Advances in Genome Editing for Genetic Hearing Loss

Abstract

According to the World Health Organization, hearing loss affects over 466 million people worldwide and is the most common human sensory impairment. It is estimated that genetic factors contribute to the causation of approximately 50% of congenital hearing loss. Yet, curative approaches to reversing or preventing genetic hearing impairment are still limited. The clustered regularly interspaced short palindromic repeats-associated protein 9 (CRISPR-Cas9) systems enable programmable and targeted gene editing in highly versatile manners and offer new gene therapy strategies for genetic hearing loss. Here, we summarize the most common deafness-associated genes, illustrate recent strategies undertaken by using CRISPR-Cas 9 systems for targeted gene editing and further compare the CRISPR strategies to non-CRISPR gene therapies. We also examine the merits of different vehicles and delivery forms of genome editing agents. Lastly, we describe the development of animal models that could facilitate the eventual clinical applications of the CRISPR technology to the treatment of genetic hearing diseases.

1. Introduction

Hearing loss (HL) is defined as the inability to hear at a threshold of lower than 25 decibels in at least one ear. As the most common sensory disorder in humans, HL currently affects over 5% of the global population. It is estimated that by 2050, more than 900 million people, or one in every ten, will be affected by HL (World Health Organization, Deafness and Hearing Loss, 2019). HL can be either congenital (present at birth) or acquired (occurs after birth), and both of them can be caused by genetic and/or environmental factors. Congenital HL impacts about 1 in 500 newborns, and over half of the cases have been attributed to genetic factors, most frequently by a single gene defect, with the rest being caused by environmental factors (National Center on Birth Defects and Developmental Disabilities, Genetics of Hearing Loss, 2019) [1] (Figure 1). To date, cochlear implantation (an electronic device placed in the inner ear to provide electrical stimulation directly to the hearing nerve) is the most available option for severe to profound HL, while hearing aids (an externally-worn electronic device for conversion and amplification of sound waves) can assist moderate HL patients [2]. However, no pharmacological therapies are currently available to treat genetic HL (GHL). In recent years, gene therapy has received increasing attention as a promising method for the treatment of inner ear dysfunction [1, 3–5]. One of the gene therapy approaches that has been used over the last two decades involves delivering functional complementary DNA (cDNA) into the inner ear, complementing the genetic defect of hearing [6, 7]. This approach, known as gene replacement, is useful for the treatment of many diseases caused by a faulty gene with loss-of-function defect (recessive mutation). However, it fails to restore gene functions of GHL caused by dominant mutations [8]. Alternatively, RNA therapies, including antisense oligonucleotides (ASOs) and RNA interference (RNAi), which silence the messenger RNA (mRNA) transcribed from mutant alleles by Watson-Crick base pairing, have also been applied in GHL research [9]. However, neither ASOs nor RNAi could overcome the barrier of transitory effects and thus fails to offer a permanent solution for genetic defects. This drawback necessitates new strategies for a one-time, precise, and permanent approach for the treatment of GHL.

Figure 1. Distribution of deafness-associated variation type.

The left pie chart shows that genetic factors account for over half of the identified deafness. Of those, 80% are associated with single nucleotide mutation, followed by deletion, insertion, and duplication depicted in the right chart. The pie charts were drawn with data adapted from ClinVar (ncbi.nlm.nih.gov/clinvar) [22].

The recently developed gene-editing platform, the clustered regularly interspaced short palindromic repeats-associated protein 9 (CRISPR-Cas9) system, has been used to edit single or multiple genes simultaneously in different species [10]. Many efforts have been devoted to improving the genome-editing efficacy and specificity of the CRISPR-Cas9 systems [11, 12]. The ability of the programmable CRISPR-Cas9 systems to disrupt or repair targeted gene loci offers great potential for treating genetic diseases, including GHL [13]. The newly emerging genome-editing tools, called “base editors”, being made by fusing a catalytically impaired Cas9 (usually denoted as nickase Cas9, nCas9) to different deaminases, enable conversion of cytosine (C) to thymine (T) (cytosine base editor, CBE) or adenine (A) to guanine (G) (adenine base editor, ABE), resulting in the direct installation or correction of point mutations in single-nucleotide variants (SNVs) [14, 15]. SNVs can be corrected by base editors with low levels of unwanted insertions or deletions (indels) as long as the targeted nucleotides are located in a suitable editing window nearby a protospacer adjacent motif (PAM) sequence [14]. Since more than 80% of deafness-associated variants are SNVs (Figure 1), base editors are anticipated to be promising toolkits for the treatment of GHL.

With the clinical trial of gene therapy for GHL currently in progress, for example, replacement therapy of delivering cDNA of Atoh1 gene through recombinant adenovirus (AdV) serotype 5 (ClinicalTrials.gov Identifier: NCT02132130), the translation from animal studies to humans’ treatment has begun to show the possibility for success. One of the main determining factors for eventual clinical translation of CRISPR-Cas9 systems would be how to deliver the genome editing agents into specific cells efficiently. Different forms of CRISPR-Cas9 agents (DNA, mRNA, and protein) have been delivered into the inner ear cells by viral and non-viral methods [16–20]. Viral vectors, while capable of efficiently delivering genetic materials, result in a limited usage in humans due to concerns over their potential safety risks. Non-viral approaches appear to have the potential to address some of the shortcomings of viral vectors, in terms of safety, packaging capacity and manufacturing. However, their comparatively low delivery efficacy limits their applications in vivo [11, 21].

Herein, we first summarize the genes that are the most commonly found to be associated with human deafness. Based on these deafness-associated genes, we review the development of gene therapy and intracellular delivery methods of genome-editing agents for GHL. We describe the current progress of CRISPR-Cas9 based strategies applied to hearing treatments and highlight the benefits of different delivery methods. Finally, we discuss applications of different animal models, particularly mouse models, which are currently available for proof-of-concept studies before applying CRISPR technology to clinical treatment of GHL in humans.

2. Deafness-Associated Genes and Variants

The advent of the high-throughput DNA sequencing technology has dramatically expanded our capability to identify deafness-associated genes and variants [23]. The demand for gene and variant interpretations has spurred the creation of public databases, such as ClinVar, Human Gene Mutation Database (HGMD) and Deafness Variation Database (DVD), which contain all the identified genomic and clinical data associated with GHL [22, 24, 25]. These databases provide comprehensive resources for basic research and relevant information about clinical needs.

2.1. Classification

119 genes to date have been identified to be responsible for non-syndromic HL [26]. Among these 119 genes, 47 and 76 are associated with autosomal dominant and recessive HL (the numbers given include overlap genes because some can cause both dominant and recessive HL), respectively, while the rest is associated with X-linked, modifier, Y-linked genes, or auditory neuropathy [26]. Deafness-associated variants can be also classified based on the variation types as single nucleotide (~82%), deletion (~10%), duplication (~3%), or insertion (~5%) [22] (Figure 1). These classifications provide the relevant disease distribution as well as important information of selecting the appropriate strategy for gene therapy (Figure 2). In principle, the genetic approaches for treating dominant and recessive inherited HL are different. In the case of recessive HL, a therapy known as gene replacement supplements the original function of a defective gene by directly delivering the wild-type gene. On the contrary, gene silencing for the therapy of dominant diseases could be more effective in suppressing the dominant-negative effects of the mutant alleles. Gene editing might be superior to gene replacement and gene silencing because of its potential to treat both dominant and recessive HL through direct gene correction, or reversion of the disease-causing mutation. To evaluate the clinical significance of each variant, five categories (pathogenic, likely pathogenic, uncertain significance, likely benign and benign) are classified based on the experimental, diagnostic, and computational data, demonstrating the likelihood that one mutation is disease-causing [27]. Since GHL with pathogenic mutations are the ultimate targets for treatments, correct identification and characterization of those pathogenic mutations leading to hearing dysfunction are essential.

Figure 2. Phenotype of deafness-associated genes and relative gene therapy strategies.

The lists of autosomal dominant and autosomal recessive HL genes are from the Hereditary Hearing Loss Homepage (hereditaryhearingloss.org) [26]. †: GJB2, GJB6, MYO6, MYO7A, TECTA, TMC1, CEACAM16, COL11A2, TBC1D24, PTPRQ, and MYO3A are identified to cause both dominant and recessive HL. ‡: MYO1A is an autosomal dominant HL gene called into question. For dominant HL, people with normal-mutant genotypes (heterozygous) develop the disease. Gene correction or gene silencing is possible to treat the disease. Whereas genetic recessive HL, which is caused by homozygous mutant-mutant genes, might be treated by gene replacement or correction.

