In a small clinical trial, methylprednisolone administered through intra-tympanic injection improved hearing sensitivity and speech discrimination for a subset of SSNHL patients (4/20) who failed to respond to oral steroids (Slattery et al., 2016). In another study, concentration of methylprednisolone in the perilymph was shown to be 126-fold higher following an intra-tympanic injection (40 mg) compared to that after i.v. injection (1 mg/kg) in human patients (Bird et al., 2007). Intra-tympanic dexamethasone also resulted in significant improvement in hearing sensitivity in a retrospective clinical study of 10 patients (Chandrasekhar, 2001). However, in a multi-center clinical trial with 250 unilateral SSNHL patients, intra-tympanic delivery of methylprednisolone (10 mg/day × 4 dose over 14 days) yielded similar hearing recovery as the orally administered prednisone (60 mg/day × 14 days, 5 days taper) (Rauch, 2011). Prednisone is only available in oral formulations and thus a similar corticosteroid, methylprednisolone, which is available in injectable form and slightly more potent, was used for intra-tympanic delivery, which could have contributed to the observed results. As such, the superiority of intra-tympanic steroid over oral steroid remains inconclusive (Stachler et al., 2012).

In a preclinical study, intra-tympanic administration of dexamethasone (0.12–0.168 mg/ear) has been shown to reduce cisplatin-induced ototoxicity (14 mg/kg intraperitoneal cisplatin) in 1–2 months old mice (Hill et al., 2008). The administration preserved auditory brainstem response (ABR) threshold to near-normal levels at lower frequencies (8 and 16 kHz) although the ABR threshold only increased by 20 dB at 32 kHz. In a cisplatin-treated (12 mg/kg intraperitoneal cisplatin) guinea pig model, similar otoprotective effect has been shown for intra-tympanic dexamethasone (0.4–1.2 mg/ear), the administration of which preserved the distortion product otoacoustic emissions (DPOAEs) amplitudes at 1–6 kHz to near-normal levels prior to cisplatin injection (Daldal et al., 2007).

Physical devices have also been designed to set up a conduit from external ear to the cochlea for drug infusion. The Silverstein MicroWick™ is a small catheter made by polyvinyl acetate, surgically inserted through an opening made on the TM to be placed just outside the RWM. One end of the catheter is accessible from the external ear canal to enable self-administered gentamicin for vertigo alleviation (Hill et al., 2006) or methylprednisolone for treatment of SSNHL (van Wijck et al., 2009). Drug application through MicroWick™ showed improvements in clinical outcome among 53 out of 69 MD patients for managing vertigo symptoms (Hill et al., 2006) and among 8 out of 12 SSNHL patients for improving the pure tone average response (van Wijck et al., 2009).

Refractory Ménière’s disease (MD) has also been managed clinically via intra-tympanic injection of corticosteroids, e.g., dexamethasone and methylprednisolone, based on their anti-inflammatory effects (Patel et al., 2016; Nevoux et al., 2018). Treatment of MD with severe vertigo sometimes resorts to intra-tympanic injection of gentamicin, termed transtympanic gentamicin (TTG) in clinical settings, despite the drug’s potential ototoxicity (Harner et al., 2001; Suryanarayanan et al., 2008).

Based on the hypothesis that diffusion across the RWM or oval window is the rate-limiting step in intra-tympanic-injection, recent research efforts have focused on delivery modalities designed to increase the membrane permeability.

Several adjuvant agents are known to temporarily upregulate RWM permeability. For example, dexamethasone co-administered with histamine onto a gelatinous hyaluronic acid disk placed outside the RWM showed enhanced drug penetration compared to dexamethasone-alone in guinea pigs (Creber et al., 2019). A few known chemical permeation enhancers, including benzyl alcohol, saponin, and N-methyl-2-pyrrolidone (NMP), have also been shown to increase the permeation of fluorescent dexamethasone into the perilymph in guinea pigs, likely through disrupting the lipid bilayers of the RWM (Li et al., 2018).

Nanocarriers, including liposome, micelles, polymeric nanoparticle, and dendrimers which are less than 1 μm in size, have emerged recently as promising delivery vehicles to permeate the RWM. These nanocarriers can mask the physical characteristics of their payloads and thus increase drug stability and permeability. Some nanocarriers can translocate across the RWM through active transport. For example, PLGA nanoparticles (NP) were reported to have permeated the RWM via transcellular pathways instead of paracellular pathways; they entered the epithelial cells predominantly through macropinocytosis and caveolin-mediated endocytosis, and were degraded by the digestive endolysosomal pathway and/or secreted via exocytosis (Zhang et al., 2018). In a guinea pig model, PLGA NPs with the diameter of 140–180 nm were placed outside the RWM and shown to enter the perilymph more effectively than systemically delivered PLGA NPs (Tamura et al., 2005). Liposomes (diameter = 85 nm) and polymersome (diameter = 90 nm) NPs have also been reported to successfully carry a neurotoxic agent, disulfiram, across intact RWM in mice (Buckiová et al., 2012).

Superparamagnetic iron oxide NPs (SPION) were placed onto RWM and magnetized by an external magnetic field for directed entry into the perilymph (Ge et al., 2016). Although SPION itself cannot carry a payload, it has been encapsulated by drug-loaded PLGA to form a composite NP for targeted delivery. Interestingly, transmission electron microscopy (TEM) results suggested that SPION-PLGA composites (diameter 160–280 nm) were distributed throughout the inner ear with and without magnetic intervention. These NPs likely crossed the RWM via simple diffusion along the concentration gradient. Another magnetic NP was developed by Otomagnetics, Inc. (Ramaswamy et al., 2017). Methylprednisolone loaded into this magnetic NP was placed intra-tympanically into the middle ear of mice treated with cisplatin (4 mg/kg daily for 10 days in total). The drug-loaded magnetic NPs were directed through the RWM by a 0.5-Tesla external magnetic field. Magnetic delivery significantly reduced the incidence of hearing loss (53% at 32 kHz) compared to intra-tympanic injection only (97% at 32 kHz) or saline control (93%), and reduced cisplatin-induced cytotoxicity in the OHCs by 3.6-fold compared to intra-tympanic injection and by 7.2-fold compared to saline. This magnetic treatment caused reversible localized inflammation in the middle ear and no adverse safety issues (Shimoji et al., 2019).

