Rescue of hearing by adenine base editing in a humanized mouse model of Usher syndrome type 1F

Abstract

Usher syndrome type 1F (USH1F), characterized by congenital lack of hearing and balance and progressive loss of vision, is caused by mutations in the PCDH15 gene. In the Ashkenazi population, a recessive truncation mutation accounts for a large proportion of USH1F cases. The truncation is caused by a single C→T mutation, which converts an arginine codon to a stop (R245X). To test the potential for base editors to revert this mutation, we developed a humanized Pcdh15^R245X mouse model for USH1F. Mice homozygous for the R245X mutation were deaf and exhibited profound balance deficits, while heterozygous mice were unaffected. Here we show that an adenine base editor (ABE) is capable of reversing the R245X mutation to restore the PCDH15 sequence and function. We packaged a split-intein ABE into dual adeno-associated virus (AAV) vectors and delivered them into cochleas of neonatal USH1F mice. Hearing was not restored in a Pcdh15 constitutive null mouse despite base editing, perhaps because of early disorganization of cochlear hair cells. However, injection of vectors encoding the split ABE into a late-deletion conditional Pcdh15 knockout rescued hearing. This study demonstrates the ability of an ABE to correct the PCDH15 R245X mutation in the cochlea and restore hearing.

Graphical abstract

The p.R245X mutation in human PCDH15 causes the deafness and blindness characteristics of Usher syndrome type 1F. Corey and colleagues created a humanized R245X mouse model of Usher syndrome type 1F and used dual AAVs to deliver adenine base editors to correct the mutation in the inner ear, partially restoring hearing.

Introduction

Hereditary hearing loss is a common sensory disorder; 1–2 in every 1,000 newborns suffers from a hearing deficit, and about half of the cases may have a genetic basis.^1,2,3,4 The majority of hereditary forms of deafness are autosomal recessive.^5,6 One of these, Usher syndrome type 1F (USH1F), is characterized by congenital deafness and vestibular dysfunction and progressive vision loss. About 50 children are born with USH1F each year in the United States, adding to the estimated 10,000–15,000 USH1F patients worldwide. Currently, treatment for USH1F is limited to cochlear implantation, with no treatment for the vestibular dysfunction or the eventual blindness. USH1F is caused by mutations in the protocadherin-15 (PCDH15) gene.⁷ PCDH15 is a large protein with 11 extracellular cadherin (EC) repeats and a single transmembrane domain. In the inner ear, PCDH15 is expressed by the sensory hair cells, and the PCDH15 protein is specifically located on their stereocilia.⁸ Reaching from one stereocilium tip toward an adjacent taller stereocilium, PCDH15 binds to cadherin-23 to form the hair cell tip link.⁸ Tip links pull directly on ion channels to initiate the electrical response to sound, making PCDH15 indispensable for hearing. Less well understood is PCDH15’s role in the retina. While PCDH15 is located on the stereocilium-like calyceal processes of photoreceptors,^9,10 its function there is still unknown. However, mutations in the PCDH15 gene cause photoreceptor degeneration and blindness in humans, mice, and zebrafish.^11,12,13

In the Ashkenazi Jewish population, a single founder base pair transition mutation (c.733 C ⋅G to T⋅A) in PCDH15 accounts for 64% of USH1F cases.¹⁴ This C-to-T mutation changes the Arg 245 codon to a stop codon (R245X) in the PCDH15 mRNA, likely initiating nonsense-meditated mRNA decay and absence of PCDH15 protein. Reversion of the mutated nucleotide to the wild-type sequence should restore full-length PCDH15 expression and rescue hearing. Because the R245X mutation is recessive, its effects cannot be ameliorated via CRISPR-based disruption of the disease locus, as demonstrated in some other hearing loss models.^15,16 Instead, either the full-length protein must be supplied in trans, or the mutated DNA region must be repaired.

Base editors are engineered genome-editing enzymes that enable mutation of targeted genomic base pairs without relying on homology-directed repair, do not require delivery of a donor template, and result in substantially reduced genotoxic double-stranded breaks relative to native Cas9.^17,18,19 Adenosine base editors (ABEs) efficiently convert A⋅T base pairs to G⋅C, enabling repair of C>T or G>A mutations.^18,20 ABEs could therefore, in principle, correct the C>T mutation that results in PCDH15 R245X.^18,20,21 Prototypical ABEs consist of a Streptococcus pyogenes (Sp) nickase Cas9 (D10A), which only cuts the strand non-complementary to the guide,²² fused to an Escherichia Coli TadA adenosine deaminase engineered to function on adenine nucleotides.¹⁸ Delivery of an ABE into the cochlea via protein transfection resulted in β-catenin gene editing,²¹ indicating that this enzyme could be utilized for deafness-associated gene editing when targeted to a suitable mutation.

Most USH1F mouse models rely on mutations not found in the human population, such as Ames-Waltzer mice, which also have a nonsense mutation that truncates the PCDH15 protein.^23,24 We sought to generate a clinically relevant model for USH1F-R245X that replaces the mouse sequence at the R245 locus with the human PCDH15 gene sequence. We developed the Pcdh15^R245X mouse model, in which 60 bases of the mouse Pcdh15 gene were replaced with the human sequence, including the R245X mutation. The resulting amino acid sequence is unchanged aside from replacement of the arginine with the stop codon. Incorporation of the human sequence into the mouse means that base editing optimized in this model could more directly translate to clinical use in humans. Here we report that the PCDH15 R245X sequence present in USH1F patients can be targeted and corrected by ABEs. We further demonstrate that delivery of an intein-mediated split ABE via dual AAV9 PHP.B into the neonatal mouse cochlea results in correction of the R245X mutation and moderate rescue of hearing in a conditional USH1F model.

Results

Development of an USH1F mouse model for gene therapy

We first developed a model of USH1F in C57BL/6J mice. This carries a portion of the human PCDH15 genomic sequence containing the R245X (c.733 C→T) truncation mutation, inserted into the orthologous mouse Pcdh15 genome location via CRISPR-Cas9 (Figure 1A). Insertion of the human PCDH15 R245X coding sequence (5′-GACCGTGCCCAAAATCTGAATGAGAGGTGAACCACCACCACCACTC TCACAGTGGATGTTC-3′) created the stop codon (TGA) and altered four other nucleotides (underlined) in the mouse Pcdh15 gene; however, these are silent mutations that do not alter the amino acid sequence (Figure 1A). There are five more amino acids in the mouse than in the human PCDH15 N terminus, so the R245 equivalent is at position R250 in the mouse, and the mutation is properly denoted as R250X. However, for simplicity and connection to the human mutation, we refer to this USH1F mouse model as Pcdh15^R245X. Sanger sequencing confirmed that the human PCDH15^R245X mutation sequence was incorporated into the proper position, making these mice applicable for translational research focusing on gene therapy in humans. Heterozygous Pcdh15^R245X/+ founders were bred with C57BL/6J mice. Mice harboring the R245X mutation, heterozygous and homozygous, were viable.

Figure 1.