2.2. Most common deafness-associated genes

Most of GHL cases are attributable to ten deafness-associated genes (GJB2, SLC26A4, MYO15A, OTOF, CDH23, TMC1, WFS1, MYO7A, KCNQ4, and COCH) [28]. GJB2 (gap junction beta 2, MIM: 121011) is the most common deafness-associated gene (~22% of all diagnoses) [29] and carries the most significant number of pathogenic or likely pathogenic variants (~70 % of coding variants) [25]. GJB2 encodes a beta class gap junction protein Connexin 26 expressed in the cochlea and epidermis, of which dysfunction leads to the failure of intracellular potassium diffusion and transmission of auditory signals [30]. The most prevalent variant at the GJB2 locus is the frameshift mutation c.35delG, which is also the most common cause of GHL worldwide (Table 1). Other two highly frequent pathogenic SNVs, especially among European and East Asian populations, c.101T>C (p. M34T) and c.109G>A (p. V37I), respectively, have shown to cause progressive mild to moderate recessive sensorineural HL [31].

Table 1.

The most prevalent mutations at the GJB2 and SLC26A4 loci.

Genes	Mutations1 (Allele frequencies2, %)
GJB2	c.35delG (0.800)	p.M34T (0.800)	p.V37I (0.659)	p.L90P (0.088)	c.167delT (0.068)	p.W24X (0.058)	c.235delC (0.036)	p.R143W (0.022)	c.−23+1G>A (0.020)
SLC26A4	c.919-2A>G (0.036)	p.L236P (0.033)	c.1001+G>A (0.028)	p.V239D (0.024)	p.V138F (0.021)	p.T416P (0.021)	p.H723R (0.020)	p.G209V (0.020)	p.T410M (0.019)

¹The most prevalent GJB2 and SLC26A4 variants worldwide were summarized in Tsukada et al. 2015 [41].

²The allele frequencies of each variant are obtained from Clinvar (ncbi.nlm.nih.gov/clinvar) [22].

Gene therapies involving the GJB2 gene have been applied to the hearing recovery in mice or guinea pigs (in which the gene is written as Gjb2). A dominant mutation c.224G>A (p.R75Q) in the Gjb2 gene was silenced by a short interfering RNA (siRNA). In a mouse model transfected with a plasmid vector expressing GJB2_R75W-enhanced green fluorescent protein (GFP), this siRNA selectively suppressed R75W expression by over 70% of control levels, while the wild-type murine Gjb2 expression was not affected [32]. Yu et al. [33] delivered a viral vector that carried the Gjb2 coding sequence into the scala media of Gjb2 knock-out mice and demonstrated an enhanced expression of Connexin-26 in the supporting cell network. However, the approach did not lead to a significant improvement in hearing. Furthermore, Iizuka et al. [34] delivered Gjb2 through the round window membrane in supporting cells in a mouse model with Gjb2 deletion. Significantly improved auditory responses and the development of the cochlear structure were observed. Takada et al. [35] delivered a brain-derived neurotrophic factor (BDNF) gene into the cochlea of Gjb2 knocked-out mouse to promote neuron survival in the cochlea. This study supported the indirect gene therapy strategy for Gjb2-deficient mice associated with the degeneration of the auditory nerve.

Mutations in the SLC26A4 gene (solute carrier family 26, member 4) are the second-most prominent contributors to worldwide GHL [36, 37]. SLC26A4 encodes pendrin protein that carries out trans-membrane anion transportation in the cochlea [38]. The most frequent mutation in the SLC26A4 gene is a splicing mutation, c.919–2A>G in all populations, while a missense mutation, c.2168A>G (p.H723R), is the most commonly found in East Asia [37] (Table 1). Lee et al. [37] found eleven pathogenic mutations that can cause aberrant splicing in SLC26A4 and fully or partially rescued the splicing using modified spliceosomal component U1 small nuclear RNA (snRNA) and ASOs. Ryu et al. [39] designed a plasmid-based CRISPR-Cas9 non-viral delivery system to edit Slc26a4 locus in a mouse cell line. The group then treated mice model containing the splicing mutation c.919–2A>G using viral delivery of CRISPR-Cas9 and achieved precise correction of the mutation back to the wild-type genotype, but its effect on hearing function had not been assessed [17]. Recently, Kim et al. [40] demonstrated that adeno-associated virus (AAV)-mediated delivery of SLC26A4 cDNA into mouse otocyst (the embryonic structure that develops into inner ear) partially restored hearing and pendrin expression. However, the restoration was shown to be unstable and fluctuating and the treated mice experienced progressive HL by age. Importantly, the study indicated that gene therapy that provides early onset of full rather than partial recovery of pendrin expression would be necessary to maintain stable auditory function [40].

2.3. Other deafness-associated genes

TMC1 (transmembrane channel-like 1) encodes a transmembrane protein that is the component for mechanoelectrical transduction of cochlear hair cells [42]. TMC1 mutations may result in either autosomal dominant or autosomal recessive non-syndromic HL, and they are one of the most frequent causes of autosomal recessive HL in consanguineous populations [43]. The mutation c.1253T>A (p.M418K) in TMC1 was identified to be responsible for autosomal dominant HL in a Chinese family [44]. This mutation is orthologous to the mutation Tmc1 c.1235T>A (p.M412K) reported in an HL mouse model named Beethoven (Bth) [45]. Given the availability of this mouse model harboring missense mutation of human genetic deafness, the TMC1 gene captured broad interests by researchers to explore gene therapies for GHL in recent years. Gene replacement [46, 47], gene silencing [48, 49], and gene editing [16, 19] have all been used to prevent or treat GHL originating from different Tmc1 mutations in mice.

MYO15A (myosin XVA) encodes unconventional Myosin 15A (MYO15A), which colocalizes with the PDZ-containing scaffold protein Whirlin (encoded by WHRN) during stereocilia elongation. These two proteins mainly regulate the growth of the stereociliary bundle [50]. Transfection of GFP-Myo15a revealed that MYO15A localizes at the tips of inner ear hair cell stereocilia, and is critical for the morphogenesis of mammalian hair bundles by the recruitment of Whirlin [51, 52]. Mutations in Whrn implicated in autosomal recessive GHL and Usher syndrome have been treated with Whrn cDNA injection in mice [53, 54].

Mutations in gene OTOF (otoferlin) cause prelingual and profound recessive HL [38]. Otoferlin primarily resides in the inner hair cells and helps synaptic vesicle formation for transmission of the acoustic information to auditory nerve [38]. Nonsense mutation c.2485C>T (p. Q829Ter) shows the highest frequency among identified OTOF mutations and creates the third most common cause of autosomal recessive HL in the Spanish population [28]. Two recent studies used a dual-vector system carrying Otof cDNA in Otof knock-out mice and partially rescued the auditory function [55, 56]. Otof (~6 kb) was split into two fragments due to the limited packaging capacity of AAVs of about 5 kb, highlighting that the split-AAV strategy can be used for other large genes, such as MYO15A and CDH23, both of which contain sizes of 11 kb [55, 56].

Lastly, mutations in gene CDH23 (cadherin 23), which encodes a transmembrane protein responsible for inter-stereocilia link and mechanotransduction of auditory stimuli, cause autosomal recessive HL [38]. For recent studies, an exon-skipping nucleotide variant of the Cdh23 gene (c.753G>A) was investigated in mice for age-related HL [18, 57, 58]. The repaired mice c.753A>G was generated using CRISPR-Cas9 technology and successfully demonstrated its influence on rescuing age-related HL [18] and syndromic HL [57]. The reciprocal c.753G>A substitution was subsequently created in embryonic stem cells via homologous recombination and used to verify the function of c.753G in preventing age-related HL [58].

