Genome-Wide Association Study of Age-Related Hearing Loss in CFW Mice Identifies Multiple Genes and Loci, Including Prkag2

Abstract

Purpose

Age-related hearing loss (ARHL) is one of the most prevalent conditions affecting the elderly. ARHL is influenced by a combination of environmental and genetic factors; the identification of the genes that confer risk will aid in the prevention and treatment of ARHL. The mouse and human inner ears are functionally and genetically homologous. We used Carworth Farms White (CFW) mice to study the genetic basis of ARHL because they are genetically diverse and exhibit variability in the age of onset and severity of ARHL.

Methods

Hearing at a range of frequencies was measured using auditory brainstem response (ABR) thresholds in 946 male and female CFW mice at the age of 1, 6, and 10 months. We genotyped the mice using low-coverage (mean coverage 0.27 ×) whole-genome sequencing (lcWGS) followed by imputation using STITCH. To determine the accuracy of the genotypes, we sequenced 8 samples at > 30 × coverage and used those data to estimate the accuracy of lcWGS genotyping, which was > 99.5%. We performed a genome-wide association study (GWAS) for the ABR thresholds for each frequency at each age, and we also performed a GWAS for age at deafness.

Results

We obtained genotypes at 4.18 million single nucleotide polymorphisms (SNP). The SNP heritability for traits ranged from 0 to 42%. GWAS identified 10 significant associations with ARHL that contained potential candidate genes, including Dnah11, Rapgef5, Cpne4, Prkag2, and Nek11. Genetic ablation of Prkag2 caused ARHL at high frequencies, strongly suggesting that Prkag2 is the causal gene for one of the associations.

Conclusions

GWAS for ARHL in CFW outbred mice identified genetic risk factors for ARHL, including Prkag2. Our results will help to define novel therapeutic targets for the treatment and prevention of this common disorder.

Supplementary Information

The online version contains supplementary material available at 10.1007/s10162-025-00994-1.

Introduction

Age‐related hearing loss (ARHL) is the most common cause of hearing loss and is one of the most prevalent conditions affecting the elderly. Twin and family studies reveal that 25–75% of the risk for ARHL is hereditary [1] with recent large-scale genetic studies strengthening the evidence for heritability [2]. There is little existing information about the genes and pathways responsible for ARHL in humans and mice, despite the evidence from our lab and others supporting its heritability and polygenic architecture [3]. Estimates suggest that 5% of the population has hearing loss [4], and approximately two‐thirds of people over the age of 70 in the United States experience ARHL [5]. ARHL has been shown to be independently associated with cognitive decline, dementia, depression, and loneliness and results in an estimated annual economic burden of over $3 billion in medical expenditures [6–8]. Recent advances in understanding the temporal progression of cochlear degeneration demonstrated that synapse loss precedes hair cell death [9, 10]. Although the use of hearing aids and/or cochlear implants has been shown to improve many of these associated conditions, ARHL remains significantly undertreated, and to date, there are no targeted therapies [7, 11].

More than 100 genes have been associated with monogenic, non-age-related deafness. However, a substantial fraction of patients with ARHL have no identifiable mutation in any known hearing loss gene, suggesting that a significant fraction of ARHL is due to a combination of environmental, unidentified monogenic, and polygenic causes [12]. There is ample evidence that the anatomical, cellular, and molecular properties of the mouse and human inner ear are highly homologous. All mammalian inner ear development begins with a thickening of the ectoderm (otic placode) [13]. The placode then invaginates to form the otocyst. Within the otocyst, Sox2-positive epithelial prosensory patches are specified, one of which gives rise to the cochlea [14]. As the cochlear duct extends, there is a wave of differentiation within and surrounding the duct. Ultimately, the cochlea consists of three fluid-filled spaces, the scala vestibuli, scala media, and scala tympani. The sensory and supporting cells exist within the organ of Corti and reside within the scala media [15]. In a recent manuscript describing mouse and human inner ear development at the single-cell level, the authors concluded: “Our analysis revealed remarkable similarity between human and mouse cell cochlear subpopulations. We observed a remarkably similar pattern of key markers of distinct subpopulations in the developing human cochlea to the developing mouse cochlea. Thus, we believe that despite the size and timing differences of cochlear development between the mouse and human, the mouse is likely a very good model of human cochlear development” [16].

Genome-wide association studies (GWAS) of hearing traits in humans, including ARHL, have identified a few genome-wide significant risk loci, but have suffered from a lack of sufficient power [17–22]. Our laboratory previously performed the first human GWAS for ARHL in which we identified a genome-wide significant risk locus within intron 2 of GRM7 [17]. GRM7 was subsequently implicated in ARHL by other genetic studies [20, 23]. Current approaches for treating hearing loss through gene therapies are reviewed in [24]. Recently, several ARHL GWAS, which used data from the UK Biobank (UKBB), which includes genotype and questionnaire (no formal audiograms) data from more than 330,000 individuals, identified several genome-wide significant associations, some of which were in or near genes that cause Mendelian deafness [25–27]. Exome analyses using the rare variant aggregate association analysis approach were also successfully used to identify candidate genes in the UKBB cohort [28]. Notably, all of the candidate genes described were significantly enriched with mouse phenotype ontologies, mostly related to mouse inner ear abnormalities and abnormal auditory brainstem response (ABR) with the authors concluding, “this finding demonstrates the shared genetic pathology in mouse and human auditory systems, supporting the use of mouse models to study human auditory function” [26]. More recently, we identified altered expression levels of Fhod3 on mouse chromosome 18, which resulted in reduced actin content in the cuticular plate, loss of the third-row stereocilia in the cochlear base, and progressive high-frequency hearing loss [29]. A recent GWAS of hearing loss diagnoses using data from the Million Veteran Program (MVP) revealed several loci containing genes associated with stereociliary structure and function [30].

