DNA-base lesions cause cancer and propagate into the genome. We use in-protein QM/MM calculations to study the repair of etheno-bridged adenine (epsilon A) by the iron(IV)-oxo species of AlkB enzymes. Recent experimental investigations, using mass-spectrometry and in crystallo isolation, suggested that epsilon A was repaired by formation of an epoxide (epsilon A-ep) that further transforms to a glycol (epsilon A-gl), ending finally in adenine and glyoxal. Theory reproduces the experimentally observed barrier for the rate-determining step and its pH dependence. However, as we show, the mass-spectrometrically identified species are side-byproducts unassociated with the repair mechanism. The repair is mediated by a zwitterionic species, of the same molecular mass as the epoxide, which transforms to an intermediate that matches the in crystallo trapped species in structure and mass, but is NOT the assumed epsilon A-gl iron-glycol complex. Verifiable/falsifiable predictions, regarding the key protein residues, follow. The paper underscores the indispensable role of theory by providing atomistic descriptions of this vital mechanism, and guiding further experimental investigations.