Try before you buy: Empirical comparison of base editing approaches

Main text

Inherited retinal disorders (IRDs) represent a heterogeneous group of diseases caused by defects in over 300 genes that ultimately lead to dysfunction and loss of retinal cells.¹ The first retinal gene therapy approval in 2017 for RPE65-associated retinal dystrophy via viral vector (voretigene neparvovec) represented a major milestone in the field, providing hope for more approvals. Alas, 7 years later it remains the only approved gene therapy for IRDs, although several promising clinical trials are ongoing. Even so, the gene augmentation approach is not possible for many disease-causing genes due to limitations in adeno-associated vector cargo capacity, dominant negative alleles, or genes with multiple functional isoforms, such as the IRD-associated gene CRB1. Base editing has emerged as a particularly promising alternative to directly edit causative variants and avoid some of the above pitfalls of gene augmentation. However, many pathogenic variants are not amenable to base editing, and even those that are have a wide range of editing efficiencies. As reported in this issue of Molecular Therapy, Tachida et al.² address this challenge by empirically validating the efficiency and specificity of predicted base editors and their target templates in USH2A. Their results highlight the utility of in vitro methods, such as expression plasmids and Sanger sequencing for initial screening, and caution against sole reliance on in silico modeling. This impactful study provides a template approach that can be used to identify top editing candidates in disease genes, which can then be tested further for therapeutic development.

Since the first description of CRISPR-Cas9-mediated genome editing in 2013,^3,4 its applications have rapidly expanded from template-based repair to single-nucleotide editing and beyond. In recent years, gene editing has become a clinical reality, in an impressively rapid transition from initial discovery to clinical trials, including for CEP290-related retinal dystrophy.⁵ The retina continues to be a favorable target tissue for gene therapy approaches, given the relative accessibility and compartmentalization of the subretinal space, and the huge impact of low vision on quality of life. Among numerous reports of in-silico-predicted outcomes to correct pathogenic variants, two focus on the CRB1 gene, which is associated with IRD.^6,7 However, neither of these studies, nor others previously reported in the literature for other target tissues, test the in vitro or in vivo accuracy of in silico predictions. Although the growing suite of gene editing technology may eventually offer customizable solutions for each genomic target, currently not every variant is equally amenable. Out of thousands of rare disease genes, each with hundreds of deleterious variants, how do we pick which to target, and by what method? Tachida et al. do just this by performing both computational and empirical analyses of pathogenic variants in the USH2A gene, which cause retinitis pigmentosa and sensorineural hearing loss (Usher syndrome type 2).

The authors first narrowed the 389 total USH2A variants linked to retinal disease to just 35 by screening in silico for a usable protospacer adjacent motif (PAM) and the target variant lying within the editing window. For these variants, the researchers performed in vitro comparison of editing efficiency with combinations of adenine and cytosine base editors (ABEs, CBEs) using 51 base pair fragments, with each fragment representing one of the 35 potential therapeutic targets in USH2A. Editing efficiency was compared in transiently transfected target fragments versus stably integrated fragments, the latter of which proved more favorable. Editing windows and bystander edits were also compared for the base editor-PAM combinations, as were proportions of “productive” edits (where only the intended amino acid is edited) to total edits. This identified multiple variants in USH2A that were amenable to highly efficient editing with low indels. The top 4 editing candidates were then evaluated further for expression of full-length USH2A (usherin) protein after base editing. Finally, favorable in vivo editing efficiency was achieved in an engineered mouse model bearing one of these promising variants.

When Tachida et al. compared their empirical observations of base editing efficiency in the selected 35 USH2A variants to in silico predictions based on BE-HIVE algorithms, the investigators found moderate agreement. This correlation was relatively higher for sites with higher empirical editing efficiency and for ABEs compared with CBEs. The results of this study highlight the importance of empirical confirmation to best match therapeutic targets with editing reagents and narrow down favorable base editing candidates. Thus, despite impressive progress in computational prediction algorithms, there may still be a role for old-fashioned “try before you buy” benchtop testing for the foreseeable future.

Given the vast number of rare disease alleles, reliable, streamlined systems to narrow down the best gene editing approaches will maximize clinical applicability and success. With rapidly evolving base editing and prime editing technologies, likely no single empirical “recipe” will stay relevant for long. For instance, the PAM constraints applied in this report can already be circumvented using novel Cas9 designs.^8,9 Moreover, prime editing or other future methods may be less susceptible to unpredictable variability in editing efficiency, and thus not require empirical validation. However, validation of functional protein production could still be important, especially in the absence of patient-derived cell lines.

Each disease gene and its variants present a unique challenge, with factors such as disease mechanism, pathophysiology, and genotype-phenotype correlations coloring therapeutic strategy, even within the retinal gene therapy space. Ideally, an individualized approach for gene therapy would be decided and applied for each patient using tools previously validated through empirical evidence such as in vitro studies, animal models, or even patient-derived human retinal organoids.¹⁰ The work described by Tachida et al.² provides a clear roadmap for benchtop testing to boost confidence in therapeutic strategies before investing in subsequent development, which could apply broadly to other genes and editing methods.

Acknowledgments

This research was supported in part by an unrestricted grant to the Department of Ophthalmology at the Keck School of Medicine of USC from Research to Prevent Blindness, New York, NY, USA (to A.N.), a National Eye Institute Career Development Award K08EY030924 (to A.N.), Las Madrinas Endowment in Experimental Therapeutics for Ophthalmology (to A.N.), and a Knights Templar Eye Foundation Endowment (to A.N.).

Declaration of interests

C.A. has no relevant financial interests to declare. A.N. holds equity in CRISPR Therapeutics and serves as a consultant for Biogen, Epicrispr, AAVantgarde, Novartis, Atsena, Eyebiotech, Ultragenyx, Alia Tx, Blue Gen, BlueRock, Endogena, and J&J.