Prime editing requires two main components: the prime editors that consist of a reverse transcriptase (RT) fused to the H840A SpCas9 nickase, and a pegRNA. The H840A mutation in the conventional SpCas9 inactivates its HNH domain generating a Cas9 nickase, which cleaves only one strand (the PAM containing strand) of the DNA instead of causing DSBs. The pegRNA is designed to extend the 3′ end of the single guide RNA (sgRNA) with a RT template (RTT) and a primer binding sequence (PBS). Thus, from the 5′ to the 3′ ends, the pegRNA contains the spacer, sgRNA scaffold, RTT, and primer binding sequence (PBS) (Figure 1B). An overview of the PE mechanism is illustrated in Figure 2. In summary, the spacer anneals with its complementary sequence, directing the Cas9 nickase to nick the PAM-containing strand of the DNA at a specific locus in the genome. The PBS then hybridizes with the 3′ end of this nicked DNA allowing the RT to carry out the reverse transcription to extend the nicked DNA according to the RTT sequence that carries the intended mutations. This process will create a 3′-flap (the newly synthesized) or a 5′-flap (the original unedited) in this locus. Since the 5′ flap is more susceptible to excision by endonucleases, such as FENI (Liu et al., 2004; Anzalone et al., 2019), the edited strand is more likely to be incorporated in the genome. The excision of the 5′ flap leads to the heteroduplex formation, which is followed by endogenous DNA mismatch repair mechanisms. Finally, the replacement of the original sequence (unedited strand) incorporates the desired mutation at the target site.

Prime editing is still in its infancy, and further studies are necessary to evaluate its full potential. Herein, we will highlight the current applications and some key limitations of the PE system.

The primary in vitro experiments in HEK293T cells showed a high efficiency and flexibly of PE, which was capable of installing different types of editions with high on-targeting rates and low off-targeting edits (Anzalone et al., 2019). Similarly, by using PE system, Schene et al. installed deletions and point mutations in patient-derived intestinal and ductal liver organoids with high editing efficiency (30–50%) and low undesired editing rates (Schene et al., 2020). However, additional studies demonstrated relatively lower rates of editing in hiPSCs, embryos, animals and plants, suggesting that there is large variability in the editing efficiency of this technique (Butt et al., 2020; Hua et al., 2020; Lin et al., 2020; Liu et al., 2020; Sürün et al., 2020). Recently, modifying and incorporating additional nuclear localization signals (NLS) at the N-terminus and C-terminus of the prime editor was found to increase the efficiency of genome editing and should be considered an important parameter in PE design (Liu et al., 2021). In addition, using two pegRNAs in trans to install the same modification into the target site, while also overexpressing the pegRNA, has shown to significantly improve the PE efficiency in plants (Jiang et al., 2020; Lin et al., 2021).

The cell type, position in the genome of the target gene, and characteristics of the prime editor—including the number of nucleotides that constitutes the PBS and RTT—all play a critical role in the efficiency of PE (Anzalone et al., 2019). Similarly to the observations made by Anzalone et al. in HEK293 cells, the PBS length showed to critically influence the editing efficiency of PE in hiPSCs (Sürün et al., 2020). In plants, those parameters also showed to be critical. In particular, designing the PBS sequence with a melting temperature of 30°C significantly increases the editing efficiency in rice (Lin et al., 2021). Taken together, these studies demonstrate that optimizing the design of the PE system is crucial and should be performed for each experimental condition. As such, this step can be time and labor consuming. With that in mind, Kim et al., 2021 developed a high-throughput screening method, using a lentivirus library to evaluate the efficiency of the PE2 system (Kim et al., 2021). Although a very interesting approach, the editing efficiency is calculated based on the edition installed on the plasmids, which may not reflect the genomic editing rate that can be complicated by the genomic DNA accessibility. Further studies are necessary to improve the screening method of the PE machinery, including methods that better reflect the in vivo efficiency of the selected PE system on the cell type of interest.

