Gene editing treatment strategies for retinitis pigmentosa assessed in Xenopus laevis carrying a mutant Rhodopsin allele

Aim: To examine the utility of gene editing therapies for retinitis pigmentosa using Xenopus laevis carrying a mutation in Rhodopsin . Methods: Xenopus laevis were genetically modified using CRISPR-Cas9 based methods and characterized by Sanger sequencing, dot blot, electroretinography, and confocal microscopy. Results: We identified genetically modified Xenopus laevis carrying a net 12 base pair deletion in the Rho.L gene. These animals have a retinal degeneration that is apparent by 14 days, with abnormal or missing rod outer segments, and a reduced electroretinogram signal. We prevented the majority of this retinal degeneration via a treatment strategy using a single sgRNA to neutralize the mutant allele via non-homologous end joining, yielding long-term improvements in histology and the electroretinogram. A second strategy using two sgRNAs to generate large deletions in the mutant allele was also successful, but did not significantly improve outcomes relative to the single-guide strategy as it was less efficient. We found limited evidence of success with a third strategy dependent on homology-directed repair; this treatment was also too inefficient to generate an outcome superior to the single-guide strategy. Conclusion: Our results demonstrate the utility of this new Xenopus laevis model for rapidly assessing and comparing multiple gene-editing based treatment strategies. We conclude that it would be technically difficult to improve on the simple single-guide based strategy, as strategies requiring multiple successive events (such as cleavage followed by homology-directed repair) are likely to be less efficient. with anti-rod opsin antibodies mabB630N or mab514-18 (at 1:10 dilution of tissue culture supernatant, W.C. The mabB630N epitope is contained within residues 3-14 [21] , while the epitope for mab514-18 includes residue F13 found in mammalian rod opsins [22] . Secondary antibody was IR-dye800-conjugated goat anti-mouse used at 1:10000 of 1 mg/mL solution (LI-COR Biosciences). A LI-COR Odyssey imaging system was used to image and quantify blot signals. Samples were prepared for confocal microscopy as previously described [23] . Fixed eyes were infiltrated with 20% sucrose for 48 h, embedded in OCT (Sakura Finetek), frozen, and stored at -80 °C for sectioning. 12 µ M slices were cut using a cryostat and labeled overnight with mAb B630N or mAb2B2 [24] antibodies (1:10 dilution of tissue culture supernatant, mAb2B2 gift of Robert Molday). Similar to mAb514-18, mammalian rod opsin residue F13 is a critical component of the mAb2B2 epitope [22] . Secondary antibody was Cy3-conjugated anti-mouse antibody (1:750 dilution, Jackson Immunoresearch). Sections were counterstained with AlexaFluor 488-conjugated wheat germ agglutinin (WGA; Life Technologies) and Hoechst 33342 (Sigma-Aldrich) and imaged using a Zeiss 800 laser scanning confocal microscope with a 40X NA 1.2 water immersion objective. Adobe Photoshop was used for post-processing and figure assembly. In order to improve image detail, non-linear adjustments were performed on Hoechst 33342 and WGA labeling channels. Antibody signal was linearly adjusted for better visualization.