The total amount of therapeutics in the inner ear can also be increased by prolonging the period over which active drug permeation across the RWM/oval window takes place, e.g., by increasing the residence time of the formulation in contact with the RWM (El Kechai et al., 2016).

Hydrogels are frequently used to prolong the release of small molecules and macromolecules. Criteria for a suitable hydrogel system for intra-tympanic delivery include low tissue toxicity or immunogenicity, biodegradability, mechanical tunability and sustained drug release profile. Fibrin, a hydrogel naturally derived from fibrinogen and thrombin, was used as a sustained-release vehicle for intra-tympanic gentamicin in chinchilla models (Balough et al., 2016). After a single dose injection of 2.5 mg gentamicin into the middle ear, gentamicin was detected in the perilymph for 72 hr at concentration above 50 μg/ml and not detected at all in the blood. The fibrin glue was also used in a prospective clinical study to treat MD patients (Casani et al., 2016). A single injection of 6 mg gentamicin into the middle ear was able to reduce the clinical signs of MD in 22/26 patients.

Dexamethasone loaded in silk fibroin-polyethylene glycol (PEG) hydrogel, applied onto the RWM of guinea pigs, helped maintain measurable drug concentration in the perilymph (100 ng/ml) for over 10 days and showed complete degradation of the hydrogel in 21 days (Yu et al., 2016). Likewise, dexamethasone impregnated in a poloxamer formulation, OTO-104, showed prolonged release for over 3 months in guinea pigs after intra-tympanic injection (Piu et al., 2011). This formulation was later marketed as Otividex by Otonomy Inc. Results from a phase III trial with 148 MD patients showed that a single intra-tympanic injection of Otividex did not effectively reduce the number of days of vertigo measured at 3 months compared to placebo (p = 0.312) (Taylor, 2021).

Hydrogels have been combined with NPs to endow sustained release kinetics to the nanocarriers. For example, interferon (IFN)-loaded PLGA NPs (diameter = 290 nm) were incorporated into a thermosensitive hydrogel composed of chitosan and glycerophosphate (Dai et al., 2017). The IFN-NP-hydrogel composite underwent sol-gel transition after intra-tympanic injection in guinea pigs and increased the IFN mean residence time in the perilymph by 3.2-fold compared to IFN solution, by 1.3-fold compared to IFN-PLGA, and by 1.6-fold compared to IFN-hydrogel. In another study, fluorescently labeled liposomes (diameter = 160 nm) were loaded into a chitosan hydrogel to increase the contact time with mice RWM, which led to the presence of liposomes in the perilymph and the cellular structures in the scala media 24 h after the injection (Lajud et al., 2015). Likewise, liposomes (diameter = 145 nm) with a dexamethasone prodrug were incorporated into a hyaluronic acid hydrogel for a single intra-tympanic injection in guinea pigs (El Kechai et al., 2016). The formulation enabled sustained release of dexamethasone in the perilymph of treated animals for 30 days after the injection (>25 ng/mL). Compared to prodrug loaded directly in the hydrogel, which achieved a maximum concentration of 39 ng/mL in the perilymph on 2-day post-injection, the liposomal hydrogel formulation increased the maximum concentration of dexamethasone to 833 ng/mL on 15-day post-injection. Confocal imaging revealed that a large proportion of these liposomes were trapped within the RWM – which the authors hypothesized to have acted as drug releasing reservoirs, and a small proportion of liposomes crossed the RWM intact (El Kechai et al., 2016).

Clinically, intra-cochlear injections are mostly used for preoperative or perioperative steroid regimens accompanying the surgical insertion of cochlear implant electrodes. A cochlear implant electrode is a neuroprosthesis used in patients who are severely hard-of-hearing to simulate hearing through stimulation of the auditory nerve. This device can be inserted through the RWM, or via a cochleostomy near the RWM (Richard et al., 2012). The immediate lesion and secondary lesion caused by the electrode insertion could be alleviated by steroid injections, which also lowers the inflammatory and fibrotic response due to surgical trauma, thus improving electrode-nerve interaction and protecting residual hearing (Paasche et al., 2006a).

In a prospective study conducted in 26 patients, injection of triamcinolone crystal (40 mg/ml) into the scala tympani before the electrode placement significantly lowered the intracochlear impedance over the first month (p < 0.05) compared to patients who did not receive the injection, suggesting better signal transmission between the electrode and inner ear tissues as a result of the steroid injection (Paasche et al., 2006a). In another study, however, triamcinolone delivered into 5 patients via a cochlear catheter prior to cochlear implant placement showed no significant difference in impedance (Prenzler et al., 2018). Due to limited numbers of clinical study available and large variations in study design and outcome measurement, currently, there is no clear consensus on the effectiveness of using preoperative and perioperative steroids in human for hearing preservation after cochlear implant (Santa Maria et al., 2014; Kuthubutheen et al., 2016; Snels et al., 2019).

Research in the pre-clinical stage has sought to leverage the cochlear implant electrode itself as a drug-eluting reservoir to alleviate the complications of cochlear placement, including deterioration in residual hearing, inflammation, and fibrosis. Many groups have coated the electrodes with dexamethasone, released in a sustained fashion, to replace the existing perioperative corticosteroid injections. To list a few examples, silicone electrodes coated with 1 or 2% dexamethasone showed statistically significant improvements in residual hearing compared to an uncoated electrode in gerbils (p < 0.05 at frequencies 0.5–16 kHz after 4–6 weeks post-implantation) (Douchement et al., 2014) and guinea pigs (p < 0.05 at frequencies 8–24 kHz after 12–24 weeks post-implantation) (Liu et al., 2015a). A follow-up study showed that a 10% dexamethasone-coated electrode had a burst release profile of dexamethasone in vivo in the perilymph of guinea pigs (>1000 ng/ml in the first 1.5 h), and the release was sustained (>100 ng/mL) for 1 week (Liu et al., 2015b).

Promising results from many small animal studies show dexamethasone-coated electrode’s protective effect on residual hearing and anti-fibrotic property (Wilk et al., 2016), despite some controversy (Stathopoulos et al., 2014). Recently, a dexamethasone-eluting electrode was tested in ten non-human primates (macaques) with normal hearing (Manrique-Huarte et al., 2020). All animals experienced ABR thresholds shift after implantation, indicative of auditory damage, and at 6 months, the dexamethasone-eluting electrode group showed significantly decreased level of electrode impedance (p = 0.005), although the ABR threshold difference was not statistically significant (p = 0.37), between drug-eluting and non-eluting electrodes.

