Mutation-agnostic RNA interference with engineered replacement rescues Tmc1-related hearing loss

Mutation-agnostic RNA interference with concomitant, knockdown-resistant gene replacement, results in robust auditory brainstem response and cochlear hair cell preservation in the murine model of TMC1-related autosomal dominant hearing loss.


Introduction
Hearing loss is the most common sensory deficit. It impacts one of every 500 newborns and in at least 50% of cases is genetic in etiology (1). Current therapies for genetic hearing loss such as cochlear implants and hearing aids are symptomatic in nature and do not restore normal auditory function (2). The advent of precision medicine, next-generation sequencing technologies, and the widespread adoption of targeted genetic testing panels has resulted in unprecedented access to molecular diagnosis and growth in understanding the diverse genetic etiologies of hearing loss. This knowledge, in turn, has spurred research in potentially curative treatments such as hair cell regeneration and gene-specific therapies.
Autosomal dominant non-syndromic hearing loss arises because of pathogenic variants with dominant-negative, gain-offunction, or haploinsufficiency mechanisms of action, and accounts for~15-20% of genetic hearing loss diagnoses (3,4). Several cochlear gene therapies have leveraged adeno-associated virus (AAV) to deliver RNAi or CRISPR/Cas9 constructs to reduce or rescue hearing impairment in neonatal mouse models of autosomal dominant non-syndromic hearing loss (5,6,7). Although these treatments represent substantial advancements toward the clinical application of gene therapy to address hearing loss, delivery before the onset of murine hearing at~P12-P15 limits direct applicability to the human cochlea, which is mature at birth (8,9,10).
TMC1 encodes transmembrane channel-like protein 1, the pore-forming subunit of the mechanotransduction complex required for conversion of the intracochlear pressure wave to an electrical signal (11). Dominant-negative and loss-of-function variants in TMC1 cause DFNA36 and DFNB7/11, respectively, which comprise~2% of genetic hearing loss diagnoses (3). Both dominant and recessive mouse models of TMC1-related deafness have been successfully treated with gene therapies (9,10,12,13,14,15,16). The Beethoven (Bth) mouse is a well-described murine model of DFNA36, which carries the semi-dominant c.1235T > A (M412K) mutation in Tmc1, resulting in a progressive hearing loss phenotype similar to the post-lingual hearing loss seen in humans segregating the orthologous M418K in TMC1 (17). We previously reported the efficacy of mutation-specific RNAi-based gene therapy using an AAV2/9 vector in neonatal and mature Tmc1 Bth/+ mice (6,10).
Traditional RNAi-based gene therapy, however, has several limitations. First, mutation-specific RNAi therapeutics (Fig 1A) must be independently developed and validated for each causative 1 variant. Allele-non-specific RNAi, in which both WT and mutant RNA transcripts are suppressed, and exon-specific strategies are proposed as potential methods to circumvent this limitation (18). However, this approach alone is only suitable for genes in which knockdown of the WT allele or exon does not result in disease. Second, RNAi-based gene therapy is primarily applicable to dominant-negative forms of hearing loss, whereas genetic hearing loss exhibits multiple modes of inheritance, of which recessive inheritance because of biallelic loss-of-function variants is the most common pattern. To overcome inherent limitations of mutation-targeted RNAi strategies, Kiang et al proposed a mutation-agnostic strategy involving RNAi suppression of both WT and mutant transcripts, with simultaneous delivery of RNAiresistant engineered gene replacement (Fig 1B) (19). This approach has since been successfully deployed in animal models of systemic and sensorineural diseases including α-1 antitrypsin deficiency and retinitis pigmentosa (20,21).
We have extended this technique to design and optimize therapeutic AAV vectors, which deliver both an RNAi construct and a custom-engineered, knockdown-resistant WT gene expression construct. The siRNA sequence targets a position present on both WT and mutant alleles of the endogenous mRNA, rather than a specific site carrying the causative mutation. An engineered WT cDNA construct altered by the inclusion of synonymous variants at the siRNA-binding site is included to escape RNAi gene silencing. This therapeutic strategy may be broadly applicable to other forms of hearing loss regardless of the mode of inheritance and mechanism of disease. Herein, we report the (B) Illustration of mutation-agnostic RNAi, wherein the vector delivers an RNAi construct to suppress endogenous (WT and mutant) transcripts of the gene, and a knockdown-resistant exogenous WT allele, which is engineered to have synonymous variants at the miRNA-targeting site. (C) qRT-PCR analysis isolated from COS-7 cells co-transfected with WT Tmc1 and either miTmc1D02-D11 or empty vector. Knockdown efficiency of miTmc1D11 was >80%. Data are the mean ± SEM, and each point represents a single biological replicate, the average of three technical triplicates. (D) Outline of transgene control and gene therapy constructs. The mU6 promoter drives miSafe or miTmc1D11 expression, and CMV promoter drives expression of eGFP or engineered Tmc1 cDNA ± WPRE. (E) miTmc1D11 sequence and diagram of knockdown-resistant exogenous Tmc1. Exogenous WT mouse Tmc1 has been engineered to have six synonymous variants in exon 17 to escape RNAi suppression.
first use of this strategy in a mature murine model of TMC1related hearing loss.

