Mouse methods and models for studies in hearing

Mice have become the dominant model for much of biomedical research. While most hearing research models are outbred, mice and rats may be inbred or outbred. Mice from any inbred strain are nearly genetically identical and are bred to be homozygous at all autosomal loci. In theory, this reduces data variance in any type of experiment, and provides the “cleanest” background for the manipulation of genes. The molecular revolution in laboratory techniques has rendered every mechanistic question of hearing and deafness an inherently “genetic” question, ushering mice into a majority of laboratories. Any inbred mouse strain is like one person for comparative purposes. This may reduce the generalizability of results from any one strain. More broadly, any commercial mouse model represents discarding of genetic diversity, relative to wild mice. Although mice have been singularly successful models for our understanding of hearing and deafness, all applications of mice are based on assumptions that may not be stated, or that the investigators may not realize they are making. Here, we offer guidelines for effective use of mice and highlight some particularly useful models.

Prior to about 1994, mice were the object of only a handful of research papers in hearing each year. Basic physiological principles of hearing were derived using larger and more easily manipulated animal models, and mouse models presented no clear advantage. To be sure, tremendously important hearing research in mice was underway by the 1970s, in the hands of Willott, Saunders, Henry, and others (e.g., Henry and Chole, 1980; Saunders and Garfinkle, 1983; Willott, 1983), but was not widely recognized. This work, characterizing the cochlear aging and noise injury in now-popular inbred strains, laid the foundation for what was to come. In the mid-1990s, papers on mouse hearing showed accelerating growth, reflecting the rapid arrival of molecular techniques and manipulation of DNA. In 1992, Friedman and Ryan published an influential review on transgenic mice (Friedman and Ryan, 1992), and the first characterizations began to appear (Rauch, 1992). Around this time, the first mouse orthologs of human deafness genes were identified (e.g., Battinelli , 1996; Birkenmeier , 1989; Hughes , 1994; Steel and Smith, 1992; Tassabehji , 1994) and mouse deafness genes were recognized as encoding proteins critical for human hearing (Avraham , 1995; Gibson , 1995). Prior to 1999, most mouse studies applied “forward genetics,” whereby useful defects were blindly identified in commercial inbred strains and then mapped by crossing with strains having divergent phenotypes. Parallel mapping endeavors in mice and humans facilitated the discovery of deafness genes. Now gene discovery through blind comparisons of extant mouse strains is receding under funding pressures. Such studies are deemed risky since there is no guarantee that any useful genes will be identified. Yet the search for new deafness genes always begins with the discovery of useful phenotypes with regard to hearing thresholds, noise vulnerability, or age-related hearing loss. Recent impactful discoveries, such as Kujawa's seminal synaptopathy findings in aging noise-exposed CBA/CaJ mice (Kujawa and Liberman, 2006, 2009), might never have been made if they depended solely upon risk-averse funding mechanisms.

Around 1999, “reverse genetics” notably accelerated, wherein one makes an educated guess about a gene, disables it, then determines the effect on form and function (Crawley, 2000; Jackson and Abbott, 2000). The initial goal was stable, constitutive insertion of exogenous DNA, which required transfection of embryonic stem cells or fertilized oocytes. Stem cell transfection worked best in 129-related inbred strains (e.g., 129SvJ, 129SvEv, 129/Ola), while oocyte injection worked well in inbred C57BL/6 (B6) mice. Several substrains of both B6- and 129-related inbred strains show progressive hearing loss, and are problematic for hearing research (Johnson , 1997; Ohlemiller and Gagnon, 2004). Mice (129-related) also came in a confusing variety of substrains, some of which were determined to be genetically contaminated (Simpson , 1997). So that the gene-modified animals could be bred and sustained, germline positive animals for the inserted sequence were often crossed to B6 or CD-1 mice. The latter is an outbred strain (see below) that also features progressive hearing loss (Shone , 1991). As a result, up to the present, applying transgenic or knockout models of interest has often forced investigators to deal with confounding polygenic background hearing loss. To minimize this problem, one might test young mice to avoid the worst hearing loss, or one could attempt to move the engineered allele to a “good hearing” genetic background by serial backcrosses. The latter approach might require 2 years, but can be shortened to <5 generations using “speed congenic” methods, in which mice carrying the fewest alleles from the undesired background are selected as breeders. At the end, the inserted DNA may still include linked genes from the original embryonic cell strain that may affect the phenotype. Randomly inserted transgenes are also prone to hidden knockout effects, wherein the insertion alters the function of a gene.

Constitutive knockouts (all cells, all ages) rarely address whether the mutation of a specific gene produces a particular human disease phenotype. Only a subset of human genetic disease will reflect completely inactivated alleles. More often, the inactivation will be hypomorphic, yielding less of the gene product or less robust protein function. Often the disease-causing allele will not be present in homozygous form. Thus, where possible, heterozygous mice should be examined for abnormal phenotypes. Constitutive gene knockouts may also cause confounding morbidity or lethality. A key refinement was to render expression of knockout alleles conditional by age or cell type using Cre-LoxP recombination (Crawley, 2000; Jackson and Abbott, 2000). This allows for testing of knockouts or transgenes at particular times in development or in specific cell populations. While this approach does not mimic natural germline mutations, which are typically constitutive, it facilitates hypothesis testing and avoids complications of lethal mutations.