In addition to cochlear implant coating which permits passive drug elution, many pumping systems have been developed, alone or hybridized with cochlear implants, to enable a more active and controlled drug delivery option for the inner ear. An external pump connected to a microcatheter was tested in 23 patients who suffered from acute SNHL and failed systemic corticosteroid therapy to receive a local delivery of steroids for 4 weeks (Plontke et al., 2005). The microcatheter was surgically implanted into the posterior bony canal. Patients received methylprednisolone (40 mg/ml, 10 μl/h, n = 6) or dexamethasone (4 mg/ml, 5 μl/h, n = 17). At 3 weeks after the start of therapy, mean ABR threshold improved by 15 dB from 103 dB to 87 dB, showing a significant improvement compared to a historical control group (p < 0.001) who failed systemic corticoid and did not receive salvage treatment. This improvement was maintained at 1-year follow-up (Plontke et al., 2005).

Osmotic pumps have also been tested in animal models for continuous drug infusion into the inner ear but has not been applied to clinical study yet. For example, dexamethasone (100 ng/ml) delivered through a mini-osmotic pump connected to a cannula that was surgically implanted into the perilymph in noise-deafened guinea pigs attenuated ABR threshold shift by 10–20 dB (Takemura et al., 2004). In another study, guinea pigs with kanamycin-induced deafness were treated with dexamethasone infusion (1 ng/ml) through a microcannulation osmotic pump system (Himeno et al., 2002). The treated animals demonstrated a ∼ 20 dB shift in their ABR threshold and significantly higher OHC survival compared to untreated animals.

Implantable peristaltic pump has also been studied in vitro for the drug release and pharmacokinetic profile. Such pump system was first tested in guinea pigs for delivering a model drug, FITC-Dextran at concentrations of 40 mg/ml with 0.6 μl/h flow rate or at 4 mg/ml with 6 μl/h. The implantable peristaltic pump (iPRECIO) was connected to a cochlear implant electrode with an inbuilt cannula and surgically placed in the subcutaneous space of the animal (Sokolowski et al., 2017). Perilymph sample retrieved at 2, 24 h, and 7 days revealed a longitudinal gradient with higher drug concentration measured at the base compared to the apex, which persisted throughout the study period. This electrode-pump system was later studied in macaques with similar settings to measure the pharmacokinetic profile of FITC-Dextran (40 mg/ml, 2 μl/h) (Manrique-Huarte et al., 2021). Results from this study revealed that similar drug concentrations from the base to apex in the perilymph was achieved within 2–24 h after infusion, and the uniform drug distribution was maintained for 7 days. Discrepancies between these two studies provide a good example of how the anatomical and physiological differences between species, and likely between individuals too, could lead to variable pharmacokinetics in the cochlea.

The Contour cochlear implant electrode has also been modified to incorporate a drug delivery channel connected to a mechanical pump (Paasche et al., 2006b). Testing in a cochlear-shaped plastic model showed that drug distribution was diffusion-driven at low drug flow rate (1 μl/h) and flow-driven at high flow rate (100 μl/h). Another commercial cochlear implant (MED-EL PULSAR) has been modified to include a drug eluting channel connected to an external infusion pump (Hochmair et al., 2016). In vitro testing showed no septum leakage and fluid flow was maintained at 5 μL/h. Furthermore, a multichannel electrode array with polyimide tubing has been designed for connecting an external osmotic pump to the implant. This pump was tested in normal-hearing guinea pigs (Shepherd and Xu, 2002). Neomycin was perfused into the electrode with a flow rate of 0.25 μl/h for over 28 days. The electrode assembly remained patent with no breakage at the end of 28 days.

Compared to the aforementioned continuous perfusion systems which allow one-way flow, reciprocating perfusion systems can recirculate inner ear fluids through a valved drug reservoir to enable zero-net flow in the cochlea, thus achieving higher drug delivery rate and potentially lower frequency of drug refills (Sewell et al., 2009). The perilymph volume in humans is estimated to be about 160 μl, making it challenging to accommodate additional fluid in the cochlea in large volumes (Buckingham and Valvassori, 2016). Instead of introducing additional fluid into the cochlea, a reciprocating microfluidic device recycles the cochlear fluid to achieve continuous delivery. At the tested infusion rates (8.6–21 μL/min), drug distribution kinetics of this reciprocating perfusion system was also reliant on diffusion, based on evidence from computational modeling in guinea pig cochlea using a small molecule hair cell neurotransmitter blocker (6,7-dinitroquinoxaline-2,3-dione) (Pararas et al., 2011). The reciprocating flow system was later optimized to incorporate all of the fluidic components into a single, compact microfluidic device, where a digital control system was also introduced to manipulate the drug dosing pattern (e.g., switching between drug and water delivery) in guinea pigs to provide steady drug infusion (Tandon et al., 2016).

Transgene delivery for replenishing copies of functional genes into the cochlea can be harnessed to treat autosomal recessive SNHLs which are typically caused by loss-of-function mutations and, to a lesser extent, autosomal dominant SNHLs involving gain-of-function mutations.

A transgene encoding the vesicular glutamate transporter-3 (VGLUT3), which is involved in hair cell glutamate release for synapses, has been delivered via AAV capsids in a knockout mouse model (Akil et al., 2012). Plasmid vectors delivered to P10–P12 mice restored VGULT3 expression in 40% of IHCs and repaired auditory synaptic transmission. ABR thresholds were near normal within 2 weeks after treatment (click stimulus and at 8–32 kHz) and remained within 10 dB of normalized range for 7 weeks.

A gene encoding the otoferlin protein involved in Ca²⁺-triggered sensory synapse, Otof (∼6 kb), is large enough to entail a dual AAV system (Akil et al., 2019). Two AAV2 vectors, each carrying one half of the murine otoferlin complementary DNA (cDNA) modified with a recombinogenic bridging sequence (inverted terminal repeats) (Figure 4A), have been delivered across the RWM of profoundly deaf Otof^–/– mice at P10. Otoferlin expression was detected in 64% of the IHCs of treated animals. The ABR thresholds (at 8, 16, 32 kHz) were restored to near-normal levels and maintained for 30 weeks post-treatment (Figures 4B,C). The same treatment applied to P19 and P30 Otof ^–/– mice resulted in similar transgene expressions in the IHCs at 82 and 85%, respectively, and near-normal ABR threshold.

