Research ArticleGENE EDITING

In vivo base editing restores sensory transduction and transiently improves auditory function in a mouse model of recessive deafness

See allHide authors and affiliations

Science Translational Medicine  03 Jun 2020:
Vol. 12, Issue 546, eaay9101
DOI: 10.1126/scitranslmed.aay9101

At the base of deafness

Mutations in the transmembrane channel-like 1 (TMC1) gene can cause hereditary hearing loss. Current treatments do not target the cause of the disease, focusing mainly on sound amplification. Here, Yeh et al. used base editing methods to treat a mouse model of genetic deafness caused by a point mutation in Tmc1. The authors included a cytosine base editor into a dual–adeno-associated virus (AAV) system and injected it into the inner ear of the animals with hearing loss. The treatment reversed the mutation, improved sensory transduction, and partially restored hearing in mice. These results suggest that base editing might be effective for rescuing recessive genetic hearing loss.


Most genetic diseases arise from recessive point mutations that require correction, rather than disruption, of the pathogenic allele to benefit patients. Base editing has the potential to directly repair point mutations and provide therapeutic restoration of gene function. Mutations of transmembrane channel-like 1 gene (TMC1) can cause dominant or recessive deafness. We developed a base editing strategy to treat Baringo mice, which carry a recessive, loss-of-function point mutation (c.A545G; resulting in the substitution p.Y182C) in Tmc1 that causes deafness. Tmc1 encodes a protein that forms mechanosensitive ion channels in sensory hair cells of the inner ear and is required for normal auditory function. We found that sensory hair cells of Baringo mice have a complete loss of auditory sensory transduction. To repair the mutation, we tested several optimized cytosine base editors (CBEmax variants) and guide RNAs in Baringo mouse embryonic fibroblasts. We packaged the most promising CBE, derived from an activation-induced cytidine deaminase (AID), into dual adeno-associated viruses (AAVs) using a split-intein delivery system. The dual AID-CBEmax AAVs were injected into the inner ears of Baringo mice at postnatal day 1. Injected mice showed up to 51% reversion of the Tmc1 c.A545G point mutation to wild-type sequence (c.A545A) in Tmc1 transcripts. Repair of Tmc1 in vivo restored inner hair cell sensory transduction and hair cell morphology and transiently rescued low-frequency hearing 4 weeks after injection. These findings provide a foundation for a potential one-time treatment for recessive hearing loss and support further development of base editing to correct pathogenic point mutations.


Hearing loss (HL) is one of the most prevalent chronic conditions among children (1). In developed countries, the majority of cases have a genetic etiology (1). Several hundred genes have been identified that cause hereditary HL when mutated. There are currently no biological treatments for any form of hereditary HL. Rather, current HL treatments focus on sound amplification or implanted electrodes that stimulate the auditory nerve, neither of which restores native function. In recent years, gene replacement therapies have yielded partial restoration of auditory function in mouse models of human recessive HL (25). Scientists have also treated genetic HL in mice using gene silencing approaches, including antisense oligonucleotides, adeno-associated virus (AAV)–mediated delivery of microRNAs, and genome editing approaches (610). Our groups, in collaboration with others, recently reduced progressive HL in the Beethoven mouse model of human genetic deafness by disrupting a dominant deafness allele in transmembrane channel-like 1 (Tmc1 in mice, TMC1 in humans) using ribonucleoprotein complexes (11) or viral delivery (9) of CRISPR-Cas9 nuclease. Although gene disruption and gene silencing may be viable strategies for treating dominant pathogenic alleles, the majority (~80%) of genetic HL cases are the result of recessive loss-of-function mutations that require correction, rather than disruption or silencing, of the pathogenic allele to prevent HL (1, 12, 13).

Gene disruption strategies that use programmable nucleases such as Cas9 introduce mixtures of insertions and deletions (indels) at the target site, and in general cannot rescue recessive loss-of-function mutations (1417). Providing exogenous coding sequences of the wild-type gene offers a potential treatment strategy for recessive genetic diseases, but this approach may result in complications arising from overexpression or ectopic expression of the wild-type gene in treated cells (1820). In addition, such therapies may require repeated administration, which increases the likelihood of inducing side effects, including clinically relevant immunogenicity (21).

Alternatively, base editors offer a potential one-time treatment strategy to efficiently and permanently correct pathogenic mutations without disrupting the target gene (2224). Since base editors (~5.2 kb) are too large to fit into a single AAV, we used a split-intein, dual-AAV base editor delivery system that divides a cytosine base editor (CBE) into in two halves, each delivered by a separate AAV (25, 26). Dual transduction and protein splicing reconstitute the active, full-length base editor (25).

To evaluate base editing in a mouse model of recessive HL, we selected the Baringo (Tmc1Y182C/Y182C;Tmc2+/+) mouse (27), which harbors a recessive loss-of-function T•A-to-C•G mutation in Tmc1 (c.A545G) that substitutes Tyr182 for Cys (p.Y182C) and results in profound deafness by 4 weeks of age. TMC1 protein is required for sensory transduction in hair cells of the cochlea (28). Mutations in human TMC1 account for 4 to 8% genetic deafness in some populations (29, 30), which result in dominant and recessive nonsyndromic HL, DFNA36, and DFNB7/11, respectively (31). In humans, 59 of 63 (94%) identified missense mutations in TMC1 result in recessive deafness (9, 32). Thus, the ability to correct recessive loss-of-function mutations could offer broad clinical utility for genetic HL.

We constructed and evaluated combinations of four optimized CBEs and three guide RNAs for their ability to directly correct the pathogenic Tmc1Y182C/Y182C C•G base pair to the wild-type T•A base pair in cultured Baringo-derived mouse cells. We then packaged the most promising combination of CBE and guide RNA into a split-intein, dual-AAV in vivo base editor delivery system and injected it into the inner ears of postnatal day (P)1 Baringo mice. This approach successfully corrected the Tmc1 point mutation back to the wild-type base in vivo, restored sensory transduction in most inner hair cells (IHCs), preserved hair cell morphology in the cochlear apex, and partially rescued auditory function. These results suggest further development of base editing as a potential treatment for recessive HL.


