NEW YORK (GenomeWeb) – A team led by researchers at the Broad Institute and Brigham and Women's Hospital have developed a machine learning model that can predict insertions and deletions from CRISPR-Cas9 genome editing with high accuracy, showing that template-free Cas9 editing is capable of repair to a predicted genotype.
As they reported this week in Nature, the Broad's David Liu and David Gifford, Brigham and Women's Richard Sherwood, and their colleagues were looking to see if they could upend the conventional wisdom that CRISPR-Cas9 editing randomly generates insertions and deletions in a gene unless a repair template is used along with it.
Non-homologous end joining (NHEJ) and microhomology-mediated end joining (MMEJ) processes are major pathways involved in the repair of Cas9-mediated double-stranded breaks that can result in highly heterogeneous repair outcomes comprising hundreds of repair genotypes. Although end-joining repair of Cas9-mediated double-stranded DNA breaks has been harnessed to facilitate knock-in of DNA templates or deletion of intervening sequence between two cleavage sites, NHEJ and MMEJ are not generally considered useful for precision genome editing applications, the researchers noted.
They developed a high-throughput Streptococcus pyogenes Cas9 (SpCas9)-mediated repair outcome assay to characterize end-joining repair products at Cas9-induced double-stranded breaks using 1,872 target sites based on sequence characteristics of the human genome.
They then used the resulting library of guide RNAs to train a machine learning model they named inDelphi to predict genotypes and frequencies of 1- to 60-base-pair deletions and 1-base-pair insertions with high accuracy in five human and mouse cell lines. InDelphi predicted that 5 percent to 11 percent of Cas9 guide RNAs targeting the human genome would be what the researchers termed "precise-50" — yielding a single genotype comprising greater than or equal to 50 percent of all major editing products.
Through various experiments, they confirmed precise-50 insertions and deletions in 195 human disease-relevant alleles, including correction in primary patient-derived fibroblasts of pathogenic alleles to wild-type genotype for Hermansky–Pudlak syndrome, which causes blood clotting deficiency and albinism, and Menkes disease, which results in copper deficiency.
They used inDelphi to design 14 gRNAs for high-precision template-free editing yielding predictable 1-bp insertion genotypes in endogenous human disease-relevant loci and experimentally confirmed highly precise editing in two human cell lines. "We used inDelphi to reveal human pathogenic alleles that are candidates for efficient and precise template-free gain-of-function genotypic correction and achieved template-free correction of 183 pathogenic human microduplication alleles to the wild-type genotype in ≥50 percent of all editing products," the authors wrote. "Finally, we integrate these developments to achieve high-precision correction of five pathogenic low-density lipoprotein receptor (LDLR) microduplication alleles in human and mouse cells, as well as correction of endogenous pathogenic microduplication alleles for Hermansky–Pudlak syndrome (HPS1) and Menkes disease (ATP7A) to the wild-type sequence in primary patient-derived fibroblasts."
Importantly, the researchers concluded that the ability to predict Cas9-mediated products could enable new precision genome editing research applications and facilitate existing applications, such as performing efficient bi-allelic gene knockout and predicting end-joining by-products of HDR.
"Moreover," they added, "we present evidence that suppressing NHEJ augments repair of pathogenic microduplication alleles, suggesting that temporary manipulation of DNA repair pathways could be combined with Cas9-mediated editing to favor specific editing genotypes with high precision." The investigators also believe that if inDelphi is given appropriate training data, it could also learn to accurately predict repair genotypes from other designer nucleases.