Results
Design of mutation-agnostic RNAi with gene replacement strategy Tmc1 siRNA and engineered gene replacement design siRNAs that do not target the sequence including the M412K mutation were designed to target exons 7,9,11,15,17, and 19 of the endogenous mouse Tmc1 mRNA (NM_028953.2) using siSPOTR ( Fig  S1) (22). The miRNA hairpins were prioritized based on predicted binding affinity and potential off-target score and cloned into the multiple cloning site of the pFBAAVmU6mcsCMVeGFP expression plasmid obtained from the University of Iowa Viral Vector Core Facility. Knockdown efficiency was tested in vitro in COS-7 cells, a Tmc1-deficient cell line, by co-transfection of Tmc1 and miRNA D02-11 expression plasmids followed by qRT-PCR. miRNA D11 (miTmc1D11) resulted in the highest in vitro knockdown efficiency (>80%) relative to empty vector controls ( Fig 1C; knockdown efficiency across contributing individual experiments is summarized in Fig S2). miTmc1D11 was therefore selected for inclusion in the final viral vector design. miTmc1D11 driven by the mU6 promoter and the Tmc1 gene replacement expression construct driven by the CMV promoter were combined into a single plasmid vector (Fig 1D). Tmc1 cDNA (referred to as mTmc1) was engineered to have synonymous variants at degenerative wobble positions within the miTmc1D11targeting site in exon 17 to escape RNAi suppression ( Fig 1E).
All vectors tested resulted in ≥94.5% transduction of inner hair cells (IHCs) throughout all turns of the cochlea (Fig 2D). However, the transduction of outer hair cells (OHCs) varied by AAV serotype and across the tonotopic axis: AAV2/9 eGFP transduced 6.5% of OHCs in the apical turn and 0.5% of OHCs within the basal turn, whereas AAV2/2 eGFP transduced 82.8% and 24.3% of OHCs within the apical and basal turns, respectively ( Fig 2E). AAV2/2 eGFP w/ WPRE resulted in increased transduction efficiency, with 74.8% of OHCs in the basal turn expressing eGFP and eGFP fluorescence detectable within the spiral ganglion region (Fig 2C and E). No hair cell loss was seen at 4 wk in any mice regardless of the type of injected vectors, suggesting that none of the vectors trialed was immediately damaging to the cochlea. At 10 wk of age, loss of 34-74% of IHCs was observed only with AAV2/2 eGFP w/WPRE; no OHC loss was observed ( Fig S3).
Auditory brainstem response (ABR) recordings were obtained at 4 and 10 wk of age after P18 RWM + CF injection. At 4 wk, AAV2/9 eGFP-and AAV2/2 eGFP-injected control mice showed no evidence of hearing impairment because of viral injection (Fig 2A and B). AAV2/2 eGFP w/ WPRE treatment resulted in a slight, variable threshold elevation of 5-15 dB compared with uninjected contralateral ears (Fig 2C). At 10 wk, AAV2/9 eGFP and AAV2/2 eGFP treatment resulted in no hearing impairment (Fig 2A and B), whereas ears treated with AAV2/2 eGFP w/ WPRE exhibited severe hearing loss (Fig 2C). Based on these results, both AAV2/2 and AAV2/ 9 serotypes were selected for packaging of the therapeutic constructs.
Bth/+ controls exhibited hearing loss that progressed to profound levels by 20 wk of age, consistent with prior reports (5,8,15). In light of previous successful rescue with an allele-specific strategy using the AAV2/9 serotype (6), rescue of mouse models of DFNB7/11 using gene replacement, and partial preservation of auditory function in Tmc1 Bth/+ mice with Tmc2 gene replacement alone (12,14), we compared treatment efficacy of AAV2/ 9.miSafe.mTmc1 with AAV2/9.miTmc1D11.mTmc1 to test the efficacy of Tmc1 expression augmentation (gene replacement alone) and with RNAi + gene replacement ( Fig 3B). Until 12 wk of age, thresholds were comparable between mice that received injections with AAV2/9 replacement alone and AAV2/9 RNAi + replacement, with both groups exhibiting mean click ABR thresholds 25 dB better than Bth/+ controls. After 16 wk of age, thresholds of mice that received AAV2/9 gene replacement alone exhibited significantly worse hearing than AAV2/9 RNAi + replacement across mid-and high frequencies. By 20 wk of age, click and 8-kHz tone-pip ABR thresholds were~10-20 dB better in AAV2/9 RNAi + replacement relative to AAV2/9 replacement alone, and~40 dB better relative to untreated Bth/+ controls. Treatment efficacy was limited at 16-32 kHz but was significantly better with RNAi + replacement than with replacement alone. Regardless of therapeutics, DPOAE thresholds of all Bth/+ animals were significantly elevated related to WT controls; DPOAE thresholds did not differ among AAV2/9 replacement alone, AAV2/9 RNAi + replacement, and untreated Bth/+ mice at 20 wk ( Fig 3C).
We hypothesized that improved transduction of OHCs and increased transgene expression might further improve ABR and DPOAE outcomes. We compared ABR thresholds of AAV2/9 RNAi + replacement, AAV2/2 RNAi + replacement, and AAV2/2 RNA + replacement with WPRE at each tested time point to evaluate the difference between AAV2/2 and AAV2/9, and with the use of the WPRE to enhance transgene expression ( Fig 3D). AAV2/2 vectors were delivered at 4-4.8 × 10 12 vg/ml, whereas the AAV2/9 vector was delivered at the highest available titer (2.1 × 10 13 vg/ml) to achieve the best possible transduction with each vector.
At 32 kHz, treatment efficacy was limited regardless of the vector delivered. DPOAE thresholds at 20 wk of age were significantly improved at 32 kHz with AAV2/2 RNAi + replacement relative to Bth/ + controls (Fig 3E), however, suggesting partial preservation of OHC function at this frequency.
Comparing the performance of AAV2/2 RNAi + replacement and AAV2/9 RNAi + replacement with WT and Bth/+ controls, ABR wave 1 recordings were of shorter latency and greater amplitude in mice that received AAV2/2 relative to Bth/+ mice that received AAV2/9 or no treatment (Fig 3F-H). Treatment of Bth/+ mice with AAV2/ 2.miSafe.eGFP did not result in any changes in ABR thresholds ( Fig S4).

