholar
- Shen L, Kondo Y, Guo Y, Zhang J, Zhang L, Ahmed S, Shu J, Chen X, Waterland RA, Issa JP: Genome-wide profiling of DNA methylation reveals a class of normally methylated CpG island promoters. PLoS Genet. 2007, 3: 2023-2036. 10.1371/journal.pgen.0030181.View ArticlePubMedGoogle Scholar
- Sakamoto H, Suzuki M, Abe T, Hosoyama T, Himeno E, Tanaka S, Greally JM, Hattori N, Yagi S, Shiota K: Cell type-specific methylation profiles occurring disproportionately in CpG-less regions that delineate developmental similarity. Genes Cells. 2007, 12: 1123-1132. 10.1111/j.1365-2443.2007.01120.x.View ArticlePubMedGoogle Scholar
- Down TA, Rakyan VK, Turner DJ, Flicek P, Li H, Kulesha E, Graf S, Johnson N, Herrero J, Tomazou EM, Thorne NP, Backdahl L, Herberth M, Howe KL, Jackson DK, Miretti MM, Marioni JC, Birney E, Hubbard TJ, Durbin R, Tavare S, Beck S: A Bayesian deconvolution strategy for immunoprecipitation-based DNA methylome analysis. Nat Biotechnol. 2008, 26: 779-785. 10.1038/nbt1414.PubMed CentralView ArticlePubMedGoogle Scholar
- Rauch TA, Wu X, Zhong X, Riggs AD, Pfeifer GP: A human B cell methylome at 100-base pair resolution. Proc Natl Acad Sci USA. 2009, 106: 671-678. 10.1073/pnas.0812399106.PubMed CentralView ArticlePubMedGoogle Scholar
- Beck S, Rakyan VK: The methylome: approaches for global DNA methylation profiling. Trends Genet. 2008, 24: 231-237. 10.1016/j.tig.2008.01.006.View ArticlePubMedGoogle Scholar
- Shiota K, Kogo Y, Ohgane J, Imamura T, Urano A, Nishino K, Tanaka S, Hattori N: Epigenetic marks by DNA methylation specific to stem, germ and somatic cells in mice. Genes Cells. 2002, 7: 961-969. 10.1046/j.1365-2443.2002.00574.x.View ArticlePubMedGoogle Scholar
- Tompa R, McCallum CM, Delrow J, Henikoff JG, van Steensel B, Henikoff S: Genome-wide profiling of DNA methylation reveals transposon targets of CHROMOMETHYLASE3. Curr Biol. 2002, 12: 65-68. 10.1016/S0960-9822(01)00622-4.View ArticlePubMedGoogle Scholar
- Tran RK, Henikoff JG, Zilberman D, Ditt RF, Jacobsen SE, Henikoff S: DNA methylation profiling identifies CG methylation clusters in Arabidopsis genes. Curr Biol. 2005, 15: 154-159. 10.1016/j.cub.2005.01.008.View ArticlePubMedGoogle Scholar
- Buckley PG, Jarbo C, Menzel U, Mathiesen T, Scott C, Gregory SG, Langford CF, Dumanski JP: Comprehensive DNA copy number profiling of meningioma using a chromosome 1 tiling path microarray identifies novel candidate tumor suppressor loci. Cancer Res. 2005, 65: 2653-2661. 10.1158/0008-5472.CAN-04-3651.View ArticlePubMedGoogle Scholar
- RepeatMasker Open-3.0. [http://www.repeatmasker.org]
- Feinberg AP, Vogelstein B: A technique for radiolabeling DNA restriction endonuclease fragments to high specific activity. Anal Biochem. 1983, 132: 6-13. 10.1016/0003-2697(83)90418-9.View ArticlePubMedGoogle Scholar
- Ge X, Yamamoto S, Tsutsumi S, Midorikawa Y, Ihara S, Wang SM, Aburatani H: Interpreting expression profiles of cancers by genome-wide survey of breadth of expression in normal tissues. Genomics. 2005, 86: 127-141. 10.1016/j.ygeno.2005.04.008.View ArticlePubMedGoogle Scholar
- Laitinen PJ, Brown KM, Piippo K, Swan H, Devaney JM, Brahmbhatt B, Donarum EA, Marino M, Tiso N, Viitasalo M, Toivonen L, Stephan DA, Kontula K: Mutations of the cardiac ryanodine receptor (RyR2) gene in familial polymorphic ventricular tachycardia. Circulation. 2001, 103: 485-490.View ArticlePubMedGoogle Scholar
- Wehrens XH, Lehnart SE, Huang F, Vest JA, Reiken SR, Mohler PJ, Sun J, Guatimosim S, Song LS, Rosemblit N, D'Armiento JM, Napolitano C, Memmi M, Priori SG, Lederer WJ, Marks AR: FKBP12.6 deficiency and defective calcium release channel (ryanodine receptor) function linked to exercise-induced sudden cardiac death. Cell. 2003, 113: 829-840. 10.1016/S0092-8674(03)00434-3.View ArticlePubMedGoogle Scholar
- Morita H, Seidman J, Seidman CE: Genetic causes of human heart failure. J Clin Invest. 2005, 115: 518-526.PubMed CentralView ArticlePubMedGoogle Scholar
- Wu X, Chang B, Blair NS, Sargent M, York AJ, Robbins J, Shull GE, Molkentin JD: Plasma membrane Ca2+-ATPase isoform 4 antagonizes cardiac hypertrophy in association with calcineurin inhibition in rodents. J Clin Invest. 2009, 119: 976-985.PubMed CentralPubMedGoogle Scholar