Generation of a Pcdh15^R245X Usher 1F mouse model with a human target sequence

(A) An oligonucleotide containing 60 bases of the human PCDH15 sequence bearing the USH1F R245X (c.733 C→T) mutation, flanked by homologous sequences of mouse Pcdh15 in exon 9, was co-injected into C57BL/6 mouse embryos with the SpCas9 enzyme and gRNAs targeting the mouse Pcdh15 locus (5′-CCGTGCACAAAATCTGAA-3′ and 5′-CCTCACAGTAGATGTTCTAGATG-3′). Aside from the mutation, there are four nucleotide differences in this segment between human and mouse, but they are silent, so the encoded protein is the same. (B) Auditory brain stem response (ABR) thresholds of Pcdh15^R245X/+ heterozygous and Pcdh15^R245X/R245X homozygous mice at P30 in response to an 8-kHz tone. Homozygous mice have no response at the highest intensity presented. (C) ABR of Pcdh15^+/+, Pcdh15^R245X/+, and Pcdh15^R245X/R245X mice for pure tone and click stimuli. Heterozygotes hear normally, but homozygotes have no response at any frequency. (D) Vestibular deficit in Pcdh15^R245X/R245X mice at age P30, observed as circling in an open field locomotion assay. Mice were placed in a featureless box and monitored for 5 min, and their track was analyzed for rotation, distance, and angular velocity using Ethovision software. Homozygous mutant mice circled at more than seven times the rate of heterozygotes. (E) Rotational velocity in the open field assay. Homozygous mutant mice circled twice as quickly as heterozygotes. Mice homozygous for the R245X mutation are deaf and display circling behavior. Error bars are standard error of the mean. ∗p < 0.05, ∗∗∗p < 0.001.

To test whether the mouse model replicated symptoms of USH1F disease, we measured auditory function with auditory brain stem response (ABR) recording. Thirty-day-old (post-natal day 30 [P30]) adult Pcdh15^R245X/R245X mice were profoundly deaf (Figure 1B; n = 5), with hearing thresholds not detectable up to 110 decibels at frequencies ranging from 4–45 kHz, while heterozygous littermates were capable of hearing at thresholds similar to wild-type C57BL/6 mice (Figure 1C; n = 4). Additionally, homozygous Pcdh15^R245X/R245X mice exhibited circling behavior, indicative of vestibular dysfunction (Figures 1D, 1E, and S1A). To quantify circling, mice were placed in an illuminated square box, their paths were recorded, and the videos were analyzed using Ethovision software. Homozygous mice exhibited a significantly greater number of rotations (defined as a >200° movement within 2 s) than their heterozygous littermates (13.3 ± 5.9 vs. 1.8 ± 1.1 rotations/min; homozygous, n = 13; heterozygous, n = 22; p < 0.00002) while still traveling the same total distance. Homozygous mice also circle at significantly greater velocities than heterozygous littermates (316 ± 72 °/s vs. 165 ± 22 °/s; homozygous, n = 5; heterozygous, n = 7; p = 0.012). In addition, homozygous mice demonstrated an inability to swim with their heads above water (Video S1), consistent with a lack of gravity perception like that of human USH1F patients.

Video S1. Swim test to assess vestibular function

Wild-type mice swim well, keeping their heads above water. The saccule and utricle of their vestibular system sense the acceleration of gravity, to allow the mouse to swim up for air. Heterozygous Pcdh15^R245X/+ mice swim well, similar to wild-type. Homozygous Pcdh15^R245X/R245X mice lack PCDH15 so they do not form tip links and their vestibular hair cells cannot sense gravity. We found that they are unable to swim normally.

Histochemistry using phalloidin to label actin filaments revealed that by P6 stereocilium bundles were disorganized in the apex, middle, and base of the cochlea (Figure 2A) and in the utricle (Figure 2B), like Pcdh15^Av-3J mice, which also bear a truncation mutation.^23,25 The lack of Pcdh15 expression was also confirmed using an antibody to PCDH15 on P6 cochleas in homozygous Pcdh15^R245X/R245X mice (Figure 2C). Scanning electron microscopy confirmed the regular arrangement of stereocilia and the presence of tip links in heterozygous mice (Figure 2D) but showed disorganization of bundles and absence of tip links in homozygotes (Figure 2E). Last, we found that cochlear hair cells in heterozygous mice rapidly accumulate the fluorescent dye FM1-43 (Figure 2F), which enters hair cells through their open transduction channels and requires tip links to open them, but there was no FM1-43 accumulation in homozygotes (Figure 2G).

Figure 2.

Severe functional and morphological deficits Pcdh15^R245X/R245X KO mice at P6

(A) Bundle morphology in phalloidin-stained Pcdh15^R245X/+ and Pcdh15^R245X/R245X cochlear hair cells. Severely disorganized hair cell bundles were observed in homozygous mutants (right) compared with heterozygote controls (left) of P6 mice. (B) Bundle morphology in vestibular hair cells from the utricle. Homozygous mutants had disorganized stereocilia. (C) Anti-PCDH15 labeling (magenta) of P6 OHCs in control mice along with phalloidin co-staining (cyan) demonstrated normal PCDH15 localization to the stereocilium tips in Pcdh15^R245X/+ mice (left), which is absent in Pcdh15^R245X/R245X mice (right). (D) Scanning electron micrographs of Pcdh15^R245X/+ mouse organ of Corti controls show largely normal OHC and IHC stereocilia morphology. Bundles have a regular appearance, with stereocilia aligned in rows. Tip links connect the tips of adjacent stereocilia along the hair bundles’ axis of sensitivity (arrows). (E) In Pcdh15^R245X/R245X mutants, the bundle structure was disrupted, with shortened stereocilia and gaps in the rows. While the lateral links were present, no tip links were detected. (F and G) At P6, FM1-43 dye uptake was abolished in Pcdh15^R245X/R245X mice, indicating no open transduction channels. Scale bars: 5 μm (A–C), 1 μm (D and E, top panels) and 200 nm (D and E, bottom panels), and 10 μm (F and G). Mice homozygous for the R245X mutation display disorganized hair bundles and lack MET channel functionality.

Editing PCDH15 in vitro

Several iterations of ABEs have been developed, including improved versions with enhanced editing potency. Each base editor relies on a Cas protein that utilizes a guide RNA (gRNA) and a protospacer adjacent motif (PAM) site to recognize and bind to a specific target DNA strand. To determine the most efficient editor, we compared ABE7.10,¹⁸ ABEmax,²⁰ ABE8e,²⁶ ABE8.20m,²⁷ and the ABE8e version of the “PAMless” Sp SpRY variant.²⁸ We first tested editing on several previously validated genomic targets (sites 5, 13, and 16) in the genomes of normal HEK293T cells.²¹ HEK293T cells were transiently transfected with plasmids encoding a base editor expressed under a cytomegalovirus enhancer/chicken β-actin (CBA) promoter along with gRNA plasmids, and the cells were harvested 3 days later.

Genomic DNA was isolated and sequenced using an Illumina MiSeq and analyzed using pysamstats (a module of pysam; materials and methods) to detect editing. Interestingly, all base editors performed statistically similarly to one another at each genomic target when compared at position +14, which is shared by all genomic sites (Figure 3A). Base editing occurred most efficiently between positions +13 and +17 base pairs from the nucleotide in the spacer most proximal to the PAM (positions +4 to +8 in the conventional base editing nomenclature using the PAM as positions +21–23). Here we refer to the last spacer nucleotide before the PAM as position +1, allowing a naming convention that applies to spacer sequences larger than 20 nt. Notably, the ABE8.20m edited the R245X mutation less efficiently than ABE8e but exhibited less editing at bases outside of the targeted editing window 13–17 bp upstream of the PAM (Figure 3B). We observed editing by ABE8e at position 19 at similar rates as positions 13–17 upstream of the PAM site (+19, +13–17), indicating that ABE8e could be used to edit sites outside of the optimal editing window. Compared at each individual editable site (sites 5, 13, and 16 and the R245X site), there were differences among editors in editing efficiency (Figure 3B).

Figure 3.