Allele-non-specific silencing and engineered replacement prevent hair cell degeneration
Cochlear hair cells labeled with anti-Myo7a antibody were counted at 20 wk of age, and all tested therapeutic vectors were compared with untreated Bth/+ controls. Cells were counted in 400-μm segments 10-20% from the apical tip (apex), 40-50% from the apical tip (middle), and 60-70% from the apical tip (base). There was no evidence of hair cell loss in WT controls ( Fig 4A, leftmost panel), whereas Bth/+ controls exhibited~40% IHC loss in the apical turn and nearly complete IHC loss in the middle and basal turn. OHC loss was limited in the apical and middle turns, with~50% of OHCs lost in the basal turn consistent with previous reports (Fig 4A, rightmost panel) (16).
Hair cell survival was improved with the gene therapy construct using both AAV serotypes. With both AAV2/9 RNAi + replacement and AAV2/2 RNAi + replacement, IHCs and OHCs were almost completely preserved throughout the apex and middle turn ( Fig 4B). However, only AAV2/2 RNAi + replacement resulted in enhanced OHC survival in the basal turn relative to Bth/+ controls (Fig 4C). AAV2/2 RNAi + replacement with WPRE generally resulted in poorer HC preservation than other RNAi + replacement vectors (Fig 4B and  C).
Single-cell RNA-Seq implicates increased transgene expression in hair cell degeneration seen with WPRE-containing vectors Single OHCs were isolated from WT controls, Bth/+ controls, and Bth/+ mice that received AAV2/2 RNAi + replacement with or without WPRE, using a manual micropipetting-based isolation strategy based on visual morphologic identification for single-cell RNA sequencing (scRNA-Seq) as previously described ( Fig 5A) (25). 34 cells, which passed quality filters (Fig 5B), were used for unbiased clustering (n = 7-12 for each group). Cells from our laboratory's scRNA-Seq dataset (previously reported in Ranum et al) were used to confirm gene expression patterns concordant with the morphologically determined cell-type identity (Fig 5C and D) (25). Read counts of Tmc1 were adjusted to account for variability in per-cell sequencing depth, and reads containing the entire exogenous or endogenous sequence at the miRNA target site were quantified.