3. Application of CRISPR-Cas9 Systems for Genetic Hearing Loss

Profiling genes or variants attributing to different degrees or types of GHL has been a significant research focus of GHL for decades. This work has revealed that mutations leading to GHL are highly heterogeneous. Mutations in the same gene can cause different phenotypes of GHL (e.g., GJB2 and TMC1), while similar symptoms can arise from mutations in different genes that may affect their transcription, expression, or function of the auditory-associated proteins [59, 60]. Therefore, treating these impairments requires a precise yet versatile approach to make accurate modifications and overcome the heterogeneity in GHL. Here, we discuss the recent CRISPR-Cas gene-editing methods regarding their therapeutic potentials for GHL, as well as the technical challenges regarding their editing efficiency and target specificity. We also explore the recent base editing approach for correcting a single base mutation causing GHL and its advantages over traditional nuclease-based genome-editing agents. Finally, we compare the characteristics of gene replacement and gene silencing with those of CRISPR-Cas9 gene editing.

3.1. CRISPR-Cas9 genome editing

3.1.1. Non-homologous end joining

Dominantly inherited GHL involves a heterozygous mutation that results in a distinct mutant allele and non-affected wild-type allele. Around 20% of GHL is dominantly inherited and mostly develop as postlingual deafness [43]. To efficiently target the dominant-negative mutation, the genome-editing approach requires selective targeting of the mutant allele over the wild-type allele. Additionally, considering the majority of mutations are acquired from a single base, high selectivity, and sensitivity for targeting the mutant alleles are essential (Figure 2).

Cas9 enzyme, guided by a single-guide RNA (sgRNA), performs double-stranded cleavage on a specific DNA sequence [61]. The feasibility of re-defining the complementary sequences on RNA to the different target DNA sites makes the system highly programmable and precise.

Non-homologous end-joining (NHEJ), which results in random indels, is the major post-cleavage pathway in mammalian cells, as it can effectively disrupt a functional gene [62]. Benefiting from these capabilities of the CRISPR-Cas9 systems, Gao et al. [19] first demonstrated selective disrupting of the dominant-negative mutant allele in a Bth mouse model. The authors successfully showed improved cochlear hair cell survival and auditory response through injecting ribonucleoprotein (RNP) of the Cas9:sgRNA complex into heterozygous Bth/+ mice [19]. Importantly, using the chosen mutant allele-targeting sgRNA showed minimal indel formation on the wild-type allele both in vitro and in vivo, suggesting high selectivity of the sgRNA for the mutant sequence. Moreover, using different sgRNAs resulted in varied outcomes of auditory improvements, indicating that prior screening of a selective and efficient sgRNA might be essential [19].

The PAM sequence was then exploited for distinguishing the mutant from the wild-type allele. Studies have suggested that PAM is a prerequisite for unwinding the target double-stranded DNA (dsDNA) and base pairing with the sgRNA [63]. If sgRNA is designed such that the mutation becomes one part of PAM, selective cleavage of the mutant allele might be executed. György et al. [16] used the same Bth mouse model, in which the mutant sequence (5’-GGAAGT-3’) is within the context of ‘NNNRRT’ (R denotes A or G) PAM motif recognized by a variant of Cas9 from Staphylococcus aureus (SaCas9), called SaCas9-KKH. High indels were observed in the mutant allele for all cases, while less than 2% and 0.0075% indels were found in the wild-type allele from the heterozygous Bth/+ fibroblasts and mice, respectively [16]. Therefore, a PAM-based allele-specific targeting strategy is another promising approach.

3.1.2. Homology-directed repair

Homology-directed repair (HDR) is the less-preferred pathway in mammalian cells to repair double-stranded breaks and is nearly absent in post-mitotic cells such as auditory sensory cells. However, HDR produces desired genomic modifications when a donor template to the target locus is delivered [64], thus holding a great promise for safe and highly precise gene therapy [62, 65]. Substantial efforts have been made into overcoming the low HDR efficiency [66, 67], however, with limited success in auditory sensory hair cells. For example, Yeh et al. [20] evaluated the HDR efficiency by introducing a mutation c.98C>T (p.S33F) in Ctnnb1 gene encoding the β-catenin protein of sensory hair cells inside the cochlea to promote hair cell regeneration. The rate of random indels surpassed that of the desired knock-in mutation in Ctnnb1, with the successful HDR rate being only approximately 5%. In another study, Mianné et al. [18] aimed to cure age-related progressive HL through HDR by microinjecting nCas9 mRNA, dual sgRNAs, and single-stranded oligonucleotide donor into homozygous mouse embryos carrying the mutation c.G753A in Cdh23 [68]. Normal auditory brainstem response (ABR) thresholds across different frequencies were observed from successfully repaired mice, but the percentage of pups containing the corrected genotype was less than 1% of the total number of embryos injected. Recently, Ryu et al. [17] also demonstrated that the splicing mutation c.919–2A>G of Slc26a4 gene could be corrected both in vitro and ex vivo through the HDR-mediated pathway but with the success rate being only 14% in primary mouse embryonic fibroblast. Thus, low efficiency remains a significant barrier to HDR-mediated approaches. Unclear endogenous factors still limit its applicability, suggesting that a deeper understanding of the cellular repair pathway after DNA cleavage is necessary to maneuver the pathway better [69]. These studies further suggest that efficient correction of mutations associated with GHL via HDR might still be challenging.

3.2. Base editor

Of all the types of deafness-associated variations (insertion, deletion, duplication, and single nucleotide), SNVs occupy the broadest distribution of all coding variants at ~82% (Figure 1). The base editor directly generates precise single nucleotide mutations in genomic DNA or cellular RNA without requiring double-stranded breaks [14, 70–74]. Fusing nCas9 to a cytidine deaminase enzyme forms a CBE that can be programmed with sgRNA to mediate a C to uracil (U) conversion, ultimately effecting a substitution of C to T (or G to A) with relatively high efficiencies, while minimizing the introduction of random indels to the genome [70]. The activity window for efficient deamination has been typically observed between positions four to eight of the spacer, counting the PAM as positions 21 to 23 (Figure 3a), although different versions of CBE may possess different activity windows [14]. While the largest distribution required for reversing pathogenic point mutation lies in A>G (or T>C) substitution, ABE has also been developed by fusing the nickase to adenine deaminase, which turns A to inosine that gets read as G by polymerase during the cellular repair pathway [72]. Together with CBE and ABE, the four transition mutations have become achievable (Figure 3).

Figure 3. Strategy for using base editors to treat GHL.

(a) Schematic of a base editor. DNA containing the target cytosine (C) or adenine (A), denoted as X, at a specified locus guided by RNA is bound by nCas9. Target C or A located within the activity window of five nt, typically from positions four to eight within the protospacer, counting the protospacer-adjacent motif (PAM) as positions 21 to 23, can be deaminated and changed to uracil or inosine, respectively (not represented), by cytosine base editor (CBE) or adenine base editor (ABE), respectively. nCas9 nicks the complementary strand, and cells favor using the edited strand as a template for DNA repair. The uracil is then read as thymine (T) and the inosine as guanine (G), effecting the conversion of a C to T or an A to G, respectively. (b) Representation of the four different codons Thr34, Ala1209, Ile37, and Ter1975 responsible for causing HL. Nucleotide conversion generating disease is represented in red, and mutation corrected by the base editor is described in blue.

Given the low efficiency of HDR in post-mitotic cells, base editors are great alternatives to correct a single base substitution efficiently. The editing efficiency is typically much higher than that of HDR, reaching ~75% or ~50% of the total cellular DNA when transfected by CBE or ABE, respectively, while undesirable indel, genomic rearrangement, and translocation are minimized down to below 1% [70, 72]. Nevertheless, the in vivo study of GHL using base editors is currently limited due to the lack of animal models harboring representative single base mutation. To date, only one study using base editor has introduced S33F mutation in β-catenin of sensory cells to block its phosphorylation and ubiquitination, thereby upregulating a signaling activity that can promote growth and transdifferentiation of supporting cells into hair cells [20]. The extracted tissue samples treated with the base editor showed editing efficiency of about 13%, while the HDR method resulted in only about 5% mutation of the total sequencing reads. The study demonstrated that base editing could restore the regenerative capacity of hair cells, providing more insights into the cellular reprogramming in vivo to improve the auditory function [20]. Despite this conclusion, no study to date has demonstrated applying base editors to rescue hearing function in mouse models by reversing point mutations.