Base editing Tmc1 in vitro

To develop a base editing strategy capable of correcting the Baringo mutation (Tmc1 c.A545G), we began by searching for protospacer sequences at the target site. We identified three protospacer-adjacent motifs (PAMs) that allow binding of Streptococcus pyogenes Cas9 (SpCas9, AGG PAM) or the engineered VRQR SpCas9 variant (33, 34) (GGA or TGA PAM) to the target locus in a manner that positions the target Tmc1 nucleotide within the cytosine base editing activity window (about protospacer positions 4 to 8, counting the PAM as positions 21 to 23) (23, 24, 3335). Three candidate guide RNAs position this target C•G base pair at protospacer positions 7, 8, and 10 (Fig. 1A).

Fig. 1 Evaluating different base editor and guide RNA combinations to correct the Tmc1Y182C/Y182C allele in Baringo MEFs.

(A) Schematic of the Tmc1 locus highlighting the c.A545G mutation (red), silent bystander bases (purple), and three candidate guide RNAs that position the target C (in red) at different protospacer positions (C8, C7, and C10) and use different PAMs (AGG, GGA, and TGA, in blue). (B) Base editing efficiencies for the four CBE-P2A-GFP variants tested with sgRNA1 (where the four CBEs are APOBEC1-BE4max, CDA1-BE4max, evoCDA1-BE4max, or AID-BE4max). Base editing values (blue bars) reflect correction of the Baringo mutation to the wild-type TMC1 protein coding sequence, with no other nonsilent changes or indels. Three days after nucleofection into Baringo MEFs, GFP-positive (GFP+) cells were sorted, and genomic DNA was characterized by HTS. (C) Base editing efficiencies for three different guide RNAs tested with AID-BE4max variants: AID-BE4max + sgRNA1, AID-VRQR-BE4max + sgRNA2, or AID-VRQR-BE4max + sgRNA3. Three days after nucleofection of these plasmids into Baringo MEFs, GFP-positive cells were sorted and sequenced by HTS. (D) Base editing efficiencies in Baringo MEFs after a 14-day incubation with dual AAV encoding AID-BE3.9max + sgRNA1 at high (N terminus, 6.1 × 108 vg; C terminus, 8.3 × 108 vg) and low (N terminus, 3.1 × 107 vg; C terminus, 4.2 × 107 vg) doses. Dots, colored bars, and error bars represent individual biological replicates, mean values, and SEM, respectively (n = 3 to 5).

We also considered potential bystander edits near the target nucleotide in Tmc1, which is located in the sequence 5′…AACAGGAAGCACGAGGCCAC…3′. When the target nucleotide is at protospacer position 8 (C8), no other C nucleotides lie within the canonical CBE activity window (24). Bystander edits at the nearby C10 would result in a silent mutation at the third position of the Ser181 codon. The nearest nonsilent Cs are located at C−8 and C15, well outside the base editing activity window encoded by any of the three candidate single guide RNAs (sgRNAs) described above (Fig. 1A). Thus, anticipated products of base editing should revert Cys182 back to Tyr, with minimal other nonsynonymous amino acid changes, demonstrated in a map showing possible base editing outcomes from different CBE and sgRNA combinations (fig. S1A).

The target Tmc1 nucleotide is in an AGC sequence context. We previously noted that APOBEC1-derived CBEs (including the commonly used BE3 and BE4 variants) (23, 36) edit GC targets less efficiently (23, 37), consistent with the known DNA sequence preferences of APOBEC1 deaminase (38). In contrast with APOBEC1, the Petromyzon marinus CDA1 deaminase (39) and human AID deaminase (4044) both deaminate GC substrates efficiently. To compare the activity of CDA1- and AID-derived base editors at the Baringo mutation site, we constructed nuclear localization-optimized, codon-optimized BE4max (also known as APOBEC1-BE4max) that replaces APOBEC1 with CDA1 (resulting in CDA1-BE4max), with a highly active laboratory-evolved CDA1 variant we recently described (37) (resulting in evoCDA1-BE4max), or with human AID deaminase (resulting in AID-BE4max).

Next, we isolated cells from Baringo mouse embryos to compare the editing efficiency of APOBEC1-BE4max, CDA1-BE4max, evoCDA1-BE4max, and AID-BE4max for targeting Tmc1. Mouse embryonic fibroblasts (MEFs) were extracted from Baringo embryos at day 13.5. We evaluated the ability of APOBEC1-BE4max, CDA1-BE4max, evoCDA1-BE4max, and AID-BE4max to convert the target Tmc1 base pair from pathogenic C•G to wild-type T•A using sgRNA1.

To minimize variability from base editor expression differences among cells, we constructed plasmids encoding each base editor as a P2A–green fluorescent protein (GFP) fusion and isolated GFP-positive cells by flow cytometry before determination of editing efficiency by high-throughput DNA sequencing (HTS). Because P2A is a self-cleaving peptide that couples GFP production with full-length base editor translation, GFP-positive cells must also express base editor. Baringo MEFs were nucleofected with two-plasmid mixtures in which one plasmid expressed sgRNA1 and the other expressed deaminase-BE4max–P2A–GFP (Fig. 1B). After 3 days, we isolated and sequenced the GFP-positive cells only.