INTRODUCTION
Retinitis pigmentosa (RP) is a vision-threatening disorder characterized by loss of functional and viable photoreceptors and retinal degeneration (RD) [1] . RP affects 1 in 3500-4500 people worldwide, making it one of the most common inherited retinal diseases [2] . RP is rod-cone dystrophy, in which initial symptoms of night blindness associated with rod degeneration progress to secondary cone death, resulting in the loss of daylight peripheral vision, potentially leading to complete blindness [3] .
Although most cases of RP are monogenic, thousands of different mutations in more than 50 genes underlie this disorder [4] . This genetic heterogeneity has made treatment development complex, and RP is untreatable for more than a million affected individuals worldwide [5] . In 30%-40% of cases, RP is inherited as an autosomal dominant trait (adRP) [5] . Notably, 25% of adRP cases stem from mutations in the rhodopsin gene (RHO), making RHO mutations the most common cause of adRP. To date, more than 150 distinct missense/nonsense RHO mutations have been associated with adRP [6] . Mutations in RHO can also cause recessive RP; in these cases, the mutations are typically null alleles [6,7] .
CRISPR/Cas9 is a two-component gene editing system found in bacteria species that has been co-opted by molecular biologists to permit a diverse range of genetic manipulations [8] . The CRISPR/Cas9 complex is made up of a DNA-cleaving enzyme called Cas9 and an RNA molecule that guides Cas9 to a specific target site, referred to as single-guide RNA (sgRNA) [9] . SgRNAs recognize unique ~20 nucleotide target sequences flanked by a protospacer-adjacent motif (PAM) e.g., NGG in the case of S. pyogenes Cas9 [10] . Cas9 introduces double-strand breaks 3bp upstream from the PAM site. When the CRISPR/Cas9 system is employed in vertebrate cells, the resulting DNA cleavage initiates either non-homologous end joining (NHEJ) or homology-directed recombination (HDR) DNA repair pathways [11] . NHEJ is an error-prone response that results in insertions and deletions (indels) of short DNA sequences at the break site. Within RHO genes, NHEJ can create frame-shifting indels that promote nonsense-mediated decay (NMD) of the mRNA, generating loss-of-function mutations and a knockout (KO) phenotype [12,13] . However, in-frame indels in RHO can cause photoreceptor degeneration via gain-of-function phenotypes such as rhodopsin misfolding or instability [12,13] .
In contrast, the HDR pathway utilizes a homologous DNA sequence as a template for repairing doublestranded break sites [14] . In experimental settings, the homologous template sequence can be specifically designed donor sequences of interest, allowing the introduction of novel DNA sequences. Hence, accurate gene repair can be carried out utilizing the HDR pathway [15] .
A previous study has shown that heterozygous Rho KO mice do not develop RD [16] , although they have half the normal complement of Rho, and their photoreceptor OS are shorter compared to wildtype (WT) controls [16] . Furthermore, parents of human patients with autosomal recessive RP caused by RHO null alleles (i.e., heterozygous carriers) do not develop RP, although they have slightly reduced rod sensitivity [7,17] . Thus, phenotypes associated with single RHO null alleles are significantly milder than the phenotypes associated with adRP RHO alleles. Therefore, a gene-based treatment strategy for autosomal dominant disease involving genes such as RHO that do not have an associated haploinsufficiency disorder is to develop allele-specific therapeutics that knock out the mutant allele, leaving the endogenous WT allele unaffected [18] . Another potential solution is to replace the mutant gene with an exogenous WT copy. In this paper, we introduce and characterize an X. laevis model of adRP caused by a mutation in the Rho.L gene developed by our research group, and subsequently use it to model treatments of these types. We explore three different CRISPR-based geneediting strategies to prevent RD, in which CRISPR/Cas9 can efficiently induce double-stranded breaks resulting in pan-retinal genomic alterations. First, we use a single sgRNA to target and introduce indels into the mutant Rho.L allele. Second, we use a pair of sgRNAs targeting the mutant Rho.L allele and an upstream sequence to introduce large inactivating deletions in the mutant allele. Third, we use a single sgRNA targeting the mutant Rho.L allele and a single-stranded DNA oligonucleotide repair template to restore the mutant allele. Each approach is designed to have minimal effects on the remaining WT allele, ideally converting the phenotype to that of a heterozygous null (1 and 2) or WT (3). Our system may provide insight into the potential of these therapeutic approaches by illustrating an idealized scenario of very high editing efficiency combined with very early treatment.

In-vitro transcription of sgRNA
Procedures were essential, as previously described [19] . Briefly, oligonucleotides corresponding to sgRNA variable regions were cloned into the PDR274 vector (Addgene 42250, gift of Keith Joung). SgRNAs were in vitro transcribed from plasmids containing inserts of interest using the HiScribe in vitro transcription kit (New England Biolabs), purified using the miRNeasy kit (Qiagen), quantified using a Nanodrop spectrophotometer, assessed for integrity by agarose gel electrophoresis, and stored at -80 °C prior to use.

Gene editing
Microinjections were performed according to the methods described by Feehan et al. [12] , except that Cas9 protein (New England Biolabs) was used in place of Cas9 mRNA. Briefly, a WT/ Rho.LΔ11Δ1 female X. laevis was injected with human chorionic gonadotropin (HCG) to induce ovulation. Oocytes were fertilized in vitro using sperm from an HCG-primed WT male. Utilizing a Hamilton syringe pump with continuous flow (36 μL/h), plates of ~125 embryos were injected with mixtures of one or two sgRNAs (0.5 μ L of each guide at 875 ng/μL), Cas9 protein (0.5 μL at 20 μM), 0.25 μL eGFP (enhanced green fluorescent protein) mRNA and nuclease-free H 2 O to make a final volume of 2.5 μL. Each embryo was injected for 1 s, equating to 20 nL of editing reagent mixture. At 36 h post-fertilization, successfully injected embryos were identified by eGFP fluorescence using an epifluorescence-equipped Leica MZ16F microscope.

