21Genome editing using nucleases such as CRISPR-Cas induces programmable DNA damage at a 22 target genomic site but can also affect off-target sites. Here, we develop a powerful, sensitive assay 23 for the unbiased identification of off-target sites that we term DISCOVER-Seq. This approach 24 takes advantage of the recruitment of endogenous DNA repair factors for genome-wide 25 identification of Cas-induced double-strand breaks. One such factor, MRE11, is recruited precisely 26 to double-strand breaks, enabling molecular characterization of nuclease cut sites with single-base 27 resolution. DISCOVER-Seq detects off-targets in cellular models and in vivo upon adenoviral gene 28 editing of mouse livers, paving the way for real-time off-target discovery during therapeutic gene 29 editing. DISCOVER-Seq is furthermore applicable to multiple types of Cas nucleases and provides 30 an unprecedented view of events that precede repair of the affected sites. 31 strengths, they also have certain weaknesses. Naïve prediction algorithms are for the most part 41 based on sequence similarity and currently have limited predictive power with very high false-42 positive rates (10). Assays that induce DSBs in vitro, such as Digenome-Seq (5), CIRCLE-Seq (6) 43 and SITE-Seq (7), have high sensitivity but dramatically under-or overestimate the number of 44 target sites that are actually modified in cellular models or in vivo (11). Nuclease concentration 45 within the cell (7), delivery method (ribonucleoprotein (RNP) vs. plasmid) (7, 12, 13) as well as 46 more complex cellular properties such as chromatin accessibility (14, 15) have been shown to 47