Discussion
In this study, we propose mutation-agnostic RNAi with exogenous engineered replacement for the treatment of genetic hearing loss. Using this strategy, we successfully prevented progression of hearing loss in a murine model of DFNA36. Treated Bth/+ mice showed dramatic hearing preservation throughout the observation period, with nearly normal thresholds in the best performers. Both click ABR and DPOAE in mice that received AAV2/2 RNAi + replacement were significantly improved relative to untreated mice. Immunohistochemistry at 20 wk showed near-complete IHC preservation and partial OHC preservation, consistent with audiometric findings. For readability, a shorthand description of each experimental group is used throughout this study.
Several authors have reported rescue of autosomal dominant hearing loss models by targeting specific single-nucleotide variants (6,26), but given the extreme heterogeneity of hearing loss, the widespread translation of variant-specific approaches that do not represent founder mutations is likely to be challenging (27). As a result, there is growing interest in exon-and gene-specific strategies. These approaches have been successfully applied to animal models of retinitis pigmentosa (20,28), and herein, we demonstrate that mutation-agnostic RNAi combined with gene replacement is effective in treating hearing loss. This approach obviates the need for costly validation of mutation-or exon-specific gene therapies.
Both hearing and hair cell preservation are markedly improved, even relative to earlier studies in our laboratory that leveraged RNAi alone, suggesting that adding gene replacement is more effective than mutation-specific knockdown alone (6). As previous work has established that gene replacement is an appropriate strategy for the treatment of the mouse model of DFNB7/11 (12), the incorporation of gene replacement makes this strategy applicable (G, H) Wave 1 latency (G) and amplitude (H) at 90 dB SPL of WT control, AAV2/2 RNAi + replacement, AAV2/9 RNAi + replacement, and Bth/+ control were compared. Latency comparison: one-way ANOVA, P < 0.0001, and Tukey's multiple comparisons test, P < 0.0001 for all between-group comparisons; amplitude comparison: Welch's ANOVA, P < 0.0001, and Dunnett's T3 multiple comparisons test: P = 0.003 for AAV2/2 RNAi + replacement versus Bth/+ control, P = 0.02 for AAV2/2 RNAi + replacement versus AAV2/9 RNAi + replacement, and P = 0.2 for AAV2/9 RNAi + replacement versus Bth/+ control. ****P < 0.0001; ***P < 0.001; **P < 0.01; and *P < 0.05. Data are means ± SEM. N mice per experimental condition are indicated in parentheses.
to both dominant and recessive modes of inheritance. Further study, however, is needed to validate the efficacy of this therapeutic strategy in other murine variant models of DFNA36 and DFNB7/11. Although a treatment response was observed in click ABR and at 8-24 kHz, there was limited therapeutic efficacy at 32 kHz, a limitation that may reflect the point of injection (P18), at which time Tmc1 Bth/+ mice already exhibit severe hearing loss at this frequency ( Fig 3C). In earlier studies, our laboratory and others have shown limited efficacy in treatment of the Tmc1 Bth/+ model's severe highfrequency hearing loss, suggesting that damage to HCs may be irreversible (6,12,13). Reduced IHC ribbon synapses have also been observed in the basal turn of Tmc1 Bth/+ mice as early as P28, suggesting defects in mechanotransduction currents early in life may impair maturation of basal IHC synapses (29). However, transduction efficiency may also be contributory, for although this delivery method achieves robust transduction throughout the cochlea, it is least efficient in OHCs within the basal turn (30,31).
The cis-regulatory enhancer element WPRE has been used in previous studies delivering both control and gene therapy constructs to the inner ear (12,13,31,32). Surprisingly, in this study, we observed that WPRE-enhanced transduction of eGFP caused severe hearing and hair cell loss in long-term follow-up (Figs 2C and S3). Although eGFP itself has immunogenic and cytotoxic effects (33), we attribute the observed ototoxicity to WPRE-associated increase in eGFP expression, as WPRE-lacking eGFP expression vectors did not cause hearing or hair cell loss (Figs 2B and S3). Similarly, WPREcontaining RNAi + gene replacement vectors were associated with hair cell loss and minimal improvement in ABR thresholds, in spite of the fact that scRNA-Seq data showed an increase of 248% in Tmc1 expression in OHCs of animals treated with the WPRE-containing vector relative to untreated WT controls (Fig 5G). These findings are consistent with previous reports of cytotoxicity associated with WPRE-induced transgene overexpression (34,35) and suggest that the level of transgene expressed in target cells is important in optimizing hearing outcomes with gene therapy for Tmc1. In contrast, Wu et al recently reported successful rescue in the murine model of Tmc1-related autosomal recessive hearing loss, using WPRE-driven Tmc1 gene replacement delivered by an AAV2/9.PHP.B vector (13). Given the many differences between the present study and the experiments described by Wu et al, the mechanisms underlying these discrepant results are uncertain. Although augmented transgene expression with WPRE is attractive and  Tables S5 and S6. ****P < 0.0001; ***P < 0.001; **P < 0.01; and *P < 0.05. efficacious in some contexts, our results raise long-term safety concerns for its use in cochlear gene therapy.
There are several limitations to this study. Although this mutation-agnostic approach should be equally applicable to other models of TMC1-related hearing loss, further work is necessary to confirm its broad-based applicability. In addition, we have not yet performed studies to determine the mechanism of hair cell death and hearing loss with WPRE-containing vectors, nor to determine the precise level of Tmc1 expression that is cytotoxic. Whether overexpression of the exogenous Tmc1 results in cytotoxicity via direct disruption of cation homeostasis and membrane potentials or an alternative mechanism is at play remains to be determined. The number of cells in the scRNA-Seq experiments reported in this study is small and insufficient to thoroughly characterize the transcriptional response to gene therapy or decisively determine in vivo knockdown efficiency of endogenous Tmc1 by miTmc1D11. Thoroughly cataloguing transcriptional responses to the vectors used in the study herein might reveal the mechanisms underlying failure to rescue high-frequency hearing and the etiology of hearing loss with WPRE-carrying vectors. Lastly, the packaging capacity of AAV is limited to~5 kb, which has spurred application of myriad strategies including dual-vector therapies and alternative vectors to deliver large packages to the murine cochlea (36,37). Simultaneous delivery of an RNAi construct, gene replacement, and associated promoter and regulatory elements is likely to be infeasible for larger hearing loss genes, mandating the use of such techniques. (G) Tmc1 expression was increased in OHCs of Bth/+ animals that received AAV2/2 RNAi + replacement and AAV2/2 RNAi + replacement with WPRE relative to WT control and Bth/+ animals (Kruskal-Wallis test, P = 0.001, and Dunn's test: P = 0.001 for AAV2/2 RNAi + replacement with WPRE versus Bth/+ control and AAV2/2 RNAi + replacement with WPRE versus WT control, P = 0.04 for AAV2/2 RNAi + replacement versus AAV2/2 RNAi + replacement with WPRE and AAV2/2 RNAi + replacement versus Bth/ + control, and P = 0.045 for AAV2/2 RNAi + control versus WT). *P < 0.05 and **P < 0.01.