Performing GWAS and other genetics studies in mice has several advantages: the environment can be more carefully controlled, allowing a greater proportion of the heritability to be captured. In addition, findings can be followed up using experimental manipulations. The Knockout Mouse Project/International Mouse Phenotyping Consortium (KOMP-IMPC) has identified 62 novel monogenic causes of early onset hearing loss by testing Auditory Brainstem Response thresholds in 14-week-old mice (http://www.mousephenotype.org) [31]. While this valuable resource may assist with modeling candidate genes discovered in this work, and we see some overlap, these mice are too young for ARHL and model Mendelian forms of congenital deafness rather than polygenic forms of ARHL that are common in the elderly. Furthermore, KOMP-IMPC will miss embryonic lethal genes; however, less drastic polymorphisms in those genes could be viable while still inducing susceptibility to ARHL. Thus, there is a need for GWAS studies of polygenic forms of ARHL in mice because they are complementary to the efforts of KOMP-IMPC. The Hybrid Mouse Diversity Panel (HMDP) model has been used to identify several known and novel loci associated with age-related hearing loss [32, 33].

We performed a GWAS of ARHL in Carworth Farms White (CFW) mice by measuring auditory brainstem response (ABR) thresholds at about 1, 6, and 10 months of age [34]. CFW mice are an outbred commercially available population [35] that have reduced linkage disequilibrium (LD) [36] which provides finer-scale mapping resolution as compared to F₂ crosses, panels of inbred strains, or other commercially available outbred mice [31, 37]. The heterogeneity of ABR thresholds in genetically diverse CFW mice provides an opportunity to study the polygenic causes of ARHL.

Methods

Animals

The detailed procedures used in this study have been previously described [34]. Briefly, the animals were obtained from the Crl:CFW(SW)-US_P08 (CFW) stock of outbred mice maintained by Charles River Laboratories (Portage, MI). The mice arrived at 3 weeks of age. They were tested when they were approximately 1, 6, and 10 months old (Table 1, Fig. 1A). Some but not all mice were tested at two or all three time points. We requested that only one mouse from one litter be shipped to avoid using siblings, which reduces the power of GWAS. However, subsequent genetic analysis suggested that about 252 out of 946 mice used in this work were siblings (Fig. 2E).

Table 1.

Demographic description of subjects

	Males			Females			Total
Age group	N*	Age mean, days	Age st. dev, days	N*	Age mean, days	Age st. dev, days	N*
1 month	97	46.5	9.6	102	47.2	9.8	199
6 months	385	197.8	14.1	363	194.5	11.9	748
10 months	337	313.4	13.1	317	315.0	13.2	654

*Some but not all mice were tested at two or all three timepoints

Fig. 1.

Mice and phenotypes used for genetic analysis. A Mice used in the experiment. Not all animals were tested at all ages; see Table 1 for the summary. B The number of CFW mice that became deaf or retained hearing at different frequencies and different time points. C Ridgeplot of the ABR thresholds in CFW mice measured at three ages at different frequencies. Females are shown above the X-axis, and males are shown below the X-axis for each frequency. ABR > 100 was considered “deaf” and was not included in this plot

Fig. 2.

Genetic architecture of CFW mice. A Density of SNPs called in autosomes 1–19; bin size is 500 kb. B LD decay in CFW mice used in this study. C Histogram of SNPs with MAF > 0. D Heatmap of the GRM of the genotypes of CFW mice in this study demonstrates an absence of a noticeable genetic structure; samples clustered by genetic relatedness. The color represents the pairwise genetic correlation between traits (ranging from 0 to 1). E Scatterplot of GRM values against IBS values for all pairs of CFW mice in this study shows that most of the mice used for the study are not closely related; however, there were some pairs that appear to be siblings

Animals were housed three per cage with a low level of ambient noise, on a standard 12:12 h light–dark cycle, standard laboratory chow, and water ad libitum. Phenotyping occurred during the light phase. Spleens were harvested after the mice were sacrificed and used as a source of DNA for genotyping.

Prkag2-deficient mice were obtained from Taconic, strain C57BL/6-Prkag2^{tm1.2 mrl} (Model 9703—KO) and housed as described above.

Ethics Approval Statement

All procedures were performed in accordance with guidelines from the National Institutes of Health and the Association for the Assessment and Accreditation of Laboratory Animal Care and approved by the Institutional Care and Use Committee at the University of California San Diego (PHS Assurance Number D16-00020; AAALAC Accreditation Number 000503).

Auditory Brainstem Response Testing

The ABR thresholds were measured at three time points: 4–8 weeks (denoted as “1 month” in this paper), 6 months, and 10 months of age, as described earlier [34]. Briefly, the mice were anesthetized using ketamine (80 mg/kg) and xylazine (16 mg/kg) intraperitoneal injection. All hearing tests were performed in a soundproof acoustic chamber. Stimuli were provided by a custom acoustic system consisting of two miniature speakers, with sound pressure measured by a condenser microphone. Auditory signals were presented as tone pips with a rise and a fall time of 0.5 ms and a total duration of 5 ms at the frequencies 4 kHz, 8 kHz, 12 kHz, 16 kHz, 24 kHz, and 32 kHz. These tone pips started at 20 dB and then increased in 5 dB increments up to 100 dB SPL and were presented at a rate of 30/s. The responses were recorded and then filtered with a 0.3 to 3 kHz pass-band. 350 waveforms were averaged for each stimulus intensity. Hearing thresholds were determined by visual inspection; if no waveform was detected at 100 dB SPL, the hearing threshold was recorded as “no response” indicating that a mouse is deaf at this frequency.