Animal husbandry
Embryos were housed in glass dishes in an 18 °C incubator on a 12 h dark: 12 h light cycle. Embryos and tadpoles were reared in10 mM NaCl, 0.2 mM KCl, 0.1 mM MgCl 2 , 0.2 mM CaCl 2 . Post-metamorphic frogs were raised in 10 mM NaCl.

Dot blot assay, immunohistochemical labeling and confocal microscopy
At 14 days post fertilization (dpf) (developmental stage 48) normally developed tadpoles were sacrificed by pithing. The left eye was solubilized in 100 μL of a 1:1 mixture of phosphate-buffered saline and SDS-PAGE loading buffer containing 1 mM ethylenediaminetetraacetic acid and 100 μg/mL phenylmethylsulfonyl fluoride, while the contralateral eye was fixed in 4% paraformaldehyde buffered with 0.1 M sodium phosphate pH 7.4. Dot blot assays were performed on solubilized eyes as described by Tam et al. [20] and probed with anti-rod opsin antibodies mabB630N or mab514-18 (at 1:10 dilution of tissue culture supernatant, gifts of W.C. Smith). The mabB630N epitope is contained within residues 3-14 [21] , while the epitope for mab514-18 includes residue F13 found in mammalian rod opsins [22] . Secondary antibody was IR-dye800-conjugated goat anti-mouse used at 1:10000 of 1 mg/mL solution (LI-COR Biosciences). A LI-COR Odyssey imaging system was used to image and quantify blot signals. Samples were prepared for confocal microscopy as previously described [23] . Fixed eyes were infiltrated with 20% sucrose for 48 h, embedded in OCT (Sakura Finetek), frozen, and stored at -80 °C for sectioning. 12 µM slices were cut using a cryostat and labeled overnight with mAb B630N or mAb2B2 [24] antibodies (1:10 dilution of tissue culture supernatant, mAb2B2 gift of Robert Molday). Similar to mAb514-18, mammalian rod opsin residue F13 is a critical component of the mAb2B2 epitope [22] . Secondary antibody was Cy3-conjugated anti-mouse antibody (1:750 dilution, Jackson Immunoresearch). Sections were counterstained with AlexaFluor 488-conjugated wheat germ agglutinin (WGA; Life Technologies) and Hoechst 33342 (Sigma-Aldrich) and imaged using a Zeiss 800 laser scanning confocal microscope with a 40X NA 1.2 water immersion objective. Adobe Photoshop was used for post-processing and figure assembly. In order to improve image detail, non-linear adjustments were performed on Hoechst 33342 and WGA labeling channels. Antibody signal was linearly adjusted for better visualization.

Genomic DNA extraction, PCR, and sequencing
We designed four PCR primers to specifically amplify Rho.L regions of interest without amplifying other X. laevis Rho genes [Supplementary Material 1]. Genomic DNA was extracted from tadpole tail clips as previously described [19] . Genomic DNA samples were PCR amplified and analyzed by gel electrophoresis. PCR products were treated with an exoSAP protocol [25] and Sanger sequenced using a commercial service (Genewiz).

Electroretinogram recordings
Electroretinograms (ERGs) were recorded from dark-adapted animals at age 9 months as previously described [26] . Following the recordings, blood samples were collected to identify genotypes (WT vs. WT/ Rho.LΔ11Δ1) via Sanger sequencing. An example of analysis is shown in Supplementary Material 2.

Statistical analyses
GraphPad Prism software (version 9.0.0) was used to perform statistical analyses and generate figures. Specific tests and numbers of animals used are given for each experiment in the results section.

