Total number of T-DMRs and DS-DMRs
MeDIP methylation analysis in conjunction with NimbleGen promoter + CpGi microarrays were used to identify adult tissue specific differentially methylated regions (T-DMRs) and developmental stage specific DMRs (DS-DMRs). We used relatively conservative criteria to identify methylation peaks requiring both independently derived biological replicates to contain the same or overlapping methylation peaks. Other investigators have indicated that MeDIP may miss differentially methylated regions in regions of low CpG density . Thus we may underestimate the total number of T-DMRs and DS-DMRs to some extent. Nevertheless, our study identified almost 5,000 adult T-DMRs and 10,000 DS-DMRs that represent a total of over 15,000 T-DMRs in four adult tissues (brain, heart, liver, and testis; Tables 1 and 2). Since tissues are made up of many cell types and it is likely that there are cell type specific epigenetic differences, the methylation peaks we identified probably correspond to the major cell types in the tissue, while many methylation peaks corresponding to minor cell types would be missed. Considering that we only sampled 4 tissues and 3 developmental stages (E15, NB, AD), and that the analysis does not include methylation peaks in regions outside promoter and CpGi regions (except in tiling regions), it is quite likely that there are many additional T-DMRs and DS-DMRs that were not identified in this study. These results demonstrate that there are numerous DNA methylation differences between adult tissues (T-DMRs) and within the same tissue at different stages of development (DS-DMRs), indicating that alterations in DNA methylation are a major feature of development.
Genomic locations of T-DMRs and DS-DMRs
Overall most T-DMRs and DS-DMRs (about 70%) are located within non-CpGi promoter regions (Table 1 and 2), which is consistent with results of other studies [12, 28, 30, 33, 39]. However, we also found that about 30% of T-DMRs (1,440/4,686) and DS-DMRs (3,014/9,336) are located in CpGi regions (i.e. the sum of all CpGi-associated DMRs vs. the total number of DMRs; Table 1 and 2). This corresponds to more than 15% of the 15,979 CpGi regions on the NimbleGen array and is a relatively large number considering the limited number of tissues and developmental stages surveyed and other factors cited above. In fact, among the 15,979 CpGi regions annotated in UCSC mm8 freeze, a total of 5,523 (34.6%) were partially or entirely methylated in one or more of the 24 samples analyzed in this study (4 tissues each with duplicated samples for 3 developmental stages; data not shown). Thus, tissue- and developmental stage-specific CpGi methylation may be a very common event during development.
Analysis of methylation in the few tiling regions available on the NimbleGen array made it possible to determine whether there was significant tissue specific methylation outside of promoter and CpG island regions. Our results indicate that as much as 30% of the T-DMRs are missed by restricting analysis to promoter and CpGi regions (Additional file 3b).
The location of methylation peaks differs according to tissue distribution. About 50% of the adult common methylation sites are within CpGi regions whereas only 30% of the T-DMRs are located within CpGi regions (Table 1). Similarly, about 48% of the methylation peaks that are common among all developmental stages within a tissue are located in CpGi regions (Table 2
). The significance of this difference between common methylation and differential methylation is not presently clear. Possibly this reflects a difference between transient changes in methylation that are tissue specific and more stable changes in methylation that are common to tissues.
The distribution of the location of T-DMRs and DS-DMRs in testis is dramatically different from somatic tissues. Almost 94% of the adult testis T-DMRs are associated with non-CpGi promoter regions (Table 1 and 2). About 68% of the testis DS-DMRs are associated with non-CpGi promoter regions, which represents an average of E15, NB, and AD developmental stages. Unlike somatic tissues there appears to be a clear pattern shift in the distribution of testis DS-DMRs during development. In E15 testis only about 37% of the DS-DMRs are located in non-CpGi promoters whereas the percentage increases in NB (49%) and AD (86%). This may reflect post-natal onset of meiosis and other developmental changes. This general bias of DMRs towards non-CpGi promoter vs. CpGi promoter is significant since the majority (10,915 or 68.5%) of 19,528 promoters (all RefSeqs promoters included in the array) are CpGi promoters (based on our criteria region of 1kb flanking each side of transcription start site).