Conclusion
In this study, we have proposed a mutation-agnostic approach to cochlear gene therapy leveraging allele-non-specific RNAi and exogenous replacement and demonstrated this strategy's feasibility for TMC1-related hearing loss in a mature murine model of DFNA36. This approach demonstrates superior long-term hearing and hair cell preservation as compared to previously reported RNAi allele-specific strategies or gene replacement alone. Mutation agnosticism allows this strategy to be leveraged for a large group of patients, obviating the need to develop and validate constructs on a variant-or exon-specific basis. In future studies, we plan to apply this mutation-agnostic RNAi strategy to other models of genetic deafness.

Ethics approval
All experiments were approved by the University of Iowa Institutional Biosafety Committee (IBC; rDNA Committee; rDNA Approval Notice #100024) and the University of Iowa Institutional Animal Care and Use Committee (IACUC; Protocol #06061787) and were performed in accordance with the NIH Guide for the Care and Use of Laboratory Animals.

Mice
Mice were housed in a controlled temperature environment on a 12-h light/dark cycle. Food and water were provided ad libitum. Isogenic heterozygous Beethoven mice (Tmc1 Bth/+ ) maintained on a C3HeB/FeJ (C3H) background were obtained as a gift from Dr. Karen Steel. Inbred WT C3H mice were obtained from the Jackson Laboratory. Crossbred homozygous Tmc1 Bth/Bth mice were caged with WT C3H mice to generate Tmc1 Bth/+ animals. Genotyping was done on DNA from tail-clip biopsies extracted using the Proteinase K method and amplified with forward (59-CTAATCATACCAAGGAAACA-TATGGAC-39) and reverse (59-TAGACTCACCTTGTTGTTAATCTCATC-39) primers in a 20-ml volume containing 40 ng of DNA, 24 pmol of each primer, and BioLase DNA polymerase (Bioline) to generate a 376-bp amplification product in Tmc1 Bth/+ mice. Amplification conditions included an initial 5-min denaturation at 95°C followed by 35 step cycles of 1 min at 95°C, 1.5 min at 58°C, and 1.5 min at 72°C, with a final elongation of 5 min at 72°C. PCR products were purified and sequenced on an automated sequencer (ABI PRISM 3130xl Genetic Analyzer; Applied Biosystems). An approximately equal number of mice of each sex were used for all experiments, with eight groups receiving seven viral vectors as described in Table 1. siRNA design siRNA sequences were designed using siSPOTR to query possible siRNA-binding positions in mouse Tmc1 exons 7, 9, 11, 15, and 17 (22). The top 16 siRNA sequences were selected based on their predicted binding affinity and potential off-target score.

