MMEJ targeting improves CRISPR precision

CRISPR-Cas9 editing in embryos is widely used to generate genetically modified mice, but the unpredictability of double-strand break repair remains a major limitation. Repair is typically dominated by non-homologous end joining, leading to heterogeneous insertions and deletions across cells. In this study, the authors instead prioritised microhomology-mediated end joining (MMEJ), a repair pathway that produces more stereotyped deletions. Using the inDelphi machine-learning model, they selected guide RNAs predicted to favour MMEJ outcomes and validated these experimentally in mouse embryonic stem cells (mESCs) before embryo editing. Notably, mutation patterns observed in mESCs more closely matched those seen in blastocysts and post-implantation embryos than predictions alone, highlighting the value of this intermediate screening step.

Application of this pipeline to the Tyr and Fgf10 genes showed that guides with high predicted microhomology strength yielded dominant, recurrent deletion alleles and reduced allelic diversity. This translated into F0 animals with largely uniform genotypes and consistent phenotypes, including albino and limbless traits corresponding to the targeted loci. By contrast, guides with weaker microhomology signals produced more variable outcomes. The work also shows that editing efficiency can remain high while improving genotypic predictability, and that commonly used in silico tools alone are insufficient to capture in vivo repair dynamics. The authors argue that integrating prediction with empirical validation provides a practical route to more reproducible genome editing and may help reduce animal use by limiting the need for breeding to achieve uniform genotypes.

The study was led by Knut Woltjen and Masatsugu Ema at Kyoto University. It was published today 23 March 2026 in Communications Biology.