As anticipated, APOBEC1-BE4max + sgRNA1 showed inefficient (mean ± SEM of 2.0 ± 0.7%) editing at GC8, likely due to the disfavored sequence context of the target C. In contrast, CDA1-BE4max resulted in 12-fold (P < 0.0001) improved target base editing efficiency (23 ± 1.4%), AID-BE4max resulted in 21-fold (P < 0.0001) more efficient editing (43 ± 0.6%), and evoCDA1-BE4max resulted in 25-fold (P < 0.0001) higher editing (50 ± 2.8%), compared with APOBEC1-BE4max (Fig. 1B). Overall, we observed significant (P < 0.0001) differences in editing efficiencies comparing APOBEC1-BE4max to CDA1-BE4max, AID-BE4max, and evoCDA1-BE4max, respectively. APOBEC1-BE4max, CDA1-BE4max, and AID-BE4max all induced low (up to 1.9%) indels at the target locus, whereas evoCDA1-BE4max resulted in a much higher (18 ± 1.9%) indel frequency (Fig. 1B), consistent with our previous findings (37). The ratio of desired base edit:indels for AID-BE4max (ratio of 23) was much more favorable (P = 0.0001) than for evoCDA1-BE4max (ratio of 2.7) (fig. S1B). Because indels in Tmc1 might lead to hair cell degeneration as observed in homozygous Tmc1 knockout mice (28) and homozygous Baringo mice (27), we chose to proceed with AID-BE4max for base editing of Tmc1.

We tested the effect of varying the position of the Baringo mutation among sgRNA1, sgRNA2, and sgRNA3, which place the target C at protospacer positions 8, 7, or 10, respectively (Fig. 1A). We used SpCas9-based AID-BE4max with sgRNA1 to access its AGG PAM and used AID-VRQR-BE4max, which contains the VRQR variant of SpCas9 that is compatible with NGA PAM sites (34), with sgRNA2 and sgRNA3 to access their TGA and GGA PAMs, respectively. We transfected plasmids encoding each pair of base editor–P2A–GFP:sgRNA variant into Baringo MEFs, sorted for GFP-positive cells, and analyzed them by HTS. We observed 43 ± 0.6% editing from AID-BE4max + sgRNA1, 39 ± 1.4% editing from AID-VRQR-BE4max sgRNA2, and 23 ± 1.4% editing from AID-VRQR-BE4max + sgRNA3 (Fig. 1C). Since the AGG PAM accessed by sgRNA1 resulted in the highest editing efficiency, consistent with sgRNA1 placement of the target nucleotide into the canonical CBE activity window (positions 4 to 8), we chose to move forward with AID-BE4max + sgRNA1 using our dual-AAV delivery system.

Dual-AAV delivery of Tmc1-targeted base editors in vitro

To prevent mutant Tmc1–mediated HL, the Baringo mutation must be corrected in cochlear hair cells in the inner ear. To deliver base editors, we used inner-ear injection of AAV vectors, selecting an ancestrally reconstructed capsid, Anc80L65, hereafter referred to as Anc80, for its demonstrated safety and efficacy in the mouse inner ear (45). To validate the ability of Anc80 to deliver genes into IHCs and outer hair cells (OHCs) of Baringo mice, we performed inner-ear injection of 9.7 × 108 viral genome (vg) of Anc80 AAV encoding GFP driven by the chicken β-actin hybrid (Cbh) promoter into the inner ear of P1 Baringo mice. This viral dose, corresponding to 1.8 × 109 vg/kg, is well within the range of AAV known to be tolerated in human retina in clinical applications (46). We observed higher viral transduction efficiency in IHC (41.7% in apex and 22.6% in base of cochlea) than in OHC (8.3% in apex and 2.6% in base of cochlea) (P = 0.0097) (fig. S2).

Because the coding sequence of base editors (~5.2 kB) exceeds the DNA capacity of AAVs, we modified AID-BE4max in two ways to enable AAV-mediated delivery. First, we divided the base editor into two halves and fused each base editor half with one half of the Npu trans-splicing split intein (25, 47). Second, we removed the second uracil glycosylase inhibitor (UGI) domain in each, yielding AID-BE3.9max. These two changes enabled the base editor along with sgRNA1 and all necessary promoter and regulatory sequences to fit within two AAVs [≤4849 base pairs (bp) each].

To test whether this split-intein dual-AAV strategy–mediated base editing of Tmc1, we transduced Baringo MEFs with dual AAVs encoding AID-BE3.9max + gRNA1 at two dosages. The high dose of the N-terminal half was 6.1 × 108 vg and the low dose was 3.1 × 107 vg; the high dose of the C-terminal half was 8.3 × 108 vg and the low dose was 4.2 × 107 vg. After applying the dual AAV encoding AID-BE3.9max + sgRNA1 to MEFs, we cultured the cells for 2 weeks before analyzing editing outcomes using HTS (Fig. 1D). Treatment of Baringo MEFs with the high dose of AID-BE3.9max AAV resulted in 57% editing (with 4.6% indels) of pathogenic C•G to wild-type T•A at Tmc1Y182C/Y182C in unsorted cells. Treatment of the MEFs with the low dose of AID-BE3.9max AAV resulted in 5 to 10% editing (Fig. 1D). Given the higher editing efficiency from high-dose AAV treatment, we selected dual AID-BE3.9max + sgRNA1 for subsequent in vivo experiments.

Off-target analysis of Tmc1 base editing

Although some base editors are capable of low-efficiency but widespread deamination of RNA (48, 49) that can be minimized by using recently described CBE variants (4851), we did not observe any off-target cytosine deamination within the sequenced region of the Tmc1 mRNA of treated mice (fig. S3). We note, however, that since the above experiments were conducted before base editor variants that minimize RNA editing were reported (4851), potential off-target RNA edits in other transcripts are possible.

Next, we investigated base editing at off-target genomic loci bound by the Cas9:sgRNA1 complex. Previous reports using unbiased genome-wide off-target detection methods for base editors (52) have observed that off-target substrates of CBEs are predominantly a subset of Cas9 nuclease off-targets using the same guide RNAs (52, 53). To maximize the capture of potential off-target substrates from Cas9:sgRNA1, we used circularization for in vitro reporting of cleavage effects (CIRCLE-seq) (54), an unbiased and highly sensitive off-target detection method, to identify potential off-target editing sites associated with Cas9 and sgRNA1. We extracted and fragmented genomic DNA from Baringo MEFs, ligated the ~500-bp DNA fragments into circles, and incubated Cas9 with sgRNA1. After Cas9 incubation, we ligated the cut circles to adaptors and identified the location of DNA cleavage events by HTS (Fig. 2A). This process applied to sgRNA1 resulted in the identification of 28 candidate off-target sites with notable CIRCLE-seq signals (>10 reads).