Efficiency of editing in HEK293 cells transfected with plasmids encoding full-length editors

(A) Comparison of base editing by five different editors at three genomic sites in normal HEK293T cells. (B) Base editing window efficiency of editors on the genomic sites used in editor selection. (C) Coding and complement sequences for the human PCDH15 knockin segment with the C>T nonsense mutation (G>A on the complement strand) underlying R245X. Three gRNAs are shown that target an ABE to the complement strand, where the mutated base is adenosine. Numbering is shown for gRNA1 with the 5'-TGG-3' PAM. (D) Comparison of base editing of the PCDH15 R245X mutant nucleotide in HEK293T cells harboring the R245X target sequence. ∗∗∗∗∗∗∗p < 10⁻⁷. (E) Editing efficiency at different times post transfection (normalized to efficiency at 48 h). (F) Effect of gRNA:editor molar ratio on editing efficiency. There were no significant differences. (G) Effects of lengthening (+) or shortening (−) the spacer length on editing efficacy of ABEmax and ABE8e for the R245X target. Numbers indicate the lengthening or shortening of the g1 20bp gRNA. The ABE8e editor performed significantly better on the R245X target sequence despite all editors working similarly on the target sequences used to isolate and develop the original editors. One-way ANOVA detected no significant difference between ABEmax and ABE8e guides (compared within same base editors); the g2 and ALT are significantly different. (H) Editing efficiency at the R245X mutation by editors with different Cas9 variants, using the most efficient gRNA for the targeted PAM. Although the editors had similar editing efficiencies at the genomic sites on which they were first selected, they showed a different editing ability on the R245X site. All comparisons were significant at p < 10⁻⁴. (I) Editing window efficiencies of ABE8e-SpRY using guides that place the A17 targeted base at positions 12–16+ PAM. gRNA1 vs. other gRNAs was significant at p < 0.01. All error bars indicate standard error of the mean.

To test editing of the PCDH15 R245X mutation, we created a model human cell line by transducing HEK293T cells with a lentivirus encoding the N terminus of PCDH15 with the R245X mutation and selecting a stable line. We initially evaluated a gRNA targeting the R245X mutation (spacer-1: 5′-TTCACCTCTCATTCAGATTT-3′) which has the adenine to be edited at position +17 (A₁₇) upstream from the NGG PAM (Figure 3C). Editor and guide plasmids were transfected into HEK293T cells carrying the mutant PCDH15 sequence, then DNA was harvested and subjected to PCR, and the amplicon was sequenced. Editing of the R245X mutation was less efficient with all tested ABEs than editing of the validation sites; however, as expected, the updated ABE8 constructs outperformed the older-generation editors (Figure 3D). In particular, ABE8e (41% ± 12%) significantly improved editing compared with ABEmax (5% ± 4%). Experiments to determine optimal in vitro editing conditions showed the greatest number of edits 48 h post transfection with an equimolar ratio of gRNA:editor (Figures 3E and 3F). Additionally, editing with spacer-1 (g1) was more efficient than with spacer-2 (g2) (5′-TCACCTCTCATTCAGATTTT-3′), which placed the R245X target at position A₁₈ (data not shown; editing of g2 = 0.4% ± 0.2%, n = 4). Because of the low editing observed with ABEmax, we tested whether shortening or lengthening the spacer sequence could increase editing efficiency at the R245X target. However, we observed no significant difference between the standard 20-nt spacers and 17- to 23-nt spacers with either ABEmax or ABE8e (Figure 3G). We also observed that ABE8e and ABEmax using SpCas9 recognized and edited at one-fifth of the efficiency when using an alternative gRNA (Figure 3G; g1 vs. ALT) targeted to a site with an NAG PAM (which positions the target A in position +10), as observed previously.²⁹

The sequence near the PCDH15 R245X mutation contains several PAMs that should enable targeting with other Cas9 enzymes, permitting evaluation of other editing approaches. We observed significantly lower editing with ABE8e-SpCas9-VRQR^30,31 (gRNA3, NGA PAM targeting A₁₁) and ABE8e-SaCas9-KKH³² (NNNRRT PAM targeting A₉) than with ABE8e-SpCas9 (NGG PAM targeting A₁₇), likely because of positioning of the target base outside or near the border of the optimal ABE edit window (Figure 3H). To determine whether moving the targeted R245X adenosine at position A₁₇ within this window could increase the efficacy of ABE8e editing, we tested several gRNAs (termed gRNA1 +12 to +16 because of the position they move) with an ABE8e-SpRY, which can recognize any 3 nt as a PAM, although with a lower efficiency than those of WT SpCas9 on NGG PAMs.²⁸ While we observed substantial editing with ABE8e-SpRY and gRNAs that placed the R245X adenine within the optimal editing window, editing with the gRNA1 NGG PAM A₁₇ gRNA remained the most efficient approach (Figure 3I).

Base editing results in protein synthesis in vitro

We next sought to determine whether correcting the R245X mutation would result in proper PCDH15 protein synthesis. To accomplish this, we developed two lentivirus reporter systems in HEK293T cells that allow quick and cost-effective detection of A-to-G editing. The first platform, PCDH15-FLAGMYC, used a truncated form of PCDH15 with an N-terminal FLAG tag and a C-terminal MYC tag (Figure 4A). The wild-type (R245WT) construct expresses FLAG and MYC tags (∼75 kDa), while the R245X mutant construct (∼37 kDa) lacks MYC expression (Figure S2B). We found that co-transfection of ABE8e-SpCas9 and gRNA plasmids into HEK293T cells carrying PCDH15-FLAGMYC resulted in editing, as assayed by western blot (Figure 4B). As observed with sequencing (Figure 3D), MYC expression showed that ABE8e led to much higher levels of editing compared with ABE7.10 or ABEmax, which were undetectable by western blotting.

Figure 4.

Base editing in HEK293T cells transduced with PCDH15 R245X reporter constructs

(A) Diagrams of the PCDH15-FLAGMYC and FlashLight constructs. (B) Western blot of PCDH15-FLAGMYC after transfection with full-length base editors and gRNAs targeting the R245X mutation. (C) Fluorescence of FlashLight HEK293T cells after editing. Many cells express mCherry, which does not require editing. Some cells—especially with ABE8e—express GFP as well, which does require editing. (D) Western blot of PCDH15-FLAGMYC after transfection with split-intein base editors and gRNAs targeting the R245X mutation. (E) Fluorescence of FlashLight HEK293T cells after editing. (F) Flow cytometry of FlashLight R245X cells after editing. Unedited cells express just mCherry (red); edited cells express mCherry and GFP (green). The proportion of cells edited was highest when the editor was transfected as either full-length (ABE8e) or intein-linked (Int-ABE8e) ABE8e. Significance: ∗p < 0.05, ∗∗p < 0.01, others N.S. (G) Quantification of editing in FlashLight HEK293 cells by sequencing. Editing was highest at the targeted adenine at position A17, although A10 was also edited. For the PCDH15 coding sequence, editing at A10 causes a silent mutation. ABE editor effects were quickly observable using the reporter cell lines, which paralleled the results of NGS. Error bars indicate standard error of the mean.

The second reporter system, termed FlashLight and built on previous assays that detect spectral shifts in fluorescence upon editing,^33,34,35 inserts a short PCDH15 R245X or R245WT target sequence within the coding sequence of GFP. Upstream of GFP is an mCherry reporter linked to GFP via a P2A self-cleaving peptide sequence (Figure 4A). The R245WT sequence does not interrupt GFP expression, whereas R245X prevents it (Figure S2C). Transfection of full-length editor and gRNA plasmids into FlashLight HEK293T cells resulted in editing, as assayed via GFP expression (Figure 4C). Again, ABE8e exhibited much better editing than ABE7.10 or ABEmax, as determined by GFP fluorescence.