T-DMRs correspond to sites of methylation that vary between tissues, whereas DS-DMRs correspond to sites of methylation within a single tissue that differ according to developmental stage. We determined the extent of overlap between DS-DMRs from different tissues by combining all DS-DMRs for the same tissue and excluding common methylation sites within a tissue (Table 3 and Additional file 5a and 5b). We reasoned that the degree of overlap of DS-DMRs among tissues would shed light on the level of tissue specificity of DNA methylation during development. We found a very low level of overlap among different tissues (1,927/9,784 or 13.2%), indicating that most DS-DMRs are tissue-unique and tissue-specific (Table 3). Even for DMRs that are shared among all three developmental stages in a tissue (Additional file 5b), almost 50% are tissue unique and almost all are tissue specific (95.2%). In addition, a large fraction of the tissue specific DS-DMRs were unique to a single tissue (7,857/9,784 or 80%). However, we would expect this percentage to decrease as more tissues are added to the analysis. These results suggest that almost all DNA methylation in non-repetitive regions is tissue specific. We have previously shown, using RLGS, that methylation of some genomic regions containing repetitive sequences is also tissue-specific .
Recent studies [22, 25] identified 16,379 T-DMRs in four human adult tissues (brain, liver, spleen, and colon) using a method termed "Comprehensive High-throughput Arrays for Relative Methylation" or CHARM. Although the methods, species and tissues used for this analysis were different from those presented here, the total number of T-DMRs is surprisingly similar (15,271 vs. 16,379), emphasizing extensive tissue specific DNA differences in DNA methylation. However, the CHARM analysis of human tissues found that 76% of the T-DMRs were located within 2 kb of CpGi regions that were denoted CpGi shores. The array design used in our study limits analysis to promoter, CpGi regions, and very limited tiled genomic regions and would appear to exclude most CpGi shores. However, we reason that many CpGi shores may be within the promoter regions in our studies. For both the promoter and tiling regions, we observed a slightly higher density of DS-DMRs in CpGi shores than in CpGi regions (data not shown). Since at the genome scale, the CpGi shore region is much larger (~6X) than the CpGi region, we can expect to have a larger number of DS-DMRs in the CpGi shores than in the CpGi, supporting the conclusion of Irizarry et al  in principle that there are more T-DMRs in CpGi shores.
Gene ontology of T-DMRs and DS-DMRs
Gene ontology (GO) analysis was performed to identify any common theme among the identified T-DMRs. In this analysis, we divided all DMRs associated with genes into three groups (non-CpGi promoter, CpGi promoter, and intragenic CpGi), since as noted by others [37, 38], methylation in different gene locations may impact the gene expression regulation differently. We found more or less similar enrichment categories for different tissues, especially among somatic tissues. For both T-DMRs and DS-DMRs, the most consistent GO enrichment pattern is seen among those in non-CpGi promoters (Additional file 11 and Additional file 12) with enrichment for "membrane proteins, G-proteins, olfactory proteins" among UnMe-AD (DS-DMRs) and "extracellular space/region" for both T-DMRs and DS-DMRS (Me-AD). Apparently, these enrichment patterns reflect a pattern that is observed for all genes with non-CpGi promoters (data not shown). Additional GO enrichment was observed for somatic DS-DMRs in intragenic CpGi regions on "regulation of biological and cellular processes, ion-binding and transport". This data supports the notion that intra-genic CpGi methylation tends to promote the expression of the involved gene [46, 47], and may participate in gene regulation during differentiation and development.
"Demethylation" of previously methylated sites (DS-DMRs) is a common feature of tissue differentiation
Quite surprisingly, our studies indicate that many DS-DMRs that are methylated at early stages of development (E15 and NB) are unmethylated in adult tissues (Table 2 and Figure 2). This is particularly evident in brain and liver where there are almost 3 times as many DS-DMRs that become "unmethylated" in adult as become "methylated" in adult. This is somewhat contrary to expectations that differentiation into adult tissues would reflect promoter methylation and silencing of genes not associated with the final gene expression pattern. In contrast, it suggests that the final gene expression pattern depends on extensive demethylation events during differentiation. Although methylation of gene promoter regions is associated with gene repression, gene body methylation is associated with gene expression [37, 38, 48]. In addition, recent reports indicate that there are extensive allelic differences in gene expression in human that result from allelic differences in methylation, due to imprinting or DNA sequence variation [49, 50]. Thus, demethylation of previously methylated sites could reflect either increased or decreased gene expression, depending on site location. Our observations that demethylation is a common feature of tissue differentiation are consistent with the recent report that liver development in human is characterized predominantly by demethylation .