Since PE does not induce DSBs, it is expected to have a lower indel rate at the target locus in comparison to NHEJ and HDR. In fact, Anzalone et al. tested this assumption in vitro (Anzalone et al., 2019). On the other hand, high levels of unexpected outcomes were installed by the double nicking of PE3 in mouse zygotes (Aida et al., 2020), and a higher frequency of unwanted mutations at target loci were also induced by PE in mice (Liu et al., 2020). On-targeting rates and indel formation are both usually evaluated by next generation sequencing (NGS), followed by bioinformatics analyses, generating individual reads of the region of interest for quantification of on-targeting and indels rates. The parameters used for normalization as well as the number of output reads can significantly affect the calculations of those rates. As such, further in vivo studies exploring indel formation at the on-targeting position, including the testing of different bioinformatics parameters, are required.

In terms of off-targeting ratio, researchers frequently identify susceptible off-targeting sites and perform the analysis using techniques such as NGS (Jang et al., 2021), circularization for in vitro reporting of cleavage effects by sequencing (CIRCLE-seq) (Levy et al., 2020; Suh et al., 2021), and nickase-based Digenome sequencing (nDigenome seq) (Kim et al., 2020). With that, the off-targeting rates can be often underestimated since only few predicted off-targeting sites are studied. Therefore, it is still necessary to evaluate the in vivo PE off-targeting frequency in a genome-wide respect.

Until now, only a few studies exploring PE in vivo have been conducted with its efficiency notably lower than it is in vitro (Liu et al., 2020; Liu et al., 2021; Gao et al., 2021). Due to the large size of PE machinery, it’s in vivo delivery to the target site can be a limitation that significantly impacts the editing efficiency of this technique. Recently, Liu et al. tested a dual adeno associated virus (AAV)-mediated delivery of a split-intein prime editor and showed its applicability for in vivo gene editing in the mouse liver (Liu et al., 2021). However, this strategy still requires the use of dual-AAV vectors, which causes concern regarding the expression of undesired truncated products and low efficiency. By using dual vectors system, both vectors must reach the target site at the same time in order to guarantee a successful outcome.

Prime editing is a cutting-edge technique that holds great promise to the advance of genome engineering. Future research should focus on applying this technique in a plethora of human disease-relevant cell types, organoids and animal models to support the clinical potential of PE in treating human genetic diseases.

CRISPR/Cas systems are a promising avenue for the treatment of IRDs. Specifically, given the eye’s unique immune-privileged nature, this organ presents an opportunity for CRISPR/Cas-mediated treatment of IRDs with limited systemic effects (Benhar et al., 2012). The significant progress being made in this field is exemplified by the ongoing phase I/II clinical trial for EDIT-101 (AGN-151587), a treatment for Leber congenital amaurosis (LCA) type 10 (Accessed on September 2021: https://clinicaltrials.gov/ct2/show/NCT03872479). EDIT-101 removes the aberrant splice donor created by the CEP290 IVS26 c.2991 + 1655 A > G mutation in the CEP290 gene through AAV5-mediated delivery of dual gRNAs in conjunction with the Cas9 ortholog from Staphylococcus aureus (Maeder et al., 2019). However, the development of DSB-independent CRISPR/Cas systems—such as PE and BE—that significantly reduce off-targeting effects as well as the introduction of indels at the editing site, will be a significant future step for IRD therapeutics. Here, we provide a brief update on DSB-independent retinal therapeutics. For a detailed overview, Gallego et al. discuss CRISPR/Cas-based gene editing approaches for IRDs (Gallego et al., 2020).

A recent retrospective analysis conducted by Fry et al. examined the prevalence of single nucleotide pathogenic variants with the potential to be corrected by BE. Specifically, this study identified 6 autosomal recessively inherited genes associated with IRDs; namely, ABCA4, CEP290, CDH23, EYS, MYO7A, and USH2A whose long coding sequence prevents them from being corrected with a classical single AAV-mediated gene augmentation strategy but have variants amenable for correction by BE (Fry et al., 2021). Prime editing, due to its potential to rectify 89% of pathogenic genetic variants, would further expand the therapeutic editing possibilities for the aforementioned genes (Anzalone et al., 2019). Base editing has been applied for the treatment of retinal degeneration 12 (rd12) mice, a representative model of humans with RPE65 mutations (Suh et al., 2021). Here, Suh et al. corrected the homozygous C > T nonsense mutation in exon3 of the Rpe65 gene, finding as high as 29% editing efficiency, through lentiviral-mediated delivery of an adenine base editor (ABE) and sgRNA. Importantly, they found minimal indels or off-target mutations and the mice had restored RPE65 expression. They additionally found the recovery of retinoid isomerase activity after BE, with a substantial increase in 11-cis-retinal in treated eyes leading to functional visual recovery as measured by electroretinography (ERG), optomotor responses (OMRs) and visually evoked potentials (VEPs) (Suh et al., 2021).