miRNA design
Oligonucleotides were obtained for each siRNA design and cloned into the multiple cloning site (mcs) of the pFBAAVmU6mcsCMVeGFP plasmid obtained from the University of Iowa Viral Vector Core Facility.

Engineered WT Tmc1 expression construct
The engineered WT Tmc1 construct was created by inserting synonymous variants in Tmc1 at the position of the siRNA-binding site in exon 17 (Fig 1D). Oligonucleotides for this engineered Tmc1 were cloned into the pFBAAVmU6mcsCMVeGFP plasmid in the position of the eGFP construct, replacing eGFP with engineered Tmc1 driven by the CMV promoter.
qRT-PCR 11 of the 16 designed miRNA hairpins were successfully cloned into the multiple cloning site of pFBAAVmU6mcsCMVeGFP expression plasmids obtained from the University of Iowa Viral Vector Core Facility. Ten of the 11 plasmids were successfully evaluated for knockdown performance in an in vitro co-transfection assay using COS-7 cells, a Tmc1-deficient cell line, grown in DMEM (Invitrogen, Thermo Fisher Scientific) with 10% FBS at 37°C with 5% CO 2 . In vitro miRNA screening consisted of co-transfection of the above miRNAexpressing plasmids with a mouse Tmc1-expressing plasmid pAcGFPmTmc1ex1WT provided by Dr. Andrew Griffith. The transfection mix was made using Lipofectamine 2000 (Invitrogen, Thermo Fisher Scientific) according to the manufacturer's protocol. RNA was extracted from cells using TRIzol (Invitrogen, Thermo Fisher Scientific).
Initial qRT-PCRs were performed using single biological replicates (technical triplicate) to enable evaluation of higher numbers of miRNAs simultaneously. Top candidates from this initial screening were carried forward and evaluated alongside other topperforming and untested candidate miRNAs in both biological and technical triplicates. Expression levels were assessed in triplicates by qRT-PCR (StepOne Plus, ABI) using intron-spanning Tmc1 forward (59-GTTCGCCCAGCAAGATCCTGA-39) and reverse (59-GGATGGTAATCTTC-CAGTTCAGCA-39) primer sets and One-step SYBR PrimeScript RT-PCR Kit 2 (Takara-Clontech), normalizing results to β-actin forward (59-TGAGCGCAAGTACTCTGTGTGGAT-39) and reverse (59-ACT-CATCGTACTCCTGCTTGCTGA-39) primer sets. Knockdown was assessed in comparison with a control transfected with pAcGFPmTmc1ex1WT and miRNA-deficient empty vector pFBAAV-mU6mcsCMVeGFP. Results were normalized to β-actin with the ΔΔCt algorithm.

Animal surgery
A round window membrane combined with canal fenestration (RWM + CF) injection was performed at P16-18, as described previously (30). Mice were anesthetized with an intraperitoneal injection of ketamine (100 mg/kg) and xylazine (10 mg/kg). Body temperature was maintained with a heating pad during the surgical procedure. The left post-auricular region was shaved and cleaned, the site was prepped and draped, and the surgical site was observed under an operating microscope. A post-auricular incision was made to access the temporal bone. After exposing the facial nerve and the sternocleidomastoid muscle (SCM), a portion of the muscle was divided to expose the cochlear bulla ventral to the facial nerve, and the posterior semicircular canal (PSCC) dorsal to the cochlear bulla. A 0.5-to 1.0-mm-diameter otologic drill (Anspach Emax 2 Plus System; DePuy) was used to make a small bullostomy, which was then widened with forceps to visualize the stapedial artery and RWM. A canalostomy was also created in the PSCC with a 0.5-mm-diameter diamond drill; slow egress of perilymph confirmed a patent canalostomy.
1.0 μl of AAV vector with 2.5% fast green dye (Sigma-Aldrich) was loaded into a sharpened borosilicate glass pipette (1.5 mm outer diameter [OD]~0.86 mm inner diameter [ID]; Harvard Apparatus) pulled with a Sutter P-97 micropipette puller to a final OD of~20 μm and affixed to an automated injection system pressured by compressed gas (Harvard Apparatus). Pipettes were manually controlled with a micropipette manipulator. Gene therapy vectors were delivered at titers specified in Table 1. The center of the RWM was gently punctured, and AAV was microinjected into the scala tympani for 3-5 min. Successful injections were confirmed by visualizing the efflux of green fluid from the PSCC canalostomy. After removal of the pipette, the RWM niche was sealed quickly with a small plug of muscle to avoid leakage. The bony defect of the bulla and canal was sealed with small plugs of muscles and Vetbond tissue adhesive (3M). 6-0 absorbable sutures and 6-0 nylon monofilament sutures were used to close the SCM and skin, respectively. Total surgical time ranged from 20 to 30 min. After all procedures, mice were placed on a heating pad for recovery and rubbed with bedding. Pain was controlled with buprenorphine (0.05 mg/kg) and/or flunixin meglumine (2.5 mg/kg) for at least 2 d. Recovery was closely monitored daily for at least 5 d post-operatively. All animals were operated on by one surgeon (Y.I.).