Genotyping

DNA was extracted from mouse spleen tissue using the DNAdvance kit (Beckman Coulter). Multiplexed sequencing libraries were prepared using the Twist 96-Plex Library Prep kit (TWIST Bioscience) and then sequenced on a NovaSeq 6000 or NovaSeq X or X plus (Illumina). The average of ~ 2.7 million reads per sample was obtained (paired end, 150 bp). The reads were aligned to the mouse reference genome GRCm38 (GCA_000001635.2). To generate genotypes at single nucleotide polymorphisms (SNPs), we used the following steps: First, we used STITCH [38] without a reference panel, with the “niterations” parameter set to 40 and a position file, to create a reference panel. To construct the position file, we used low-coverage sequencing data from earlier generations of CFW mice [36, 38, 39] and from 8 CFW individuals from the current study that were sequenced at 30 × using whole genome library prep KAPA HyperPrep Kit (Roche) followed by paired-end, 150-bp sequencing on NovaSeq X Plus Illumina instrument. See below for details on position file construction. Then we ran STITCH again with the “niterations” parameter set to 1, using the above result as a reference panel. We removed SNPs that were missing (genotype probability called by STITCH was < 0.9) in > 10% of animals and then performed imputation using BEAGLE [40]. The SNPs on X, Y, and MT chromosomes were not called, in part because only autosomes were available in the reference datasets. After genotypes were called using this procedure, the following SNPs were removed from the analysis: (1) monomorphic SNPs, since they are not useful for further GWAS; (2) SNPs that deviate from Hardy–Weinberg equilibrium (HWE) with − log₁₀(p) > 7, where p is the p-value of the HWE test at a SNP; and (3) SNPs that have genotype missingness rates > 0.1 based on the results generated by STITCH. The filtered dataset contained ~ 4.18 million SNPs on 19 autosomes (Fig. 2A). The data is available from UC San Diego Library Digital Collections (10.6075/J0ZP46BH). An additional 35 mice with SNP missingness rates > 0.2 were removed from the analysis. The final dataset for genetic analysis consisted of 946 CFW mice. A subset of eight of those CFW mice was also sequenced at > 30 × coverage. Genotypes for the animals with > 30 × coverage were called by bcftools and filtered using bcftools to exclude the following SNPs: (1) not biallelic, (2) QUAL < 20 (QUAL score is a Phred-scaled probability that the variant call is wrong, a QUAL score of 20 means a 1-in-100 chance of an incorrect call), (3) GQ < 20 (genotype quality (GQ) score represents the Phred-scaled confidence in the called genotype for a given sample. GQ score of 20 means a 1-in-100 chance of the genotyped allele assignment being incorrect), and (4) genotype missing rate ≥ 0.15, resulting in 8.6 million SNPs. The eight deeply sequenced samples were used for the construction of the position file for STITCH (see above) and as a “truth set” to determine the accuracy of the genotyping. The error rate was estimated by comparing genotypes of eight samples sequenced with > 30 × coverage and the same samples sequenced using low-coverage. A total of 3.89 million SNPs were present in both sets and were used for the error rate estimate, which was 0.45% (e.g., accuracy of > 99.5%).

The extent of LD (r²) decay rates in CFW mice was estimated as follows. First, the SNPs with MAF < 0.05 were excluded to match the SNPs that are used for genetic analysis. Next, bcftools was used to randomly select 3000 SNPs within 5-Mb intervals, and then vcftools was used to calculate the LD metric r² for each pair of selected SNPs that are within the 5-Mb window. The relationship curve between physical distance and r² was fitted by LOESS (locally estimated scatterplot smoothing).

Genetic Analysis

For genetic analysis, each quantitative trait was quantile-normalized. Sex was used as a covariate if it explained more than 2% of the variance. Based on this standard, there were two traits where sex was used as a covariate: ABR threshold at 4 kHz and at 12 kHz at 1 month of age; sex explained approximately 3% of the variance for these traits. The SNP heritability was estimated using GCTA-GREML [41]. Genetic correlations between traits were computed through the bivariate GREML analysis performed with GCTA [42]; this method estimates the genetic correlation by leveraging the genomic relationship matrix (GRM), which is constructed from genome-wide SNP data. GWAS analysis was performed using a linear mixed model, as implemented in GCTA [43], with the GRM used to account for the complex family relationships within the CFW population, and the leave-one-chromosome-out (LOCO) method to avoid proximal contamination [44, 45]. LOCO, which was coined by [46] and first proposed by [44], is a computationally efficient strategy to address the concern of “proximal contamination” [47] that can reduce the statistical power of GWAS. To control for the type I error, the significance threshold was estimated by a permutation test [48]. We used permutation to establish the genome-wide significance thresholds. The genome-wide thresholds for − log10(p) at the significance levels 0.05 and 0.10 were 5.58 and 5.0, respectively. Quantitative trait loci (QTL) were determined by at least one SNP that exceeded the permutation-derived threshold of − log₁₀(p) > 5.0, which was supported by a second SNP within 0.5 Mb of this SNP that had a p-value that was within 2 − log₁₀(p) units of the most significant SNP. Regional association plots were generated using LocusZoom software [49]. The genes that contain SNPs with r² > 0.8 with the top SNPs are considered candidate genes and included in Table 3.

Table 3.