New applications of using base editors in GHL treatments are being discovered. For example, we expect to develop a base editing strategy that could convert C to T in mutation c.101T>C (p. M34T) in GJB2, one of the most common deafness-associated variants (Figure 3b). As detailed in Table 2, more deafness-associated variants with relatively high allele frequencies can be edited. However, base editors cannot carry out deletions, insertions, or exchanges of purines into pyrimidines, such as mutations c.35delG, c.167delT, and c.235delC, which are all common pathogenic deafness-associated variants (Table 1).

Table 2.

List of base-editable gene variants associated with GHL with an NGG or NG PAM positioned appropriately.

dbSNP#	Genotype	Protospacer and PAM sequence1	Alleles frequency2	Pathogenicity	Ref.
CBE-EDITABLE GENE VARIANTS (AUTOSOMAL RECESSIVE)
rs35887622	GJB2: c.101T>C (p. Met34Thr)	CGCATTAXGATCCTCGTTGTGG	0.00868	Pathogenic	[31]
rs200104362	GJB2: c.503A>G (p. Lys168Arg)	GCACXTCACCAGCCGCTGCATGG	0.00005	Pathogenic	[88]
rs80338945	GJB2: c.269T>C (p. Leu90Pro)	GCTCCXAGTGGCCATGCACGTGG	0.00065	Pathogenic	[89]
rs80338848	SLC26A4: c.707T>C (p. Leu236Pro)	CAGCXAAAGATTGTCCTCAATG	0.00027	Pathogenic	[28]
rs121908362	SLC26A4: c.2168A>G (p. His723Arg)	TCAXGGACCGTCAAAAAGAATG	0.00012	Pathogenic	[37]
rs200532919	MYO15A: c.1454T>C (p. Val485Ala)	CCAGGXGAAGCTGTTTGGGAAGG	0.00101	Pathogenic	[25]
rs200451098	MYO15A: c.8090T>C (p. Val2697Ala)	AGGXGTTTTACCCCAAGGACAG	0.00026	Pathogenic	[90]
rs370155266	USH2A: c.7679A>G (p. Asn2560Ser)	CCCCAXTGGATTTTCTAGGATG	0.00006	Likely pathogenic	[25]
CBE-EDITABLE GENE VARIANTS (AUTOSOMAL DOMINANT)
rs797044967	KCNQ4: c.808T>C (p.Tyr270His)	CCTCXTACGCCGACTCGCTCTGG	0.00000	Pathogenic	[91]
rs121908934	COCH: c.1535T>C (p. Met512Thr)	AGATAXGGCTTCTAAACCGAAGG	0.00001	Pathogenic	[92]
ABE-EDITABLE GENE VARIANTS (AUTOSOMAL RECESSIVE)
rs72474224	GJB2: c.109G>A (p. Val37Ile)	CCACAAYGAGGATCATAATGCG	0.00772	Pathogenic	[31]
rs111033196	GJB2: c.380G>A (p. Arg127His)	TCCYCATCGAAGGCTCCCTGTGG	0.00140	Pathogenic	[88]
rs145254330	SLC26A4: c.349C>T (p. Leu117Phe)	GAGTAGAYACCATATCCGACAGG	0.00033	Likely pathogenic	[25]
rs727504303	SLC26A4: c.2003T>C (p. Leu668Pro)	TCCYGGACGTTGTTGGAGTGAG	0.00005	Pathogenic	[25]
rs181611778	CDH23: c.5131G>A (p. Val1711Ile)	GATCYTCACTGCCACAGACCAG	0.00030	Likely pathogenic	[25]
rs140884994	CDH23: c.9524G>A (p. Arg3175His)	TTTGGGCYTGAGCCAGCAGCTG	0.00028	Pathogenic	[25]
rs111033536	CDH23: c.9569C>T (p. Ala3190Val)	GATGGCAYCCCGCAGGTATCGG	0.00034	Likely pathogenic	[25]
rs397517356	CDH23: c.8257G>A (p. Ala2753Thr)	AGGCYCAGTGGATGCAGATGAGG	0.00004	Pathogenic	[25]
rs375290498	MYO15A: c.5925G>A (p. Trp1975Ter)	GTGYAGGTGCCAGGTGGAGGGGG	0.00200	Likely pathogenic	[25, 93]
rs117071200	MYO15A: c.5978G>A (p. Arg1993Gln)	CAAGCYGGAGGTAGTCGCTGTGG	0.00016	Pathogenic	[25]
rs80356593	OTOF: c.2485C>T (p. Gln829Ter)	GTTCTYGCACAGCCTCAGCTTG	0.00017	Pathogenic	[28]
ABE-EDITABLE GENE VARIANTS (AUTOSOMAL DOMINANT)
rs28931593	GJB2: c.224G>A (p. Arg75Gln)	CATCCYGCTATGGGCCCTGCAG	0.00001	Pathogenic	[94]
rs28938175	COCH: c.151C>T (p. Pro51Ser)	CCCCCTGYGCAGAGGACATCTG	0.00001	Pathogenic	[28]
rs147834269	WFS1: c.2209G>A (p. Glu737Lys)	ACGGCYAGGCCTACCCTGCCTG	0.00073	Pathogenic	[25]
rs35932623	WFS1: c.2452C>T (p. Arg818Cys)	CTGGCYCAGGCTGAGCAGCACG	0.00502	Pathogenic	[25]

¹“X” indicates target “C” of CBE; “Y” indicates target “A” of ABE. Underlined letters indicate an active window.

²Alleles frequency data are from gnomAD Database (gnomad.broadinstitute.org) shown in ClinVar (ncbi.nlm.nih.gov/clinvar) [22].

CBE, cytosine base editor; ABE, adenine base editor.

Off-target activities of genome targeting platforms are major concerns in therapeutic applications [75]. For base editors, their ability to edit unwanted Cs or As within their editing windows, called bystander editing, can potentially have deleterious effects [14]. To improve the specificity of base editors, researchers have engineered the deaminase domain or used sequence-specific deaminase variants and achieved precise single desired nucleotide change with minimized bystander effects [76–78]. Moreover, because deaminases can recognize DNA or RNA as substrates for non-specific deamination of C>U or A>I, which could generate genomic or transcriptome-wide off-target editing, studies have identified key residues in the deaminase domain that could be mutated to substantially reduce the off-target editing and dramatically improve the base editing fidelity [79–81]. Yet, it should be noted that while the core activity of base editors is performed by the deaminase, nCas9 which guides the deaminase to the target could also lead to off-target sites. Since the discovery phase of the CRISPR-Cas9 system, studies have allowed understanding the mechanisms underlying the specificity of Cas9 through structural analysis and unbiased genome-wide off-target detection methods such as GUIDE-seq [82–84]. Today, several engineered Cas9 variants that display high on-target specificity have been characterized, with one even showing undetectable off-target editing by the current detection method [85–87]. These improvements in precision demonstrate the greatly increased targeting fidelity of rapidly evolving genome editing tools. Through continued progress and characterization of these tools including base editors, concerns regarding off-target effects have the potential to be alleviated, which may facilitate moving towards the clinic.

3.3. Comparison of gene replacement and gene silencing with gene editing

Gene replacement has been the most widespread gene therapy strategy of GHL over the past decade. It supplements a functional gene copy to override the mutant gene by delivering cDNA with the wild-type gene sequence [3]. The applications of gene replacement have been exhaustively reviewed [1, 3, 4, 95], thus in this section, we will only indicate the strengths and weaknesses of this approach as compared with gene editing (Table 3). In contrast to CRISPR-Cas9, gene replacement has no requirement for the PAM motif, which limits the targeting scope of CRISPR-Cas9 cleavage activity in the genome [96]. Nonetheless, the usability of gene replacement is still limited since its application is restricted to recessive diseases that require re-installation for the loss-of-function gene. Autosomal-dominant GHL is less likely to be corrected by gene replacement [97].

Table 3.

Comparisons of gene therapy strategies for GHL.