Auditory testing
ABRs were recorded as described previously (30). All mice were anesthetized with an intraperitoneal injection of ketamine (100 mg/ kg) and xylazine (10 mg/kg). All recordings were conducted from both ears of all animals on a heating pad using electrodes placed subcutaneously in the vertex and underneath the left or right ear. Clicks were square pulses 100 ms in duration, and tone bursts were 3 ms in length at distinct 8-, 16-, 24-, and 32-kHz frequencies. ABRs were measured with BioSigRZ (Tucker-Davis Technologies) for both clicks and tone bursts, adjusting the stimulus levels from 90 dB SPL to 5 dB SPL in a 5 dB decrement. Electrical signals were averaged over 512 repetitions. ABR thresholds were defined as the lowest sound level at which a reproducible wave 1 could be observed. For transduction efficiency analysis, ABRs were measured at 4 and 10 wk; for gene therapy experiments, they were measured at 4-wk intervals for 20 wk. Responses from either the left ear in WT or untreated Tmc1 Bth/+ mice, or the contralateral ear that did not undergo surgery were used as controls.
Distortion product otoacoustic emissions (DPOAEs) were measured after ABR recording. DPOAEs were measured at the following center frequencies (Fc): 4, 8, 16, and 32 kHz. Primary signals (F1 and F2) were calculated using Fc where F1 is Fc ×0.909 and F2 is Fc ×1.09. F1 and F2 were generated from two separate MF1 speakers at an intensity level from 80 to 20 dB SPL in a 10 dB decrement. The level of two primary tones was set equal (L1 = L2). DPOAE threshold was defined as the lowest level of F2 where the DPOAE amplitude at 2F1 -F2 was at least 3 dB above the average level of noise floor.

