The main outcome of efficient CRISPR-Cas9 cleavage in the chromosome of bacteria is cell death. This can be conveniently used to eliminate specific genotypes from a mixed population of bacteria, which can be achieved both in vitro, e.g. to select mutants, or in vivo as an antimicrobial strategy. The efficiency with which Cas9 kills bacteria has been observed to be quite variable depending on the specific target sequence, but little is known about the sequence determinants and mechanisms involved. Here we performed a genome-wide screen of Cas9 cleavage in the chromosome of E. coli to determine the efficiency with which each guide RNA kills the cell. Surprisingly we observed a large-scale pattern where guides targeting some regions of the chromosome are more rapidly depleted than others. Unexpectedly, this pattern arises from the influence of degrading specific chromosomal regions on the copy number of the plasmid carrying the guide RNA library. After taking this effect into account, it is possible to train a neural network to predict Cas9 efficiency based on the target sequence. We show that our model learns different features than previous models trained on Eukaryotic CRISPR-Cas9 knockout libraries. Our results highlight the need for specific models to design efficient CRISPR-Cas9 tools in bacteria.

Genome-wide CRISPR-Cas9 screen in E. coli identifies design rules for efficient targeting