Paper of the Month February 2019

Predictable and precise template-free CRISPR editing of pathogenic variants

DNA repair upon Cas9 cleavage without a donor template is a heterogeneous process. Non-homologous end joining (NHEJ) and microhomology-mediated end joining (MMEJ) are major pathways involved in the repair of double-strand breaks following CRISPR-Cas9, characterized by a template-free mechanism. This article illustrates a new method able to predict the cut-sites of genotypic products resulting from double-strand breaks originated from Cas9 enzymatic process.

In order to characterize Cas9-induced double strand-strand breaks in DNA through end-joining repair processes, a high-throughput method used Streptococcus pyogenes (SpCas9) targeting single genotypes allowed to predict the frequencies of the substantial majority of template-free Cas9-induced insertions and deletions events in a range of repaired products. A library containing comparable 1872 gRNAs (lib-A) was designed to target these genomic sites in mouse embryonic stem cells (mESCs) and human U2OS cells facilitating the design of a machine learning algorithm named inDelphi that is able to predict MH, MH-less and 1-bp insertions. inDelphi is able to discriminate MH (simulates MMEJ pathway) from MH-less deletions (resembled by NHEJ), by attributing a phi score based on GC content (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1538862/) and length of neural network. inDelphi learned that strong microhomologies are longer and richer in GC bases as well as MH-less frequencies decrease as length increases. It was revealed a distribution of deletions (MH), constituting between 63 and 87% of all edited products, from which 39-58% containing microhomology-less (MH-less) deletions and also single-base (1bp) insertions ranging between 30 to 9% in mESCs and U2OS cells, respectively. This discriminates the competitive process between MMEJ and NHEJ and therefore predicts the deletion genotype frequencies in both MH and MH-less deletions. k-value predicts 1-bp insertions and found that the majority are duplications in the -4 position (from PAM site) and the precision is higher when an A or T base is present in that position (lower %GC). By defining a gRNA based on precision-X, it’s possible to select a number of gRNAs able to produce a single genotypic outcome in a X percentage of all the editing outcomes. 14 gRNAs predicted to induce 1-bp insertions in 40% of edited products were delivered to U2OS cells and 10 out 14 showed significant high precision when comparing with baseline. Moreover, 10 gRNAs were also found to show high-precision deletions.

To predict template-free correction of pathogenic alleles, above 1500 pathogenic human loci with high predicted rates of frame correction or microduplication correction to the wild-type were selected for targeting with another library (lib-B). 183 pathogenic alleles from microduplication were corrected to wild-type and 508 from frameshift to in-frame in more than 50% of edited products in mESCs, comparably correlated with U2OS cells. Similar approach was taken for 4 NHEJ-deficient conditions, where MH deletions increased, enabling pathogenic alleles repair to wild-type with significant higher precision, raising the numbers to 286 for more than 50% and 153 for more than 70% from all the edited products, respectively. 

LDLR, which is involved in hypercholesterolemia, was conferred with 87% of correction of microduplications for wild-type in a template-free NHEJ mechanism and constitutes one of the pathogenic alleles that might be addressed with this strategy. To assess this, the authors delivered Cas9 and a gRNA designed from inDelphi specific to each of the pathogenic allele, leading to correction of the five microduplicated alleles to wild-type, restoring the LDL uptake in 79% of mESCs. Delivery of Cas9 and an inDelphi gRNA specific to the pathogenic microduplication allele in HPS1 gene achieved high-efficiency correction (88%) to the wild-type in primary cells of patients with Hermansky-Pudlak syndrome as well as for ATP7A (94%) in Menkes disease.

Overall, this paper presents inDelphi as a method that is able to select gRNAs with significantly more precise editing than the conventional population of gRNAs and the findings show great potential in using template-free on therapeutic settings, particularly when pathogenic microduplication alleles are involved.