Split and conventional ABEs exhibit comparable editing efficiencies in vitro

Several groups have shown that delivery of base editors in dual AAVs is possible using intein linkage of two editor protein fragments after synthesis.^36,37,38 One AAV does not have the capacity to encode any of the SpCas9-based ABEs tested here, making a dual AAV strategy necessary for in vivo delivery of these editors. We tested the editing efficiency of intein-linked ABE8e and ABEmax constructs, the latter of which was previously demonstrated to edit in vivo³⁹ (Figure S3A). Of note, the C-intein base editor plasmid can be used interchangeably with either the ABEmax or ABE8e N-terminal intein plasmids to generate the full-length base editor in vitro. We first transfected normal HEK293T cells with the ABE C-intein plasmid and with either the ABEmax or ABE8e N-intein plasmid. When the N- and C-intein halves were expressed together, they constituted the full-length ABE, detected with a western blot using antibodies to the hemagglutinin (HA) and FLAG tags (Figure S3B).

We then transfected PCDH15-FLAGMYC HEK reporter cells with the ABE C-intein plasmid and with either the ABEmax N-intein or ABE8e N-intein plasmids. When the N- and C-intein halves were expressed, the PCDH15 R245X target sequence in the reporter cell line was edited, allowing expression of the full-length reporter, as detected with a western blot using antibodies to the FLAG and MYC tags (Figure S3C). To complement sequencing-based data, we also performed flow cytometry on FlashLight R245X cells transfected with single or split-intein editors. We observed similar editing efficiencies of the intein-linked editors and the single construct editors for ABE8e and ABEmax (Figure 4F). ABE8e split-intein outperformed the ABEmax construct, as seen previously via sequencing, demonstrating that these systems are useful for performing quick and cost-effective editor/gRNA screens for base editing.

We further verified that intein-extein formation occurred by generating intein-detection plasmids independently expressing either an mCherry or BFP marker protein and a split-intein GFP (Figure S3D). Transfection with N- or C-intein marker plasmids individually resulted in mCherry or BFP single-positive, GFP-negative cells, while co-transfection of both plasmids led to robust triple-positive cells (Figure S3E). Similarly, we were able to separately detect N- and C-terminal Cas9 halves upon co-expression of intein-editor plasmids by using antibodies that detected the N or C terminus of SpCas9 (data not shown).

One concern about the improved efficacy of the ABE8e editor is expansion of the editing window to include adenines not involved in the R245X mutation. We observed low but non-zero editing of adenine at position PAM+10 (A₁₀) (Figures 4G and S3F). In the case of the PCDH15 gene, however, this results in a silent mutation that does not alter the amino acid sequence (Figure 3C).

To predict genomic off-target editing sites, we used CCTop⁴⁰ and Cas-OFFinder⁴¹ prediction software to locate 14 possible locations in the human genome that could be targeted by our spacer-1 gRNA. These sites were chosen based on the criteria of containing three or fewer total mismatches from the spacer sequence, two within the core of the target sequence, and up to two in the 5′ or 3′ ends of the target sequence. PCR amplification and sequencing of these sites after transfection of all editors into HEK293T cells revealed no significant off-target editing (Figure S3G).

To test base editing efficacy on a genomic PCDH15 mutation, we obtained human induced pluripotent stem cells (iPSCs) isolated from an USH1F patient homozygous for the PCDH15 R245X mutation (provided by Drs. Steven Tsang and Barbara Corneo, Columbia University). Patient iPSCs were cultured into organoid spheres and transiently transfected with a single plasmid containing the gRNA along with ABEmax or ABE8e. After 3 days, PCDH15 stem cell DNA was harvested and sequenced (Figure S4). Because iPSCs are resistant to transfection, editing occurred at a much lower efficacy than observed in HEK293T cells. As expected, however, the ABE8e editor performed much better than the ABEmax.

Delivery of the split-intein base editor into the mouse cochlea via AAV

After observing that the split-intein versions of the ABEmax and ABE8e editors were able to edit the R245X mutation in vitro, we tested an intein-mediated, dual AAV base editor therapy in vivo. We selected the AAV9-PHP.B capsid, which we and others have found to be efficient in transducing hair cells.^42,43,44 Split-intein editor AAVs were generated by co-transfecting HEK293T cells with helper plasmids and the same intein-editor plasmids as used for in vitro editing. Virus was purified with an iodixanol gradient and titered by qPCR (ABEmax N-intein, 2.54 × 10¹³ genome copies [GCs]/mL; ABE8e N-intein, 3.18 × 10¹³ GCs/mL; C-intein, 1.01 × 10¹⁴ GCs/mL). To test whether dual AAV delivery resulted in intein-extein formation and editing, we first transduced the HEK293T FlashLight cells at an MOI of 10⁶. We observed GFP-positive cells, indicating that base editing had occurred (Figures 5A and 5B).

Figure 5.

R245X DNA editing in HEK293 FlashLight cells and in the Pcdh15^R245X mouse cochlea by dual AAV delivery of intein-linked base editors

(A and B) Editing in HEK293 cells carrying the FlashLight reporter. Dual AAVs encoding ABEmax or ABE8e were added to cultures at a multiplicity of infection of 10⁶. The number of GFP-fluorescent cells, indicating successful editing, was much larger with ABE8e. (C) DNA editing in the Pcdh15^R245X mouse cochlea after injection of intein-linked ABEmax or ABE8e. DNA from whole cochlea was sequenced. Editors performed equally well, with no significant difference. (D) cDNA sequencing to detect editing in hair-cell Pcdh15 mRNA. ABE8e performed about twice as well as ABEmax. Significance: ∗p < 0.05. (E) Failure to rescue ABR thresholds with dual-AAV editor delivery in the Pcdh15^R245X/R245X mouse model. Mice receiving dual AAVs encoding intein-linked ABEmax or ABE8e showed auditory sensitivity, as assessed by ABR at P30, no better than the deaf untreated mice (n = 3 ABE injected, n = 4 uninjected). All error bars indicate standard error of the mean.

We then injected the AAV vectors into newborn (P0–P1) Pcdh15^R245X/R245X mice through the round window membrane. Round window membranes were surgically exposed on cryo-anesthetized P0–P1 pups, and 1 μL of AAV vectors (3.4 × 10¹⁰ GC each) was injected using a controlled-injection micropipette. Mice were sacrificed at P6 to determine editing efficacy or maintained until P30 for ABR hearing tests. By harvesting DNA from whole cochleas, we observed that the PCDH15 R245X mutation was edited in whole genomic DNA (hair cells and support cells) by either ABEmax (1.40% ± 0.62%) or ABE8e (1.52% ± 0.36%) delivered in dual vectors (Figure 5C). This assay tests editing in all cochlear cells, but not all may have been transduced by the AAVs, and so the editing may be underestimated. To determine whether editing was occurring specifically within the hair cells, we additionally harvested total RNA from injected whole cochleas and synthesized and sequenced cDNA. In the cochlea, Pcdh15 should only be expressed in hair cells, which are robustly transduced by AAV9-PHP.B. Sequencing of cochlear cDNA revealed that the R245X mutation was corrected (TGA→CGA) by both editors but significantly more in mice injected with ABE8e (3.3% ± 2.2%) than in those injected with ABEmax (1.60% ± 0.58%, p < 0.01) (Figure 5D). qPCR using primers specific for the AAV vector revealed that 1.67 × 10⁶ GCs were present in the injected cochlea (Figure S5A). Interestingly, an average of 0.23 × 10⁶ GCs was found in the uninjected ear, consistent with the common observation in neonatal mice that AAV vectors delivering a marker protein to one ear can be detected in the other ear and in the brain. Presence in the contralateral ear was also observed when measuring the levels of AAV present by qPCR in comparison with genomic GAPDH (Figure S5B).