Description of QTLs. For each trait, the top SNP within the QTL is listed, along with its characteristics: trait with the age (in months) and frequency (in kHz) information; top SNP: the genomic position (chromosome and base pair) of the SNP most significantly associated with the trait; AF of the top SNP: the allele frequency of the top SNP; effect size: the magnitude of the SNP’s effect on the trait; effect size SE: the standard error of the effect size; − log10(p) for the top SNP, the genome-wide thresholds for − log10(p) at the significance levels 0.05 and 0.10 were 5.58 and 5.0, respectively; number of genes located within the identified QTL region

Trait	Top SNP	AF of the top SNP	Effect size	Effect size SE	− log10(p) for the top SNP	Number of genes
deaf at 06 mo 08 kHz	chr12:117,951,339	0.140	0.181	0.037	5.99	4
deaf at 06 mo 16 kHz	chr2:49,860,512	0.183	− 0.153	0.034	5.24	2
deaf at 06 mo 16 kHz	chr12:91,900,313	0.951	− 0.274	0.059	5.52	2
deaf at 10 mo 04 kHz	chr1:95,110,035	0.844	0.190	0.040	5.56	0
deaf at 10 mo 16 kHz	chr10:67,242,708	0.315	0.141	0.031	5.24	5
deaf at 10 mo 24 kHz	chr4:155,998,026	0.051	0.283	0.064	5.00	6
deaf at 10 mo 24 kHz	chr10:67,229,071	0.315	0.142	0.032	5.11	5
ABR threshold at 06 mo 08 kHz	chr5:24,688,891	0.146	0.434	0.090	5.87	35
ABR threshold at 06 mo 08 kHz	chr9:105,380,885	0.057	0.605	0.133	5.29	5
ABR threshold at 10 mo 32 kHz	chr11:16,309,495	0.132	− 0.578	0.119	5.90	2

Immunohistochemistry

Prkag2-deficient mice were obtained from Taconic, strain C57BL/6-Prkag2^{tm1.2 mrl} (Model 9703—KO). These mice carry a deletion of exon 3 of the isoform NM_145401.2 (Taconic, personal communications), which causes a frameshift and loss of function. Cochlear samples from 2-year-old Prkag2-deficient mice and their WT littermates were intracardially perfused and post-fixed with 4% PFA for 1 h. Fixed samples were rinsed extensively in phosphate-buffered saline (PBS), dissected under a microscope into three half-turns, and permeabilized in a 0.3% PBS Triton X-100 solution for 15 min at room temperature. The specimens were washed three times with PBS, blocked with 10% goat serum for 1 h at room temperature, and incubated at 4 °C overnight in the dark with polyclonal rabbit anti-myosin VIIa at 1:200 (25–6790; Proteus Biosciences, RRID:AB_10015251). The following day, after three 15-min PBS washes, the tissues were incubated with the Alexa Fluor 594-conjugated secondary antibody at a concentration of 1:1000 for 1 h in the dark at room temperature. Following the final washes after secondary incubations, samples were mounted on slides using ProLong Glass antifade mountant and left to dry in the dark for at least 24 h before image acquisition. Frequency regions corresponding to 32 kHz were located through their distance from the cochlear apex, based on the place-frequency map [50] and imaged with a 20 × 0.8 NA on a Zeiss 880 LSM Airyscan confocal microscope (Carl Zeiss, Oberkochen, Germany). After acquisition, images were processed using the Airyscan technique [51].

Results

Age-Related Decrease in Auditory Brainstem Response

ABR thresholds were measured in 946 outbred CFW mice (Fig. 1A). As expected, an increasing proportion of mice became deaf as they aged (Fig. 1B), and even for mice who retained hearing, the ABR thresholds increased as they aged (Fig. 1C). A detailed description of the hearing loss patterns in CFW mice was published by [34] for a subset of the mice that were used in this paper; the description and conclusions of [34] can also be applied to the cohort used here. We performed a two-way ANOVA to determine whether there was a difference between males and females across frequencies for each time point. The analysis shows that females have a small, but statistically significant increase in ABR thresholds across all frequencies at all time points (Supplemental Table 1, Supplemental Fig. 1). For the genetic analysis, we considered two types of phenotypes: “deaf vs. not deaf at each frequency” and “ABR threshold at each frequency.”

Genetic Architecture of the CFW Mice

Polymorphic SNPs were unevenly distributed across the autosomes; we identified several large regions with few or no polymorphic SNPs, which is consistent with prior studies using CFW mice [36] and likely reflects both a true lack of diversity, as well as regions that may be highly repetitive, contain segmental duplications, or are otherwise difficult to genotype (Fig. 2A). LD decay in this population supports its suitability for high-resolution mapping (Fig. 2B). The distribution of minor allele frequencies (MAF) of SNPs shows that 85.6% of polymorphic SNPs had MAF > 0.05 (Fig. 2C), which is consistent with the history of CFW mice. The genetic structure of a population can be altered due to the breeding schema, which can impact genetic analyses [52]; therefore, we checked that the CFW population used in this study was genetically homogeneous. A heatmap of the genetic relatedness matrix (GRM) demonstrates an absence of a noticeable genetic structure (Fig. 2D). The scatterplot of the GRM metric and relatedness, calculated as Identical by State (IBS) score between all pairs of CFW mice in this study, shows that most of the mice used for the study are not closely related, but that some closely related mice were included (Fig. 2E).

Heritability estimates for hearing thresholds and hearing loss traits range between 0 and 42% (Table 2). The SNP heritability is expected to be lower than the heritability estimated in twin studies or by using inbred strains. In this dataset, the heritability was moderate, but is more than sufficient for GWAS for most of the traits.

Table 2.