- Ladd-Acosta C, Pevsner J, Sabunciyan S, Yolken RH, Webster MJ, Dinkins T, Callinan PA, Fan JB, Potash JB, Feinberg AP: DNA methylation signatures within the human brain. Am J Hum Genet. 2007, 81: 1304-1315. 10.1086/524110.PubMed CentralView ArticlePubMedGoogle Scholar
- Illingworth R, Kerr A, Desousa D, Jorgensen H, Ellis P, Stalker J, Jackson D, Clee C, Plumb R, Rogers J, Humphray S, Cox T, Langford C, Bird A: A novel CpG island set identifies tissue-specific methylation at developmental gene loci. PLoS Biol. 2008, 6: e22-10.1371/journal.pbio.0060022.PubMed CentralView ArticlePubMedGoogle Scholar
- Oakes CC, La Salle S, Smiraglia DJ, Robaire B, Trasler JM: A unique configuration of genome-wide DNA methylation patterns in the testis. Proc Natl Acad Sci USA. 2007, 104: 228-233. 10.1073/pnas.0607521104.PubMed CentralView ArticlePubMedGoogle Scholar
- Kapranov P, Cheng J, Dike S, Nix DA, Duttagupta R, Willingham AT, Stadler PF, Hertel J, Hackermuller J, Hofacker IL, Bell I, Cheung E, Drenkow J, Dumais E, Patel S, Helt G, Ganesh M, Ghosh S, Piccolboni A, Sementchenko V, Tammana H, Gingeras TR: RNA maps reveal new RNA classes and a possible function for pervasive transcription. Science. 2007, 316: 1484-1488. 10.1126/science.1138341.View ArticlePubMedGoogle Scholar
- Branco MR, Pombo A: Intermingling of chromosome territories in interphase suggests role in translocations and transcription-dependent associations. PLoS Biol. 2006, 4: e138-10.1371/journal.pbio.0040138.PubMed CentralView ArticlePubMedGoogle Scholar
- Zilberman D, Gehring M, Tran RK, Ballinger T, Henikoff S: Genome-wide analysis of Arabidopsis thaliana DNA methylation uncovers an interdependence between methylation and transcription. Nat Genet. 2007, 39: 61-69. 10.1038/ng1929.View ArticlePubMedGoogle Scholar
- Cooperative Human Tissue Network. [http://chtn.nci.nih.gov]
- Fiegler H, Carr P, Douglas EJ, Burford DC, Hunt S, Scott CE, Smith J, Vetrie D, Gorman P, Tomlinson IP, Carter NP: DNA microarrays for comparative genomic hybridization based on DOP-PCR amplification of BAC and PAC clones. Genes Chromosomes Cancer. 2003, 36: 361-374. 10.1002/gcc.10155.View ArticlePubMedGoogle Scholar
- Cleveland WS, Grosse E, Shyu WM: Local Regression Models. Statistical Models. Edited by: Chambers JM, Hastie TJ. 1992, Pacific Grove, CA: S. Wadsworth & Brooks/ColeGoogle Scholar
- Locke DP, Sharp AJ, McCarroll SA, McGrath SD, Newman TL, Cheng Z, Schwartz S, Albertson DG, Pinkel D, Altshuler DM, Eichler EE: Linkage disequilibrium and heritability of copy-number polymorphisms within duplicated regions of the human genome. Am J Hum Genet. 2006, 79: 275-290. 10.1086/505653.PubMed CentralView ArticlePubMedGoogle Scholar
- Ihaka R, Gentleman RR: R: a language for data analysis and graphics. J Comput Graph Stat. 1996, 5: 299-314. 10.2307/1390807.Google Scholar
- UCSC Genome Browser. [http://genome.ucsc.edu]
- Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U, Speed TP: Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics. 2003, 4: 249-264. 10.1093/biostatistics/4.2.249.View ArticlePubMedGoogle Scholar
- SCGAP Hematopoietic Stem Cells Project. [http://www.cbil.upenn.edu/SCGAP/resources.html]
- Halekoh U, Højsgaard S, Yan J: The R package geepack for generalized estimating equations. J Stat Softw. 2005, 15: 1-11.Google Scholar
- Lorincz MC, Schübeler D, Goeke SC, Walters M, Groudine M, Martin DI: Dynamic analysis of proviral induction and De Novo methylation: implications for a histone deacetylase-independent, methylation density-dependent mechanism of transcriptional repression. Mol Cell Biol. 2000, 20: 842-850. 10.1128/MCB.20.3.842-850.2000.PubMed CentralView ArticlePubMedGoogle Scholar
- Tusnady GE, Simon I, Varadi A, Aranyi T: BiSearch: primer-design and search tool for PCR on bisulfite-treated genomes. Nucleic Acids Res. 2005, 33: e9-10.1093/nar/gni012.PubMed CentralView ArticlePubMedGoogle Scholar
- Ewing B, Green P: Base-calling of automated sequencer traces using phred. II. Error probabilities. Genome Res. 1998, 8: 186-194.View ArticlePubMedGoogle Scholar
- Ewing B, Hillier L, Wendl MC, Green P: Base-calling of automated sequencer traces using phred. I. Accuracy assessment. Genome Res. 1998, 8: 175-185.View ArticlePubMedGoogle Scholar
- Gordon D, Abajian C, Green P: Consed: a graphical tool for sequence finishing. Genome Res. 1998, 8: 195-202.View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.