Detection and analysis of CpG sites with multimodal DNA methylation level distributions and their relationships with SNPs

DNA methylation levels at cytosine-phosphate-guanine (CpG) sites with multimodal distributions among different samples have been reported recently. One possible explanation for such variability is that genetic variants might affect epigenetic variation. One obvious case is that mutations such as single-nucleotide polymorphisms (SNPs) interrupt CpG sites, resulting in different DNA methylation levels for different genotypes. However, the relationship between genetic variations and epigenetic differences has not been studied thoroughly, partially because of the lack of powerful and robust methods to survey genome-wide CpG sites with multimodal methylation level distributions (mmCpGs). In this article, we develop a Gaussian mixture-model clustering (GMMC)–based approach to systematically detect all mmCpGs across the genome based on the GAW20 data set. In total, 3785 and 3847 mmCpGs have been identified in pre- and posttreatment data sets, respectively. Result analysis shows that approximately 68 to 70% of mmCpGs detected from unrelated individuals either have direct overlaps with SNPs or have associations with nearby SNPs, suggesting a strong correlation between SNPs and mmCpGs. Comparison with an existing approach illustrates that our GMMC-based method is more consistent when the number of samples decreases. In conclusion, mmCpGs may reveal important connections between genetics and epigenetics and they should be carefully identified and evaluated.