Heritability of traits. SNP heritability and standard error of the mean are shown for the ARHL traits. Nominally significant p-values are shown in bold text

Trait	N	SNP heritability	SNP heritability SEM	p-value
deaf_06 mo_04kHz	748	0.106	0.055	0.02
deaf_06 mo_08kHz	748	0.054	0.052	0.157
deaf_06 mo_12kHz	748	0.1	0.056	0.034
deaf_06 mo_16kHz	748	0.148	0.059	0.003
deaf_06 mo_24kHz	748	0.19	0.06	0
deaf_06 mo_32kHz	745	0.187	0.059	0
deaf_10 mo_04kHz	654	0.132	0.063	0.01
deaf_10 mo_08kHz	654	0	0.048	0.5
deaf_10 mo_12kHz	654	0.06	0.054	0.11
deaf_10 mo_16kHz	653	0.047	0.055	0.18
deaf_10 mo_24kHz	651	0.032	0.053	0.263
deaf_10 mo_32kHz	636	0.033	0.055	0.264
thr_01 mo_04kHz	190	0	0.154	0.5
thr_01 mo_08kHz	195	0.346	0.166	0.014
thr_01 mo_12kHz	191	0.212	0.171	0.094
thr_01 mo_16kHz	186	0.18	0.198	0.222
thr_01 mo_24kHz	191	0.417	0.169	0.005
thr_01 mo_32kHz	176	0.22	0.193	0.142
thr_06 mo_04kHz	493	0.086	0.074	0.109
thr_06 mo_08kHz	523	0.208	0.076	0.001
thr_06 mo_12kHz	497	0.109	0.072	0.044
thr_06 mo_16kHz	464	0.123	0.082	0.048
thr_06 mo_24kHz	453	0.035	0.074	0.312
thr_06 mo_32kHz	427	0.2	0.099	0.023
thr_10 mo_04kHz	390	0.089	0.097	0.18
thr_10 mo_08kHz	437	0.226	0.096	0.006
thr_10 mo_12kHz	427	0.106	0.094	0.144
thr_10 mo_16kHz	405	0.176	0.097	0.024
thr_10 mo_24kHz	382	0.23	0.108	0.014
thr_10 mo_32kHz	329	0.188	0.126	0.078

Genetic Correlations

To examine the genetic relatedness among thresholds and deafness, genetic correlations were computed (Fig. 3). Correlations were not calculated for the “deaf at 1 month” traits because of an imbalanced number of deaf mice and a small sample size. The deafness at 6 months for any frequency had strong genetic correlates with deafness at 6 months for all frequencies. Similarly, ABR thresholds at 6 months at any frequency had a strong genetic correlation with ABR thresholds at 6 months at other frequencies. The ABR thresholds at 1 month tend to have a negative correlation with ABR thresholds at 10 months. One explanation is that the mice whose hearing was not good at 1 month were already deaf at 6 and 10 months and were dropped from the comparison.

Fig. 3.

Genetic correlation for the hearing traits measured at three time points. The color of the circle indicates genetic correlation. The size of the circle is proportional to one minus the standard error of the mean of the genetic correlation metric; therefore, a larger circle indicates a lower standard error and thus a more reliable correlation. The values on the diagonal show heritability for each trait

GWAS Results

GWAS was performed to identify loci that were significantly associated with hearing thresholds or with deafness, for each threshold, at each time point. We identified ten QTLs for seven traits. QTLs, the top SNP for each QTL, their effect sizes, and the number of genes located within each QTL are shown in Table 3. The chromosomal locations for identified regions of interest are shown as a porcupine plot (Fig. 4). The full genetic report is available in Supplemental Materials.

Fig. 4.

Porcupine plot for all measured ARHL traits. The red line indicates a threshold for genome-wide alpha of < 0.10 (− log₁₀(p) > 5.0); the blue line indicates a threshold for genome-wide alpha of < 0.05 (− log₁₀(p) > 5.58). Circles indicate top SNPs for hearing traits measured at the 6-month time point, triangles indicate top SNPs for hearing traits measured at the 10-month time point; colors are showing a specific trait

Due to the high resolution of genetic mapping in the CFW population, most QTLs contained only a few genes. Each QTL interval was examined at the Mouse Genome Informatics portal (MGI) [53] which aggregates data on previously reported QTLs and mutant phenotypes, as well as gene expression. To identify candidate genes within each QTL, we considered several criteria: whether the gene was located within an interval that contains SNPs in high LD with the top SNP (r² > 0.8), the presence of moderate or high-impact variants located within the gene, as predicted by SnpEff [54], and the expression in the tissue of interest, cochlea, in the publicly available datasets that are available in the MGI database and the gEAR portal (umgear.org). In addition, the dataset from [55] was obtained from 48 10-month-old CFW mice with and without hearing loss; 45 of those mice [55] were included in the current study. Five of the genes that were found in the QTL regions identified in this study were also detected in [55] and are discussed below.

We detected three QTLs for elevated thresholds at 6 months (Table 3).

QTL on Chromosome 12 at Around 118 Mb

This QTL (Fig. 5A) is associated with being deaf at 6 months at 4 kHz, and it also shows a trend towards an association with elevated thresholds at 6 months at 12 kHz (− log₁₀(p) = 4.6). There are 4 genes in this locus: Dnah11 (dynein, axonemal, heavy chain 11), Cdca7 l (cell division cycle associated 7 like), Sp4 (trans-acting transcription factor 4), Rapgef5 (Rap guanine nucleotide exchange factor 5). No hearing-related mutations or QTLs have been reported for this locus previously.

Fig. 5.

QTLs for the traits of being deaf at 6 months. A QTL on chromosome 12 for the trait of being deaf at 6 months at 4 kHz. B QTL on chromosome 2 for the trait of being deaf at 6 months at 16 kHz. The x-axis shows the position on a chromosome (in Mb); the y-axis shows the significance of the association (− log₁₀(p) value). The individual points represent SNPs. The SNP with the most significant p-value (“top SNP”) is highlighted in purple. The color indicates the LD between the top SNP and the other SNP. The red line indicates the threshold for genome-wide alpha of < 0.10 (− log₁₀(p) > 5.0); the blue line indicates the threshold for genome-wide alpha of < 0.05 (− log₁₀(p) > 5.58). The effect plots for the top SNP are shown on the right, with minor allele frequency (MAF) indicated

Dnah11 has no previously reported connection to hearing/vestibular phenotypes in mice, according to the MGI database. In humans, mutations in DNAH11 cause diseases related to the dysfunction of the cilia during embryologic development, such as situs inversus (abnormal distribution of the major visceral organs within the chest and abdomen). Dnah11 is expressed in hair cells in 10-month-old CFW mice [55]. Other genes that encode dynein heavy chains are known to cause hearing dysfunction: DNAH2 is predicted to cause ARHL in humans, as determined by the presence of a rare variant that affects protein function in a cohort of hearing loss patients [56]; another related gene, Dync1 li1, is required for the survival of cochlear hair cells in mice [57]. In CFW mice, Dnah11 has two missense mutations, both of which are in strong linkage disequilibrium with the top SNP: 12:118,190,825 (c719 A > G; Glu240Gly, r² with the top SNP 0.982) and 12:118,198,712 (c121 C > T; Arg41 Cys, r² with the top SNP 0.972), suggesting they could be the cause of this QTL.