Now, CRISPR/Cas9-based gene editing, which co-opts mechanisms evolved by bacteria for removing viral DNA (Gaj , 2013), is replacing prior methods. By greatly extending the types of cells and conditions under which cells can be transfected, and by boosting success rates, CRISPR/Cas9 is revolutionizing the production of transgenic mice and the possibilities for gene therapy. CRISPR/Cas9 gene editing may be rendered conditional, is far more efficient than prior methods, and allows direct transfection of fertilized embryos, eliminating the need for embryonic stem cells (Wang , 2013; Yang , 2013). CRISPR/Cas9 methods permit the simultaneous editing or deletion of multiple loci. While it is not always desirable, CRISPR/Cas9 can alter both alleles of the target gene, so that in a single generation one could therefore produce homozygous carriers for multiple engineered alleles. However, off-target gene modifications remain a problem. While the number of commercially-generated CRISPR/Cas9-gene edited models is growing fast, it is still common to find that the knockout or transgenic model one needs is still available only on a B6 or mixed B6/129/CD-1 background. It is just a matter time before CRISPR/Cas9, or even newer and cheaper methods, render previous gene engineering methods obsolete (Skarnes, 2015). CRISPR/Cas9 methods also have greatly expanded the range of species that can be genetically altered, but the resulting phenotypes will often be problematic to detect and characterize on non-standardized genetic backgrounds.

Mice offer many advantages for modeling human hearing and deafness (Bowl and Dawson, 2015; Schughart , 2013). Their inner ear is functionally mature by 3 weeks of age and they are sexually mature by 6 weeks. Most commercial mouse strains reproduce well, providing large numbers in a short period of time. The typical mouse lifespan is just 2 to 3 years, enabling collection of a lifetime of data within some grant cycles. Foremost, except for an emphasis on high frequencies, their inner ears and brains work in a typical non-specialized “mammalian” mode. Compared to larger mammals, the mouse brain features different functional weightings and interconnectivity of sub-cortical nuclei (e.g., Cryan and Holmes, 2005; Hofman, 2012; Schreiner and Winer, 2007). The lissencephalic cerebral cortex in mice comprises far less of total brain mass, and is less than half as thick as that in primates (Gilman , 2016). Moreover, the shape and function of cortical pyramidal cells is sufficiently different in mice and primates that they may form a qualitatively different type of columnar circuit (Gilman , 2016). The main organizational feature that distinguishes mice and other rodents from primates appears to be the degree of “encephalization” of coding in the latter. Consequently, in rodents particular stimulus characteristics may be extracted at lower levels in the CNS (e.g., Piscopo , 2013).

The popularity of mouse models fundamentally relies on the preponderant similarity of the genetics of human and mouse hearing. At least 140 loci and 100 identified genes are known to cause deafness in humans (Girotto , 2014). According to Steel (2014), more than 1000 genes may ultimately be involved, either by promoting deafness in a Mendelian fashion, or by magnifying the effects of environmental or personal risk factors. About 99% of mouse genes have human orthologs (Bowl and Dawson, 2015). In most cases, mutations of mouse genes promote deafness phenotypes similar to those associated with comparable human mutations. The incomplete overlap of human and mouse deafness genes or phenotypes suggests that inner ear function can involve functionally related genes or different gene modifiers. About two-thirds of ∼340 mouse genes known to cause inner ear dysfunction have not yet been linked to deafness in humans, but this most likely simply reflects how far mouse work has outpaced the human gene search (Steel, 2014). Nevertheless, many mouse knockout models for suspected hearing-related genes yield negative results. Before concluding that the result of a knockout experiment is universally negative, the next step should always be to repeat the experiment on genetic backgrounds quite different from the one initially tested. Just as no single person can be taken to exemplify the human genome, no inbred mouse strain represents all mice, nor is really asserted to model all humans in any respect. Conversely, a clear phenotype observed on a specific inbred background indicates only what is possible and might occur in humans if orthologous genes and modifiers act similarly. Thus, even a positive result should be checked on multiple backgrounds prior to making broad statements.

The most stark human/mouse genetic differences derive not from changes in protein-coding genes, but from more rapid divergence in regulatory sequences that impact gene transcription (Vierstra , 2014). Probably as a result, humans and mice have developed different gene redundancies by tissue, so that humans may solve a particular functional problem with a different protein isoform, or an entirely different protein. A recent comparison of cochlear gene expression patterns in marmosets and C57BL/6 mice concluded that discrepancies in deafness phenotype for 20 major deafness genes are best accounted for by species differences in gene expression, not coding differences (Hosoya , 2016). Another recent study determined that the cellular distributions of key components of endocochlear potential (EP) generation are different in mice and humans (Liu , 2016). These include expression of Kir4.1 (KCNJ10) and the Na⁺/K⁺/Cl^- co-transporter (SLC12A2). This re-distribution of critical elements may alter the respective roles of some cell types in EP generation and its pathophysiology. While the roles of their respective epithelia are likely similar, some functions may be shifted among cell types, so that cell-by-cell roles are not exactly the same.