	Gene replacement	Gene silencing		Gene editing
		RNAi	ASOs	CRISPR-Cas9	Base editor
Mechanism	Overriding the faulty gene with the correct copy	Post-transcriptional RNA degradation	Inducing RNase H cleavage of targets	NHEJ/HDR	Deaminating activity
Phenotype and genotype	Recessive homologous	Dominant heterozygous	Dominant heterozygous/recessive homologous	Dominant heterozygous/recessive homologous
Molecular target	None	RNA	RNA	DNA	DNA
Target modulation	Knockin	Knockdown	Knockdown	Knockout or knockin	Base substitution
Off-target rate	None	High	High	Low or moderate	Low or moderate
Sustained time	Short	Short	Short	Permanent	Permanent
Delivery form	cDNA	dsRNA	ssDNA/ssRNA	cDNA/mRNA/protein	cDNA/mRNA/protein
Delivery strategy	Viral vectors	Viral vectors, lipid and polymer nanoparticles.		Viral vectors, electroporation, microinjection, lipid and polymer nanoparticles.
Examples	Gjb2, Tmc1, Whrn, Otof, Clrn1, Ush1c, Lhfpl5, MsrB3, Kcnq1, VGLUT3	Gjb2, Tmc1	Slc26a4, Ush1C	Slc26a4, Tmc1, Cdh23	Ctnnb1
Reference	[33, 34, 46, 47, 53–56, 101–108]	[32, 48, 49]	[9, 37, 100]	[16–19]	[20]

Gene silencing is a strategy to “switch off” the expression of the mutant gene. For dominant GHL, it could be sufficient to restore hearing when the mutant allele is silenced, while the wild-type allele is functionally expressed [3] (Figure 2). As mentioned earlier, ASOs and RNAi are two useful strategies for gene silencing in GHL treatment (Table 3). Both ASOs and RNAi use short synthetic nucleotides that bind target RNAs through Watson-Crick base pairing, but they differ in composition and modes of action. ASOs are composed of short, synthetic, single-stranded DNA (ssDNA) or RNA (ssRNA) oligodeoxynucleotides and function in gene silencing via targeting RNA for degradation, preventing the translation of a specific RNA into a protein, and altering the splicing of pre-mRNA [98, 99]. Differently, RNAi employs a double-stranded RNAs (dsRNAs) to induce post-transcriptional silencing of a gene. The guide strand of dsRNAs can bind with and activate the RNA-induced silencing complex to cleave the target mRNA [99].

The first application of ASOs in the treatment of GHL was the restoration of vestibular function in a mouse model of Usher syndrome type 1C [100]. Lentz et al. [100] first created a homozygous (216AA) Usher syndrome mouse model by gene replacement. When it was treated with a single intraperitoneal injection of ASO-29, hearing at low frequencies was entirely rescued to the wild-type levels; however, the rescue was transient with a gradual decline after two months [100]. In a recent study, two splicing sites c.1001+5G>C and c.1001+5G>T in the SLC26A4 gene in HeLa cells were silenced with efficiencies of ~22.8% and ~81.8%, respectively when ASOs were co-transfected with the vectors harboring modified U1 snRNA [37]. The transient effect of ASOs was also seen in the RNAi treatment [32]. In a proof-of-principle study, the maximum effect of using RNAi for suppressing the expression of the R75W allele variant of Gjb2 to prevent GHL occurred at 3-day post-transfection, but returned to the baseline within 168 hours [32]. In another study, mice carrying the Bth missense mutation were treated with an artificial microRNA (miRNA) expressed in an AAV vector. A single injection of Tmc1-targeted miRNA slowed the progression of HL in Bth/+ mice for up to 21 weeks [49]. Subsequently, a recent study used the artificial miRNA to bring the same effect but only accomplished to a small degree, suggesting the treatment outcome was impacted by the age of mice [48].

Gene silencing has enabled researchers to achieve the knockdown of specific gene targets in a relatively quick and straightforward manner. Nevertheless, it has critical limitations, notably that RNAs are inherently unstable and easily degraded by nucleases. Therefore, it is challenging to deliver unmodified ASOs and RNAi in high concentrations to achieve complete gene silencing. While the gene interference by ASOs and RNAi is partial and transient, the gene editing mediated by CRISPR-Cas9 is permanent. Therefore, from this perspective, CRISPR-Cas9 could be advantageous over ASOs and RNAi. However, the delivery of CRISPR-Cas9 systems is more challenging due to their relatively large size, while only small RNAs need to be delivered in ASOs and RNAi-based therapies.

4. Delivery of CRISPR-Cas9 Systems

Different approaches could be employed for in vivo gene delivery based on different targeted cell types, routes of administration, and therapeutic potency. The labyrinthine structure of the inner ear leads to the difficulty of reaching target cells and slow distribution of genome-editing agents. As illustrated in Figure 4, routes of gene delivery into inner ear cells include: (a) canalostomy, (b) trans-round window membrane, and (c) cochleostomy to the scala tympani and scala media [1, 109]. To transport Cas9 into the inner hair cells to achieve efficient genome editing, many viral and non-viral approaches have been tested for the delivery of genome editing agents into the inner hair cells of mice. Also, three forms of Cas9 are available as delivery cargoes, including Cas9 DNA, mRNA, and protein, with different strengths and weaknesses.

Figure 4. Inner ear schematic showing established delivery approaches for CRISPR-Cas9 gene-editing system.

Established approaches of gene delivery into inner ear cells include canalostomy (a), trans-round window membrane (b), and cochleostomy to the scala tympani and scala media (c) [109]. Viral vectors can be used for the delivery of Cas9 DNA or RNA with sgRNA into the inner hair cells, while non-viral vectors can be additionally used for Cas9 protein delivery (d). Viral vectors and non-viral vectors enter cells through membrane fusion and endocytosis, respectively (e). Delivery of either DNA or mRNA can lead to the expression of Cas9 protein inside the cell, and Cas9 ultimately functions as protein form in genome editing combined with sgRNA (f).

4.1. Delivery vehicles

4.1.1. Viral delivery

Viral vectors have been commonly used in all gene replacement, silencing, and editing strategies for the GHL treatment. During the past decade, tremendous efforts have been devoted to refining vector systems, improving production, enhancing transduction efficiency, and evaluating safety profiles [110]. Three main types of viruses, which have been extensively reviewed elsewhere [1, 3, 4], have been used for delivery of genetic material to treat GHL in mammalian models: AdVs, retroviruses especially lentiviruses (LVs) and AAVs. A summary of the viral vectors used in gene delivery for GHL, including the comparison of their respective pros and cons, is shown in Table 4. Among them, AAV, which belongs to the parvovirus family and are small nonenveloped viruses capable of binding to different receptors, has demonstrated to be the most effective delivery vehicle for the recovery of auditory functions in mouse models of GHL by explicitly targeting the auditory sensory cells and hair cells [5, 12]. AAV also achieves long-term transgene expression that helps avoid the need for repetitive administration [111]. The characterized twelve serotypes of human AAV exhibit distinct transduction efficiencies in individual tissues. As for inner ear cells, AAV1, 2, 6, 8, and Anc80L65 show greater transfection efficiency and have more usability in inner ear delivery accordingly [4] (Table 4). In general, different AAV serotypes have displayed relatively uniform and efficient transduction on inner hair cells but lower efficiencies on outer hair cells [112]. For non-sensory supporting cells, which play an essential role in maintaining ion homeostasis, AAV1, 6.2, and rh.43 displayed higher transduction efficiencies compared with other serotypes [112]. Recently, a variant named AAV-ie (AAV-inner ear) was developed by attaching a short modified cell-penetrating peptide into the capsid of AAV-ie and demonstrated highly efficient transduction efficiencies for all types of cochlear supporting cells in early postneonatal mice [113]. Given that GJB2 is the most common cause of GHL and is primarily expressed in supporting cells, AAV-ie offers a potential gene delivery system that is efficient, safe, and durable [113].

Table 4.

Viral vectors used in gene delivery for GHL studies.