RESULTS
In a previously published study [12] , we used CRISPR/Cas9 to generate indels in the X. laevis Rho genes. We determined that X. laevis have three Rho genes (Rho.L, Rho.2.L, and Rho.S), and that in-frame indels in the X. laevis Rho genes cause RD, while frame-shifting indels have minimal phenotype. From this initial set of animals, we identified a line carrying a complex deletion comprised of adjacent 11bp and 1bp deletions that immediately follow the start codon of the Rho.L gene. The complex deletion could be formally designated Chr4L:131405006_131405016delAACGGAACAGA,131405030delT using HVGS nomenclature, but is hereafter referred to as "Rho.LΔ11Δ1" [ Figure 1A, Supplementary Materials 1 and 3]. Preliminary data derived from X. laevis Rho.L knockout tadpoles [Supplementary Material 4] suggests that Rho.L is the highest expressing of the three X. laevis Rho genes, and that each Rho.L allele contributes no more than 42% to the total retinal rod opsin.

Initial characterization of the Rho.LΔ11Δ1 X. laevis model for adRP
We characterized X. laevis tadpoles aged 14 dpf generated from a cross between WT and heterozygous WT/ Rho.LΔ11Δ1 animals. Genotypically, the offspring were expected to be half WT and half WT/Rho.LΔ11Δ1. A Corresponding predicted cut sites are indicated by scissors icons. (B) Dot blot assay for total rod opsin in a population derived from a mating between WT and WT/Rho.LΔ11Δ1 genotypes shows that two distinct groups are present with high and low rod opsin levels. Detection was with mAbB630N. Each data point represents a different animal. Five animals with high rod opsin levels and five with low rod opsin levels were genotyped (colored dots). Green were identified as WT, red were identified as WT/Rho.LΔ11Δ1. Black: not genotyped. (C) Confocal micrographs comparing retinal structure between WT and WT/Rho.LΔ11Δ1 genotypes. RD was observed in WT/Rho.LΔ11Δ1 animals relative to WT. Green: mabB630N (rod opsin); red: WGA; blue: Hoechst dye. dot blot assay indicated markedly lower levels of rod opsin (6.6% ± 0.6%) in half the tadpole eyes [ Figure 1B]. Five subjects with high rod opsin levels and five with low rod opsin levels were randomly selected and genotyped by Sanger sequencing. All five animals with high rod opsin levels were WT (green data points, Figure 1B) while all five animals with low rod opsin levels were WT/Rho.LΔ11Δ1 (red data points, Figure 1B). The results indicate that the low rod opsin phenotype is associated with the WT/Rho.LΔ11Δ1 genotype (P = 0.001, chi-squared test). Based on results shown in Supplementary Material 4 , one would expect > 50% of Rho remaining in heterozygous animals with a single functional allele. In the case of our adRP model, the very low rod opsin levels apparent by dot blot (6.6% ± 0.6% of WT) indicate a dominant RD phenotype in heterozygous WT/Rho.LΔ11Δ1 animals.
Note that it is likely that the mAbB630N antibody used in these assays does not detect the mutant protein, as the epitope is located at residues 3-14 [21] and residues 2-8 are missing or altered by the Rho.LΔ11Δ1 deletion. However, western blot experiments using the anti-rod opsin C-terminal antibody mAb11D5 indicated that the mutant protein is present at very low levels relative to the WT protein (not shown). Therefore it is unlikely that the dot blot assay significantly underreports rod opsin levels in these animals.
Immunohistochemical labeling and confocal microscopy of contralateral eyes from genotyped tadpoles further demonstrated that low rod opsin levels were due to an RD phenotype. In WT animals, healthy retinas were observed with extended Rho-expressing rods and rod outer segments. In contrast, WT/Rho.LΔ11Δ1 animals had significant RD apparent by confocal imaging, with disturbed and shortened rod outer segments [ Figure 1C].

sgRNA design
In order to target precise regions within Rho.L, we designed and generated sgRNAs as previously described [12] . We identified 20bp target sequences adjacent to PAM sites appropriate for CRISPR editing [Supplementary Material 1]. We identified two target sites that incorporate the Rho.LΔ11Δ1 mutation sequence in exon 1 (Sg5 and Sg6). In addition, we identified one site 1218 bp upstream of the start codon in a less-conserved upstream promoter region of Rho.L that was not found in Rho.2.L or Rho.S (Sg2). In order to guard against off-target effects, we confirmed that the Sg2, Sg5, and Sg6 target sites were unique by using nblast to scan the X. laevis genome sequence available online at Xenbase.org [27] . Notably, the Sg5 recognition sequence only differs from WT by a single nucleotide [ Figure 1A and Supplementary Material 1].