Decisions about what age of mice to study implicitly aligns human and mouse life stages. The stages we might want to align include early development, adolescence, middle age, menopause/estropause, and old age. Early developmental stages are difficult to align between humans and mice since mice are altricial while humans are precocial. This difference has proven useful, however, as it has allowed testing in neonatal mice of therapies that must ultimately be applied in utero in humans, facilitating proof-of-concept. Pre-adolescence is important because of mouse data suggesting that the cochlea is more vulnerable to ototoxins during roughly the first month of life (Henry , 1981; Prieve and Yanz, 1984). Adolescence is important for a large body of animal data suggesting that the cochlea is more vulnerable to noise, beginning at the onset of adult-like sensitivity, peaking around the time of sexual maturity (∼6 weeks in mice), and ending in early adulthood (∼4 months in mice) (Henry, 1982a,b, 1983). Exactly how and whether distinct early vulnerability windows for ototoxicity and noise apply to humans is unclear (Henley and Rybak, 1995; Henry and McGinn, 1992; Pujol, 1992). Aligning human/mouse middle- and old-age is of interest for relating hearing loss to age-associated risk factors in humans, such as cardiovascular disease, diabetes, obesity, hypertension, and hypercholesterolemia (Agrawal , 2009; Lin , 2011). Menopause in females offers a definitive event. Women generally reach menopause at 48–55 yr, while female mice undergo estropause (the mouse analog) at 11–16 months (Syed , 2010). Unlike humans, wherein females tend to outlive males (Iachine , 2006), a survey of 31 inbred mouse stains showed a high degree of strain-dependence in longevity by sex (Yuan , 2009). Thus, the human pattern of males being more affected by aging and age-related disease may be flipped in some strains, and should be investigated prior to using any strain for aging research. Universal statements likening or contrasting human and mouse aging should be avoided, as there appears to be no single age normalization that applies across organ systems or health issues (Geifman and Rubin, 2013). This may complicate studies of interactions between age-associated co-morbidities and sensory loss. For example, mice are not prone to complex human diseases such as Alzheimer's or cardiovascular disease unless they are “humanized” with predisposing mutations or special diets (Vanhooren and Libert, 2013). Even then, they may only manifest some hallmark characteristics (Kokjohn and Roher, 2009). Likewise, numerous mouse genetic models of Parkinson's disease produce some aspects of the human disease, yet none reproduce the definitive feature, loss of dopaminergic neurons (Blesa and Przedborski, 2014). Unlike humans, mice can synthesize their own vitamin C, a difference with wide-ranging implications for aging processes. Like other rodents, mice have comparatively long chromosomal telomeres and high levels of telomerase activity (Vanhooren and Libert, 2013). This may protect organs and tissues that self-renew through mitosis. Mouse models have been developed that usefully model several accelerated aging or progeroid syndromes (Vermeij , 2016). Although the hearing in many of these has not been tested, several extant models bear mention for multi-systemic accelerated aging profiles that include hearing loss. One is the SAMP8 inbred line, which features broad cochlear degeneration by 6 months of age (Menardo , 2012). The causative gene(s) are unknown. Another model is a line of mice engineered to carry an inactivating mutation of mitochondrial DNA polymerase gamma (Kujoth , 2005). Mice lacking αKlotho show phosphate imbalance and shortened life span due to impaired urinary phosphate excretion and other complications (Kuro-o, 2018). Klotho KO mice show progressive hearing loss that may primarily reflect strial dysfunction (Kamemori , 2002; Yuan , 2018). Finally, DBA/2J mice show rapidly progressive polygenic hearing loss (Johnson , 2008). One of the driving genes is suggested to be mitochondrial (Someya and Prolla, 2010).

Because of their high metabolic rate, mice must be given much larger systemic doses of any ototoxic or therapeutic drug than might be given to larger animals. Alternatively, they may be given the drug at an earlier age, when vascular barriers are not fully developed. With aminoglycosides and platinum-based agents, this has meant applying them only to mice less than 1 month old (Henry , 1981; Wu , 2001), boosting dosing by 4–8× (Poirrier , 2010), or combining them with a potentiator such as furosemide (Hirose and Sato, 2011; Xia , 2014). For some therapeutics, systemic doses in mice must be larger by a factor of up to 10× versus other animals or humans (Davis , 2010; Dowdell , 2009). Other complicating factors include species differences in cytochrome P450 enzymes, which metabolize about 75% of drugs, as well as differences in gastro-intestinal microbiome (Uhl and Warner, 2015). Overall, rats and mice accurately predict human drug toxicity in about 43% of cases. This is not a rodent weakness, but rather an animal model limitation: When all animal models are considered, the success increases only to 71% (Uhl and Warner, 2015). Any mouse study of drug toxicity or therapeutic value should be undertaken in more than one inbred strain.

Mouse hearing extends to 100 kHz, well beyond the upper frequency limit of human hearing. This could be associated with significant anatomic and physiologic differences not yet appreciated. The micro-architecture of the organ of Corti and lateral wall are difficult to define and quantify, and prone to fixation artifacts, so that fine anatomic differences are difficult to evaluate. Recent quantitative analyses of the conformation of the organ of Corti (Soons , 2015) determined the mouse to be typically mammalian. Anatomic features specific to the base of non-ultrasonic animals (e.g., Boettchers' cells, Hensen's stripe, elaboration of root cells) also appear only in the cochlear base of mice. There are also quantitative differences. For example, humans and some mice differ with regard to the density of root cells (a subset of outer sulcus cells extending root processes into the spiral ligament) (Santi , 2016).