Although the truncation mutant ablates PCDH15 protein function, we did not detect a significant difference in Pcdh15 RNA cochlear expression between Pcdh15^R245X/+ and Pcdh15^R245X/R245X mice with or without injection of the editors (Figure S5C). Despite the presence of AAV in the cochlea and editing of the R245X mutation in the mRNA, we did not observe hearing restoration in Pcdh15^R245X/R245X mice (Figure 5E). Unlike the limited human off-target editing, we did observe substantial editing in three of the off-target sites predicted by CCTop (Figure S5D); however, these sites are intergenic or located within non-protein-coding regions.

Rescue of hearing in a conditional Pcdh15 knockout model

The observation that Pcdh15^R245X homozygous mice have disordered bundles by P6 (Figure 2) suggested that dual AAV delivery of the editor at P0 may not be early enough to rescue tip links. We sought to create a delayed deletion model that could be more amenable to AAV base editor rescue by crossing Pcdh15^R245X/R245X mice with a conditional Pcdh15 knockout line that deletes the critical transmembrane domain of PCDH15 using Cre recombinase under the hair-cell-specific Myo15 promoter.⁴⁵ Myo15 is expressed around P0, so conditional Pcdh15^fl/fl, Myo15-Cre mice have all PCDH15 isoforms ablated, but only at birth, avoiding the early morphogenic defects observed in constitutive Pcdh15^R245X/R245X mice. Crossing Pcdh15^R245X/+ mice with Pcdh15^fl/fl, Myo15-Cre mice yielded Pcdh15^R245X/fl, Myo15-Cre+ mice. Deletion of the floxed allele by Cre recombinase leaves the R245X allele; editing of that allele should then restore function.

ABR measurements demonstrated that untreated Pcdh15^R245X/fl, Myo15-Cre+ mice were completely deaf at P30 (Figure 6B). At that age, hair bundles were disorganized, but the hair cells survived (Figure S6). Surprisingly, they had no discernable vestibular deficits, as assayed with behavioral tests (Figures S1B–S1D).

Figure 6.

Rescue of hearing by ABEmax and ABE8e in the late-deletion Pcdh15^R245X/fl, Myo15-Cre mouse model

(A) Normal ABR responses at P30 to 8-kHz tones in an untreated Cre-control mouse (Pcdh15^R245X/fl, Myo15-Cre-). The threshold is indicated with a green arrowhead. (B) ABR in an untreated KO mouse (Pcdh15^R245X/fl, Myo15-Cre+). (C) ABR in a Cre-control mouse (Pcdh15^R245X/fl, Myo15-Cre-), treated with dual AAV delivery of ABEmax. (D) ABR in a Cre-control mouse, treated with dual AAV delivery of ABE8e. (E) Rescue of a Cre+ KO mouse (Pcdh15^R245X/fl, Myo15-Cre+) with dual AAV delivery of ABEmax. The threshold, at about 80 dB, is about 40 dB improved from the untreated mouse. (F) Rescue of a Cre+ KO mouse with dual AAV delivery of ABE8e. The threshold, at about 100 dB, is slightly improved from the untreated KO. (G) ABR threshold plots for all conditions. Overall, ABEmax delivered the best rescue. In Cre- and WT control mice, it showed some elevation of the threshold at high frequencies. (Cre+, uninjected n = 4; Cre+, ABE8e n = 15; Cre+, ABEmax n = 15; WT, ABEmax n = 6; Cre−, ABEmax n = 15; Cre−, ABE8e n = 7; Cre−, uninjected n = 4). In Pcdh15^R245X/fl, Myo15-Cre+ mice, ABRs from ABEmax- and ABE8e-injected Cre+ mice were significantly different at each frequency measured, except for 22.6, 32, and 45 kHz (p < 0.05). (H) Durability of rescue in older mice. Mice for the durability study were selected as the best-rescued mice of (G). Rescue of hearing by ABEmax and ABE8e persisted at ages P60–P80, although to a lesser extent than at P30. n = 4 mice for each condition. ABRs were measured in the injected ears. All error bars indicate standard error of the mean.

Unlike the lack of rescue in Pcdh15^R245X/R245X constitutive knockout mice, hearing was rescued by injection of Pcdh15^R245X/fl, Myo15-Cre+ late deletion mice at P0–P1 with dual AAVs encoding base editors ABE8e or ABEmax. Improvement was statistically significant (p < 0.05) for ABEMax at all frequencies except 22.6 kHz and for ABE8e at all frequencies except 5.6, 16, 32, and 45 kHz. On average, a 20-decibel (dB) rescue in hearing from the 110-dB ceiling was observed (Figure 6; n = 15 for Cre⁺, editor-injected mice). Rescue was mainly observed in the 4.0- to 22.6-kHz range. This includes the best responder to treatment, which had a hearing threshold of 55 dB at 16 kHz. The limited high-frequency rescue could reflect differences in editor efficiency between apical and basal hair cells; the later developmental stage of basal, high-frequency hair cells; or the lower transduction of basal hair cells by the AAV9-PHP.B vector.

Contrary to the efficacy of editors in vitro, mice receiving the ABEmax editor regained hearing significantly more than ABE8e-injected mice at 8, 11.2, 16, and 32 kHz and for clicks (p < 0.05). Injections of AAVs encoding base editors had no deleterious effects in Pcdh15^+/+ wild-type mice (Figure 6G, black circles). Follow-up ABRs at P60+ revealed that hearing rescue was sustained, albeit less pronounced than at P30, in about half of the mice that had hearing rescue (Figure 6H). Sustained rescue of hearing using gene editors has not been observed previously.

Discussion

Hereditary deafness is one of the most common hereditary sensory disorders. The majority of these cases are due to recessive mutations, which can be treated using AAV delivery of the wild-type gene. For some deafness genes, especially those that cause Usher syndrome, the coding sequence is too long to fit in a single AAV capsid; for some Usher genes, the coding sequence will not even fit in two capsids. An alternate strategy for treatment is to deliver a gene editor that corrects rather than replaces the mutant gene. Here we show success at in vivo editing of a mutation causing USH1F in a mouse bearing a segment of the human PCDH15 genomic sequence with the common USH1F R245X mutation. While base editors have been used to edit genes in the inner ear and have transiently rescued mouse deafness,³⁶ this study is the first to edit a commonly mutated human deafness gene in a mouse model and modestly but durably rescue hearing in a way directly applicable to human deafness.