Rapgef5 (Rap guanine nucleotide exchange factor 5) has not been previously associated with hearing/vestibular phenotypes in any species. An indel mutation in RAPGEF5 causes epilepsy in dogs [58]. The function of RapGEFs can vary depending on the specific isoform and the cellular context; they regulate cell adhesion, cytoskeletal dynamics, and tissue morphogenesis. These processes are important for the development and maintenance of the structures of the inner ear, making Rapgef5 a plausible novel candidate gene. In 10-month-old CFW mice, this gene was found to be expressed in endothelial cells, border cells, and pillar cells [55].

QTL on Chromosome 2 at 49.7 Mb

This locus was associated with being deaf at 6 months at 16 kHz (Fig. 5B) and also showed a trend towards associations with hearing loss at 6 months for all other tested frequencies, although these associations do not reach the significance threshold (− log₁₀(p) range from 4.01 to 4.55). There are two genes in this locus: Kif5c (kinesin family member 5 C) and Lypd6b (LY6/PLAUR domain containing 6B).

Kif5c (kinesin family member 5 C) encodes a protein that has microtubule motor activity and is located in the related cellular components, including ciliary rootlet. Although we are not aware of any previous reports that Kif5c is involved in hearing in any species, other kinesin motor proteins are known to be associated with hearing loss. For example, Klc2 (kinesin light chain 2) knockout mice have early hearing loss at low frequencies, and KLC2 binds KIF5 C in the mouse cochlea, as shown in co-immunoprecipitation experiments [59]. This gene was not detected in 10-month-old CFW mice [55], but is expressed in inner hair cells in young mice [60]. Lypd6b encodes a protein that is predicted to regulate the activity of the acetylcholine receptor and is not known to be associated with hearing function.

QTL on Chromosome 12 at 92 Mb

This locus is associated with being deaf at 6 months at 16 kHz (Table 3 and Supplementary Materials). The most strongly associated SNP for this QTL also showed a trend towards an association with hearing loss at 6 months for 12 kHz and 24 kHz (− log₁₀(p) = 4.4 and 5.02). This locus contains the genes Ston2 (stonin 2) and Sel1 l (sel-1 suppressor of lin-12-like). These two genes are not known to be associated with hearing loss and were not expressed in 10-month-old CFW mice [55]; however, Sel1 l was expressed in cochlea cells in several of the gEAR datasets [61, 62].

We detected three QTLs for being deaf at 10 months (Table 3).

QTL on Chromosome 1 at 95.2 Mb

This locus was associated with hearing loss at 10 months at 4 kHz (Table 3 and Supplementary Materials), and also showed a trend towards an association with being deaf at 10 months for the 16 kHz and 32 kHz frequencies (− log₁₀(p) = 4.26 and 4.13 correspondingly). This locus does not contain any known genes, but may contain unknown genes or transcripts or regulatory sequences that influence the expression of genes outside the associated region. As far as we know, this locus has not been previously associated with hearing loss in any species.

QTL on Chromosome 10 at 67 Mb

This chromosomal region contains QTLs for hearing loss at 10 months for both 16 kHz and 24 kHz (Fig. 6). It contains three genes: Reep3 (receptor accessory protein 3), Jmjd1c (jumonji domain containing 1 C), and the predicted gene Gm31763 for a long non-coding RNA.

Fig. 6.

QTL for the trait of being deaf at 10 months at 16 kHz. The regional association plots are shown on the left. The x-axis shows the position on a chromosome (in Mb); the y-axis shows the significance of the association (− log10 p-value). The individual points represent SNPs. The SNP with the lowest p-value (“top SNP”) is highlighted in purple. The colors represent the correlation between the top SNP and the other SNPs. The red line indicates a threshold for genome-wide alpha of < 0.10 (− log10(p) > 5.0); the blue line indicates a threshold for genome-wide alpha of < 0.05 (− log10(p) > 5.58). The effect plots for the top SNP are shown on the right, minor allele frequency (MAF) indicated

Reep3 is predicted to play a role in tubular network organization. It is expressed in cochlea [60, 63, 64], but was not detected in 10-month-old CFW mice [55], and is not known to be associated with hearing loss in any species. Reep3 has a missense variant in CFW mice in high LD with the top SNP of the QTL (c50 T > C; Phe17 > Ser, r² with the top SNP 0.911).

Jmjd1c is a predicted histone demethylase and coactivator for transcription factors. It is expressed in inner and outer hair cells in embryonic and young mice [60, 64, 65], but was not detected in 10-month-old CFW mice [55], and is not known to be associated with hearing loss in any species.