Single cochlear nerve fiber studies in mice (e.g., Taberner and Liberman, 2005) indicate prevailing similarities to other mammals. Compared to larger mammals, mice showed broader frequency tuning (Farrahi , 2018). However, this is likely the result of conventional traveling wave mechanics in a smaller cochlea that must also accommodate a larger frequency range. The most striking differences pertain to input/output and spike timing characteristics that are thought to be influenced by phase-locking in mammals with good low-frequency hearing (<4 kHz). Cats, guinea pigs, and especially gerbils tend toward apical-basal divergence of single fiber characteristics that suggests an emphasis on spike timing in the cochlear apex and spike rate in the base (Huet , 2016; Ohlemiller and Siegel, 1994). Mice hear poorly at the frequencies where phase-locking is prominent, and may have no cochlear neurons tuned below ∼2 kHz (Taberner and Liberman, 2005). The entire mouse cochlea may mirror the cochlear base of other mammals.

Human presbycusis is a complex condition, reflecting both genetic and environmental contributions (Ohlemiller and Frisina, 2008). These influences can be separated in mice, which recapitulate many specific aspects of the human disease (Bowl and Dawson, 2015; Kikkawa , 2012). All human cochlear cell pathologies noted by Schuknecht and others (Schuknecht, 1964; Schuknecht and Gacek, 1993) can be found in mice, as well as pathology that corresponds to all Schuknecht's suggested types of presbycusis (Ohlemiller, 2006). Like most humans, most inbred strains do not model these in isolation, and in neither case does the shape of the audiogram offer a definitive diagnostic (Landegger , 2016). While Schuknecht suggested severity criteria for diagnosis of particular types of presbycusis, there is ambiguity across the animal literature as to when a particular form or aspect of presbycusis is usefully modeled. Almost any aged human or animal specimen will show loss of nearly any cochlear cell type, as well as reduction in levels of key proteins. The correlation between hearing thresholds and cellular changes is strong with regard to outer hair cell loss and dysfunction, but weak for loss of cochlear neurons, strial cells, fibrocytes, and a host of other cochlear cells. It seems fair to ask how informative mice or other animal studies showing modest cell or protein loss that may have no impact on cochlear function are. Roughly a dozen genes have been implicated in human presbycusis, although most of these have not been associated with a particular type (Dawes and Payton, 2016). Likewise, a variety of natural and engineered mutations produce a progressive sensory hearing loss phenotype in mice (Bowl and Dawson, 2015). These show little overlap with human candidate presbycusis genes, although some categories of genes (e.g., antioxidant and ion homeostasis) appear common to humans and mice.

Most mouse age-associated cochlear pathology appears to fall under the heading of “sensory” presbycusis, wherein hair cell loss or function become limiting for hearing (Schuknecht, 1993; Schuknecht and Gacek, 1993). Among hair cells, outer hair cell loss appears more critical, since even pronounced inner hair cell loss may not impact thresholds (Lobarinas , 2013). Inner hair cell dysfunction is difficult to diagnose and may more readily impact neural function related to Schuknecht's “neural” presbycusis. While little is yet known about human genetic predisposition to neural presbycusis, both engineered mutations and insults can magnify this type of lesion in mice (e.g., Kujawa and Liberman, 2009; Lang , 2006). Schuknecht's severity criteria for neural presbycusis (loss of at least 50% of neurons) led to an early estimate of 15%–30% of all presbycusis cases, yet this seemed at odds with common complaints about sound quality in aging (Frisina, 2009). Kujawa's recent work in CBA/CaJ mice showing extensive loss of cochlear neurons after even mild noise exposure has caused excitement for potentially linking common levels of environmental noise to age-related neural loss and neural presbycusis (Kujawa and Liberman, 2006; Kujawa and Liberman, 2009). However, comparative studies indicate that CBA/CaJ mice may represent an extreme case and that the capacity for synapse repair varies widely (Shi , 2015, 2016; Song , 2016; Valero , 2017). Subsequent studies have not tied human deficits to noise exposure (Plack , 2014; Guest , 2017, 2018).

Since the EP in humans has been measured only anecdotally (Kobayashi , 1996; Tran Ba Huy , 1989), human “strial” (metabolic) presbycusis remains somewhat theoretical. The best evidence derives from temporal bones showing such extensive strial pathology that the EP was almost certainly reduced (Pauler , 1988; Schuknecht , 1974), in conjunction with studies in gerbils relating EP measures to estimates of functional stria remaining (Gratton , 1997; Schulte and Schmiedt, 1992). Strial presbycusis may disproportionately appear in women (Gates , 1999), particularly after menopause (Hederstierna , 2010). Notably the mouse analog (estropause) is associated with EP reduction in female CBA/CaJ mice (Ohlemiller , 2010), perhaps through mechanisms that may operate in post-menopausal women (Guimaraes , 2004; Price , 2009). Moreover, strial marginal cell pathology constitutes both the primary initial age-associated pathology in human stria (Schuknecht , 1974), and the pathology most predictive of EP decline in aging mice (Ohlemiller, 2009; Ohlemiller , 2010, 2016). While mutations of many genes could impair strial function (Wangemann, 2006), none clearly promote age-related EP reduction. Work in inbred mice suggests some underlying principles, however. B6 mice show pronounced age-related changes in both stria and lateral wall, yet without EP reduction. These mice exhibit progressive outer hair cell (OHC) and hearing loss, so that strial dysfunction was initially predicted as part of their broad pathology (Hequembourg and Liberman, 2001). Thus, strial presbycusis is not universal and need not accompany pronounced hair cell loss or even moderate lateral wall degeneration. This does not mean that lateral wall degeneration cannot affect hair cell function in ways other than EP reduction, but this falls outside the realm of strial/metabolic presbycusis as it was defined. In addition, albino congenics to B6 mice show age-associated EP decline (Ohlemiller , 2009), suggesting that melanin is needed to maintain strial function with age. This finding fits well with work linking skin pigmentation to cochlear pigment levels and preservation of hearing with age (Lin , 2012). It bears emphasis that albinism (lack of melanin) and genetic mutations that impair the migration and function of melanocytes are different conditions. The latter “spotting mutations” may yield asymmetrically sparse or absent strial intermediate cells (Steel , 1987; Steel and Barkway, 1989). While the effects of albinism appear subtle, spotting mutations can eliminate the EP and promote extensive hair cell loss. Another key finding in mouse strial presbycusis models is that age-related EP reduction only occurs in some animals, suggesting that it is a result of genetic, stochastic, and environmental factors. All three remain poorly understood. Finally, mapping efforts in BALB/cJ mice, which show EP decline with age, have identified a quantitative trait locus on chromosome 12 (Maced QTL) for both strial marginal cell density and EP (Ohlemiller , 2016). No gene has yet been identified.