Comparing single-guide and double-guide CRISPR-based treatments
In our first experiments of gene-editing therapy in WT/Rho.LΔ11Δ1 animals, we utilized the Rho.LΔ11Δ1targeting sgRNAs (Sg5 and Sg6) either alone or in combination with the Rho.L promoter-targeting sgRNA (Sg2) to evaluate the utility of these therapeutic strategies and compare their efficacy. The sgRNAs, along with Cas9 protein and mRNA encoding eGFP (as a tracer to confirm successful delivery of RNA) were injected into single-cell X. laevis embryos derived from a WT/Rho.LΔ11Δ1 female and a WT male. After 36 h, non-fluorescent embryos were eliminated from the experiment. At 14 dpf, the tadpoles were sacrificed, genomic DNA was collected for PCR analysis, and the animals were genotyped by Sanger sequencing. One eye was solubilized for a rod opsin dot blot assay, while the other was processed for confocal microscopy.

RD was prevented in animals treated with single and double-guide approaches
As shown in [ Figure 2A and C], the single and double-guide treatments had minimal effects on WT animals, with a mild toxic effect in the case of Sg6. However, treatment of WT/Rho.LΔ11Δ1 animals with sgRNAs targeting the Rho.LΔ11Δ1 mutation resulted in significantly higher levels of rod opsin compared to untreated animals [ Figure 2B and D]. This effect was observed with both Sg5 and Sg6 guides [untreated (n = 9), Sg5-treated (n = 19), P = 0.000056; untreated (n = 11), Sg6-treated (n = 11), P = 0.0013 by Dunn's test for multiple comparisons]. However, average rod opsin levels were still approximately 1/3 lower than WT animals. Combining the Sg5 and Sg6 guides with the Sg2 guide (targeting the upstream promoter) also resulted in higher levels of rod opsin compared to the untreated WT/Rho.LΔ11Δ1 group, but was not superior to the single-guide approach [untreated (n = 9), Sg2+5-treated (n = 10), P = 0.018; untreated (n = 11), Sg2+6-treated (n = 7), P = 0.032 by Dunn's test for multiple comparisons]. There was no significant difference between treatments with a single guide vs. two guides.
As shown in [ Figure 2E], the Rho.LΔ11Δ1 allele results in significant RD and loss of rods. RD was prevented in animals treated with either the Sg5 or Sg6 guides. Single-guide treatment of WT embryos with Sg5 and Sg6 did not have detrimental effects, confirming the specificity of the guides for the mutant allele, as NHEJ of the WT allele would be expected to cause RD [12] .
RD was also significantly prevented in animals simultaneously treated with two sgRNAs, in both Sg2+Sg5 and Sg2+Sg6 groups. Overall, although the rod opsin levels were increased [ Figure 2B and D] and RD was prevented in the group treated with two guides [ Figure 2E], this treatment approach was not statistically superior to the single-guide approach.

Simultaneous treatment with Sg2 and Sg5 resulted in large deletions in 6 out of 11 cases
The novel approach of attempting to knock out a mutant gene by inducing DSBs on both sides of the mutant allele start codon was successfully carried out in over half of the WT/Rho.LΔ11Δ1 animals simultaneously injected with Sg2 and Sg5. In those animals, a 1248bp DNA fragment between the Sg2 and Sg5 recognition sites containing the Rho.L transcription start site and start codon was removed. This large edit was verified via PCR product sizes on 1% agarose gels [ Figure 3A] and Sanger sequencing [ Figure 3B]. Animals with large deletions were protected from RD [ Figure 3C]. However, as noted above, the overall histological outcomes and protein levels were not significantly improved by this approach relative to the single-guide approach.

Editing a non-conserved region of the Rho.L promoter does not cause RD
In a separate experiment, we looked for detrimental effects NHEJ induced by the upstream sgRNA might have on the function of WT alleles. We injected WT embryos with Cas9 and Sg2. A dot blot assay demonstrated no difference in rod opsin content in retinas from un-injected and Sg2-injected WT tadpoles, suggesting that minor disruptions of the promoter region caused by Sg2 editing do not affect rhodopsin levels (n = 15 per group; Mann Whitney U test, Figure 3D). Additionally, retinas of Sg2-injected WT animals were indistinguishable from retinas of untreated controls via confocal microscopy, and we did not observe any RD [ Figure 3E]. Extensive chimeric editing of the target site, the typical result of similar injections in this system [12,27] was confirmed by Sanger sequencing of a subset of the animals (n = 5 from each group, not shown). (Example shown in Supplementary Material 5).