In an Usher syndrome type IIIA (USH3A) mouse model, where Clrn1 mutation adversely affects the stereocilia morphogenesis and synapse of hair cells, recombinant AAV 2/8 (denoting a hybrid vector where a recombinant AAV2 genome is packed into an AAV8 capsid) carrying clarin-1 cDNA was injected via the RWM at P2–P3. The treatment demonstrated 90% transduction efficiency in IHCs and 20% in OHCs (Dulon et al., 2018). This gene therapy corrected stereocilia morphogenesis and preserved the hearing sensitivity of conditional Clrn1 knockout animals (ABR threshold = 20 dB at 10 kHz) compared to the untreated group (ABR threshold = 67 dB at 10 kHz).

Transmembrane channel-like (TMC) proteins TMC1 and TMC2 are involved in stereocilia mechanotransduction machinery. Mutations on TMC1 comprise about 3% of genetic hearing loss in human, including the autosomal recessive deafness DFNB7/11 and the autosomal dominant DFNA36 (Sloan-Heggen et al., 2016). Trans-RWM injection of AAVs has been explored to introduce Tmc1 and Tmc2 genes into TMC1-knockout mice (Tmc1^Δ/Δ) and Beethoven (Bth) mice at P0–P2 (Askew et al., 2015). AAV2/1-Tmc1 vectors injected into Tmc1^Δ/Δ mice (model for DFNB7/11) induced Tmc1 protein expression in the IHCs (∼65%) and the OHCs (∼5%) with base-to-apex gradient, and showed improvements in ABR response (85–100 dB at 5–16 kHz) compared to untreated group (>115 dB at 5–32 kHz). However, the DPOAEs of treated animals were similar to the untreated group, indicating little recovery of OHC function. Using the same vector construct, AAV2/1-Tmc2 injected into Bth mice (model for DFNA36) resulted in a similar partial recovery of response of 90–110 dB at 5–16 kHz compared to untreated Bth mice (>115 dB at 5–32 kHz) (Askew et al., 2015).

Adeno-associated virus vectors were also used to deliver Gjb2, a gene encoding the gap junction beta-2 protein (GJB2, also known as connexin 26) in the sensory cells of Gjb2-deficient mice (Iizuka et al., 2015). RWM injection of AAV vectors into neonatal P0 mice restored the formation of the tunnel of Corti, preserved SGNs, and lowered ABR threshold by 20–30 dB at 12 and 24 kHz, compared to untreated ear. Again, the restored morphology was more evident in the basal turn compared to the apical turn, hinting at the effect of slow diffusion or tissue tropism of AAVs. This treatment did not restore hearing in adult Gjb2-deficient mice, however, although GJB2 expression was detected, revealing a common limitation of SNHL gene therapies concerning the narrow treatment window for restoring normal auditory organ development (Iizuka et al., 2015).

Many AAV-mediated gene therapies for treating SNHL were conducted in neonatal mice. This was partly due to AAV uptake in hair cells becoming less effective as neonatal mouse mature (Akil et al., 2012; Shu et al., 2016; Suzuki et al., 2017; György et al., 2019a; Lee et al., 2020) and also because mouse cochlea continues to develop until P15 (Kopecky et al., 2012). When the same treatment was carried out in non-neonatal mice, the therapeutic effects usually diminishes. For example, AAV1-mediated injection of Kcne1 gene into Jervell and Lange-Nielsen syndrome type 2 (JLNS2) mice at P3 failed to preserve hearing while injections performed at P0–P2 preserved auditory function in 80% of mice (Wu et al., 2021).

Another complication for gene replacement therapy is that hearing loss rescue may not be sustained for the lifetime as the introduced exogenous gene tend to diminish over time (Iizuka et al., 2015) while protein recycling is a continuous process (Schneider et al., 2002). In one instance, Myo15a transgene introduced to mouse zygote was able to correct inner ear structure and function for up to 6 months before partial hearing loss started to recur (Kanzaki et al., 2006). AAV-mediated delivery of the gene encoding VGLUT3, after restoring near normal ABR threshold in 100% of mice, witnessed decline in auditory rescue after 7 weeks (Akil et al., 2012).

Around 20% of genetic SNHL in human are caused by gain-of-function mutations (Yeh et al., 2020), the treatment of which could be accomplished through silencing the expression of mutated allele or through the aforementioned transgene delivery to increase expression of the correct copies of the gene. Therapeutic application of RNA interference is achieved through the introduction of synthetic, short non-coding RNA (20–30 nucleotides), among which small interfering RNAs (siRNAs) and microRNAs (miRNAs) have attracted considerable interest (Carthew and Sontheimer, 2009). The siRNAs are double-stranded and highly specific in targeting a gene of interest, while the single-stranded miRNAs could have multiple gene targets, but both siRNAs and miRNAs can activate the formation of RNA-induced silencing complexes (RISCs) to prevent messenger RNA (mRNA) translation to achieve gene silencing (Lam et al., 2015).

A single intracochlear injection of artificial miRNAs enclosed in AAV2/9 vectors in Bth mouse on P0–P2 suppressed the expression of a semi-dominant point mutation on Tmc1 (c.1235T > A) (Shibata et al., 2016). Levels of mRNA expression of the mutant allele was suppressed by >88% compared to untreated ear. Consequently, hearing loss was slowed for approximately 21 weeks, at which point click ABR threshold shifted by >20 dB compared to wild-type control. The AAV2/9 showed 74% transduction efficiency in IHCs at the apical turn but only 7% efficiency in the OHCs. Consequently, IHC cell count was significantly improved compared to untreated ear at the apical turn (p < 0.005) but OHC survival showed no improvement.

Small interfering RNAs (siRNAs) were also tested in mice which were infected with a transgene carrying a dominant Gjb2 mutant allele (Maeda et al., 2005). The siRNAs were complexed with liposomes and placed against the RWM of adult mice. Expression of the exogenous mutant Gjb allele was suppressed by over 70% while endogenous Gjb expression was unaffected. Click ABR threshold was recovered to similar level as control animals (14.5 dB higher than control) although infected animals were not exposed to a severe shift in ABR threshold in the first place (23 dB higher than control).