Immunohistochemistry, cell counts, and transduction efficiency analysis
All injected and non-injected cochleae were harvested after animals were euthanized by CO 2 inhalation. Temporal bones were locally perfused and fixed in 4% PFA for 2 h at 4°C, rinsed in PBS, and stored at 4°C in preparation for immunohistochemistry. Specimens were visualized with a dissection microscope and dissected for whole-mount analysis. In all cochlear whole mounts, GFP was detected by intrinsic fluorescence. After infiltration using 0.3% Triton X-100 for 30 min and blocking with 5% normal goat serum for 30 min, tissues were incubated with rabbit polyclonal Myosin-VIIA antibody (#25-6790; Proteus Biosciences) diluted 1:200 in PBS for 1 h. Subsequently, fluorescence-labeled goat anti-rabbit IgG Alexa Fluor 568 (#A-11036; Thermo Fisher Scientific) at 1:500 dilution was used as a secondary antibody for 30 min. Specimens were mounted using ProLong Diamond Antifade Mountant with DAPI (#P36965; Thermo Fisher Scientific) and observed with a Leica TCS SP8 or Leica DMi8 confocal microscope (Leica Microsystems). Cell counts and transduction efficiency analysis were performed as described previously (30,31). Z-stack images of whole mounts were collected at 10×-20× on a Leica SP8 confocal microscope. Each turn of the cochlea was analyzed as follows: 10-20% from apical tip (apex); 40-50% from apical tip (middle); and 60-70% from apical tip (base), corresponding to approximate frequencies of 8, 16, and 24-32 kHz, respectively (38). Maximum-intensity projections of Z-stacks were generated for each field of view, and images were prepared using LAS X (Leica Microsystems) to meet equal conditions. IHCs and OHCs were counted per 400-μm cochlear sections for each turn in each specimen with ImageJ Cell Counter (NIH Image). For transduction efficiency, IHCs and OHCs with positive eGFP and with immunostaining for Myo7a were counted; the total numbers of HCs and eGFP-positive HCs were summed and converted to a percentage.
Single-cell RNA sequencing scRNA-Seq was performed as previously described, with minor modifications (25). After euthanasia via CO 2 at P57-P69, murine temporal bones were harvested and transferred to a culture dish containing ice-cold 1× DPBS for visualization under a dissecting microscope (Model M165FC; Leica Microsystems). Bone covering the apical turn was removed with #5 forceps (Fine Science Tools) to access the membranous labyrinth, which was transferred to a 1.5-ml tube containing Accumax (Innovative Cell Technologies) for digestion at RT for 5 min. Gentle pipetting was used to mechanically dissociate single hair cells, and the disrupted tissue and digestion buffer were then transferred to a glass microscope slide (Superfrost Plus 25 × 75 × 1.0 mm; Thermo Fisher Scientific). Hair cells were visualized on an inverted microscope (Model DMI3000B; Leica Microsystems) equipped with 20× and 40× DIC objectives and a 3D micromanipulator (Model MN-153; Narishige) driving a pulled glass micropipette attached to a nitrogen gas-powered pico-injector (Model PLI-100; Harvard Apparatus) used to control aspiration pressure. A wash slide of 1× DPBS was immediately adjacent. Candidate cells were identified by scanning the field of view and aspirated into the first glass micropipette in a slow and controlled manner. Cells were transferred to the adjacent 1× DPBS wash slide and washed to remove extracellular debris, then reaspirated using a second clean glass micropipette. The washed cell was expelled into a 0.2-ml tube containing RNase inhibitor-containing lysis buffer (Takara Bio). The tubes were briefly centrifuged and flash-frozen on dry ice, then stored at −80°C before RT-PCR and amplification using Smart-Seq HT (Takara Bio) according to the manufacturer's protocol. Illumina library preparation was performed with Nextera XT (Illumina) before equimolar pooling for sequencing on a single lane of an Illumina NovaSeq 6000 using 250-bp paired-end read chemistry.

Illumina sequencing data analysis
Raw sequence FASTQs were pseudo-aligned and transcripts quantified by Kallisto 0.48.0 (39) using the Ensembl 106 Mus musculus GRCm39 transcriptome and 100 bootstrap samples. Transcript counts were summarized as per-cell gene counts using tximport 1.24.0 (40). Expression patterns were confirmed to be concordant with OHCs by unbiased clustering analysis using Seurat 4.1.1 (41). First, gene counts were combined with counts computed in the same way for known Deiters' cells, IHCs, and OHCs at P15 from Ranum et al (25). Genes were filtered to those present in at least three cells, and cells were filtered for quality to those with at least 3,000 such genes expressed and mitochondrial reads ≤7%. Normalization, feature selection, and scaling were performed per defaults. Cells were then clustered using the first three principal components and resolution 0.05-parameters empirically chosen to form three celltype clusters. Finally, cluster membership for each cell in this study was examined and cells belonging to the cluster with all the known outer hair cells were retained for subsequent analyses.

Statistical analysis
Two groups were compared using Welch's t test. For comparisons between more than two groups, one-way ANOVA was performed followed by post hoc analysis with Tukey's multiple comparisons test; when SDs were significantly different by Bartlett's test, Welch's ANOVA with post hoc Dunnett's T3 multiple comparisons test was performed. P < 0.05 was considered statistically significant. Statistical analysis of Tmc1 expression was performed using R 4.2.0 (42) and the FSA 0.9.3 R package Dunn's test function (43). To estimate endogenous and exogenous Tmc1 expression and knockdown efficiency in vivo, reads containing exact matches for the endogenous (GACGATCATGTACAGGACCGG) or the exogenous (CACTATCATATA-TAGTACTGG) sequence, or the reverse complement thereof, at the RNAi-binding site were counted to estimate the percentage of endogenous and exogenous reads in all cells containing reads at the siRNA-binding site. Per-cell Tmc1 expression was then multiplied by these percentages to generate a rough estimate of the expression of exogenous and endogenous Tmc1 on a per-cell basis. Tmc1 expression levels were compared using the Kruskal-Wallis test followed by post hoc analysis using Dunn's test with the Benjamini-Hochberg adjustment.

Data Availability
RNA-sequencing data reported in this study are available at https://umgear.org/p?s=c65e0b22 and will be submitted to the Gene Expression Omnibus database. Software resources used are summarized in the Reagents and Tools Table. Requests for additional data will be distributed upon request.