Fig. 2 In vivo base editing of Tmc1Y182C/Y182C in Baringo mice, in vitro off-target analysis for sgRNA1, and in vivo analysis of hair cell stereocilia bundle morphology.

(A) The 10 most abundant genomic DNA cleavage products (which include the on-target site and nine potential off-target sequences) from Cas9 nuclease + sgRNA1 as identified in vitro by CIRCLE-seq, aligned to the on-target Tmc1 sequence. (B) Editing analysis of the nine candidate off-target sites identified by CIRCLE-seq in MEFs treated with dual AAV encoding AID-BE3.9max + sgRNA1. The on-target locus, plus the top nine off-target sites identified by CIRCLE-seq, was sequenced by HTS. Dots and bars represent biological replicates and mean ± SEM (n = 3). (C) Efficiency of AID-BE3.9max + sgRNA1–mediated editing in treated Baringo (Tmc1Y182C/Y182C;Tmc2+/+) mice. Mouse inner ears were injected at P1 with 1 μl (3.1 × 109 vg of each AAV) of dual AAV encoding AID-BE3.9max + sgRNA1. After 14 days, cochleae were microdissected into base, mid, and apex samples. Genomic DNA was extracted from each sample and sequenced by HTS. Each dot represents the efficiency of generating Tmc1 alleles with wild-type TMC1 protein sequence and no other nonsilent mutations or indels, averaging all samples sequenced from one injected cochlea. To obtain Tmc1 mRNA from the cochlea, we extracted the cochlea at P30, isolated RNA, reverse transcribed into cDNA, and analyzed by HTS. Each dot represents the mRNA from one injected cochlea. (D to F) Representative SEM images at the apical turn of OHCs and IHCs of (D) wild-type (Tmc1+/+;Tmc2+/+) mice, (E) untreated Baringo (Tmc1Y182C/Y182C;Tmc2+/+) mice, and (F) Baringo mice treated with dual AAV encoding AID-BE3.9max + sgRNA1. The organs of Corti were imaged by SEM at 4 weeks. Scale bar, 10 μm.

We then performed amplicon sequencing to measure base editing at the 10 genomic sites with the largest number of CIRCLE-seq reads, including the on-target site and the top nine off-target sites (Fig. 2A). The on-target base editing efficiency we observed for the Baringo allele (from Baringo MEFs transduced with AAV in vitro) was 57% (Fig. 2B). HTS of the candidate off-target amplicons revealed no off-target editing at any protospacer position (Fig. 2B) above that of an untreated control sample (≤0.1% mutation frequency above the untreated control) at any of the nine tested off-target sites tested (Fig. 2B and fig. S4). Collectively, these data suggest that base editing of Tmc1Y182C/Y182C by AAV-delivered AID-BE3.9max and sgRNA1 occurred efficiently and was not accompanied by substantial editing at candidate off-target sites identified by the highly sensitive CIRCLE-seq method. Additional off-target characterization of base editors that target corresponding human loci, however, is needed before such editing agents could be considered for clinical applications.

Characterizing sensory transduction currents in Tmc1Y182C/Y182C;Tmc2∆/∆ mice

The Tmc1Y182C mutation is known to cause deafness in Baringo mice by 4 weeks of age (27); however, the consequence of this mutation on hair cell function is unclear. To determine the effect of the Baringo mutation on sensory transduction currents, we dissected the cochlea from Baringo mice at P7 and recorded sensory transduction currents from hair cells on the same day of dissection. We observed robust hair cell transduction current amplitudes (fig. S5).

On the basis of previous reports (28, 55), we hypothesized that the robust currents in P7 mice were the result of transient expression of Tmc2, which encodes TMC2 and is redundant with Tmc1 in neonatal mice (P8 or younger) (28). After the first postnatal week, Tmc2 mRNA and TMC2 protein expression declines (28, 55), which likely accounts for the decline in current amplitudes we noted between P12 and P27 (fig. S5). To isolate the consequences of the Y182C substitution on transduction current, we crossed Baringo mice with Tmc2 knockout mice to generate Tmc1Y182C/Y182C;Tmc2Δ/Δ mice. Hair cells from Tmc1Y182C/Y182C;Tmc2Δ/Δ mice lacked sensory transduction currents entirely (fig. S5), even during the first postnatal week. Collectively, these findings indicate that the Baringo mutation results in a complete loss of TMC1 function. We conclude that after early postnatal expression of Tmc2 has declined to near zero, the loss of sensory transduction in mature hair cells due to the c.A545G point mutation is the proximal cause of deafness in Baringo mice. These results also suggest that successful base editing of the Tmc1Y182C/Y182C mutation might restore hair cell sensory transduction and auditory function.

Tmc1 base editing in vivo

After establishing that AAV-mediated base editing can directly correct the Tmc1Y182C/Y182C mutation in cultured Baringo MEFs and that hair cells from Tmc1Y182C/Y182C;Tmc2Δ/Δ mice lack sensory transduction, we next tested whether inner-ear injection of dual AAV encoding AID-BE3.9max + sgRNA1 can correct DNA encoding Tmc1Y182C/Y182C. The injection was performed at P1, and the organ of Corti (the part of the cochlea containing hair cells) was extracted from bulk cochlear tissue of treated Baringo mice at P14. We sequenced DNA from cochlear tissue of injected Baringo mice and observed base editing at the Tmc1 locus in the organ of Corti from all three treated mice examined (Fig. 2C). Although the fraction of hair cells in the dissected organ of Corti is estimated to be less than 2% of total cells harvested for HTS (11), the whole organ of Corti from treated mice contained the desired base edit in Tmc1 at an average frequency of 2.3 ± 0.4% (Fig. 2C).