Long term functional outcomes assessed by electroretinography
Retinal degenerations are progressive disorders that, in late stages, may be driven by the altered retinal environment caused by missing photoreceptors. Therefore, a number of animals from the experiments described above were set aside and allowed to reach maturity to examine long-term outcomes. We compared rod photoreceptor function between the WT/Rho.LΔ11Δ1 and WT/WT animals using scotopic ERG measurements. The a-wave and b-wave amplitudes were measured in response to different flash intensities. A-wave amplitude was found to be significantly reduced (P = 0.0059; two-way ANOVA) in Rho.LΔ11Δ1 animals (n = 10) compared to the WT (n = 6). The average b-wave amplitude was not found to be statistically different between the two groups [ Figure 4B]. The effects on a-wave were most apparent at higher flash intensities [ Figure 4A]. Subsequently, we assessed long-term effects of the Sg5-guide treatment on retinal function in WT/Rho.LΔ11Δ1 animals. We found no difference in mean b-wave amplitudes between the treated and untreated groups [ Figure 4D]. However, mean a-wave amplitude was significantly increased (P = 0.0001; two-way ANOVA) in Sg5-treated animals (n = 5) compared to the untreated adRP animals (n = 8) [ Figure 4C]. Our results indicate long-term improvement in retinal function in the Sg5-treated group.
We further assessed these long-term outcomes by histology. Nine-month-old animals from untreated and Sg5-treated groups were sacrificed and genotyped. Eyes were dissected, embedded, sectioned, labeled, and imaged as previously described to assess long-term effects of the Rho.LΔ11Δ1 mutation and gene editing treatments. As shown in [ Figure 4E], significant RD and loss of rods are observed in older Rho.LΔ11Δ1 animals. In agreement with the functional recovery seen in our ERG analysis, RD was prevented in Sg5treated animals, again indicating long-lasting benefits of treatment.

CRISPR-mediated HDR as a treatment approach for adRP
Although knocking out the Rho.LΔ11Δ1 allele by inducing NMD via indels introduced by NHEJ was effective in preventing RD in our animal model, we expected the strategy to be successful only in the twothirds of photoreceptors that would carry frame-shifting mutations. In the remaining one-third of cases, inframe mutations would be neutral or aggravating, and this is likely reflected by the total rod opsin levels in treated groups averaging about 2/3 of WT levels [ Figure 2B and D]. Therefore, we attempted to utilize a CRISPR-induced HDR repair mechanism to improve treatment outcomes, as HDR should produce errorfree repairs. We designed an experiment to compare the efficacy of our single-guide-based therapy to HDRbased repair. We designed a 120-nucleotide single-stranded repair template that spanned the Sg5 cut site, with the Rho.LΔ11Δ1 sequence restored to WT; recombination would revert the Rho.LΔ11Δ1 sequence to WT and eliminate the Sg5 target site. In addition, we included a mutation that induces the amino acid change M13F; this innocuous change generates a binding site for anti-mammalian rod opsin monoclonal antibodies 2B2 and 514-18, allowing us to identify cells that have undergone recombination by immunolabeling [20,22] . Finally, we included a silent mutation that introduces an ApoI restriction site [Supplementary Material 2].
To test this strategy, we injected embryos from a WT male and a WT/Rho.LΔ11Δ1 female with Cas9 and Sg5 alone to induce the NHEJ, or Cas9 and a combination of Sg5 and ssDNA template to induce HDR and compared them to untreated controls. The amount of ssDNA template injected (200 pg per embryo) was separately determined to be the maximum amount that could be injected without high levels of toxicity [Supplementary Material 4]. At 14 dpf, the animals were sacrificed, and genomic DNA samples were collected for PCR analysis and Sanger sequencing. One eye was solubilized for dot blot assays, while the other was fixed for confocal microscopy.
Similar to our findings in the previous experiment, both rod opsin dot blot and histology were consistent with RD in untreated heterozygous WT/Rho.LΔ11Δ1 animals that were prevented by the Sg5 treatment ( Figure 5A; n = 32 per group; P = 0.017 by Kruskal-Wallis test). Hence, the results of the previous experiment were replicated. Significant RD was also prevented in the group of animals treated with Sg5 in combination with the ssDNA repair template [ Figure 5A and B]; n = 32 per group; P = 0.027 by Dunn's multiple comparisons. There was no statistically significant difference between the Sg5 and Sg5 + ssDNA treatments. The genotypes of a subset of animals were identified via Sanger sequencing ( Figure 5A, colored data points). As anticipated, we found higher levels of rod opsin in WT animals compared to mutants in both treated and untreated groups. Similar to the experiment shown in [ Figure 2], we observed evidence of mosaic editing by Sg5 in the sequencing traces, and no evidence of HDR repair.