Small antisense oligonucleotides (ASOs) have been used to correct Usher syndrome type 1C (USH1C) in a mouse model, by redirecting the splicing of the mutant Ush1c allele (216G > A) (Lentz et al., 2013). The ASOs were dissolved in saline and injected intraperitoneally at a dose of 50 mg/kg – 300 mg/kg into P3–P5 mice. The ASOs partially rectified the pre-mRNA splicing of Ush1c, enabled the production of harmonin which is involved in stereocilia morphogenesis and restored cochlear hair cell morphology. ABR response at lower frequencies (8, 16 kHz) were similar to those of normal ears but ABR threshold at 32 kHz was similar to untreated ears, suggesting high-frequency hearing was not rescued.

Together, these findings demonstrate that various gene silencing approaches can be harnessed in vivo, to selectively inhibit the expression of the mutant alleles that lead to dominant gain-of-function mutations. These therapies almost always require packaging vehicles to protect the genes or small oligonucleotides from degradation catalyzed in part by nucleases and lysozymes, but these vehicles may also elicit varying degrees of immunological response, especially viral vectors which can initiate both adaptive and innate immune responses (Kanasty et al., 2012; Ronzitti et al., 2020). Gene silencing therapies may also be limited by potential innate immune responses mounted against the double-stranded siRNAs, resulting in increased secretion of interferons and proinflammatory cytokines, which may be detrimental to the inner ear (Meng and Lu, 2017). Such immune responses can also be modulated, or nearly abrogated by careful design of siRNAs, such as, avoiding the use of uridine and guanosine motif or the use of synthetic nucleoside replacements (Judge et al., 2005, 2006; Sajid et al., 2020).

It has been a recognized challenge that in vivo AAV transduction in the inner ear is mostly limited to IHCs, even when experiments with ex vivo cochlear explants show similar transduction efficiencies between IHCs and OHCs. Efforts to improve viral vector uptake in the OHCs include (1) creating synthetic AAV capsids with higher OHC-targeting abilities and (2) exploring different intra-cochlear or intra-labyrinthine injection methods.

Several synthetic AAV variant capsids with differed or selective cell-targeting abilities have been tested. For example, AAV2/Anc80L065, a synthetic vector which approximates the common ancestor of AAV serotypes 1, 2, 6, 8, and 9, was tested to infect IHCs and OHCs and showed better transduction efficacy compared to conventional AAV serotypes in neonatal mice following RWM injection (100% IHCs and ∼90% OHCs, compared to <5% OHCs in natural serotypes) (Landegger et al., 2017). The synthetic capsid AAV/Anc80L065 was later used to deliver wild-type Ush1c into a mouse model (Ush1c c.216G > A) through RWM on P0–P1. Plasmid injection induced a significant recovery of mechanosensitivity in both OHCs and IHCs (p < 0.001) and a prominent recovery of ABR threshold at lower frequencies (25–30 dB at 5.6–16 kHz). However, when injected on P10–P12, the treatment did not provide therapeutic effects (Pan et al., 2017). AAV9-PHP.B, a synthetic vector first engineered for transport across the BBB (Deverman et al., 2016), has also shown high transduction efficacy in IHCs (50% - 70%) and OHCs (30–40%) via RWM injection in neonatal mice, but did not transfect supporting cells in the sensory epithelium (György et al., 2019a). The synthetic vector AAV2.7m8 was identified from an in vivo directed evolution screening for mouse retinal gene delivery (Dalkara et al., 2013). When applied to cochlear gene delivery, AAV2.7m8 loaded with GFP was shown to infect IHCs and OHCs in mice, in addition to the inner pillar cells and inner phalangeal cells. AAV2.7m8 has been shown to infect murine OHCs at higher efficiency (83%) than AAV/Anc80L65 (67%) when injected via canalostomy into mice (Isgrig et al., 2019). The three aforementioned synthetic vectors, i.e., AAV2/Anc80L5, AAV9-PHP.B, and AAV2.7m8, have been compared in mice cochleae, where AAV9-PHP.B showed the highest transduction efficiency, at nearly 100%, for both IHCs and OHCs following utricle injection (Lee et al., 2020).

The exact route of injection may also influence AAV uptake in the organ of Corti. AAV delivered via trans-RWM injection into the perilymph were largely undetectable in the OHCs (Stone et al., 2005; Konishi et al., 2008; Dulon et al., 2018; Kim et al., 2019). In other studies where AAVs were delivered directly into the mouse endolymph via cochleostomy (Kilpatrick et al., 2011) or canalostomy (Tao et al., 2018), uptake for a few AAV serotypes (e.g., AAV2, AAV8) was observed in the OHCs (0–20%), but still at a much lower incidence rate compared to IHCs (40–100%). Utricle injection, which also directly access the endolymph, of a synthetic vector AAV9-PHP.B in mice supported a higher OHC transduction rate (100%), compared to that observed after RWM injection (40–70%) (Lee et al., 2020), suggesting that endolymphic delivery may present an advantage over perilymphic delivery for the purpose of transducing sensory cells.

Non-homologous end joining generates multiple random nucleotide insertion or deletions (indel) at the repair junction (e.g., in a mouse embryonic stem cell line, indels can account for ∼50% of all NHEJ events) (Guo et al., 2018b) which can disrupt the open reading frame, therefore NHEJ can be harnessed to silence dominant alleles involved in hearing loss, which accounts for about 20% of all genetic deafness (Angeli et al., 2012).

The use of Cas9-sgRNA complex to disrupt the dominant point mutation on Tmc1 (p.M418K, c.T1235A) has been demonstrated in Bth mouse (a model for DFNA36 hearing loss) (Gao et al., 2017). In this study, Cas9 and sgRNA were delivered as a ribonucleoprotein (RNP) complex using Lipofectamine 2000, a cationic liposome formulation, via cochleostomy into the scala media of Tmc1^Bth/+ Tmc2^+/+ mice on P1. Allele-specific disruption occurred in up to 10% of sampled organ of Corti cells from treated mice. IHC and OHC survival was enhanced by ∼80% and ∼30%, respectively, at location corresponding to 32 kHz and stereocilia bundles were preserved at 16 and 32 kHz whereas untreated ear suffered almost complete loss of stereocilia at these locations. In terms of hearing function, the treated mice showed reduction in ABR threshold by ∼15 dB compared to untreated mice, with the difference being most evident at lower frequency ranges (8–23 kHz). The treated ears also demonstrated slight change in DPOAE (∼20 dB shift at 16 kHz), suggesting possible OHC damage from the surgical procedure of cochleostomy.