To more directly assess the base editing efficiency of hair cells within organ of Corti samples, we sequenced cochlear Tmc1 mRNA of treated mice by reverse transcription of total mRNA and amplicon sequencing using primers specific to Tmc1. Given that Tmc1 in the cochlea is only expressed among hair cells (28, 56), base-edited Tmc1 complementary DNA (cDNA) observed in the cochlea likely reflects base editing in hair cells. We observed 10 to 51% editing efficiency of Tmc1 mRNA, 5- to 25-fold higher than DNA editing efficiencies measured in bulk organ of Corti (Fig. 2C). Together, these observations confirm in vivo base editing of the Tmc1 locus from treatment with dual AAV encoding AID-BE3.9max + sgRNA1, with likely higher editing efficiency in hair cells.

AAV-mediated in vivo base editing preserves IHC stereocilia morphology

IHCs and OHCs of Baringo mice begin to die around 4 weeks of age, progressing from the base of the cochlea toward the apex (27). To investigate the ability of AAV-delivered AID-BE3.9max + sgRNA1 to preserve hair cells and hair bundle morphology, we injected Baringo mice at P1, euthanized them at P28, and excised inner ear tissue for histological examination. Cochleas were harvested and the entire organ of Corti was dissected, mounted, and stained. Given the lack of high-quality anti-TMC1 antibody to visualize TMC1 directly, we stained with an anti-Myo7A antibody to label surviving hair cells. Confocal microscopy analysis of the immunostained organ of Corti revealed no difference in overall OHC or IHC survival between untreated and treated Baringo mice (fig. S6). First, we observed no overt evidence of inflammation or tissue damage in any of the injected ears within the normal outline of hair cells marked with Myo7A antibody (fig. S6). Second, both groups showed loss of OHCs, especially in the basal region of the cochlea, where almost no surviving OHCs were observed (fig. S6). The IHCs of both groups appeared, by confocal microscopy, to be mostly intact in both apical and basal turns of the cochlea, consistent with prior characterization of Baringo mice (27).

To analyze hair bundle morphology, we used scanning electron microscopy (SEM). High-resolution SEM images revealed substantial morphological differences between treated and untreated Baringo hair bundles, particularly in the cochlear apex. Representative images from a Baringo mouse injected with AAV-AID-BE3.9max + sgRNA1 show visible IHC and OHC bundles from the apical end of the cochlea (Fig. 2, D to F). At the basal end of the cochlea, IHC hair bundles with partially preserved morphology were observed in the treated Baringo mouse (fig. S7). These morphological differences suggest that treatment with AID-BE3.9max + sgRNA1 promotes preservation of normal hair bundle morphology, which is otherwise disrupted in untreated Baringo mice. Because intact hair bundle morphology is a prerequisite for normal hair cell function, these findings raise the possibility that preservation of hair bundles from base editing with AID-BE3.9max + sgRNA1 might render Baringo hair cells functional.

Base editing Tmc1 in vivo restores hair cell sensory transduction current

After establishing that AAV-mediated base editing can directly correct the Tmc1Y182C/Y182C mutation in vivo and that hair cells from Tmc1Y182C/Y182C;Tmc2Δ/Δ mice lack sensory transduction, we next tested whether inner-ear injection of dual AAV encoding AID-BE3.9max + sgRNA1 can rescue sensory transduction currents in auditory hair cells of Tmc1Y182C/Y182C;Tmc2Δ/Δ mice. To identify hair cells with functional sensory transduction, we visualized uptake of FM1-43, a styryl dye that enters hair cells through sensory transduction channels (57, 58). Hair cells lacking functional TMC1 and TMC2 proteins do not internalize FM1-43, whereas cells with functional sensory transduction channels readily take up FM1-43 (28).

We imaged FM1-43 uptake in two groups of Tmc1Y182C/Y182C;Tmc2Δ/Δ mice: an untreated control group and a treated group that received an inner-ear injection of 1 μl of 6.2 × 109 vg total of dual AAV encoding AID-BE3.9max + sgRNA1 at P1. After 5 to 7 days of treatment, we dissected the cochlea from both groups of mice (Tmc1Y182C/Y182C;Tmc2Δ/Δ), cultured the cochleas in vitro for 7 to 10 days, and applied FM1-43. We observed no FM1-43 uptake in the IHCs or OHCs of untreated mice, but robust FM1-43 uptake among 75 ± 10% of IHCs of treated mice, and very little FM1-43 uptake in OHCs of treated mice (Fig. 3, A and B). These results suggest restoration of sensory transduction in IHCs of base editor–treated mice, but not in untreated mice.

Fig. 3 Inner-ear injection of dual AAV encoding AID-BE3.9max + sgRNA1 restores sensory transduction in Tmc1Y182C/Y182C;Tmc2∆/∆ IHCs.

(A) Confocal images of midturn cochlear sections excised from P5 Tmc1Y182C/Y182C;Tmc2∆/∆ mouse cochleae. A representative untreated mouse (top panel) or a representative mouse treated with 1 μl (3.1 × 109 vg of each AAV) of dual AAV encoding AID-BE3.9max + sgRNA1 (bottom panel) are shown. The tissue was cultured for 9 to 13 days and treated with 5 μM FM1-43 for 10 s followed by three full bath exchanges to wash out excess dye. The tissue was mounted and imaged for FM1-43 uptake (green) in IHCs and OHCs. All images are 500 × 100 μm. Scale bars, 50 μm. (B) Quantification of FM1-43–positive IHCs (P = 0.0072) and OHCs (P = 0.026) from untreated (gray) and treated (purple) mice represented as mean ± SD. (n = 3 to 4 different mice in each group). (C) Representative families of sensory transduction currents evoked by mechanical displacement of hair bundles recorded from apical IHCs of untreated Tmc1Y182C/Y182C;Tmc2∆/∆ mice at P8 (untreated, orange), from Tmc1Y182C/Y182C;Tmc2∆/∆ mice treated with dual AAV encoding AID-BE3.9max + sgRNA1 at P14 (blue) and P18 (purple) and from wild-type (WT) Tmc1+/+;Tmc2 +/+ mice at P14 to P16 (gray). Horizontal lines and error bars reflect mean values and SD of three to four independent mice and four to eight hair cells (indicated on top of x axis), with each dot representing one IHC.