Viral vector	Packaging capacity	Advantages (+) and limitations (−)	Example	Reference
Adenovirus (AdV)	∼8 or ∼36 kb	+ Efficiently transduce both mitotic and non-mitotic cells − High immunogenic effects	Ad5	[112, 114, 115]
			Ad28	[116]
			Ad-GFP-Baylor, Ad-GFPVB, Ad-Cre-GFP-Baylor	[117]
Lentivirus (LV)	∼8 kb	+ Long-term transgene expression + Large packaging capacity − Minimal transduction of inner and outer hair cells. − Possible ototoxic effects.	HOX-GFP, WOX-GFP	[118]
			Lentivirus-EGFP	[119]
			Lenti-VSVG	[120]
			LV-CMV-eGFP, LV-Ub-eGFP	[121]
Adeno-associated virus (AAV)	∼5 kb	+ High transduction efficiency + Long-term transgene expression + Safety + Broad cell tropism − Limited packaging capacity − Difficulty in production	AAV1	[108]
			AAV1, 2, 3, 4, 5, 7, 8	[122]
			AAV1, 2, 5, 6, 8	[123]
			AAV1, 2, 5, 6, 6.2, 7, 8, 9, rh.8, rh.10, rh.39, rh.43	[124]
			AAV1, 2, 6.2, 8, 9, rh.39, rh.43, AAV2/Anc80L65	[112]
			AAV1, 2, 6, 8, Anc80L65	[125]
			AAV1, 6, 8, 9, PHPeB, Anc80L65, AAV-DJ, AAV-ie	[113]
			AAV1, 9	[105]
			Recombinant AAV21	[16, 17, 33, 40, 47, 54, 56, 104, 106, 109, 120, 121, 126–129]
			AAV2. 7m8, AAV8BP2	[130]
			AAV2/DJ, DJ8, PHP.B	[131]
			AAV2 quadY-F	[55]
			AAV5	[132]
			AAV8	[53, 133]
			AAV9/PHPB	[101]

¹AAV2 rep gene is fused with the cap genes of other AAV serotypes.

On the other hand, AAV’s limited cargo capacity prevents its application to deliver genes over 5 kb, such as the OTOF gene (6 kb). The most commonly used Cas9 from Streptococcus pyogenes (SpCas9) is a large nuclease (~4.5 kb), and packaging SpCas9 with its sgRNA in one AAV vector for functional expression is challenging, and the same for even bigger base editors [21]. Ryu et al. [17], and György et al. [16] packed Cas9 DNA and sgRNA into AAV to edit Slc26a4 and Tmc1, respectively. SaCas9 plasmid has been chosen to pack into AAV with sgRNA due to its relatively smaller size (~1 kb smaller than SpCas9). Recently, Al- Moyed et al. [56] and Akil et al. [55] adopted a dual-AAV gene therapy approach to split the oversized six kb- long Otof into two fragments for delivery, leading to the otoferlin expression of 50% and 60% in inner hair cells, respectively and partial restoration of auditory function. These works revealed the possibility that big-size deafness-associated genes, such as MYO15A (11 kb), CDH23 (11 kb) and MYO7A (7 kb), could also be efficiently delivered using dual vectors and provide promising gene therapy tools to treat GHL.

4.1.2. Non-viral delivery

Unlike viral vectors, non-viral approaches have lower immunogenicity and less packaging restrictions. By using non-viral approaches, the delivery of recombinant CRISPR-Cas9 protein and in vitro-transcribed Cas9 mRNA and sgRNAs are relatively easy to achieve [21]. Non-viral delivery can be further classified into two methods: physical methods (microinjection and electroporation) and chemical methods (lipid-based, polymer-based, and gold nanoparticles). Both of the methods have been applied for GHL treatment with showing various advantages and limitations (Table 5), while only microinjection [18, 19] and lipid-based nanoparticles [19, 20] have been demonstrated for CRISPR-Cas9 delivery to date (Table 6). Microinjection allows direct injection of the CRISPR-Cas9 and sgRNA complex into embryonic cells using a glass micropipette at a microscopic level with high reproducibility and specificity [11]. Mianné et al. [18] microinjected nCas9 mRNA into mice embryos to correct a single mutation of Cdh23 through HDR. Although the method enables highly localized delivery, approximately 10% of cells were lysed due to the applied physical stress on embryos, indicating as the major limiting factor of microinjection [18].

Table 5.

Non-viral methods used in gene delivery for GHL studies.

Non-viral methods		Advantages (+) and limitations (−)	Example	Reference
Chemical	Lipid/polymer nanoparticles	+ Enables dose-dependent delivery	Liposomes	[134]
		+ Easy to produce	Lipofectamine 2000	[17, 19, 20, 135]
		+ Versatile for delivering different forms of CRISPR-Cas9 agents	Polybrene	[136]
		+ Biodegradable	Dendritic polymers	[137]
		− Potential cytotoxicity	Polyethylenimine	[39, 138]
	Gold nanoparticles	+ High chemical stability	Gold nanoparticles	[51, 52]
		+ Efficient delivery
		− Potential cytotoxicity
Physical	Microinjection	+ Highly specific delivery into a single target cell	Microinjection	[18, 19, 58, 117, 119, 124]
		+ Versatile for delivering different forms of CRISPR-Cas9 agents
		− Mechanical damage of cells
		− Low-throughput
	Electroporation	+ Broad applicability of cell types	Electroporation	[6, 139–141]
		+ Efficient delivery
		+ Versatile for delivering different forms of CRISPR-Cas9 agents
		− Potential cytotoxicity
		− Nonspecific delivery

Table 6.

Different forms of Cas9 delivery used in GHL studies.

CRISPR-Cas9 forms	Advantages (+) and limitations (−)	Gene	Delivery strategy	Reference
CRISPR-Cas9 plasmid	+ High chemical stability	Tmc1	AAV	[16]
	+ Easy to produce and store
	− High off-target effects	Slc26a4	Lipofectamine 2000, AAV	[17]
	− Potential immunogenic effects
	− Delayed onset
Cas9 mRNA and sgRNA	+ Transient expression	Cdh23	Microinjection	[18]
	− Poor stability
Cas9 protein and sgRNA	+ Quick onset	Tmc1	Lipofectamine 2000, microinjection	[19]
	+ Low off-target effects
	− Low efficiency	Ctnnb1	Lipofectamine 2000	[20]
	− High cost of production

Cationic lipids have been used for the delivery of CRISPR-Cas9 and sgRNA RNPs with relatively high transfection efficiency. The highly anionic RNPs can interact with the cationic lipids, and the resulting nanoparticles could efficiently enter cells through endocytosis (Figure 4). Gao et al. [19] applied the Cas9-sgRNA-lipid complexes to target the Bth allele within the cochlea of Bth/+ mice, and significantly enhanced the survival of inner and outer hair cells. Improved ABRs were also observed after injection, which successfully demonstrated that the lipid-based delivery system enabled to ameliorate HL in the mouse model [19].

4.2. Delivery cargoes

Compared to the gene replacement and gene silencing strategies, which can be implemented by delivering DNA and/or RNA, CRISPR-Cas9 gene-editing systems enable the more flexible design of the delivery cargoes (Table 3). Cas9 has been successfully delivered as DNA, mRNA, or protein with sgRNA to achieve the therapeutic genome editing for GHL in mice (Table 6). Delivering either Cas9 DNA or mRNA form needs to undergo the process of transcription and/or translation to ultimately function as a protein (Figure 4).

As discussed regarding ASOs and RNAi, delivering RNA is challenging owing to their extreme susceptibility to RNase-mediated degradation. This shortcoming also exists in the delivery of Cas9 mRNA [142]. To address this problem, some lipoplexes and polyplexes, such as polyethyleneimine, protamine, TransIT and MegaFectin, have been adapted to the Cas9 mRNA delivery in recent years [21, 143]. Instead of using these protective carriers, Mianné et al. [18] directly microinjected nCas9 mRNA into the zygotes from mice carrying homozygous recessive mutants. To resist rapid clearance by the enzymatic degradation, the mRNA was in vitro transcribed by capping and polyadenylation reactions. The correction of this HL phenotype was highly specific with no off-target editing at any predicted sites [18].

In contrast, delivering Cas9 DNA could achieve better genome editing efficiency but trigger a longer lag time to drive the expression, redundant transgene products causing more off-target effects, and probably stronger immunogenic response (Table 6) [142]. Both ex vivo and in vitro approaches have been used to demonstrate that delivering Cas9 DNA can efficiently induce HDR-mediated genome editing at the Slc26a4 locus by using viral and non-viral vehicles, respectively [17].

The Cas9:sgRNA RNP complex is highly negatively charged and could be efficiently encapsulated into cationic lipid reagents. Gao et al. [19] and Yeh et al. [20] delivered Cas9:sgRNA and base editor:sgRNA RNP complexes, respectively, into newborn mouse cochlea and successfully induced hair cell recovery. When the complexes were delivered into murine cells in vitro, only a few off-target editing was observed based on the unbiased genome-wide quantification method [19, 20]. Delivery of the RNP complex can increase editing specificity through transient expression but could bring lower on-target editing efficiency [142]. This relatively moderate effect of RNP has also been suggested by György et al., when they achieved the prevention of GHL up to one year in Bth mice by delivering Cas9 DNA packed in AAV, while only a short-term auditory improvement was observed when delivering RNP in the same GHL mouse model [16, 19].