HDR inefficiently repaired the Rho.LΔ11Δ1 allele
In order to assess the efficiency of HDR in the 36 animals treated with Sg5 + ssDNA, three different assays were carried out: (1) Rho.L exon 1 was amplified by PCR from each animal, and restriction analysis was used to detect the novel ApoI restriction site included in the ssDNA repair template. However, we were unable to detect any ApoI cleavage. (2) A second dot blot assay was probed with anti-mammalian rod opsin antibody mab514-18 to detect rod opsin containing the M13F point mutation that we included in the ssDNA template. However, we found no evidence of increased mab514-18 binding in this assay. (3) Retinal cryosections from all 36 animals were labeled with mab2B2 anti-mammalian rod opsin antibody and imaged using confocal microscopy. Only a single rod in one of the sections was found to be 2B2-positive [ Figure 5C]. Since 2B2 labeling was not detected in any other animal, we concluded that HDR was extremely inefficient. Considering the lack of evidence for HDR repair from multiple modes of testing, we concluded that the improved phenotype in the Sg5-ssDNA treated group was likely due to the NHEJ mechanism induced by the Sg5 edit and not HDR.

DISCUSSION
In this study, we have described a new X. laevis model for adRP, for which we have demonstrated utility for testing CRISPR-based gene editing strategies as treatment approaches. This model, which carries a 12 bp deletion in the Rho.L gene, was previously developed in our laboratory using the CRISPR/Cas9 gene editing system [12] . The in-frame deletion in the first exon of the X. laevis Rho.L gene results in a dominant RD phenotype that is likely caused by defects in the biosynthesis of Rho protein, similar to previously described mutations associated with adRP in humans (e.g., P23H) [28,29] . In support of this, we were unable to convincingly detect the truncated mutant protein in the background of WT rod opsin by western blot using the anti-C-terminal antibody mab11D5 (not shown). Successful gene editing therapies in this model have similar requirements to those for human autosomal dominant disease, in that the mutant allele would either need to be eliminated or restored [18] . Thus, this model provides us with the means to rapidly survey a variety of gene-editing treatment strategies for adRP. Due to the relative ease of gene editing in embryonic X. laevis, the therapies were uncomplicated by issues such as delivery to the retina, or timing of delivery, both currently the subject of extensive research. Therefore, our experiments model idealized scenarios in which these issues have been solved.
Treating animals with a single sgRNA uniquely targeting the Rho.LΔ11Δ1 allele was overwhelmingly successful. We expected NHEJ to result in out-of-frame mutations and nonsense-mediated decay in approximately two-thirds of photoreceptors. Hence, loss-of-function mutations in the malfunctioning alleles were projected to prevent the majority, but not all, of the RD. The treatment significantly prevented RD in almost all animals, assessed by both dot blot for rod opsin and histology. Additionally, we were able to demonstrate that a single-base discrepancy is sufficient for Cas9 to distinguish between the WT and mutant alleles. This is particularly important for potential clinical use of gene editing strategies, as the majority of adRP-associated RHO mutations reported in human patients are alterations of a single nucleotide. Most recently, Diakatou et al. [30] have published their work using a similar editing strategy to treat G56R and NR2E3 adRP mutations in iPSCs. They designed unique sgRNAs to specifically target mutant alleles with a single-base difference and successfully utilized CRISPR/Cas9 to introduce NHEJmediated indels that KO the mutant allele [30] . Latella et al. [31] have also used a similar gene-editing approach in mouse retinas carrying the human P23H rhodopsin gene, through subretinal electroporation of plasmidbased Cas9. There are also limitations associated with NHEJ-based treatment for adRP. Although we have shown that Cas9 is able to discriminate minor differences between disease and WT alleles, sgRNAs cannot be designed for every known mutation due to the requirement for an adjacent PAM site. Our findings emphasize that even an imperfect treatment strategy can be highly effective.