To directly assess the effect of in vivo base editing on IHC function, we recorded sensory transduction currents from IHCs. We injected into the inner ear of P1 Tmc1Y182C/Y182C;Tmc2Δ/Δ mice 3.1 × 109 vg of each AAV encoding AID-BE3.9max + sgRNA1 and extracted the organ of Corti at P5. Extracted P5 organ of Corti was maintained in culture and incubated for an additional 7 to 10 days before cellular recording. In agreement with the FM1-43 uptake data (Fig. 3, A and B), IHCs of Tmc1Y182C/Y182C;Tmc2Δ/Δ mice injected with dual AAV encoding AID-BE3.9max:sgRNA1 displayed robust sensory transduction at both time points tested (P14 and P18) (Fig. 3C, in blue and purple). Nine of 14 IHCs from treated mice exhibited current amplitudes that were not significantly distinct (P = 0.1731) from those of wild-type (Tmc1+/+;Tmc2+/+) mice. In contrast, untreated Tmc1Y182C/Y182C;Tmc2Δ/Δ mice showed no transduction currents in any of the 12 hair cells tested (Fig. 3C and fig. S5).

Collectively, the transduction current measurements and FM1-43 data demonstrate that in vivo delivery of dual AAVs encoding AID-BE3.9max and sgRNA1 restored sensory transduction in a substantial fraction (64 to 75%) of IHCs from treated Tmc1Y182C/Y182C;Tmc2Δ/Δ mice, which was absent without treatment (Fig. 3C).

In vivo base editing partially and transiently rescues auditory function

The rescue of IHC morphology and restoration of IHC sensory transduction in base-edited Baringo mice suggest that these mice may exhibit rescued cochlear function compared with untreated Baringo mice, which are profoundly deaf at 4 weeks. To test this possibility, we measured auditory brainstem responses (ABRs) at P30 in untreated Baringo mice and Baringo mice injected at P1.

The ABR threshold is the lowest sound pressure level (in decibel units) needed to generate identifiable auditory brainstem waveforms. Representative families of ABR waveforms recorded in response to 5.6-kHz tone bursts of varying sound intensity are illustrated (Fig. 4). The waveform families in Fig. 4 (A and B) were selected to illustrate representative responses of wild-type (Tmc1+/+;Tmc2+/+) control mice with or without treatment with dual AAV encoding AID-BE3.9max + sgRNA1 inner-ear injection (6.2 × 109 vg) (Fig. 4B and table S1), and Baringo mice with or without the same AAV treatment (Fig. 4A). The ABR threshold for a 5.6-kHz tone burst for wild-type control groups (injected or uninjected) was 30 dB (Fig. 4B). In contrast, all untreated Baringo mice showed no detectable ABR thresholds at the maximum sound amplitude tested (110 dB), indicating profound deafness (Fig. 4A). Treated Baringo mice had ABR thresholds as low as 60 dB (Fig. 4A), representing at least 50 dB of improvement compared with untreated Baringo mice.

Fig. 4 Dual AAV base editor treatment partially restores auditory function in Baringo (Tmc1Y182C/Y182C;Tmc2+/+) mice.

(A) Representative sets of ABR waveforms recorded in response to 5.6-kHz tone bursts of varying sound intensity for untreated Baringo mice (left) and Baringo mice treated with dual AAV encoding AID-BE3.9max + sgRNA1 (right). (B) Same as (A), but with untreated wild-type mice (left) and wild-type mice treated with 1 μl (3.1 × 109 vg of each AAV) of dual AAV encoding AID-BE3.9max + sgRNA1 (right). (C) Mean ABR responses for all four groups (untreated and treated, Baringo and wild-type mice) across all tested frequencies. Untreated Baringo mice (black, n = 10) are shown with no detectable ABR threshold (>110 dB, indicated by the upward arrows). Among the treated Baringo mice (n = 15) injected with dual AAV encoding AID-BE3.9max + sgRNA1, nine showed ABR response improvements of up to >50 dB indicated with purple lines, while six did not show any ABR threshold changes shown with gray lines. Untreated wild-type mice in blue (n = 6). Wild-type mice injected with dual AAV encoding AID-BE3.9max + sgRNA1 in orange (n = 4). (D) The same mice in (C) were subjected to DPOAE testing. All recordings were done at mice age of 4 weeks (P30). (E) Mean ABR responses for three animals tested at two time points (4 and 6 weeks) across all tested frequencies. Baringo mice (n = 3) injected with dual AAV encoding AID-BE3.9max + sgRNA1 at P1 and tested at 4 weeks showed (up to 65 dB) hearing improvement (yellow, blue, and purple solid lines). The same animals were retested at 6 weeks (dashed lines). Solid and dashed lines of the same color denote the same animal tested at 4 and 6 weeks, respectively. Values and error bars reflect mean ± SEM for the numbers of mice specified above.

A summary plot of ABR thresholds as a function of frequency for all four groups is illustrated in Fig. 4C. Of the 10 untreated Baringo mice, none showed detectable ABR across all frequencies tested, even at 110 dB. In contrast, of 15 Baringo mice injected with AAV encoding AID-BE3.9max + sgRNA1, 9 showed some recovery of auditory function, with ABR thresholds at 5.6 and 8.0 kHz averaging ~90 dB, and ABR thresholds at higher frequencies of 11.3, 16.0, 22.6, 32.0 kHz averaging ~95 to 100 dB (Fig. 4C). Thus, across all Baringo mice treated with dual AAV-AID-BE3.9max + sgRNA1, ABR thresholds were significantly improved by 5 to 50 dB across all frequencies tested (P < 0.005; table S2).