Split-intein ABEs and cytosine base editors (CBEs) have also been developed and showed favorable transduction efficiency when delivered by dual AAVs (either the evolved PHP.B or Anc80 AAV capsids) to the retina (Levy et al., 2020). Rhodopsin-Cre mice were crossed with Ai9 mice to generate mice that expressed tdTomato only in rod cells. These mice were injected at 2-weeks-old with the AAV-ABE and AAV-CBE constructs to target the Dnmt1 locus, in addition co-injection with their corresponding reporter constructs PHP.B-CBh-GFP–KASH or Anc80-CBh-GFP–KASH with the nuclear membrane-localized Klarsicht/ANC-1/Syne-1 (KASH) homology driven by chicken-beta hybrid (CBh) promoters. At 3-weeks post-injection, efficiency was determined by assessing editing in sorted cells. The authors found 48 ± 5.9% C•G-to-T•A editing with PHP.B-CBE and 37 ± 22% A•T-to-G•C editing with Anc80-ABE in GFP⁺/tdTomato⁺ transduced rod photoreceptors. However, while ABE delivery led to the generation of minimal indels in retinal cells, CBE delivery to retinal cells generated substantial indels of up to 34% (Levy et al., 2020). Interestingly, there was minimal overlap between base-edited and indel-containing alleles. One possibility proposed by the authors is that CBE-mediated indels may occur at a higher rate in retinal cells due to mutual exclusivity between uracil excision pathways and those pathways required for CBE-mediated editing outcomes (Levy et al., 2020). Recently, AAV8-mediated delivery of split-PEs has also been shown to successful edit the Dnmt1 locus in the mouse retina using a CMV promoter (Zhi et al., 2021). In this study, Zhi et al. sub-retinally co-injected their split-intein PE along with an AAV8-CMV-GFP reporter at 6 weeks of age and found expression limited to photoreceptors and retinal pigment epithelium (RPE). At 6 weeks post-injection, genomic DNA of mouse retina was collected (no cell sorting). This revealed an average editing efficiency of 1.71 ± 1.35% and average indels of 0.17 ± 0.01% in Dnmt1 locus (Zhi et al., 2021). Similarly, a preprint by Jang et al. showed an editing efficiency of 1.87% in the Atp7b locus in transduced mouse retina (no cell sorting) and no detectable indels were found. Jang and others used a trans-splicing AAV8 vector, which allows the expression of a single transcript encoded by two independent vectors coadministered to the same tissue to deliver the PE to the retina via intravitreal injection, in conjunction with an additional AAV8 construct to deliver the pegRNA and sgRNA (Jang et al., 2021).

Lastly, the rd10 mouse model mimics autosomal recessive retinitis pigmentosa (RP) and is caused by the Pde6b ^rd10 c.1678C > T (p.Arg560Cys) mutation. To date, neither the PE nor the BE approach has been used to correct the rd10 model. However, Vagni and others illustrated the amenability of the rd10 model to in vivo treatment using CRISPR/Cas9-mediated HDR (Vagni et al., 2019). Their findings revealed a higher visual acuity compared to the untreated eye 3 months post gene editing (Vagni et al., 2019). Recently, our laboratory demonstrated the applicability of PE for the successful installation and correction of the Pde6b ^rd10 c.1678C > T mutation in vitro using the Neuro-2a (N2a) mouse neuroblastoma cell line (Tsai et al., 2021). We hope this proof-of-concept work will pave the way for future in vivo studies on the applicability of PE for IRDs.

Taken together, this information continues to point towards the advancement of CRISPR genome editing techniques and their application in the field of ophthalmology. As we continue down this path, BE and PE approaches will ultimately be at the forefront of ophthalmic gene therapy.