Here we demonstrate a comprehensive methylation map of human gastric tissue at high resolution. Our data provide a global view of a mammalian methylome, along with several intriguing findings, some of which are novel and worth further investigation. First, we found that hypermethylation of CGIs in promoters is an important epigenetic feature that dictates gene expression changes in cancer. Second, hypermethylation of the 5'-end of coding exons arises in cancer and appears to play an important role in cancer progression. Third, cancer-induced methylation changes in younger repetitive elements and LRES have potential clinical implications in terms of early detection and therapeutic design.
Among the genes analyzed in this study was MDM2, which encodes an important negative regulator of p53. MDM2 and p53 are known to regulate one another through a feedback loop . MDM2 overexpression is frequently detected in many human cancers, suggesting that MDM2 overexpression may be one of the common features of tumorigenesis. In this study, we showed that the upstream region of MDM2 is hypermethylated in most cancer samples. We also found that, in some cancer samples, hypomethylation occurred along with MDM2 amplification at the same site, suggesting that there is major dysregulation of the MDM2-mediated pathway at both the genetic and epigenetic levels. This appears to cause aberrant early tumor cell development and subsequently cancer.
HOX genes cluster on chromosomes 2, 7, 12, and 17, and they are frequently inactivated by CpG hypermethylation in several cancers [40, 45, 46]. Accordingly, we found that many HOX genes were hypermethylated, indicating that HOX gene clusters may be general targets of epigenetic alterations during tumorigenesis. However, hypomethylation of a few HOX genes was also detected in gastric cancer. Therefore, the methylation of HOX genes may be regulated in a tissue-specific manner in cancer. In addition, the regulation of HOX expression is significantly correlated with histone modifications and interaction with Polycomb group genes [47, 48]. Our data point to another possible mode of regulating HOX gene expression-- DNA hypermethylation of the promoter region of histone genes such as H2B. Further investigation of histone genes might offer new insights for cancer studies.
DNA hypomethylation of repetitive elements as a major contributor to genome size is one of common features in cancer and the methylation changes in SINE, LINE, or LTR-retrotransposon with possess transcriptional activity are critical for cellular functions. A number of SINEs close to CpG islands retain a high proportion of CpG sites and frequently hypomethylated in cancer. Because younger Alu elements are usually closer to active chromatin regions , the hypomethylation of them has more biological significance than that of older Alu elements. Additionally, these hypomethylation of repeat elements, such as Alu and LINE1 might also affect the inactivation of X chromosome .
The findings in this study are based on an intuitive and efficient normalization method for comparative analysis. Unlike bisulfite sequencing, MIRA and MeDIP simulate the in vivo behavior of methyl-CpG binding domain (MBD) proteins, which recognize both the methylation level and concentration of individual CpG sites . For example, it has been shown that MBD binding is not sufficient for gene repression at low CpG densities, even when individual sites are highly methylated. In this case, the common practice is to use MeDIP or MIRA outputs as measures for the functional consequences of methylation of all CpGs in a given region [26, 30, 32, 38, 51–53]. However, a few studies have attempted to use the spatial density of CpGs to normalize the experimental readout of MeDIP-chip [25, 54]. Therefore, the MES normalization used here provides several advantages over other methods. First, using logarithmic transformation, we can scale down raw read counts four orders of magnitude to obtain MESs. This provides mathematical benefits such as variance stabilization. Second, MES normalization allows us to use the correct background distribution. As a probability function for the number of events in a given time interval, a Poisson distribution can be used to assess the statistical significance of read counts in a given genomic interval. However, we found that our MES index, particularly when normalized with local methods rather than raw read counts, better illustrates the nature of the Poisson event. Third and most importantly, it enables comparative analysis of independent samples. In the normalization step, direct subtraction of MESbg proves to be an efficient correction method for background sequencing.
Although the methods used here have clear advantages, the biological and technical limitations of these methods should also be mentioned. Since our methods are based on affinity purification, methylation changes and karyotypic alterations cannot be distinguished. However, this can be overcome by comparing normal and cancer genomes following measurement of background enrichment. Thus, this comparative analysis scheme should be of value for future clinical epigenomics investigations.