We also measured the function of OHCs using distortion product otoacoustic emissions (DPOAEs; Fig. 4D). DPOAE analysis revealed that none of the 15 treated Baringo mice showed recovery of DPOAEs relative to untreated mice. The lack of DPOAEs suggest a lack of OHC recovery, consistent with the lack of sensory transduction in individual OHCs and the lack of OHC bundles in the base (fig. S7). To follow hearing function over longer periods of time, we assessed ABR thresholds for three injected Baringo mice at two time points: 4 and 6 weeks. For all three mice, we observed a decline in auditory function at 6 weeks relative to 4 weeks, although two of the three mice retained detectable hearing function at 6 weeks, in contrast with profound deafness observed among all untreated mice (Fig. 4E). We speculate that this decline may arise from incomplete base editing (Fig. 2C). Indeed, a well-appreciated mechanism of progressive HL (5961) involves the death of dysfunctional (in this study, nonedited) hair cells, which negatively affect neighboring healthy (edited) hair cells.

We analyzed hair cell survival by immunohistochemistry at 4 and 6 weeks of age. Given that more than 75% of hair cell survival is thought to be necessary to maintain stable hearing thresholds based on previous reports (9, 62), our observations of 46 ± 6% hair cell survival at 4 weeks (fig. S8A) is too low to prevent progressive HL, consistent with the continued loss of hearing function in the treated mice. To further characterize the relationship between auditory function and hair cell survival, we plotted the mean ABR threshold as a function of surviving hair cells (fig. S8B) and observed that a lower percentage of surviving hair cells correlated with higher ABR thresholds (R2 = 0.43) and further hair cell loss over time. These results suggest that more efficient base editing is needed to permanently restore auditory function.

Last, to rule out any possible adverse effects of the injection procedure, AAV transduction, or postsplicing intein peptide in the ABR or DPOAE tests, we injected AAV encoding AID-BE3.9max + sgRNA1 into the inner ears of four wild-type mice (Fig. 4, C and D). ABR and DPOAE thresholds of treated wild-type mice were not significantly different (each frequency has a P > 0.1) than those of the untreated wild-type mice (Fig. 4, C and D), confirming that the injection technique, viral capsid, AID-BE3.9max, and sgRNA1 did not have any apparent effect on auditory function in the absence of the Tmc1Y182C/Y182C mutation.

Collectively, these results demonstrate that AAV-mediated base editing of Tmc1Y182C/Y182C improves auditory function in Baringo mice, an important advance for the use of in vivo base editing to recover from recessive sensory impairment.


Recessive loss-of-function mutations are the most frequent cause of hereditary HL. In this study, we used base editing in vitro and in vivo to correct a point mutation in Tmc1 that causes profound deafness. Base editing fully restored hair cell function in the majority of IHCs, preserved hair cell morphology, and partially and transiently restored auditory sensitivity in a mouse model of human recessive deafness. These results demonstrate correction (rather than disruption) of a pathogenic mutation in the inner ear, resulting in improved auditory function, and demonstrate the promise of base editing to directly correct loss-of-function recessive mutations. Among 108 recorded human TMC1 mutations that likely cause genetic HL, 72 can, in principle, be corrected with cytosine or adenine base editors assuming the existence of accessible PAMs (table S3) (22, 23, 33, 35, 36, 63, 64). Although this study focused on a recessive loss-of-function mutation, base editing might also be used to correct dominant mutations.

In vivo delivery of AAV encoding an optimized base editor and guide RNA resulted in up to 50% base editing efficiency in restoring the wild-type coding sequence of Tmc1 in hair cells in Baringo mice. Base-edited hair cells were mostly IHCs, which upon treatment resisted morphological degeneration normally seen in untreated Baringo mice. The treated mice also exhibited normal sensory transduction currents in 65% of IHCs, unlike IHCs of untreated Baringo mice. Treated mice exhibited ABR thresholds that improved by 5 to 50 dB relative to undetectable ABR thresholds observed in untreated Baringo mice at 5.6 kHz. Given that the untreated Baringo mouse used in this study has no detectable auditory function at 4 weeks of age, these outcomes represent an important improvement. For a patient with a similar loss-of-function TMC1 mutation, a similar corresponding improvement would represent the difference between hearing nothing at all to being able to detect salient auditory cues in the environment, such as alarms, ringing phones, or sirens from an emergency vehicle. However, 5 to 50 dB improvement in low frequencies may not be sufficient to enable understanding of spoken language. Therefore, auditory function could potentially be further improved with hearing aids to extend the functional recovery. Although we observed no off-target editing using CIRCLE-seq (54) in mouse Tmc1, we note that a more comprehensive off-target analysis (52), ideally including unbiased base editor off-target profiling, should be conducted before such an editing strategy is contemplated in humans.

In this study, we packaged the base editor into Anc80 viral capsids to target hair cells. The efficiency of viral transduction with 6.2 × 109 was not optimal (23 to 42% transduction in IHCs and 2.6 to 8.3% in OHCs). We speculate that the lack of DPOAE recovery resulted from lower viral transduction efficiency of Anc80 in OHCs, as previously reported (65, 66), or from the lower efficiency of the Cbh promoter in OHCs as noted in fig. S2. Since Anc80 AAV is known to preferentially target IHCs (45, 65, 66), we reasoned that 2.3% editing in the entire organ of Corti (Fig. 2C) is consistent with substantial base editing in IHCs. Although base editing improved the pathological status of the cochlea, we observed progression of HL in treated mice between 4 and 6 weeks. The underlying pathology likely progressed because as nonfunctional hair cells died, they affected neighboring functional cells, causing their dysfunction and death (5961). To prevent progressive HL, two recent studies documented the relationship between hair cell survival and stable hearing thresholds (9, 62), suggesting that more than 75% hair cell survival is needed. In contrast, we observed 46% hair cell survival at 4 weeks (fig. S8), consistent with continued progressive HL.