Mice remain a barely tapped resource for the study of both human presbycusis and noise vulnerability, since alleles that promote injury may also promote age-associated pathology (Ohlemiller and Frisina, 2008). Few of the over 400 commercial inbred strains have been examined in detail, and there is no substitute for blind exploration of what these strains have to teach us. Unfortunately, once a useful phenotype is discovered, identification of the gene(s) involved still typically requires hundreds or even thousands of mice, and success in such ventures is never assured.

Genetic mutations, loud Gaussian noise, impulse noise, and ototoxins can alter the cochlea in a wide variety of ways. It may be worth considering whether the lesion we are creating exists in nature, or represents a type of injury that innate repair processes have ever encountered. Apparent failure of a therapeutic to correct an unnatural lesion may be misleading, and some thought should be given to how to “titrate” the lesion to its clinical counterpart. Although more strain comparison data are sorely needed, it appears difficult to achieve a cochlear hair cell wipeout in mice with noise above the basal turn without huge disruption to the organ of Corti (e.g., Wang , 2002). The relation between outer hair cell loss and permanent threshold shift (PTS) appears to differ for mice versus chinchillas, cats, rats, guinea pigs, and possibly humans (Altschuler , 1992; Bredberg, 1968; Chen and Fechter, 2003; Hamernik , 1989; Liberman and Kiang, 1978). In general, hair cell loss does not reliably account for the extent of PTS in mice. Instead, permanent hair cell damage and changes in the appearance of the organ of Corti may be more predictive (Ou , 2000a; Ou , 2000b; Wang , 2002). For a given insult, the extent of hair cell loss will nevertheless be strain dependent. Extant parametric studies cover only a few strains and noise conditions, so that one should plan pilot studies to evaluate the extent of lesions. It should be kept in mind that any single noise exposure tests only one point on a typically unknown noise energy-versus-noise-induced permanent threshold shift (NIPTS) relation. A finding of reduced NIPTS for one exposure does not mean the entire relation is right-shifted, i.e., that subjects are protected over a wide range of exposures. Finally, for any type of noise there exists a level where “micro-injury” and metabolic fatigue will give way to overt tearing of the reticular lamina. These are different modes of injury, with potentially different prospects for remediation.

Beyond the first month of life, mice are relatively impervious to aminoglycosides or cisplatin (Henry , 1981; Prieve and Yanz, 1984; Wu , 2001). Near-complete outer hair cell wipeouts have been achieved in adult mice through systemic application of kanamycin combined with the loop diuretic furosemide in a single dose (Oesterle , 2008) or sub-chronic dosing regimen (Hirose and Sato, 2011). A potentially cleaner total hair cell wipeout may be achieved using transgenic models in which hair cells can be selectively killed using an agent such as diphtheria toxin (e.g., Kaur , 2015). Another experimental ototoxin that produces relatively pure outer hair cell lesions in mice with few other complications is hydroxypropyl-β-cyclodextrin (Crumling , 2012).

The effects of aging and noise exposure are modulated by sex (Hederstierna , 2010; Henry, 2004; McFadden , 1999; Ohlemiller , 2010), so that the makeup of the sample population by sex should be considered in experimental design. The National Institutes of Health (NIH) now requires that both sexes be included in experimental plans. Features of cochlear anatomy such as total number of hair cells and basilar membrane length can be sexually dimorphic (Al-Mana , 2008). The cochlea and entire auditory system are rich with receptors for sex hormones (Caras, 2013). While the reasons for this are not clear, potential roles include adjustment of sensitivity or responsiveness to communication sounds, courtship song or mating behavior, or detection and responsiveness to pup isolation cries. A recent paper (Milon , 2018) confirmed sex-dependence of both NIPTS and its remediation in B6CBAF1/J hybrid mice.

