A discontinuous DNA glycosylase domain in a family of enzymes that excise 5-methylcytosine.

DNA cytosine methylation (5-meC) is a widespread epigenetic mark associated to gene silencing. In plants, DEMETER-LIKE (DML) proteins typified by Arabidopsis REPRESSOR OF SILENCING 1 (ROS1) initiate active DNA demethylation by catalyzing 5-meC excision. DML proteins belong to the HhH-GPD superfamily, the largest and most functionally diverse group of DNA glycosylases, but the molecular properties that underlie their capacity to specifically recognize and excise 5-meC are largely unknown. We have found that sequence similarity to HhH-GPD enzymes in DML proteins is actually distributed over two non-contiguous segments connected by a predicted disordered region. We used homology-based modeling to locate candidate residues important for ROS1 function in both segments, and tested our predictions by site-specific mutagenesis. We found that amino acids T606 and D611 are essential for ROS1 DNA glycosylase activity, whereas mutations in either of two aromatic residues (F589 and Y1028) reverse the characteristic ROS1 preference for 5-meC over T. We also found evidence suggesting that ROS1 uses Q607 to flip out 5-meC, while the contiguous N608 residue contributes to sequence-context specificity. In addition to providing novel insights into the molecular basis of 5-meC excision, our results reveal that ROS1 and its DML homologs possess a discontinuous catalytic domain that is unprecedented among known DNA glycosylases.