Although these data indicate that in vivo base editing can repair a recessive loss-of-function mutation and rescue hair cell function, they also suggest that improved efficiency will be required to further advance this strategy. Development of viral capsids and promoters that increase OHC transduction could enable delivery of editing agents to both classes of hair cells and potentially enhance long-term auditory improvement. Future improvements in base editor expression, intein-mediated splicing, and base editing efficiency would also provide opportunities to further improve allele correction. Enhanced editing efficiency could improve allele correction and yield improved outcomes, including enhanced hair cell survival, improved hearing thresholds, and durable long-term recovery of auditory function. Additional challenges remain, including identification of the temporal window for therapeutic intervention in mice and humans, and demonstration of long-term safety of editing agents delivered via viral vectors.

Last, although this study shows promise in restoring some auditory function in a mouse model of human deafness, the equivalent mutation is not yet characterized in human TMC1, and different TMC1 mutations may respond differently to correction by genome editing. Nonetheless, these proof-of-concept data support further development of base editing for correction of point mutations that cause inherited human diseases, including genetic HL.


Study design

This study aimed to use base editing in the postnatal mouse inner ear to correct a recessive loss-of-function Tmc1 point mutation that causes congenital deafness, resulting in the rescue of hair cell sensory transduction, hair cell morphology, and auditory function. We identified base editor variants that correct a recessive mutation in Tmc1 in cultured cells and in vivo. We used AAV to deliver base editors in vitro and in vivo and evaluated editing outcomes using HTS, quantitative reverse transcription polymerase chain reaction, immunolocalization and confocal microscopy, SEM, imaging of FM1-43 uptake, single-cell current transduction recording, histology and imaging of whole cochleae, and measurement of ABR and DPOAE thresholds. Noninjected Baringo mice served as controls. Each experiment was replicated as indicated by n values in the figure legends. All the experimental samples were included in the analysis, with no data excluded. Wild-type and Baringo mice were randomized into injected and noninjected groups without regard to gender. ABRs and DPOAEs were acquired by investigators blinded to the experimental conditions. Sample size was selected on the basis of prior studies (4, 7, 8) and statistical significance; a power analysis was not performed.

Statistical analysis

Statistical analyses were performed with Origin 2016 (OriginLab) or Prism 7. Data are presented as means ± SD or SEM as noted in the text and figure legend. Welch’s t test was used to determine statistical significance (P values). Statistical significance was set at P < 0.05. Error bars and n values of biological replicates for experiments are defined in the respective paragraphs and figure legends.



Fig. S1. Base editing outcomes from different CBE and sgRNA combinations.

Fig. S2. Anc80-Cbh-GFP AAV transduction in IHCs and OHCs in wild-type mice.

Fig. S3. Base editing outcomes from mRNA of treated and untreated Baringo cochlea.

Fig. S4. Base editing at on-target and off-target genomic DNA sites identified by CIRCLE-seq using Cas9 + sgRNA1.

Fig. S5. Transduction currents from IHCs and OHCs of Tmc1Y182C/Y182C;Tmc2+/+ and Tmc1Y182C/Y182C;Tmc2/ mice at different time points.

Fig. S6. Hair cell morphology in the organ of Corti from Tmc1Y182C/Y182C;Tmc2+/+ mice with and without treatment with dual AAV-AID-BE3.9max + sgRNA1.

Fig. S7. Hair bundle morphology in the basal turn of the organ of Corti from Tmc1Y182C/Y182C;Tmc2+/+ mice with and without treatment with dual AAV-AID-BE3.9max + sgRNA1.

Fig. S8. Auditory function and correlation of ABR thresholds with surviving hair cells in Baringo mice at 4 and 6 weeks.

Table S1. P value calculation on ABR and DPOAE data from untreated wild-type mice and wild-type mice treated with dual AAV encoding AID-BE3.9max + sgRNA1.

Table S2. P value calculation on ABR data (5.6 to 16 kHz) from Baringo mice treated with dual AAV encoding AID-BE3.9max + sgRNA1 that showed ABR responses and untreated Baringo mice.

Table S3. Primers used for HTS.

Table S4. CRISPResso2 output for base editing at the target locus.

Table S5. List of base editing targets to correct known pathogenic point mutations in TMC1.

Data file S1. Single data points and exact P values.

References (67, 68)


Acknowledgments: We thank A. Vieira for help in editing the manuscript, M. J. Jennings for assistance with cell sorting, C. Nist-Lund for help with the SEM imaging, and H. A. Rees for the helpful suggestions. Funding: This work was supported by U.S. NIH U01 AI142756, UG3 TR002636, RM1 HG009490, R35 GM118062, and HHMI (to D.R.L.), and the Jeffrey and Kimberly Barber Fund, the Foundation Pour L’Audition, and NIH R01 DC013521 (to J.R.H.). Author contributions: Study design: W.-H.Y., J.M.L., D.R.L., O.S.-O., and J.R.H.; in vitro and in vivo experiments: W.-H.Y., J.M.L., D.R.L., O.S.-O., B.P., and J.R.H.; CIRCLE-seq experiment: W.-H.Y., G.A.N., M.W., and J.C.C.; CRISPResso2 analysis: W.-H.Y. and D.R.L.; provided Baringo mice: R.B.; writing and editing: all authors; resources: D.R.L. and J.R.H. Competing interests: D.R.L. is a consultant and cofounder of Beam Therapeutics, Prime Medicine, Editas Medicine, and Pairwise Plants, companies that use genome editing. D.R.L. is also a cofounder and consultant of Exo Therapeutics and a scientific advisory board member of Tevard Biosciences, Wuxi Biologics, and Voyager Therapeutics. D.R.L., J.M.L., and W.-H.Y. have filed patent applications on “AAV delivery of base editors” (2018/0127780). J.R.H. holds patents on “TMC1 gene therapy” (62638697), is a scientific founder of Audition Therapeutics, and is a consultant to several companies focused on inner ear therapeutics. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials. HTS data are deposited in the NCBI Sequence Read Archive (project number PRJNA578156). CBE and AAV constructs are available from Addgene.

Stay Connected to Science Translational Medicine

Navigate This Article