Several families of base editors were tested in vitro on a human PCDH15 DNA segment bearing the R245X mutation to determine the most optimal ABE for in vivo editing. While all base editors had similarly high editing rates (∼50%) for the three genomic test sequences on which the ABEs were originally developed,^20,27 only the eighth-generation editors led to efficient editing on the PCDH15 R245X sequence, demonstrating that each target locus requires individual optimization. This benefit was not due to the target adenosine being outside of the optimal +13 to +17 nt editing window⁴⁶ because high editing was accomplished on genomic test sequences with adenosines at positions PAM +12 to +19 in ABE7.1 and ABEmax as well as the eighth-generation ABEs. With ABE8e-SpRY, we observed R245X reversion using gRNAs that positioned the target adenosine within the optimal editing window, albeit at levels lower than ABE8e on sites harboring NGG PAMs, as reported previously.²⁸ Despite lower editing efficiency, SpRY-based editors are able to edit sequences constrained by the guanosine-poor sequences upstream of target sites.²⁸ SpCas9-VRQR ABE, which recognizes an AGA PAM site,²⁰ also performed poorly on the R245X mutation. The AGA site places the R245X target adenosine at position 11 from the PAM site, outside of the optimal editing window, and we observed no editing by the VRQR editor at the PCDH15 site (Figure 3H). Finally, gRNA2—which uses the GGG PAM 1 nt downstream of gRNA1, used for the in vivo experiments—resulted in no editing, likely because of the target base being now outside of the edit window. These negative results indicate that idiosyncrasies of each target sequence must be considered when using a Cas9:gRNA. Despite this, we observed that the SpCas9 ABE does recognize an NAG PAM, resulting in editing at about 20% of the efficiency of the NGG PAM, as reported previously for Cas9 recognition for DNA cleavage, demonstrating that this also applies to Sp Cas9 base editors.²⁹

A fundamental challenge for this project was the limitation of this generation of ABEs, in that the editor coding sequence and gRNAs do not fit in a single AAV capsid.⁴⁷ Consequently, the editor coding sequence was split and encoded by dual AAVs, and the sequences for intein peptides were added to the split site. In transfected cells in vitro, we found no difference between the split-intein editors and full-length editors, which is encouraging for in vivo editing applications. Recently, an editor:gRNA system capable of delivery by a single AAV was developed,¹⁷ which may allow higher-efficiency editing and functional rescue in vivo without the need for the editor cloning optimization required for this project.

To test editing in vivo, we created a Pcdh15^R245X mouse model that recapitulates the human USH1F deafness and vestibular deficiencies and shows hair bundle disorganization like other null models, such as the Pcdh15^av−3J mouse. Despite robust editing in vitro, editing efficiency in the Pcdh15^R245X mouse cochlea in vivo was more limited. The AAV9-PHP.B vector we used has been demonstrated to efficiently transduce hair cells at the titers used; however, it is less efficacious when transducing non-hair cells in the cochlea. Likely other AAVs that are better at whole-cochlea transduction, such as AAV-S, could increase the whole-cochlea DNA editing.^44,48,49,50 The critical target cells, however, are the inner and outer hair cells. To determine whether this population was edited more thoroughly, we isolated total RNA, made cDNA, and sequenced Pcdh15 (which should only be present in hair cells). Editing of Pcdh15 mRNA (∼4%) was higher than in genomic DNA (∼1.5%) but still far less than in cells in vitro (Figures 5C and 5D).

Despite successful editing of the R245X mutation in the mouse cochlea, Pcdh15^R245X/R245X mice showed no rescue of hearing. We hypothesized that the disorganization of hair bundle stereocilia observed at P6 prevented normal formation of tip links even when the genome was edited and wild-type PCDH15 synthesized or some threshold of editing had not been met. We therefore turned to a late deletion model, Pcdh15^R245X/fl, Myo15-Cre, in which the floxed wild-type Pcdh15 is expressed up to the point when the Myo15 promoter becomes active and Cre is synthesized (∼P0). Hair bundles in Pcdh15^R245X/+ mice were normal at P6 (Figure 2), and those in Pcdh15^R245X/fl, Myo15-Cre+ mice were only partly disorganized as late as P30 (Figure S6), presumably because the transient presence of wild-type PCDH15 during early development allowed normal development. Lacking PCDH15 at later stages, these mice were completely deaf. After the floxed allele was recombined in the first post-natal week, the R245X allele remained as a target for editing. As expected, dual AAV delivery of ABEs to these mice restored hearing by an average of ∼25 dB for the best editor and as much as 40 dB in some animals. Surprisingly, ABEmax performed better than ABE8e in rescuing hearing, opposite to the editing seen in vitro and further illustrating the need to test multiple editors in the animal model itself. The rescue of hearing persisted, at least in part, up to P60 in multiple test animals, suggesting that editing the underlying mutation may lead to a more permanent rescue than gene augmentation or replacement therapy.³⁶

Although we tested adenine base editing in the mouse cochlea, which is a demanding environment for PCDH15 and one that provides a rapid and quantitative assay, PCDH15 mutations in humans also compromise vestibular and visual function. The therapy we describe could be especially effective in the vestibular system, where hair cells develop later than in the cochlea. Notably, a cochlear implant can restore some hearing in USH1F patients, but there is not yet an equivalent prosthesis for the vestibular system. Perhaps the greatest opportunity for USH1F therapy is in the retina, which degenerates over nearly 30 years, so there is a long window of intervention for therapy. Because PCDH15 is not immediately critical for photoreceptor function, even a modest editing efficiency may be sufficient to maintain sight. Although retinal photoreceptors primarily use the CD1 splice form of PCDH15, while the inner ear hair cells use CD2, the R245X mutation is near the N terminus and is common to all splice forms. Adaptation of the strategy to the retina requires only packaging in AAV capsids capable of transducing the rods and cones.

Finally, we used a ubiquitous promoter to express the editor. For some transgenes, ubiquitous expression in cells that do not normally express the protein can be detrimental. In contrast, correction of a mutation in a non-expressing cell should be of little consequence.

Materials and methods

Mice

Humanized Pcdh15^R245X mice were generated by targeting the Pcdh15 gene on mouse chromosome 10, exon 9 (GenBank: NM_001142742.1). We used CRISPR-Cas9 to replace 61 bases of the mouse sequence with the equivalent human sequence bearing a C→T mutation at position c.748, creating a stop codon (TGA) that replaces the Arg codon (CGA) at p.250. This mutation is equivalent to the human R245X mutation, and we refer to the mouse model as R245X even though the mutation in the mouse is at amino acid position 250. The endogenous mouse sequence, 5′-GACCGTGCACAAAATCTGAATGAGAGGCGAACAACCACCACCACCCTCACAGTAGATGTTC-3′, was replaced with 5′-GACCGTGCCCAAAATCTGAATGAGAGGTGAACCACCA CCACCACTCTCACAGTGGATGTTC-3′. In addition to the mutation, there are four nucleotide differences between the mouse and human genomic sequences (double underlined), but these are silent and do not change the amino acid sequence. Mouse engineering was performed by Cyagen (Santa Clara, CA, USA) via injection of Cas9 mRNA, R245X oligo, and gRNAs targeting site 1 (5′-CCGTGCACAAAATCTGAATGAGA-3′) and site 2 (5′-CCACCCTCACAGTAGATGTT-3′) into fertilized C57BL/6J eggs (strain 000664). Genotyping of mice was performed by genomic DNA isolated from the ear, which was screened by PCR using primers (F, 5′-GGAGACTGGAGGGCAGCAATCAG-3′; R, 5′-CATTCTGAGACAGGACTTCAGTGGG-3′) and performing Sanger sequencing and/or XcmI digestion, which yielded a 300-bp band for the Pcdh15 R245X allele. R245X mice were mated with inbred C57BL/6J mice to produce R245X/WT, R245X/fl, and R245X/R245X mice.

We created a Pcdh15 conditional knockout mouse with loxP sites flanking exon 31, which encodes the critical transmembrane domain of PCDH15 (described in Ivanchenko et al.⁵¹). Pcdh15^fl/fl, Myo15-Cre+ mice lack PCDH15 in the inner ear and are profoundly deaf.

