DNA Sequence Explains Seemingly Disordered Methylation Levels in Partially Methylated Domains of Mammalian Genomes

摘要

For the most part metazoan genomes are highly methylated and harbor only small regions with low or absent methylation. In contrast, partially methylated domains (PMDs), recently discovered in a variety of cell lines and tissues, do not fit this paradigm as they show partial methylation for large portions (20%–40%) of the genome. While in PMDs methylation levels are reduced on average, we found that at single CpG resolution, they show extensive variability along the genome outside of CpG islands and DNase I hypersensitive sites (DHS). Methylation levels range from 0% to 100% in a roughly uniform fashion with only little similarity between neighboring CpGs. A comparison of various PMD-containing methylomes showed that these seemingly disordered states of methylation are strongly conserved across cell types for virtually every PMD. Comparative sequence analysis suggests that DNA sequence is a major determinant of these methylation states. This is further substantiated by a purely sequence based model which can predict 31% (R~(2)) of the variation in methylation. The model revealed CpG density as the main driving feature promoting methylation, opposite to what has been shown for CpG islands, followed by various dinucleotides immediately flanking the CpG and a minor contribution from sequence preferences reflecting nucleosome positioning. Taken together we provide a reinterpretation for the nucleotide-specific methylation levels observed in PMDs, demonstrate their conservation across tissues and suggest that they are mainly determined by specific DNA sequence features. Author Summary Methylation is an essential DNA modification, which is attracting a lot of attention as a regulator of gene expression. Recent technological advances have allowed the genome-wide measurement of methylation at single-nucleotide resolution, leading to the discovery of several new types of methylation patterns. One prominent example are partially methylated domains (PMDs), which are regions with reduced average methylation, covering up to 40% of the genome. PMDs are found in only a subset of cell types, particularly in differentiated and cancer cells. An outstanding question is how methylation levels in PMDs are determined and how they can be interpreted at the single-nucleotide level. Here we provide a new model of methylation in PMDs. Single-nucleotide methylation levels in PMDs, albeit reduced on average, are highly variable along the genome. Furthermore, they are precisely set and can be predicted using DNA sequence features, establishing a new link between methylation and the underlying genetic information. This results in a high correlation of methylation levels in PMDs across different cell types. Our findings suggest that in any comparative analyses, PMDs should be analyzed as entities, strongly reducing the complexity of high-resolution DNA methylation analyses.