Base editors leverage RNA-guided, catalytically inactivated nucleases to target base-modifying enzymes to DNA or RNA. Their great success in correcting point mutations has stimulated the fast-moving field of base editing.

Examining base editors’ on-target and off-target sites. Credit: Marina Corral Spence/Springer Nature

The first DNA base editor was created by fusing the APOBEC1 cytidine deaminase to a dead Cas9, which enables C to U conversion; this class is now known as cytosine base editors (CBEs). Later there emerged adenine base editors (ABEs), which convert A to G, and RNA base editors.

Because they circumvent double strand breaks (DSBs), base editors are considered to be safer editing tools that eliminate undesired indels, translocation or rearrangements resulting from DSBs. Recent studies, however, have revealed that base editors may not be as safe as we had hoped. Sequencing studies performed on genomic DNA have confirmed that off-target mutations are induced by CBEs and ABEs.

Base editors’ specificity depends on the targeting nuclease complex and the fused deaminase. In addition to the off-target mutations induced by the nuclease complex, deaminase encounters with transient single-stranded DNA and RNA may also lead to off-target editing outside the editing window, proximally or distally.

The question then becomes how to detect off-targets efficiently and, more importantly, without bias. Whole-genome sequencing of genomic DNA offers a way to detect all types of off-targets in vitro. For example, modified Digenome-seq showed that CBEs, ABEs and unmodified Cas9 produce different off-target profiles, which underline the necessity of assessing genome-wide specificity (Nat. Biotechnol. 37, 430–435, 2019).

The question is further complicated in vivo because of genome background interference. A recent tool uses two-cell mouse embryos, one edited and the other one unedited, to infer the true mutations introduced by base editors (Science 364, 289–292, 2019). There is still a need, however, for tools that enable in vivo off-target detection at large scale. Additionally, computational tools that predict the off-target sites on the basis of sequence similarity or enzyme binding affinity will be helpful to guide sequencing studies. Comprehensive analysis of sequencing data to calculate off-targets would also benefit the community when evaluating the activity and specificity of emerging base editors.