Animals with existing cochlear injury and hearing loss are resistant to further injury. This is partly because the same cells cannot be lost twice, and partly because a pre-existing threshold shift reduces the shifts we may observe after insults. Genetic backgrounds associated with progressive hearing loss (e.g., C57BL/6, BALB/c, DBA/2J) are not suitable for most noise or ototoxic protocols after a few months of age. By 4–6 months of age, cochlear hair cell loss may envelop the basal ∼2 mm of the cochlea (Ding , 2001; Hequembourg and Liberman, 2001; Spongr , 1997; Willott , 1998), rendering much of the cochlea uninformative for other manipulations. More generally, mice with very different hearing thresholds cannot be compared after noise exposure because they will effectively receive different exposures. The confounding effect of prior injury provides a cogent rationale for reporting the animals' initial thresholds in any publication. Editors and reviewers should require this information, certainly in papers that apply mice with known progressive hearing loss or strains that are not well characterized. Inclusion of baseline thresholds also promotes confidence that threshold assessment was properly performed.

Some experimental protocols require unilateral noise exposure so that animals can still respond to sound in behavioral experiments. Unilateral exposure, in turn, may require that animals be exposed while anesthetized. Anesthesia is likely to alter local injury and repair processes, as well as top-down feedback loops that modulate cochlear responses. The result of any exposure in awake versus anesthetized animals should not be assumed to be the same, and anesthesia should be avoided if possible.

Protocols that involve repeat noise or ototoxic exposures are potentially subject to “pre-conditioning” or “toughening” effects, whereby one or more mild insults are protective against a later, more severe, insult (Li , 2017). In the mouse cochlea, such protection has been elicited by restraint, heat shock, hypoxia, noise, and aminoglycoside exposure (Yoshida and Liberman, 2000; Gagnon , 2007; Yoshida , 1999; Wang and Liberman, 2002; Fernandez , 2010). Pre-conditioning appears to work by prodding cells to up-regulate repair or protective factors such as heat shock proteins or corticosteroids (Peppi , 2011; May , 2013). While a single exogenously-applied therapeutic compound may throw protective cellular cascades out of balance, cells appear able to respond to pre-conditioning protocols in a near-optimal way. Thus, pharmacologic means of engaging pre-conditioning processes are promising. Notably, pre-conditioning appears cross-modal. That is, low-level ototoxins may protect against noise and whole-body heat shock appears broadly protective. Therefore many triggers may activate overlapping protective cascades. The robustness of some pre-conditioning varies by inbred mouse strain (Gagnon , 2007), so that the experimental result may depend on two phenomena: growth of injury and growth of resistance. These may be impacted by different genetic factors.

Some experiments may require local application of drugs to one ear, via the middle ear or inner ear fluids. Published protocols in mice have involved injection of up to 10–12 μl directly through the tympanic membrane, or more invasive surgery to open up the ventral bulla directly over the round window niche (e.g., Murillo-Cuesta , 2017; Oishi , 2013). The former is quick and minimally invasive. The latter constitutes a more time-consuming survival surgery with potential complications, and may hold no advantages over the transtympanic approach.

Drugs placed in the middle ear may not elicit purely local effects. Because of their small head size, mice may be particularly prone to the Schreiner effect, whereby drugs that reach the perilymph may reach cerebrospinal fluid (CSF) via the cochlear aqueduct (Barkdull , 2007; Ciuman, 2009; Schreiner, 1999). From there, they may reach the opposite inner ear, so that it is not clear if the opposite ear can be considered an appropriate untreated control. Also, given that the major routes of inner ear entry of drugs injected into the middle ear are likely through both round and oval windows (see AN Salt resources at http://oto2.wustl.edu/cochlea/), basal-apical dispersion patterns will vary with cochlear length. Some drugs may pass directly into the cochlea through channels in the cochlear capsule, especially in the cochlear apex (Salt and Plontke, 2009). The cochlear capsule in mice is particularly thin, as are the stapes and the bone that covers the cochlear nerve just distal to the internal auditory meatus. These features also may render the inner ear distribution of drugs applied to the middle ear of mice somewhat atypical. Finally, perhaps as a reflection of the high mouse metabolic rate and the high surface area-to-volume of the mouse middle ear, compounds injected into the mouse middle ear may have very short dwell times. These times may depend on the hydrophobicity of the drug, and whether the drug is removed by specific carriers or transporters.

An additional route that can be used for either drug application or fluid sampling from the mouse inner ear is the posterior semicircular canal (PSCC), which lies just beneath the surface immediately posterior to the pinna (e.g., Hirose , 2014). In our hands, the muscle lying over the PSCC can be quickly blunt-dissected to expose the thin bone of the canal. In some strains (e.g., B6, BALB/c), barely touching the bone will bring a steady stream of fluid, while in others (e.g., CBA/CaJ), more drilling may be required. The ratio of bone thickness to canal inner diameter also varies by strain, so that the rate of fluid accumulation can vary accordingly. Although the canal houses both perilymphatic and endolymphatic spaces, the endolymphatic space tends to collapse upon opening of the PSCC, so that relatively pure perilymph can be drawn into a micropipette. However, this is true for only the first ∼1 μl, after which the sample will contain primarily CSF.

Mice may not be the best model for some types of inner ear studies. Some manipulations require a larger bulla, more than 2½ cochlear turns [the approximate number in B6 and CBA/J mice (Muller , 2005; Saunders and Garfinkle, 1983)], or a cochlea that affords access to all three scalae in multiple cochlear turns. At no location does the mouse cochlea offer easy access to scala vestibuli, and only in the basal turn is scala tympani accessible. If mice are the best animal model for a particular study, it is important to use complete standardized nomenclature (http://www.informatics.jax.org/mgihome/nomen/) for the strain or substrain selected. Small differences in a name may represent substantial genetic divergence, and may mean the difference between success and failure in replicating a study. Reviewers and journals should not accept incomplete terminology or vagueness regarding where mice were obtained. Access to major research strains is facilitated by agreements among major suppliers (JAX, Taconic, Harlan Sprague Dawley) to sell the others' mice with the exact substrain designation.