Combinations of Cas9 variants and optimized sgRNAs have later been investigated to improve the efficiency of Cas9-mediated Tmc1 gene disruption in Bth mouse (György et al., 2019b). In this study, AAV/Anc80L65 with CMV promoter was used and the exact route of intra-cochlear injection was not disclosed. When using AAV vectors instead of liposomes, the Cas9-Tmc1 sgRNA combination used in the previous study (Gao et al., 2017) lost the reported allele specificity when editing Bth mouse fibroblasts, suggesting perhaps an unresolved influence of the route and format of Cas9 delivery. Thereafter, the authors selected a Cas9 variant which recognizes a specific protospacer-adjacent motif (PAM) site that is present in the mutant allele of Tmc1 but not in the wild-type allele. In heterozygous Bth mouse fibroblasts, 98% of all indels on Tmc1 occurred preferentially in the mutant allele, mostly as a frameshift mutation which disrupted the gene. When injected into Bth mice on P1, this Cas9 variant lowered the expression of Bth mRNA in cochlear tissue by 24%. ABR threshold improved by about 20 dB compared to the untreated group (at 8–22 kHz) while DPOAE showed near-normal function of OHCs (at 5–11 kHz).

Homology directed repair-mediated high-fidelity repair can be harnessed to correct recessive, loss-of-function mutations, which are responsible for the remaining ∼80% of genetic deafness (Yeh et al., 2020). In addition to native Cas9 enzymes, Cas9 nickase (a mutated Cas9 which only creates single-strand DNA nick) has been adopted since DNA nicks can be repaired with HDR with higher fidelity compared to the toxic double-strand breaks and incur lower off-target edits.

The HDR pathway was harnessed to repair the Cdh23^ahl allele in a C57BL/6NTac mouse model, which mimics the progressive hearing loss DFNB12 and USH1D syndrome in human caused by a single nucleotide substitution on Cdh23^ahl (c.753) (Mianné et al., 2016). In this study, a Cas9 nickase (D10A) with a pair of sgRNAs was used to produce two single strand breaks with a 5′ overhang, reducing off-target activities from double-strand producing-Cas9 enzymes. After the DNA cleavage, a 121 bp ssODN was used as the donor template for homologous recombination to repair the Cdh23 gene. The Cas9 nickase mRNAs, sgRNAs, and ssODNs were injected into one-cell state mouse embryos. Of the 456 embryos injected, 104 gave rise to live pups (22.8%), of which 15 were transgenic at the target site (14.4%). Four pups carried the correct Cdh23^753A>G repair (3.8%) while the other 11 pups carried incorrect repairs from HDR. Heterozygously repaired littermates (Cdh23^ahl/753A>G) showed normal IHC and OHC morphology throughout the cochlea, whereas untreated animals suffered from >50% OHC loss at the basal turn (hair cell morphology was normal at other regions). ABR threshold was reduced by >25 dB at 32 kHz in the treated group. ABR response at 8 and 16 kHz were within normal range for both treated and untreated group.

A base editor is a fusion protein comprised of a Cas9 nickase or a catalytically inactive ‘dead’ Cas9 (dCas9) fused onto a deaminase. Base editors can directly exchange one base pair for another at the target locus, relying on the cellular mismatch repair machinery instead of HDR, and therefore is not limited only to mitotic cells (Komor et al., 2016; Kantor et al., 2020). Prime editors further expand the scope of donor-free DNA editing to cover all possible point mutations and small insertion or deletion. Prime editors comprise a Cas9 nickase fused to a reverse transcriptase, which uses a long, single-strand prime editing guide RNA (pegRNA) to perform the desired sequence change, and are also well tolerated in post-mitotic cells (Anzalone et al., 2019; Kantor et al., 2020).

The use of base editing to repair deafness-causing point mutations was first demonstrated in a Baringo mouse model (Tmc1^Y182C/Y182C; Tmc2^+/+), which harbors a recessive, loss-of-function mutation on Tmc1 (c.A545G, p.Y182C) that mimics the congenital hearing loss DFNB7/B11 in human (Yeh et al., 2020). The selected cytosine base editor (∼5.2 kb) (Figure 4D) exceeded the maximum loading capacity of a single AAV (∼4.7), and therefore was slightly truncated and split in half to be encapsulated in a split-intein, dual-AAV system. AAV/Anc80L65 with Cbh promoter was injected into the inner ear of P1 Baringo mouse. The exact route of injection was not disclosed. On-target base editing in sampled cochlear cells occurred at an efficiency of 10 - 51% and off-target editing was minimal. Editing restored the normal IHC hair bundle morphology at the apical end and partially in the basal end (Figure 4F). Some treated mice (9/15) showed slight recovery of ABR response (5–50 dB improvement at 5.6 kHz) and the remaining 6/15 treated mice showed no detectable ABR response. DPOAE response was not seen in any treated mice, suggesting a lack of functional recovery of the OHCs. This might be in part due to the low transduction efficiency of Anc80L65 and the promoter, or a combination of both (expression of GFP was 22.6–41.7% in IHCs, and 2.6–8.3% in OHCs) (Figure 4E). Another study investigated the use of Anc80L65 with CMV promoter to deliver Cas9 into the cochlea via cochleostomy in P1-P3 mice or canalostomy in adult mice also showed that Anc80L65 had almost no transduction in the OHCs although IHCs were completely infected (Kang et al., 2020).

Cas9 can be delivered in the form of DNA plasmid, mRNA, or protein. Most of the aforementioned transgene delivery strategies, especially those deployed in vivo, also apply to the delivery of Cas9 systems in plasmid forms. Ultimately, for editing to take place, the RNP of Cas9-sgRNA must be present inside the cell and translocate into the cell nucleus. AAV vectors are commonly used for assisting Cas9 in vivo in bypassing cellular barriers and evading cellular degradation pathways. One limitation of AAV delivery is its loading capacity (∼4.7 kb). The restricted cargo size poses difficulties for packaging the Cas9 (e.g., the commonly used Streptococcus pyogenes Cas9 is about 4.2 kb) and sgRNA into a single DNA plasmid. As a solution, smaller Cas9 orthologs (e.g., the Staphylococcus aureus Cas9 which is about 3.2 kb) or separate plasmids for Cas9 and sgRNA may be used (Ran et al., 2015). Base editors and prime editors also exceed the loading capacity and therefore need dual-AAV systems. To promote nuclear transport, Simian vacuolating virus 40 (SV40) nuclear localization sequence (NLS) are often encoded into Cas9 DNA or RNA constructs or attached to the terminal of Cas9 proteins (Staahl et al., 2017).

