DNA methylation by de novo DNA methyltransferases 3A (DNMT3A) and 3B (DNMT3B) at cytosines is essential for genome regulation and development. Dysregulation of this process is implicated in various diseases, notably cancer. However, the mechanisms underlying DNMT3 substrate recognition and enzymatic specificity remain elusive. Here we report a 2.65-ångström crystal structure of the DNMT3A-DNMT3L-DNA complex in which two DNMT3A monomers simultaneously attack two cytosine-phosphate-guanine (CpG) dinucleotides, with the target sites separated by 14 base pairs within the same DNA duplex. The DNMT3A-DNA interaction involves a target recognition domain, a catalytic loop, and DNMT3A homodimeric interface. Arg836 of the target recognition domain makes crucial contacts with CpG, ensuring DNMT3A enzymatic preference towards CpG sites in cells. Haematological cancer-associated somatic mutations of the substrate-binding residues decrease DNMT3A activity, induce CpG hypomethylation, and promote transformation of haematopoietic cells. Together, our study reveals the mechanistic basis for DNMT3A-mediated DNA methylation and establishes its aetiological link to human disease.

Mammalian DNA methylation patterns are established by two de novo DNA methyltransferases, DNMT3A and DNMT3B, which exhibit both redundant and distinctive methylation activities. However, the related molecular basis remains undetermined. Through comprehensive structural, enzymology and cellular characterization of DNMT3A and DNMT3B, we here report a multi-layered substrate-recognition mechanism underpinning their divergent genomic methylation activities. A hydrogen bond in the catalytic loop of DNMT3B causes a lower CpG specificity than DNMT3A, while the interplay of target recognition domain and homodimeric interface fine-tunes the distinct target selection between the two enzymes, with Lysine 777 of DNMT3B acting as a unique sensor of the +1 flanking base. The divergent substrate preference between DNMT3A and DNMT3B provides an explanation for site-specific epigenomic alterations seen in ICF syndrome with DNMT3B mutations. Together, this study reveals distinctive substrate-readout mechanisms of the two DNMT3 enzymes, implicative of their differential roles during development and pathogenesis.