5. Disease Model

Currently, direct visualization of human hearing anatomy is impractical. Neural tissue in the auditory system must be removed from the body to be structurally analyzed and preserved, which cannot be accomplished from the alive human bodies. Additionally, studying the human hearing system with modern methods involving electrophysiological assays, including patch-clamp recording and endocochlear potential analysis, is unfeasible [50]. Therefore, evaluating the assay results of genetic manipulations from established animal models is crucial for progressing towards the clinic.

5.1. Mouse models

Mice are the most frequently used organisms in human genetic disease research, as they are behaviorally and environmentally manageable, have short life spans, and possess low genetic heterogeneity. Mice are viable experimental substitutes for humans because the underlying mechanism of mouse hearing function bears a close resemblance to that of humans. Biologists have used them as models in research concerning anatomic development and functionality of the human ear, and continue to use them to study the pathological and physiological causes of auditory impairment [50].

The complete sequence of the mouse reveals around 80% similarity with the human genome [144]. The analogy in the genetic background between mouse and human applies to deafness-associated genes both in terms of sequence and function. An increasing amount of interest has been focused on the discovery of new gene underlying mouse and human GHL [145, 146]. To achieve this goal, the International Mouse Phenotyping Consortium (IMPC) (www.mousephenotype.org) was founded with the collaboration of 19 international research institutions to develop null mutants for the entire mouse genome by characterizing the functionality of each protein-coding gene. To date, 3,006 knockout mouse lines have been generated by IMPC (https://www.mousephenotype.org/data/biological-system/hearing) for GHL studies and subjected to ABR testing to determine their hearing thresholds [145]. For common deafness-associated genes, including Gjb2, Slc26a4, Tmc1, Myo15a, Otof, and Cdh23, 274 mouse models have been established in International Mouse Strain Resource (IMSR) (findmice.org/index) (latest update in September 2019). Among these, only five genotypes correspond to those of humans (Table 7). Further data sets of mouse models for both conventional and novel deafness-associated genes are expected in the coming years.

Table 7.

List of available mouse models with mutations corresponding to human variants in common deafness-associated genes.

Gene	Species1	Accession2	Genotype3	Mouse symbol1	Reference
GJB2	Mouse (Mus musculus)	3056479	Gjb2: c.355G>T (p.Glu119Ter)	Gjb2E119stop	[154]
	Human (Homo sapiens)	552025	GJB2: c.355G>T (p.Glu119Ter)		N/A
SLC26A4	Mouse (Mus musculus)	5806435	Slc26a4: c.2118T >A (p.Cys706Ter)	Slc26a4m2Btlr	MGI4, 2016
	Human (Homo sapiens)	544449	SLC26A4: c.2118C>A (p.Cys706Ter)		N/A
SLC26A4	Mouse (Mus musculus)	5529273	Slc26a4: c.2168A>G (p.His723Arg)	Slc26a4tm2.1Dontu	[155]
	Human (Homo sapiens)	19864	SLC26A4: c.2168A>G (p.His723Arg)		[156]
TMC1	Mouse (Mus musculus)	2177263	Tmc1: c.1235T>A (p.Met412Lys)	Tmc1Mhdabth	[45]
	Human (Homo sapiens)	181568	TMC1: c.1253T>A (p.Met418Lys)		[16, 19]
CDH23	Mouse (Mus musculus)	1857520	Cdh23: c.8803C>T (p.Arg2935Ter)	Cdh23v-5J	MGI4, 1998
	Human (Homo sapiens)	611734	CDH23: c.8803C>T (p.Arg2935Ter)		N/A

¹Information of mice and human variants was collected from International Mouse Strain Resource (IMSR) (findmice.org/index) [157] and ClinVar (ncbi.nlm.nih.gov/clinvar) [22], respectively.

²Single-nucleotide variants in mice and humans are represented by Allele ID from ClinVar and MGI ID from IMSR, respectively.

³Orthologous sequences of human and mouse were analyzed using Ensembl variation resources (useast.ensembl.org) [158].

⁴MGI direct data submission.

N/A: not available.

The most significant differences between mouse and human derive not only from polymorphism in protein-coding genes but also from the tissue-by-tissue discrepancy of gene expression. Hence, disparate phenotypes could be generated from the same gene mutation in humans and mice [147]. For example, when comparing cochlear gene expression patterns in C57BL/6J mice and marmosets, differences identified in the phenotypes of 20 major deafness-associated genes were due to variations in gene expression between the two species [148]. Similar differences were observed between human/marmoset and mouse gene expression of Slc26a4 [149]. In addition to the difference with respect to gene expression level, some ear morphological differences between the human and mouse are noteworthy to mention given their important physiological implications. Comparative anatomy between mice and humans reveals that the smaller cochlea of mice results in the shorter length of basilar membrane. This rigid structure divides cochlea into two interior canals and vibrates sensitively in response to sound. The basilar membrane length negatively correlates to the audible frequency range [150, 151]. Consequently, the shorter length of the basilar membrane of mice (~7 mm) relative to that of humans (~35 mm) results in a higher range of audible frequencies of mice (1–100 kHz) than that of humans (0.02–20 kHz) [152, 153]. This variability of hearing function arising due to the difference in anatomical size makes it challenging to assign the corresponding cochlear regions in mice to humans. These apparent differences could generate confounding variables to translating the studies obtained from mice to humans, indicating that unknown effects other than the genotype itself influence the pathogenicity of GHL and studying large animal models that might possess a more similar phenotypic pattern as humans might be warranted.

5.2. Other animal models

Under the US Food and Drug Administration (FDA) recommendations, alternative suitable mammalian models should be considered for the clinical application of advanced inner ear gene therapy [3]. Although most congenital HLs are caused by recessive mutations, those that progressively damage the auditory function, which might be diagnosed on the late mature stage, are caused by dominant mutations [159]. Thus, gene therapy for dominant HLs using mature animal models may provide insights into strategies of late intervention and also assess possibilities for treating congenital HLs. The guinea pig is another potentially useful mammalian animal model for GHL studies because the anatomical similarities between human and guinea pig ears are greater than those shared between human and mouse ears [3, 160]. Kawamoto et al. [161] demonstrated early success with Math1 overexpression in nonsensory cochlear cells by delivering the Math1 gene through AdVs in a mature guinea pig model. Izumikawa et al. [7] deafened healthy guinea pigs using ototoxic drugs and then delivered the Atoh1 gene to nonsensory cells through AdVs to induce regeneration of hair cells. Substantial improvement of auditory function was achieved in the mature inner ear [7]. It was also testified that AAV vectors could be applied safely in gene transfer into the guinea pig cochlea [162, 163]. Budenz et al. [164] used AAV vectors to administer neurotrophin-3 and BDNF in the inner ears of deafened guinea pigs. Besides viral vectors, non-viral vectors such as liposomes and polyethyleneimine were also utilized in the guinea pig model with GHL [134, 138].

Table 8 lists other alternative animal models and classifies the advantages and disadvantages of each species [160]. In future investigations, larger mammalian models such as rhesus monkeys and pigs may offer alternative platforms for providing insights into translational potency of various proof-of-principle GHL studies. As for nonmammalian models, zebrafish has become an increasingly popular animal model for studying the fundamental mechanisms of hearing [165]. Some genes, such as cdh23 and ush1c, required for hair-cell function in the zebrafish have been associated with the genes in mice and humans [165]. Thus, the same genetic pathways can be utilized for the development and function of the inner ear as those of humans [166]. However, certain data obtained from zebrafish are rarely applied to mammals due to the significant genetic disparities, which makes relating the results for the purpose of understanding humans deafness challenging [5]. Furthermore, identification of a number of major deafness-associated genes in zebrafish is still under progress. For example, the ortholog of the most frequently mutated human deafness gene GJB2, which encodes the connexin 26, has yet to be found in zebrafish by mutagenesis screen [165].

Table 8.

Advantages and disadvantages of different animal models that can be used for GHL studies.

