16Mammalian de novo DNA methyltransferases (DNMT) are responsible for the establishment of 17 cell-type-specific DNA methylation in healthy and diseased tissues. Through genome-wide 18 analysis of de novo methylation activity in murine stem cells we uncover that DNMT3A prefers 19 to methylate CpGs followed by cytosines or thymines, while DNMT3B predominantly methylates 20 CpGs followed by guanines or adenines. These signatures are further observed at non-CpG 21 sites, resembling methylation context observed in specialised cell types, including neurons and 22 oocytes. We further show that these preferences are not mediated by the differential recruitment 23 of the two de novo DNMTs to the genome but are resulting from structural differences in their 24 catalytic domains. Molecular dynamics simulations suggest that, in case of DNMT3A, the 25 preference is due to favourable polar interactions between the flexible Arg836 side chain and 26 the guanine that base-pairs with the cytosine following the CpG. This context-dependent de 27 novo DNA methylation provides additional insights into the complex regulation of methylation 28 patterns in different cell types. 29 34 replication, the maintenance methyltransferase DNMT1 ensures correct propagation of the 35 methyl mark (1, 2). Numerous genome-wide studies identified the exact position and tissue-36 specific dynamics of individual methyl groups on DNA. These revealed that the majority of CpGs 37 in mammalian genomes are fully methylated, with the exception of active promoters and cell-38 type-specific enhancer elements (3, 4). In addition to CpG methylation, non-CpG (or CpH) 39 methylation has been identified in numerous tissues (3, 5, 6), with highest levels found in brain 40 where it is suggested to potentially contribute to gene regulation and neuronal function through 41 readout by the methyl-CpG-binding protein MeCP2 (7-10).
43Nevertheless, the mechanisms governing the precise deposition of DNA methylation to 44 the genome remain to be fully understood. Although the mammalian genome is almost entirely 45 methylated (3, 4), several regions rely on active recruitment of de novo DNA methylation 46 activity, including promoters, enhancers and CpG islands (11-13), repetitive elements (14, 15),
47and transcribed gene bodies (16, 17). These sites also show tissue-specific variability of DNA 48 methylation and display altered methylation patterns in many diseases (4, 9,(18)(19)(20)(21). These 49 observations suggest that pathways that recruit de novo methylation vary from cell type to cell 50 type and that the genome-wide methylation patterns are a composite picture resulting from 51 combined activity of multiple DNMT targeting mechanisms and mechanisms that actively or 52 passively remove methylation. In previous work, we and others have shown that the de novo 53 DNMTs associate with genomic regions that are marked by distinct chromatin modifications, 54 resulting in enhanced deposition of methyl marks at these sites. For example, readout of 55 4 H3K36me3 by DNMT3B targe...