Sometimes a research goal may require mice with a particular phenotype. Online resources exist for this purpose, including the Mouse Phenome Database (http://phenome.jax.org/), the International Mouse Phenotyping Consortium (http://www.mousephenotype.org/), and the German Mouse Clinic (http://www.mouseclinic.de/). Nevertheless, in designing many experiments, one often runs into unknowns. For example, which strains are more vulnerable to noise or ototoxins? Beyond a few heavily studied strains (Ohlemiller , 2011b; Wu , 2001), too little is known. Alternatively, one might pick young mice (e.g., 1 month of age) when both noise vulnerability and ototoxicity appear heightened.

Typically, when testing a functional principle or therapy in mice, the goal is to choose a background that will minimize uncontrolled variables, particularly those that might reflect multiple alleles of unknown genes. Inbred mice have generally been the obvious choice, since any two mice of the same inbred strain are, in theory, nearly clones. While there is no such thing as a “neutral” strain, eliminating unknown genetic variance has seemed clearly preferable (Festing, 2010). This is obtained despite the unsurprising result that, since extreme inbreeding and homozygosity are often deleterious, most inbred strains constitute a smorgasbord of health problems. By contrast, outbred models—that is, nearly every animal model other than inbred mice and rats—do not attempt the same level of genetic standardization, and are preferred for some experiments as more nearly the equivalent of human genetic variation. In contrast to inbred lines like B6 and CBA/CaJ, commercial outbred strains such as Swiss, NMRI, ICR, SABRA, and CD-1 are bred to foster genetic heterogeneity through deliberate heterozygosity and multiple allelic combinations. In theory, no two mice of an outbred strain are clones. However, they are often derived from a small number of strains, so that their phenotype may be dominated by alleles common to a few source strains. In that regard, they are not necessarily a source of genetic diversity. Outbred mice are bred according to protocols that may differ among commercial breeders, who make few claims about the degree of standardization one may expect. Relaxed standardization and heightened fecundity allow breeders to produce and sell these mice relatively cheaply. Breeders tend to be selected for a large litter size, which effectively co-selects for size, since larger mice have larger litters (Land, 1970).

An emerging use of commercial outbreds is in complex trait mapping (Chia , 2005), particularly using Diversity Outbred (DO) mice, which were derived from eight genetically diverse founder strains. DO mice display a broad range of phenotypes. They are maintained at JAX as a randomized breeding colony (Svenson , 2012). DO mice can be used for genome-wide association mapping of complex trait loci at sub-Megabase resolution, although large sample sizes (200–800 mice) are recommended and mice must be individually genotyped for a large array of markers (Gatti , 2014). Compared to other commercial outbreds, DO mice provide a much more genetically diverse population for generalizing findings.

If the goal is to minimize genetic heterogeneity while avoiding the limitations of inbreds, F1 hybrids are a sensible approach. F1s are the first generation progeny from a cross of two inbred strains. Every F1 mouse carries one allele from each strain at every autosomal locus, yet all F1s are genetically identical (except for sex). F1s gain robustness (hybrid vigor) from a high degree of heterozygosity, and the experimental design benefits from the use of healthier mice and increased generality of results.

Origins and general characteristics of inbred mouse strains can be found in Festing's Index of Major Mouse Strains (http://www.informatics.jax.org/external/festing/mouse/STRAINS.shtml), and in datasheets available online from distributors such as The Jackson Laboratory (https://www.jax.org/mouse-search?p=205:1:4621098353963360727) and Charles River (http://www.criver.com/find-a-model). We conclude with consideration of the two most heavily utilized inbred strains for hearing research.

Some of the earliest characterizations of hearing in inbred mice focused on the progressive hearing loss of C57BL/6 mice (Mikaelian , 1974; Mikaelian, 1979). These mice begin life with normal hearing, but soon show high-frequency hearing loss that gradually extends to lower frequencies. While more than one locus has turned out to drive the high-frequency hearing loss (Johnson , 1997; Nemoto , 2004), a significant influence is the hypomorphic Cdh23^753A allele (Noben-Trauth , 2003), which has turned out to be present in many other inbred strains (Johnson , 2000). The encoded protein forms part of the hair cell stereociliary tip-link apparatus (Siemens , 2004). The mutation contributes to a generally accelerated loss of OHCs, but intriguingly, also to NIPTS susceptibility (Davis , 2001). The CDH23 protein interacts with Ca⁺⁺ (e.g., de Brouwer , 2003), so that the Cdh23^753A allele magnifies the influence of other deafness genes that likewise encode proteins that interact with Ca⁺⁺ (e.g., Schultz , 2005), as well as other tip-link and hair bundle components (e.g., Geng , 2013). CDH23 is a known deafness gene, underlying both DFNB12 and USH1D (e.g., Pennings , 2004). B6 mice have repeatedly been proffered as an “accelerated aging” model to test therapeutics against age-related hearing loss. They have also been used to speed up aging studies to fit into NIH funding cycles. This approach will continue to have supporters and detractors among reviewers, so that one should be prepared to make a strong case for using these mice.