QTL on Chromosome 4 at 156 Mb

This chromosomal region contains QTLs for hearing loss at 10 months for 24 kHz (Table 3 and Supplementary Materials). The top SNP for this QTL also shows an association with hearing loss at 10 months for 32 kHz (− log₁₀(p) = 4.89). This region contains six genes: B3 galt6 (UDP-Gal:betaGal beta 1,3-galactosyltransferase, polypeptide 6), Sdf4 (stromal cell derived factor 4), Tnfrsf4 (TNF Receptor Superfamily Member 4), Tnfrsf18 (TNF Receptor Superfamily Member 18), Ttll10 (tubulin tyrosine ligase-like family, member 10), and Gm16008 (predicted long non-coding RNA). None of these genes is expressed in 10-month-old CFW mice [55], although they are expressed at various levels in the cochlea of E16, P0, P1, P7, and P16 mice [60, 63–67]. To our knowledge, none of these genes was associated with hearing loss in any other species.

We detected 3 QTLs for elevated ABR thresholds (Table 3).

QTL on Chromosome 5 Around 24 Mb

This QTL for elevated ABR threshold at 6 months for 8 kHz encompasses a > 1-Mb chromosomal region containing 35 genes (Fig. 7A). SNPs in this QTL also show association with hearing loss at 6 months for 4 kHz, and at 10 months for 4 kHz and 24 kHz, although these associations do not reach the significance threshold (− log₁₀(p) = 5.05, 4.84, and 4.11, respectively). This QTL contains several genes that have previously been known to affect hearing. Asic3 (acid-sensing ion channel 3) is expressed in sensory neurons and participates in neuronal mechanotransduction [67–69] and is expressed in supporting cells in adult CBA/J mice [60]. Knocking out this gene is known to disrupt hearing in mice [70]. Asic3 expression was not detected in 10-month-old CFW mice [55]. Slc4a2 (solute carrier family 4 (anion exchanger), member 2) plays a role in ion homeostasis and is expressed in the inner ear [60, 71]. Knockout mice have multiple severe phenotypes, including deafness [72]. Slc4a2 expression was not detected in 10-month-old CFW mice [55]. Crygn (crystallin, gamma N) is expressed in newborn [67] and adult mice [60]; its expression is required for post-migratory survival and proper function of auditory hindbrain neurons. Ablation of this gene does not affect ABR thresholds but causes an increase in the amplitude of wave IV [73]. Crygn expression was not detected in 10-month-old CFW mice [55]. Nos3 (nitric oxide synthase 3, endothelial cell) is expressed in inner and outer hair cells of newborn [67] and adult mice [60]. In humans, polymorphisms in NOS3 are associated with sudden sensorineural hearing loss [74]. Nos3 expression was not detected in 10-month-old CFW mice [55]. Cdk5 (cyclin-dependent kinase 5) is expressed ubiquitously in the cochlea of newborn [65, 67] and adult mice [60, 75]. The cochlea-specific inactivation of Cdk5 causes hearing loss in mice due to loss of stereocilia [75]. Cdk5 expression was not detected in 10-month-old CFW mice [55].

Fig. 7.

QTLs for the ABR threshold for 8 kHz at 6 months. A QTL on chromosome 5. B QTL on chromosome 9. The regional association plots are on the left. The x-axis shows the position on a chromosome (in Mb); the y-axis shows the significance of the association (− log₁₀ p-value). The individual points represent SNPs. The SNP with the lowest p-value (“top SNP”) is highlighted in purple. The colors represent the correlation between the top SNP and the other SNPs. The box plots on the right are showing ABR thresholds in each animal grouped by the genotype at the top SNP, with the minor allele frequency (MAF) indicated

The other genes located in this QTL were not previously reported to be associated with hearing loss. Only one of them, Prkag2 (protein kinase AMP-activated non-catalytic subunit gamma 2) is expressed in 10 month old CFW mice, mostly in hair cells and spiral ganglion neurons [55]. Our work with Prkag2 mutant mice (described below) provides support for the role of Prkag2 in hearing loss (see below).

QTL on Chromosome 9 Around 105 Mb (Fig. 7B)

This chromosomal region contains QTLs for elevated ABR threshold at 6 months for 8 kHz. It contains 6 genes: Cpne4 (copine IV), Mrpl3 (mitochondrial ribosomal protein L3), Nudt16 (nudix hydrolase 16), Nek11 (NIMA (never in mitosis gene a)-related expressed kinase 11), and Aste1 (asteroid homolog 1).

Cpne4 is expressed in supporting cells in newborn mice [67] and at low levels in pillar cells in adult mice [60]. In 10-month-old CFW mice, Cpne4 expression was detected in spiral ganglion neurons [55]. The copine family of Ca-dependent membrane adaptors is well studied in retinal ganglion cells [76, 77]. Our results raise the possibility that Cpne4 is also involved in the functioning of spiral ganglion cells in cochlea and is associated with hearing loss.

Mrpl3 is expressed in cochlear cells of newborns [65, 67] and adult mice [60]. The expression is not detected in 10-month-old CFW mice [55]. CFW mice have a missense mutation in Mrpl3 (c55G > A; Ala19 Thr, r² with the top SNP 0.822). Mutations in Mrpl3 have been previously reported to cause altered ribosome assembly and abnormal function of respiratory chain complexes [78].

Nudt16 is expressed in the cochlea of newborn [64, 65, 67] and adult mice [60]. It was not detected in 10-month-old CFW mice [55]. CFW mice have a missense mutation in Nudt16 (c457 C > A; Val153Met, r² with the top SNP 0.911).

Nek11 regulates the cell cycle. It is expressed in hair cells in newborn and adult mice [64, 65, 67]. In 10-month-old CFW mice, it is detected in hair cells at a low level, but is mostly expressed in a novel cell type characterized by the expression of Dnah12 and Rgs22 [55]. Dnah11, a candidate gene discussed above, is also expressed in this novel cell type, providing an intriguing possibility of the role of this novel cell type in hearing loss.

Aste1 is expressed mostly in hair cells in newborn and adult mice [64, 65, 67]. It is not detected in 10-month-old CFW mice [55].