We also tested more complex editing strategies, designed to address the non-productive third of expected in-frame edits. In attempting to restore the mutated allele to WT in our adRP model, we found that HDR is inefficient in X. laevis using our described protocol. Although the reasons remain unclear, similar results were also obtained by Feehan et al. [12] in our laboratory. Here, we attempted to improve HDR efficiency relative to previous attempts by using a single-stranded repair template, as opposed to the double-stranded template used by Feehan et al. [12] . However, HDR efficiency remained low relative to NHEJ. The incidence and efficiency of HDR-mediated repair are also reported to be low in mammalian cells. Maruyama et al. [32] have reported HDR repair efficiencies ranging from 0.5% to 20%, compared to 20%-60% NHEJ-mediated repair in mouse models. To improve HDR efficiency, Lin et al. [33] have proposed a novel strategy to synchronize cells into late S and G2 phases of the cell cycle, where HDR predominantly takes place. Alternatively, other research groups have attempted to suppress NHEJ by targeting components such as DNA ligase IV, forcing HDR to become the dominant repair pathway [34,35] .
We also used a gene inactivation strategy in which the transcription and translation start sites of the mutated gene were removed by introducing DSBs upstream and downstream of the Rho.L start codon to create large deletions. This also resulted in viable retinas without RD. Tsai et al. [36] previously carried out a similar treatment approach in human rhodopsin knock-in adRP mouse models using AAV injections. Utilizing double sgRNAs in combination with Cas9 protein, they introduced DSBs upstream and downstream from the rhodopsin gene's start codon (363 bp apart), resulting in large inactivating deletions in both alleles. Simultaneous with the elimination of the targeted gene, they enabled the expression of WT rhodopsin through an exogenous cDNA [36] . Our treatment approach differs in that one of the sgRNAs designed to introduce the large inactivating deletion was designed to target only the mutant allele. Our second non-allele-specific sgRNA was designed to introduce an edit upstream from the promoter regulatory region, in a non-conserved region such that typical small indels would not be predicted to significantly disrupt gene function. Therefore, our strategy did not require the delivery of an exogenous Rho cDNA.
Although the introduction of large deletions was only detected in a subset of animals (6 out of 11 tested), this strategy could potentially be optimized and may also have significant utility in creating X. laevis knockout models. One contributing factor to the efficiency of simultaneous breaks may be the physical distance between the target sites. In our experiment, the sites were 1248 bp apart. Shorter or longer distances may affect the efficiency of deletion. Another factor may be the timing of editing; unless both cuts occur within a sufficiently narrow timeframe, two separate instances of NHEJ are more likely than a large deletion.
Treatment with a single mutation-specific sgRNA conferred long-term protection against RD and loss of retinal function, as late as 9 months of age, as assessed by histology and ERG. However, benefits were primarily observed in the ERG a-wave, which is associated with the phototransduction functions of photoreceptors. The b-wave is a reflection of communication between photoreceptors and bipolar and Müller cells within the inner retina [37] . Hence, we speculate that the reduction in a-wave amplitudes in our adRP model is likely due to the reduced ability of rod photoreceptors to initiate phototransduction due to missing or dysmorphic outer segments, while the relatively unaffected b-wave amplitudes indicate that a large subset of viable rod photoreceptors remains at 9-months in WT/Rho.LΔ11Δ1 animals. Overall, our results are consistent with functional impairment of rods in older animals that are prevented by gene editing. Evaluation of long-term outcomes is another useful feature of this system, as X. laevis lifespan routinely exceeds a decade in our hands.

CONCLUSION
Our results indicate that X. laevis can be used to rapidly model and compare therapeutic gene editing strategies for adRP. From initial experiments in the system, we conclude that although more complex editing strategies may be theoretically superior to simpler strategies, ultimately, they will lack superiority if their complexity reduces efficiency. Therapies that require multiple linked events to occur in each cell (such as two adjacent cuts within a narrow time frame, or a cut followed by recombination, or transfection with two viruses) will require increased efficiencies for these independent events in order to surpass the efficacy of theoretically inferior treatments requiring only a single step. Therefore, once the challenges of efficient delivery of gene editing therapeutics are overcome, it should be possible to develop efficacious adRP therapies based simply on NHEJ repair. There is likely merit to the idea of emphasizing the development of less complex treatment strategies.