Myo15-Cre mice, in which the coding sequence for Cre recombinase is knocked into the Myo15 locus, were obtained from Dr. Ronna Hertzano with permission from Dr. Christine Petit and are as described in Caberlotto et al.⁵² Myo15-Cre+ mice were on a mixed C57BL/6–129/Sv background. Myo15 is expressed in the mouse cochlea beginning at P0 in the base. For hearing rescue studies, heterozygous Cre mice (Myo15^Cre/+) were used because homozygotes lack MYO15 and are deaf. Myo15-Cre mice were genotyped using PCR primers to detect Cre (F1, 5′-AGGGACCTGACTCCACTTTGGG-3′; R1: 5′-GGAACTGACCTTTCTTAGAGATCTTGGG-3′; R2: 5′-TGGTGCACAGTCAGCAGGTTGG-3′), where Cre+ yields a 500-bp product, and Cre-yields a 550-bp product. Pcdh15^fl/fl mice were genotyped with primers to detect the loxP sites (F, 5′-GGTTTCTCTGTTTCTTTAAC-3′; R, 5′-GGGTTTCAGTTCTAATAAGA-3′), where wild-type (WT) and R245X alleles yield a 200-bp product, and floxed alleles yield a 300-bp product.

WT C57BL/6J littermates were used for ABR calibration and ABE injection testing. Male and female mice were used for all studies. Injections were carried out on mice up to 1 day after birth. ABRs were performed on 30-day-old mice and sometimes at later ages when assessing long-lasting rescue. All studies involving animals were approved by the Harvard Medical School Standing Committee on Animals, and experiments were conducted in accordance with approved protocols (#03524). All mice used in the study were bred and housed in Harvard Medical School facilities.

Plasmids and cloning

We tested three different gRNAs with SpCas9 for the targeted PCDH15 R245X allele. Most experiments used gRNA1 with a 20-bp spacer region (5′ TTCACCTCTCATTCAGATTT-3′). We additionally tested the effect of longer (23-bp) and shorter (17-bp) spacer sequences as well as spacers that moved the target adenosine to within the canonical 12- to 16-bp editing window of Cas9-SpRY and those specific to Staphylococcus aureus Cas9-KKH and Sp Cas9-VRQR. For construction of intein editors and AAV constructs, we swapped the ABEmax with the ABE8e construct via Gibson cloning using the New England Biolabs (Ipswich, MA, USA) HiFi Assembly Mix (NEB E2621). Additionally, we inserted an R245X-targeting gRNA1 spacer into the C-terminal split-intein Cas9 plasmid so that only three plasmids were required to test ABEmax and ABE8e split-intein constructs.

To construct the PCDH15-tag lentivirus construct, we ordered a human PCDH15 gene block (Azenta Life Sciences, Cambridge, MA, USA) encoding amino acids 1–619 with a FLAG-G₄S linker tag inserted at amino acid 29 to circumvent the signal peptide cleavage sequence and a MYC tag at the C terminus. Subsequent PCR amplification allowed insertion via BmtI and NotI digestion into the CD513B1 lentivirus backbone (SBI) via restriction cloning. The original mCherry:P2A:GFP FlashLight construct was a gift from Dr. Bence György. We inserted an XhoI and AvrII site via PCR mutagenesis before cloning in the PCDH15 R245X target sequence via primer dimerization and T4 ligation using primers with XhoI and AvrII overhangs.

See Supplemental materials and methods for plasmid and spacer sequences.

Nucleic acid isolation, amplification, and sequencing

Genomic DNA was purified from cultured and flow cytometry-sorted cells using DNeasy Blood & Tissue Kits (QIAGEN, Germantown, MD, USA). Similarly, all RNA was collected using Trizol reagent following the manufacturer’s instructions. To isolate DNA and RNA from cochlear tissues, mouse cochleas were surgically extracted into buffer provided in the Quick-RNA Microprep Kit (Zymo Research, Irvine, CA, USA). Cochleas from P1–P8, P30, or P60 mice were pulverized with disposable sterile pestles in a dry-ice-cooled mortar holding a sterile 1.5-mL Eppendorf tube. DNA and RNA were then isolated using the Quick-DNA/RNA Microprep Kit (Zymo Research). cDNA was synthesized using approximately 100–1,000 ng of total isolated RNA, treated with DNAse (Promega, Madison, WI, USA) using random hexamers in the ProtoScript II First Strand cDNA Synthesis Kit (New England Biolabs). All nucleic acid isolations were performed within 10 min of euthanasia. qPCR was carried out using the QuantStudio Pro 6 (ABI) using a standard curve composed of the AAV plasmids used to create the virus prep, SYBR Select Master Mix. To quantify viral GCs, primers amplifying the WPRE or BGH poly(A) sequence in each AAV construct were used (AAV WPRE F, 5′-AATCAACCTCTGGATTACAAAAT-3′; AAV WPRE R, 5′-TTTATACAAGGAGGAGAAAATGA-3′; BGH poly(A) F, 5′-CAGCATGCCTGCTATTGTCTT-3′; BGH poly(A) R, 5′-TAGTTGCCAGCCATCTGTTG-3′).

For high-throughput sequencing of base editing, genomic DNA was isolated as above and amplified via PCR using primers for human or mouse DNA. The 250-bp amplicon sequencing was performed by the CCIB DNA Core Facility at Massachusetts General Hospital using an Illumina MiSeq platform. Editing frequencies were analyzed utilizing the pysam Python program (github URL: https://github.com/pysam-developers/pysam). Pysam was used to generate and align NGS sequencing reads to a reference genome. Follow-up statistics and editing frequency were then analyzed using Microsoft Excel. Python scripts and primers for base editing quantification are shown in Supplemental materials and methods. CCTop40 and CasFinder41 off-target sites were determined using a core length of 12 nt with up to three mismatches as long as the first and last two nucleotides were identical to the gRNA1 used for in vivo gene editing.

Cells

HEK293T cells, purchased from the ATCC (Manassas, VA, USA), were used for testing PCDH15-FLAGMYC, FlashLight, and intein constructs. Similarly, HEK293Ts were used to test AAV intein editor injection. Cells were maintained in DMEM with 10% fetal bovine serum (FBS) (Gibco, Thermo Fisher Scientific, Waltham, MA, USA). Lentivirus was generated via transfection into HEK293T cells of psPAX2, pVSV-G, and the CD513B1 lentiviral backbone harboring the tested constructs using FuGENE (Middleton, WI, USA) transfection reagent. Viral supernatant was filtered through a 0.45-μm polyethersulfone membrane and used for inoculation with target cells to create cell lines. Transfections were carried out using jetPRIME transfection reagent (Polyplus, New York, NY, USA) as specified by the manufacturer.

For culture of human PCDH15^R245X iPSCs, gamma-irradiated mouse embryonic fibroblast (MEF) feeder cells were seeded onto plates or flasks coated with gelatin prior to seeding with iPSCs. MEFs were maintained in DMEM with 10% FBS until used with iPSCs, which were co-cultured in hES medium. hES medium was composed of DMEM-F12 with 20% knockout (KO) serum replacement (Life Technologies, 10828-028), amino acid supplement (Life Technologies, 11140-050), 1× penicillin/streptomycin (Life Technologies, 15140122), 1× L-glutamine (Life Technologies, 25030081), and 1× β-mercaptoethanol (Life Technologies, 21985023). All iPSCs were thawed at 37°C before washing in DMEM/F12 + 10% FBS to remove cell-freezing medium. iPSCs were re-suspended in hES medium supplemented with 1 μM ROCK inhibitor (STEMCELL Technologies, 27632). The following day, medium was exchanged with ROCK-free hES supplemented with 10 ng/mL bFGF (R&D Systems, 233-FB). Feeder-free iPSCs were cultured on Matrigel-coated plates in mTESR1 medium (STEMCELL Technologies, 05850). iPSCs were grown to 90% confluency before transfecting with 4 μg of plasmid containing ABEmax or ABE8e with gRNA1 using GENEJUICE or FuGene.