Methylation of ES cells
ES cells are pluripotent and are derived from the blastomeres of the early embryo that are thought to be extensively demethylated due to active and passive demethylation that follows fertilization. Therefore, we decided to compare the methylation peaks from ES cells with those found in E15 embryonic tissues (brain, heart, liver, and testis) to determine the extent of methylation differences during early stages of tissue differentiation (Additional file 1). We found that the number of E15 T-DMRs (1,769) is less than half of that found in adult (4,686), which is consistent with E15 tissues being generally less methylated than adult. ES cells also had a low level of methylation with 981 T-DMRs. However, somewhat surprisingly, ES cells had more methylation peaks (981, Additional file 1) than any of the E15 tissues. This would suggest that many genomic sites that are methylated in ES cells become demethylated during early development. We previously found that many (60%) of the adult T-DMRs identified by RLGS were methylated in ES cells, suggesting these were demethylated during tissue differentiation . It is also possible that some or many of the sites that are methylated in ES cells are due to growth in tissue culture. It has previously been shown that tumor cells grown in tissue culture accumulate excessive aberrant methylation that is unrelated to tumorigenesis . Also, growth of mouse neural progenitor cells in culture after extended passages results in aberrant methylation  and investigators found extensive differences in the genomic methylation patterns of independently isolated human ES cell lines . At the present time, it is unclear whether sites of methylation in ES cells is aberrant due to extended growth in culture or whether demethylation during early differentiation in ES cells is an important process as in later stages of tissue differentiation noted above.
Methylation analysis of stem cells revealed extensive Cytosine methylation in a non-CpG context [35, 48, 51] that was mainly located within gene bodies . The non-CpG methylation appears to be mostly confined to stem cells and disappears upon differentiation. Our analysis of mouse ES cells indicated a very low level of intragenic CpGi methylation. This suggests that most of the non-CpG gene body methylation was in non-CpGi intragenic regions that were not on the NimbleGen array or that our analysis did not resolve non-CpG methylation.
Non-Promoter and non-CpGi methylation in the Hoxa gene cluster
A 100 Kb region on chromosome 6 that includes the Hoxa gene cluster is essentially a tiling array on the NimbleGen promoter plus CpGi array. This region includes 16 RefSeq genes and 22 UCSC-annotated CpGi regions (Additional file 6). Somewhat surprisingly all 8 methylation peaks were in non-CpGi regions and only one was in a promoter region (Additional file 6 Additional file 7). In addition, developmental analysis of Hoxa gene methylation indicated stage-specific methylation differences (Figure 3). These differentially methylated regions include 3' exons and intron regions. Since these regions are highly conserved and Hox genes are known to have important roles in development [41, 42], these results suggest that methylation in non-promoter, non-CpGi regions may have novel, currently undefined roles in regulating development. A recent report also indicates differential development-associated methylation within Hox gene clusters in human .
DNA methylation and alternative promoter use
Interrogation of the MeDIP/NimbleGen array data suggests that DNA methylation may be associated with alternative promoter use. Analysis of methylation within the Protocadherin gene clusters on chromosome 18 indicates extensive methylation within CpGi promoter and non-CpGi promoter regions, particularly within somatic tissues (Additional file 8 Additional file 9). Some differential tissue specific methylation in this region is also noted. Pcdha mRNAs are synthesized by the activation of one of the alternative promoters on only one of the two chromosomes resulting in monoallelic expression . Although the mechanistic basis for this selection is unknown, it is hypothesized that it provides a foundation for neuron adhesive diversity that is required for complex synaptic interactions . Recent genome-wide methylation analysis using CHARM  also indicates an association between tissue specific DNA methylation and alternative transcripts. That study indicated that most tissue specific differentially methylated regions were located in CpGi shores and that 68% of the shores were within 500 bp of alternative promoter sites.
The impact of T-DMRs and DS-DMRs on gene expression
To understand the biological function of T-DMRs and DS-DMRs, one obvious approach is to examine their impact on gene expression. Efforts to address this question at a genome scale is complicated by several factors that can obscure correlations.
First, one gene may be subjected to DNA methylation in multiple regions with the T-DMR or DS-DMR in question being just one of those. Therefore, the level of gene expression may depend on methylation status in other regions of the gene. Second, existence of multiple splicing isoforms, particularly those associated with alternative promoters, as well as the use of multiple expression probes for the same gene makes this one-gene vs. one DMR association analysis very challenging. Last but not least, DNA methylation is not the only factor affecting the gene expression. In other words, the lack of DNA methylation in one of the promoter region does not necessarily confer gene activation, as the expression can be limited by other factors, for instance the lack of the required transcriptional factor(s). These complications may be responsible for the highly diverse situations we observed between the occurrence of T-DMRs/DS-DMRs and the expression level of their associated genes on a gene-by-gene basis. Despite these complications, our preliminary analysis does seem to reveal a few novel insights. First, it appears that T-DMRs are associated with lower gene expression in non-CpGi promoter regions. Second, there may be some differences between T-DMRs and DS-DMRs from earlier developmental stages in the manner they impact gene expression. Certainly, extensive data analysis using more data sets representing more tissues, as well as experimental validation is needed to confirm these observations.