NEW YORK – Researchers in South Korea have developed computational tools that they said can predict the efficiency and outcome product frequencies of adenine base editors (ABEs) and cytosine base editors (CBEs).
In a study published on Monday in Nature Biotechnology, the researchers noted that while ABEs and CBEs are widely used to induce point mutations, determining whether a specific nucleotide can be edited can require time-consuming experiments. Further, when the editable window contains multiple target nucleotides, the editing process can generate a variety of genotypic products.
To develop computational tools that predict the efficiency and outcome product frequencies of base editing platforms, the researchers first evaluated the efficiencies of an ABE and a CBE, and the outcome product frequencies in human cells at thousands of target sequences. They found only modest asymmetric correlations between the activities of the base editors and Cas9 at the same targets. Using deep-learning-based computational modeling, the investigators then built tools to predict the efficiencies and outcome frequencies of ABE- and CBE-directed editing at any target sequence, in order to facilitate modeling and therapeutic correction of genetic diseases by base editing.
In order to assess base-editing activity in a high-throughput manner, the researchers used a lentiviral library of 15,656 guide RNA-encoding and target sequence pairs that they had previously used for high-throughput evaluation of Cas9 activities. Among the 15,656 sequence pairs in the cell library, 13,504 contained at least one target adenine and 14,157 contained at least one target cytosine. The cell library was transiently transfected with plasmids encoding either ABE7.10 or BE4.
Five days after the plasmid transfection, the researchers evaluated the conversion efficiencies and editing outcomes at the appropriate target nucleotides using deep sequencing, and observed a robust correlation between the results of the replicates for both ABE and CBE. In subsequent experiments, they also found strong correlations between ABE and CBE efficiencies at integrated target sequences and those at the endogenous sites.
When they compared the activities of Cas9 and those of either ABE or CBE at the same target sequences, the researchers found modest positive correlations — Cas9 activities were high in most cases when ABE or CBE activities were high, but ABE or CBE activities were variable when Cas9 activities were high.
The researchers next attempted to develop computational models that predict the efficiencies and outcomes of base editing. Using deep-learning frameworks and training data sets, they first developed computational models that predicted ABE- and CBE-induced base-editing efficiencies, defined as the percentage of DNA copies containing base-edited sequences within the wide editable window regardless of the number of base-edited nucleotides among the total analyzed DNA copies.
Four models, ABE_proportion and CBE_proportion as well as DeepABE and DeepCBE, showed high performance in all replicates across tested cell types, the researchers said. However, the absolute efficiencies of base editing varied depending on the cell type. When the team fine-tuned DeepCBE considering chromatin accessibility, the performance of this model, named DeepCBE-CA, was similar to that of DeepCBE. The same was true for DeepABE.
Finally, they applied the computational models they developed to predict the base-editing efficiencies and outcomes for the modeling and correction of human disease-relevant point mutations reported in ClinVar. They observed high correlations between predicted versus measured outcome proportions and significant correlations between predicted versus measured efficiencies in both modeling and correction, although the absolute efficiencies of base editing were generally lower than the predicted values.
"These results suggest that our predictions will be useful for predicting outcome proportions and relative efficiencies of disease modeling and therapy in other cell types in addition to HEK293T cells, although the absolute efficiencies of base editing are variable depending on the experimental conditions including the cell type and transfection efficiencies," the authors concluded. "We expect that these computational models, together with the information about how sequence context affects ABE and CBE activity, will greatly facilitate genome editing using ABE and CBE."
The researchers have provided web tools for DeepABE and DeepCBE and codes for incorporation of DeepABE and DeepCBE into existing tools here.