An important consideration when designing a delivery system for Cas9-based therapy is that, different from other gene therapy reagents, the long-term expression of Cas9 may not be desirable. A transient, one-time expression of Cas9 in the host cell can shorten the time window of editing and reduce off-target edits. Delivery of Cas9 and sgRNA as an RNP complex is therefore appealing due to the fast onset of the gene editing machinery and the non-replicable nature of protein agents. Comparing the gene modification specificity of Cas9-sgRNA delivered as viral plasmid vs. RNP in HEK293T cells, cationic-lipid mediated RNP delivery showed a 19-fold higher on-target/off-target editing rate compared to plasmid delivery and the on-target modification rate (10%) was comparable between the two methods (Zuris et al., 2014). Compared to viral plasmids, liposomes or other nanoparticle-based strategies are also favorable because they avoid the risk of integration of the plasmid sequence into host genome and thus likely lowers host-mounted immune reactions against Cas9 entry.

Although native Cas9 protein carries positive charge and is therefore unlikely to be encapsulated into cationic lipids, it can be combined with a sgRNA, which is rich in phosphate groups and carries multiple negative charge, to become compatible with cationic liposomes (Zuris et al., 2014). The liposomes can then interact with the negatively charged cell surface, carrying the RNP into the cell. Moreover, liposomes may facilitate endosomal escape, increasing the likelihood that RNPs can reach the nucleus for access to the transcriptional machinery (Dalby et al., 2004). As with how different virus serotypes show differed transfection efficiency, various commercialized cationic lipid transfection kits also show variable degree of efficacy. In one study, Lipofectamine 2000 showed higher efficacy (12% indel) than RNAiMAX (7.7%) or CRISPRMAX (8.9%) for delivering the Cas9-Tmc1 RNP of interest (Gao et al., 2017). Liposomes can also be used to deliver Cas9 in the form of DNA plasmids and mRNA strands.

Lipoplex has also been used to deliver Cas9 RNP. A cationic cell-penetrating peptides (CPP) containing 9 Arginine groups was used to create two groups of cationic nanoparticles from CPP-conjugated Cas9 proteins and CPP-complexed sgRNAs (Ramakrishna et al., 2014). These NPs effectively transfected HEK293T cells in vitro, but have not been tested in vivo. Inorganic NPs represent another strategy for Cas9 RNP delivery. A CRISPR-Gold complex was designed with a gold NP core conjugated to a layer of DNA, which was then complexed with ssODN and the Cas9 RNP. This complexation was then covered with a layer of cationic endosomal disruptive polymer, poly(N-(N-(2-aminoethyl)-2-aminoethyl) aspartamide) [PAsp(DET)] (Lee et al., 2017). CRISPR-Gold showed success in enabling HDR gene editing in mice through non-clathrin-mediated endocytosis, but has not been used for treating hearing loss yet. In another design, the negatively charged Cas9 RNPs self-assembled with cationic arginine gold nanoparticles to form nanoassemblies which can enter HeLa cells through a membrane fusion process instead of endocytosis and demonstrated ∼30% indel efficiency in vitro (Mout et al., 2017).

Hair cells have been hypothesized to regenerate from two pathways: (i) transdifferentiation of supporting cells without cell divisions, or (ii) mitotic regeneration from progenitor cells, which is more difficult to achieve (Shibata et al., 2020). Regenerative therapies targeting the inner ear often modulates the highly conserved Wnt and Notch signaling pathways, which are both critical for cell proliferation, differentiation, and fate determination in early morphogenesis (Samarajeewa et al., 2019). These two pathways have a complex interplay for hair cell formation (Zheng and Gao, 2000; Li et al., 2015; Cheng, 2017).

Notch signaling upregulates several effectors, including the hairy and enhancer of split-1 (Hes1) which downregulates the expression of basic helix-loop-helix transcription factors such as atonal homolog 1 (Atoh1) (Samarajeewa et al., 2019). Strategies to inhibit Notch signaling or directly upregulating Atoh1 have induced the formation of new cochlear hair cells (Woods et al., 2004; Bramhall et al., 2014). The canonical Wnt signaling pathway elevates the cytosolic concentration of β-catenin, which is an effector that can translocate to the nucleus and bind with T cell-specific factor/lymphoid enhancer-binding factor (TCF/LEF) and co-activators to initiate the transcription of target genes (Samarajeewa et al., 2019).

In the organ of Corti, a small group of supporting cells express G-protein-coupled receptor (Lgr5), and are believed to possess progenitor-like activities (Shi et al., 2013). In transgenic mice, the Lgr5-positive cells have been shown to undergo mitotic generation into hair cells following forced expression of β-catenin, a mediator of the canonical Wnt signaling pathway (Shi et al., 2013). The Lgr5-positive cells (i.e., inner pillar cells and the third row of Deiters cells) have also been shown to spontaneously transdifferentiate into hair cells after Notch inhibition in gentamicin-damaged newborn mouse cochlea (Bramhall et al., 2014).

Mounting evidence has suggested that Notch inhibition more likely promotes hair cell regeneration from supporting cell transdifferentiation rather than from mitotic regeneration (Li et al., 2015). In one study, the Atoh1 gene was carried in AdV vectors and injected into the endolymph of young adult guinea pigs. Atoh1 expression was induced in the supporting cells; these immature hair cells formed mostly in the organ of Corti and some ectopic hair cells were found in adjacent areas with evidence of neurofilament extension to these cells (Kawamoto et al., 2003). In a later study, AdV-Atoh1 was delivered via cochleostomy into the scala media of young adult guinea pigs, after complete loss of hair cells (but preservation of supporting pillar cells) due to high dose of kanamycin. Atoh1 induced the formation of new hair cells with similar surface morphology as normal hair cells. An average of 256 IHCs and 691 OHCs were detected in a 2 mm organ of Corti region compared to no cells detected in untreated ear. Some of these new cells had mixed phenotype of OHC and supporting cells, suggesting that supporting cells have transdifferentiated into hair cells. A significant increase in the number of nuclei (1548 vs. 830 nuclei in untreated ear) also suggest mitotic events for hair cell regeneration. ABR recovery in Atoh1 inoculated ear was significantly improved compared to the untreated ear (4–24 kHz, p < 0.004) (Izumikawa et al., 2005).