Species	Advantages	Disadvantages	Reference
Mouse	• Vulnerable to sound • Apparent effects of age because of short lifespan	• Limited frequency range of hearing	[152]
Rat	• Good availability of tectorial membrane, Reissner’s membrane, organ of Corti and cochlea	• Hard manipulation • Susceptible to Otitis media	[135–137, 167–169]
Guinea pig	• High structural similarity of cochlear turns, Hensen’s cells, tectorial membrane, Reissner’s membrane and organ of Corti with humans • Easy manipulation	• Few studies on vestibulocochlear	[7, 114, 134, 138, 160–162, 164]
Rabbit	• High structural similarity of the middle ear and ossicular chain with humans	• Limited animal model available	[160, 170]
Rhesus monkey (Macaca mulatta)	• High similarity of the effect of age on progressive auditory damage	• High cost • Limited animal model available	[160]
Pig	• High structural similarity of temporal bone, tympanic membrane, middle ear and ossicular chain with humans • Easy vision and manipulation	• Difficult access to the middle ear • Limited animal model available	[160]

6. Conclusion and Perspectives

A large number of studies using gene replacement and gene silencing have shown that gene therapy can be used to restore hearing in animal models of hereditary deafness. Two centuries after the death of Ludwig van Beethoven, CRISPR-Cas9 gene-editing technology successfully prevented progressive HL in Beethoven mice [19]. Before clinical translation, prolonging the gene expression duration, reducing levels of off-targeting, and improving the efficacy of the delivery system need to be addressed. When RNAi was used to knock down the mutant allele in the same Bth mouse model [19, 49], the initial hearing improvement was found to dissipate after 21 weeks. Consistently, Gao et al. [19] indicated that hearing in the treated mice had deteriorated over time, potentially owing to the lack of specificity or non-cell autonomous factors, in which non-treated mutant cells drive the nearby corrected cells to exhibit mutant phenotype [16]. The subsequent AAV-mediated SaCas9-KKH approach established a long-lasting improvement of hearing preservation through one year of age with no apparent toxicity [16]. Off-target is a significant concern because any unintended modifications are highly unwanted. Recent engineering work of the SpCas9 further introduced high-fidelity variants, such as hyper-accurate Cas9 (HypaCas9), that demonstrate genome-wide specificity with minimized off-target editing [87]. In general, delivery of the complex of Cas9 protein and sgRNA has lower off-target effects than the delivery of Cas9 plasmid and mRNA, and using non-viral approaches engenders lower off-target effects than using viral vectors. However, an ongoing challenge for delivery systems is integrated efficacy and safety.

Animal models like Bth mice, which carry an identical substitution in the orthologous position of humans, are in high demand for proof-of-concept studies of gene editing. Unlike TMC1, gene editing has been less studied in common deafness-associate genes, such as GJB2 and SLC26A4, because of the lack of representative disease models. Generating GHL models in different species should become easier using CRISPR-Cas9 technologies. Base editors can potentially be a useful tool to reverse or install deafness-associated point mutations.

In conclusion, the prospects of translating the successes with animal models to the clinical treatment of GHL becomes more feasible. We are approaching the day when patients with GHL can be treated and listen to “the Ode to Joy” led by gene therapy.

Glossary

AAV

Adeno-associated virus A member of the Parvovirus family, which is composed of small viruses with single-stranded DNA and commonly used as gene therapy vectors

ABE

Adenosine base editor A fusion of Cas9 nickase and an adenine deaminase, which can catalyze the conversion of A>G in the target site directed by single-guide RNA

ABR

Auditory brainstem response A neurologic test of auditory brainstem function using response to auditory (click) stimuli

AdV

Adenovirus A member of the Adenoviridae family, containing 36-kb double-stranded DNA genome

ASOs

Antisense oligonucleotides Small pieces of DNA or RNA that can bind to specific molecules of RNA

BDNF

Brain-derived neurotrophic factor A protein that is encoded by the BDNF gene and related to the canonical nerve growth factor in humans

Bth

Beethoven A mouse model for dominant, progressive hearing loss, that carries a missense mutation in Tmc1 gene

CBE

Cytosine base editor A fusion of Cas9 nickase with a cytosine deaminase, which can catalyze the conversion of C>T in the target site directed by single-guide RNA

cDNA Complementary DNA

DNA synthesized from a single-stranded RNA (mRNA or miRNAs) template using reverse transcriptase

CRISPR-Cas9

Clustered regularly interspaced short palindromic repeats-associated protein 9 A genome editing platform adapted from bacterial immune system that uses an RNA-guided endonuclease Cas9 to form double-stranded breaks at a specific location in the genome

dsDNA

Double-stranded DNA DNA that exists as a double-stranded molecule in which two antiparallel strands are held together by hydrogen bonds between A-T and C-G

dsRNA

Double-stranded RNA Complementary RNA strands utilized in RNA interference through leveraging endogenous mechanism that cleaves both strands to prevent gene expression

DVD

Deafness Variation Database A database that provides genetic, genomic, and clinical data underlying deafness http://deafnessvariationdatabase.org/

FDA

Food and Drug Administration The agency within the US Public Health Service that provides a number of health-related services, majorly responsible for approving foods, drugs, and other biological products based on the tested efficacy and safety

GFP

Green fluorescent protein A protein that fluoresces bright green when exposed to light

GHL

Genetic hearing loss Hearing loss that is caused by genetic factors

HDR

Homology-directed repair An endogenous repair mechanism of double-stranded DNA break that incorporates homologous piece of DNA

HGMD

Human Gene Mutation Database A database that provides published mutations underlying human inherited diseases http://www.hgmd.cf.ac.uk/ac/index.php

HL

Hearing loss A partial or total inability to hear

HypaCas9

Hyper-accurate Cas9 A variant of SpCas9 (N692A/M694A/Q695A/H698A) that displays greater genome-wide specificity with reduced off-target editing

IMPC

International Mouse Phenotyping Consortium An international scientific endeavor to create and characterize phenotypes of mouse strains

IMSR

International Mouse Strain Resource A combined catalog of worldwide mouse resources including inbred, mutant, and genetically engineered strains http://www.findmice.org/

Indels

Insertions or deletions A type of genetic variation caused by double-stranded DNA break, in which random nucleotide sequence gets inserted or deleted

LV

Lentivirus A member of the Retroviridae family that is composed of single-stranded RNA

miRNA

MicroRNA A small non-coding RNA molecule (~22 nucleotides) that functions in RNA silencing and post-transcriptional regulation of gene expression

mRNA

Messenger RNA A single-stranded RNA molecule that specifies the amino acid sequence of a protein for gene expression

nCas9

Nickase Cas9 A variant of Cas9 nuclease (D10A or H840A) that cleaves only one DNA strand

NHEJ

Non-homologous end joining An endogenous repair mechanism of double-strand DNA break that results in direct NHEJ Non-homologous end joining ligation of the broken strands without relying on the homologous template

PAM

Protospacer adjacent motif A consensus DNA sequence motif (usually 2–6 base pairs) that is positioned nearby the protospacer and recognized by the Cas9 protein for enabling targeted DNA cleavage

RNAi

RNA interference A biological process in which RNA molecules inhibit gene expression or translation, by neutralizing targeted mRNA molecules

RNP

Ribonucleoprotein A complex of RNA and RNA-binding protein

SaCas9

Cas9 from Staphylococcus aureus A Cas9 with a size of 1,053 amino acids, recognizing an NNGRRT (where R denotes A or G) PAM

sgRNA

Single-guide RNA A short piece of RNA that binds to a nuclease such as Cas9 and also to a specific DNA sequence to guide the nuclease to a specific location in the genome

siRNA

Short interfering RNA A class of double-stranded, non-coding RNA molecules (20–25 base pairs) operating within the RNA interference pathway

snRNA

Small nuclear RNA A class of small RNA molecules (~150 nucleotides) that are confined to the nucleus and involved in splicing or other RNA processing reactions

SNV

Single-nucleotide variant A variation of a single nucleotide within the genome of an individual

SpCas9

Cas9 from Streptococcus pyogenes The most commonly used Cas9 with a size of 1,368 amino acids, recognizing an NGG PAM

ssDNA

Single-stranded DNA DNA that is synthesized as a single-stranded molecule

ssRNA

Single-stranded RNA RNA that is synthesized as a single-stranded molecule