Although age and noise-induced hearing loss studies suggest that B6 mice have relatively fragile OHCs, they appear relatively resistant to aminoglycosides (Wu , 2001). Their cochlear lateral wall and endocochlear potential also resist injury from noise exposures that impose much greater injury in CBA-related and BALB/c mice (Ohlemiller and Gagnon, 2007; Ohlemiller , 2011a). In addition, B6 mice appear to be unresponsive to some pre-conditioning paradigms (Gagnon , 2007; Ohlemiller , 2011c). B6 mice carrying null alleles for catalase also showed paradoxical protection of hearing with age, and after noise (Rellinger , 2012). Purely protective paradigms like antioxidant treatments have been reported to work in B6 mice by some investigators, but not others (Davis , 2010; Heman-Ackah , 2010). What may differentiate these results is whether the experimental manipulation is stressful or protective. Kraev (2014) summarizes concerns over atypical stress responses in C57BL/6J mice (B6J; here, the J denotes the substrain of C57BL/6 produced by The Jackson Laboratory) that may result from a null genotype for the Nnt gene, which encodes the enzyme nicotinamide nucleotide transhydrogenase. This mitochondrial enzyme increases the production of reactive oxygen species under stressful conditions, so that its elimination appears protective (Nickel , 2015). As one option, experiments related to stress responses can be performed in closely related C57BL/6N (B6N, the NIH substrain of C57BL/6), although these are not congenic to B6J. According to one recent report (Kendall and Schacht, 2014) B6N mice show larger high-frequency threshold shifts with age than B6J, while B6J mice are more susceptible to noise exposure. In summary, C57BL/6J mice may not yield generalizable results in some types of experiments, and B6N mice may not offer a suitable replacement. A commercial B6J congenic line with a wild-type Nnt allele is still needed.

Amid the current heightened interest in synaptopathy and primary neuronal loss (Kujawa and Liberman, 2006, 2009), one feature reported only in C57BL/6J mice merits mention. They may recapitulate a feature of the human spiral ganglion, whereby the somata of radial afferent neurons form unmyelinated “aggregates” that may be connected electrically (Cohen , 1990; Hequembourg and Liberman, 2001; Jyothi , 2010). In humans, where most spiral ganglion cell bodies are unmyelinated (Glueckert , 2005; Tylstedt , 1997; Tylstedt and Rask-Andersen, 2001), this feature may promote survival and sound responsiveness in neurons that have become disconnected from their hair cell targets. In B6 mice, aggregates are observed principally in the apical half of the cochlea and appear to multiply with age. In older B6 mice, where most neurons have been lost as a secondary effect of the Cdh23^753A mutation, the surviving neurons tend to be those within aggregates, so that formation of aggregates may represent a means of preserving neurons in these mice. While it is not clear exactly why this feature appears in B6 mice, it is magnified in B6 congenics carrying a particular variant of the Ly5 (Ptprc, or CD45) gene (Jyothi , 2010).

B6 mice are prone to atherosclerotic lesions (Bult , 2008) that may impact their cochlear aging profile. Most notably, however, the Cdh23^753A mutation can confound normative studies in these mice. Alternative congenic lines lacking this allele are commercially available. One carries the wild type Cdh23 allele from CAST/Ei mice (Johnson , 1997), and the other from CBA/CaJ mice (Kane , 2012). Both of these lines retain normal mid- and high-frequency hearing beyond one year, although they show progressive low frequency hearing loss that probably reflects other mutations. B6 mice with a genetically engineered knock-in of the Cdh23^753G protective allele have also been produced through CRISPR/Cas gene editing (Mianné , 2016). Single nucleotide knock-in mice eliminate the possibility of confounding effects from linked genes in conventionally generated congenic strains.

CBA/J and CBA/CaJ provided “good-hearing” controls in early studies (e.g., Henry and Chole, 1980; Henry, 1982a; Li, 1992). Perhaps due to the early overlap in their uses, some papers have simply stated “CBA” as the type of mice used, which would be fine if these strains were identical, but they are not. Having diverged ∼80 yr ago, they are separated by >2000 polymorphisms (Bult , 2008). Notable among these are the Pdeb^rd1 retinal degeneration mutation, which renders CBA/J mice completely blind, plus other unknown mutations that impart to them different hearing thresholds after 1 yr, and different sensitivities to noise only during the early vulnerable period to noise (Ohlemiller , 2011b). The age-related divergence can largely be accounted for by effects of aging on the EP and cochlear stria vascularis in CBA/CaJ, so that these mice currently represent the “purest” and least confounded mouse model of Schuknecht's strial presbycusis (Ohlemiller , 2010). CBA/J mice are relatively poor breeders and are prone to middle ear and respiratory infections (McGinn , 1992), while CBA/CaJ mice do not show these complications.

Due to their economy, short life span, and high degree of genetic standardization, laboratory mice have become the dominant animal model for much of hearing research. The mouse cochlea operates according to standard mammalian principles and relies on the same cells, genes, and proteins as in other models. Accordingly, there is considerable overlap of deafness genes between mice and humans. Some functions, such as the regulation of particular ions, may be re-distributed to different protein isoforms or cell types in mice versus other models, even though the overall epithelial functions remain the same. Many research methods that can be applied to larger animals can also be performed in mice, including perilymph sampling and injection, EP measures, and single fiber recording. For any given experiment, the particular mouse model, age, and sex should be carefully considered, as all three factors can greatly influence results. Wherever possible, investigators using inbred strains should test multiple strains prior to making broad claims about results.