In another study, overexpression of Atoh1 was suggested to treat hearing loss via stabilizing injured hair cells and promoting stereocilia repair (Yang et al., 2012). In this guinea pig model of noise-induced hearing loss, injured hair cells and supporting cells were both present before the treatment of Atoh1-loaded AdV through trans-RWM injection and the repair of existing hair cell was purposed to be due to regeneration of damaged stereocilia instead of supporting cell transdifferentiation. Improvements in ABR threshold of 35–40 dB were noted at 4–20 kHz compared to untreated ear. When the same treatment was carried out at 1 month after noise exposure, none of the animals recovered hearing, supporting the notion that the presence of hair cells was necessary for the suggested Atoh1-induced repair.

Another Notch inhibition strategy involved using siRNAs (encapsulated in PLGA nanoparticles) directed against Hes1, a downstream effector of Notch signaling (Du et al., 2018). The formulation was delivered into the cochlea of young adult guinea pigs at 72 h after acoustic deafening, via a mini-osmotic pump placed through cochleostomy for 24 h infusion. New IHCs and OHCs were observed mostly at the basal and second turn of the cochlea which were proximal to the infusion site. SEM images of the organ of Corti at basal turns showed that OHCs were completely ablated in noise-injured cochlea (Figure 4G) whereas in PLGA-Hes1 treated cochlea, OHCs were present, although many were equipped with immature stereocilia showing shorter, incorrectly angled hair bundles (Figure 4H). ABR threshold recoveries at 3–9 weeks post-treatment exhibited a base-to-apex gradient, with more recovery witnessed at higher frequency (∼21 dB improvement at 16 kHz vs. ∼9 dB at 2 kHz) (Du et al., 2018).

Adenoviruses are popular among Atoh1 transgene therapies for hair cell transdifferentiation, but AAV are much less commonly used because of their low transduction rate in supporting cells. An AAV variant, AAV-ie, was made by the insertion of a CPP-like peptide derived from a neuron-targeting vector, AAV9-PHP.eB (Chan et al., 2017), into the capsid of a liver-targeting and low-immunogenic vector, AAV-DJ (Grimm et al., 2008). The newly constructed AAV-ie was shown to infect a broad range of cells in the cochlea at high efficiencies, including IHCs, OHCs, supporting cells, and SGNs, following intra-RWM injection into mouse cochlea (Tan et al., 2019). At the apical turn, AAV-ie transduced 77% of supporting cells, while PHP.eB, AAV-DJ, and AAV/Anc80L65 transduced < 20%, <55%, and <55% of supporting cells, respectively. Atoh1 delivered through AAV-ie vector via the RWM at P0 into normal C57BL/6 mice induced ectopic new hair cell formation in the sensory epithelium and the greater epithelial ridge. The new hair cells possessed hair bundles and some excitatory characteristics (Tan et al., 2019). However, due to the broad tropism of AAV-ie, it was difficult to analyze the exact action site of Atoh1 in this experiment, and this property of AAV-ie may limit its therapeutic applications, but may be solved by cell-specific promoters, such as GFAP which is only expressed in supporting cells but not hair cells.

Following the death of hair cells, spiral ganglion cells, which are the sole messengers in the cochlea for relaying auditory information from the hair cells to the central auditory pathway in the brain, also gradually degenerate as they cease to receive sensory signals and neurotrophic factors from the organ of Corti (Spoendlin, 1975; Fritzsch et al., 1997; Hardie and Shepherd, 1999). Degeneration of the SGN can limit the success of hair cell-targeted therapies or the outcome of cochlear implants which electrically stimulate any residual SGNs to restore hearing perceptions (Ahmed et al., 2017). Exogenous replenishment of several neurotrophins, including brain-derived neurotrophic factor (BDNF), glial cell line-derived neurotrophic factor (GDNF), and neurotrophin3 (NT3) have been used to rescue SGN after hair cell damage. For example, BDNF proteins delivered via a mini-osmotic pump (Alzet) for 28 days into chemically deafened rat cochleae (flow rate 0.25 μl/h) rescued SGN density to statistically similar level to that of normal rats while non-treated group suffered from significant SGN loss (p < 0.001) (McGuinness and Shepherd, 2005). Similarly, BDNF proteins delivered via osmotic pump for 28 days into chemically deafened guinea pig cochleae also showed preservation of all three elements of the SGNs (peripheral processes, somata, axon) and electrically evoked compound action potentials which were similar to that from normal hearing cochlea for the duration of BDNF infusion and at least 8 weeks after termination of treatment (Ramekers et al., 2015; Vink et al., 2021).

These neurotrophic factors can also be delivered as genetic material. GDNF transgene packed in AdV vectors were delivered to chemically deafened guinea pigs through the scala tympani 4 days after deafening. SGN cell survival was 10–30% higher in GDNF group compared to non-inoculated control ears (Yagi et al., 2000). Auditory function assessment was not reported. In another study, NT-3 cDNA packed in HSV-1 vector was inoculated into the scala vestibuli of murine cochlea. Expression of NT-3 lowered cisplatin-induced SGN apoptosis or necrosis and preserved SGN survival by more than 60% compared to control virus injection (Bowers et al., 2002).

In addition to viral vector-mediated packaging, neurotrophins were also successfully encapsulated in synthetic nanoparticles in the form of mRNA and proteins. Non-viral delivery bypasses the safety concern with exogenous viral particles and avoids the risk of DNA plasmid integration into host genome. As an example, nanoporous silica NPs was used to package BDNF proteins and demonstrated sustained release in fibroblast cell culture for up to 39 days and significantly improved SGN survival in vitro (p < 0.001) (Santos et al., 2018). A neutral lipid nanoparticle (LNP) was also developed for encapsulating BDNF mRNA (Miwa et al., 2020). The LNPs were placed onto a gelatin sponge outside the RWM of guinea pigs at day 1 or day 14 after gentamicin-induced hearing loss. Both day 1- and day 14-BDNF treatment rescued SGN survival to near normal level, but untreated groups suffered significant SGN cell loss (p < 0.001). OHC survival was similar between day 1-BDNF group and control but day 14-BDNF group suffered significant OHC loss (p < 0.01). ABR thresholds measured from day 1-BDNF group were similar to that of normal ear (4–32 kHz) but the day 14 BDNF-group showed ABR response similar to untreated group.