AAV injection

Cochlear injections of AAV split-intein through the mouse round window membrane were performed in a surgical suite using a Nikon SMZ-745 stereomicroscope. Newborn mice (P0–P1) were cryo-anesthetized for several minutes before being placed upon a 4°C ice pack. Using sterile implements, a small incision was made in the left ear bud, followed by resolution of the round window membrane, utilizing the white stripe, facial nerve, and Y-shaped vessel landmarks. Glass micropipette syringes were used to inject 3.4 × 10¹⁰ genomic copies of both split-intein vectors at a rate of 75 nL/min (total volume, 1.2 μL). After injections, surgical wounds were closed with 7-0 sutures and disinfected with iodine, and mice were placed on a heating pad to recuperate. Injected pups were placed in used bedding for 15 min prior to return to birth or foster mothers, which greatly increased survival after surgery. Surgical tools were cleaned and sterilized using a bead sterilizer before continuing with the next surgery. Entire litters were injected with PBS or with AAVs encoding editor or were left uninjected.

Viral vector generation

AAV vector design was performed using constructs developed by Wei His Yeh and Jon Levy in the Liu laboratory. Vector synthesis was performed at the Boston Children’s Hospital Viral Core using the StrataGene helper-free system: pAAV-RC and pHelper plasmids along with our split-intein constructs flanked by AAV ITRs (Supplemental materials and methods). All constructs used for injection were packaged in AAV9-PHP.B capsids and purified using an iodixanol gradient, followed by filtration concentration using a MilliporeSigma (Burlington, MA, USA) Amicon filter unit (UFC910008, 100,000 MWCO) and F68 PBS. Titering was performed by the Core and validated in our lab via qPCR using the primers described above. Virus aliquots were stored at −80°C and thawed at 4°C immediately before use. Titer stocks were as follows: JL646g1-Cintein Cas9, 1.1 × 10¹⁴ gc/mL; JL645-ABEmax N-intein, 2.54 × 10¹³ gc/mL; JL645-ABE8e N-intein, 3.18 × 10¹³ gc/mL.

Immunofluorescence labeling

Neonatal cochleas were dissected at P6 in L-15 medium, fixed with 4% formaldehyde in Hank’s balanced salt solution (HBSS) for 1 h, washed three times with HBSS, and then blocked with 10% donkey serum for 2 h at room temperature. Samples were incubated in rabbit polyclonal anti-PCDH15 antibody (DC811; diluted 1:200 in 10% donkey serum) for 24 h at room temperature, followed by several rinses in HBSS. Next, samples were incubated in a blocking solution for 30 min and overnight with a secondary antibody conjugated to Alexa Fluor 594 (Thermo Fisher Scientific) in a 1:500 dilution. To label hair bundle actin, we used phalloidin conjugated to Alexa Fluor 405 (1:20).

Adult cochleas (P30+) were fixed and washed in 1× HBSS before being incubated in 120 mm EDTA for 2 days. After the cochleas became pliable, they were carefully dissected to separate the base, middle, and apex, as described previously.⁵³ Tissue was permeabilized by incubation in donkey serum fortified with 0.3% Triton X-100 in 1× PBS or HBSS for 2 h at room temperature. We then incubated tissue with a polyclonal anti-MYO7A antibody (Proteus Biosciences, Ramona, CA, USA; 1:500) overnight at 4°C. The next day, tissue was washed and incubated with secondary antibody (donkey anti-rabbit Alexa 594, 1:500) in blocking solution with Alexa Fluor 405-conjugated phalloidin (1:40) overnight at 4°C.

Tissue was washed and mounted on glass coverslips with Vectashield mounting medium (Vector Laboratories, Newark, CA, USA) and imaged using a Carl Zeiss LSM800 confocal microscope with a 10× or 63× lens. Three-dimensional projection images were generated from Z stacks using Zen Blue (Carl Zeiss).

Scanning electron microscopy

Scanning electron microscopy (SEM) in neonatal mice was performed as described previously.⁵⁴ Neonatal organ of Corti explants were dissected at P6 in L-15 medium and fixed with 2.5% glutaraldehyde in 0.1 M cacodylate buffer (pH 7.2) supplemented with 2 mM CaCl₂ for 1–2 h at room temperature. After rinsing in sodium cacodylate buffer and water, they were dehydrated in an ascending series of ethanol and critical-point-dried from liquid CO₂. Samples were mounted on aluminum stubs with carbon-conductive tabs, sputter coated (EMS 300 T dual-head sputter coater) with 5-nm platinum, and observed with a Hitachi S-4700 FESEM.

FM1-43 loading

Organ of Corti epithelia were acutely dissected from P6 mice in Leibovitz’s L-15 medium. Following tectorial membrane removal and medium aspiration, FM1-43 solution (Thermo Fisher Scientific, 2 μM in L-15) was applied to the tissue for 30–60 s and then quickly aspirated. The explant was then quickly rinsed with L-15, and the excess dye was quenched by a 0.2-mM solution of SCAS (Biotium, Fremont, CA, USA) in L-15. Samples were observed on an Olympus upright FV1000 confocal microscope equipped with a 60 × 1.1 NA water-dipping objective lens.

Hearing tests

ABR and distortion product otoacoustic emission (DPOAE) measurements were recorded using the closed EPL acoustic system (Massachusetts Eye and Ear, Boston, MA, USA). Stimuli were generated with a 24-bit digital input/output card (PXI-4461, National Instruments) in a PXI-1042Q chassis, amplified by an SA-1 speaker driver (Tucker-Davis Technologies), and delivered by two electrostatic speakers (CUI CDMG15008-03A) in a custom acoustic system. Measurements were carried out in a soundproof room. An electret microphone (Knowles FG-23329-P07) at the end of a small probe tube was used to monitor ear-canal sound pressure. ABRs and DPOAEs were recorded from mice during the same session. Mice were anesthetized with an intraperitoneal injection of xylazine (10 mg/kg) and ketamine (100 mg/kg). Three subcutaneous needle electrodes were inserted into the skin: (1) dorsally between the two ears (reference electrode), (2) behind the left pinna (recording electrode), and (3) dorsally at the rump of the animal (ground electrode). Additional ketamine (50 mg/kg intraperitoneal) was given to maintain anesthesia when needed. DPOAEs were recorded first. The f1 and f2 primary tones (f2/f1 = 1.2) were presented, with f2 varied between 5.6 and 32.0 kHz in half-octave steps and (L1 − L2 = 10 dB). At each f2, L2 was varied between 10 and 80 dB in 10-dB increments. The DPOAE threshold was defined from the average spectra as the L2 level eliciting a DPOAE of a magnitude of 5 dB above the noise floor. The mean noise floor level was under 0 dB SPL across all frequencies. ABRs were then recorded, using either broadband clicks or pure tones of a frequency of 4–32 kHz in half-octave steps, all presented as 5-ms tone pips. The responses were amplified (10,000 times), filtered (0.1–3 kHz), and digitized with an analog-to-digital board in the data acquisition system (EPL Cochlear Function Test Suite). The sound level was raised in 5- to 10-dB steps from 0–110 dB SPL. At each level, 256–512 responses were averaged after artifact rejection, with stimulus polarity alternated. The threshold was determined by visual inspection of the appearance of peaks 1–5 in comparison with background noise. Data were analyzed and plotted using Python (gitgud program by Eric Zinn). Thresholds are presented as mean ± SD unless otherwise stated. These tests were performed before genotyping of animals to maintain a blinded approach.

Data availability

All data are available on request from the authors.