QTL on Chromosome 11 Around 16 Mb

This QTL is associated with elevated ABR thresholds at 10 months at 32 kHz (Table 3 and Supplementary Materials). It contains two genes: Vstm2a (V-set and transmembrane domain containing 2 A) and Sec61 g (SEC61 translocon subunit gamma).

Vstm2a and Sec61 g are expressed in supporting cells of newborn mice [65, 67], but were not detected in 10-month-old CFW mice [55]. We are not aware of any prior associations between these genes and any hearing-related phenotypes.

Prkag2 Deficiency Causes Age-Related Hearing Loss at High Frequency

Prkag2 was identified as a candidate gene in QTL on chromosome 5 around 24 Mb for elevated ABR threshold at 6 months for 8 kHz. Prkag2 was the only gene in this QTL that was expressed in the cochlea of 10-month-old CFW mice [55], emphasizing its possible role in age-related hearing loss. To further investigate the role of Prkag2, we used Prkag2 constitutive null mice and aged these animals, along with the littermate controls, in an effort to assess the impact of Prkag2 deficiency on hearing. The onset of high-frequency hearing loss at 20 weeks in the mutants, in comparison to the wild-type (Fig. 8) and the loss of not only outer hair cells, as expected on the C57BL/6 background, but also inner hair cells at 2 years (Fig. 8), confirms a role for Prkag2 in hearing and suggests that it may be the causal gene at the locus on chromosome 5.

Fig. 8.

Prkag2-deficient mice show increased sensitivity to ARHL. A Mice homozygous for the Prkag2 loss-of-function mutation show high-frequency hearing loss as measured by ABR thresholds in 20 weeks old mice, N = 5. B Confocal images from cochlea samples of 2-year-old mice show that wild-type littermates have higher IHC preservation in comparison to the mice homozygous for the Prkag2 loss-of-function mutation. Scale bar is 10 µm. C Quantitation of IHC in WT mice and mice homozygous for the Prkag2 loss-of-function mutation

Discussion

ARHL is a complex trait, meaning that it is influenced by many genetic variants, each having a small effect size. In humans, 153 genes that are associated with hearing loss had been described as of 2024 [79], with various effect sizes and various degrees of confidence. However, this list is not exhaustive, and the genetic architecture of hearing loss remains an active field of discovery. Animal models allow gene discovery, which can help to refine our understanding of biological pathways that contribute to hearing loss. In the current work, we use CFW mice to find the genetic underpinnings of age-related hearing loss. The variability in hearing loss in CFW mice has been reported previously, where a subset of mice did not respond to the 120-dB pulses at the age of around 4 months old [36], in agreement with the current report. This strain has been used for genetic mapping of complex traits [36, 37, 39, 80], but was not previously used for genetic studies of hearing loss; therefore, we are taking advantage of the genetic variation present in this strain. The outbred nature of the population allows for precise mapping, as reflected by the LD decay curve (Fig. 2E), meaning that identified loci often contain a small number of genes. The LD decay curve reported in this work is not directly comparable with the LD decay curve published for the CFW mice population previously [36] for multiple reasons, including the library preparation method (low-coverage sequencing versus reduced representation digest-based sequencing), number of SNPs (4.18 M SNPs versus 92,734), and genotyping error rate (0.45% versus 1.55%). Nevertheless, both analyses show rapid LD decay, supporting the utility of the CFW population for genetic mapping studies.

The SNP heritability of the traits ranged from 0 to 0.42 (Table 2). Unlike heritability estimates from inbred strain panels or twin designs, SNP heritability is expected to be lower, since it measures the proportion of phenotypic variance explained by all measured SNPs [81]. The heritability of hearing thresholds reported in human studies varies from 0.2 to 0.54 in twin studies [82], and from 0.13 to 0.7 for SNP-based heritability [83]. Similarly, the heritability of age-related hearing impairment tends to be higher in twin study designs (0 to 0.65) [84, 85] than in association studies (0.03 to 0.22) [3, 19, 26]. In animal models, using a panel of BXD mouse strains, heritability was estimated to range from 0.21 to 0.70, depending on the exact phenotype [86, 87].

We examined the genetic correlations between all possible pairs of traits (Fig. 3). Within each age, the correlations between frequencies tended to be higher, forming a characteristic “triangle.” The correlation between ABR thresholds measured at 1 month, and ABR threshold or deafness at 6 and 10 months was lower, suggesting that the genetic underpinnings of hearing loss in older mice are less similar to those observed in younger mice.

We discovered ten loci associated with seven ARHL traits. Due to the small size of most of the implicated regions (Fig. 2B), the QTLs contained from 0 to 35 genes (seven genes on average; Table 3).

The most interesting genes are those with known coding variation, those that are expressed in cochlea tissue based on several datasets publicly available in gEAR portal [55, 60, 64–67], and those that are supported by previous publications. We confirmed the role of Prkag2 in hearing loss by demonstrating the onset of high-frequency hearing loss in 20-month-old Prkag2 constitutive null mice, compared to the wild type littermates, accompanied by the loss of both inner hair cells and outer hair cells. Similar Prkag2-deficient mice (Prkag2^em1(IMPC)J, on a C57BL/6 background, constructed by JAX Laboratories) were tested for hearing loss at 14 weeks. These mice, similarly to C57BL/6-Prkag2^{tm1.2 mrl}, carry a deletion of exon 3, which causes a frameshift and loss of function. ABR testing showed normal hearing thresholds at this age (www.mousephenotype.org/data/genes/MGI:1336153#data). This confirms that Prkag2 deficiency causes age-related hearing loss, but does not cause hearing impairment at a young age.

In the future, we hope to examine eQTLs in the cochlea of CFW mice, which would offer an additional line of evidence not available in the current analysis.

Many inbred and outbred mice experience very early hearing loss, resulting in severe hearing loss as early as 9 weeks [88]. In contrast, few CFW mice showed deafness by 4–8 weeks of age, the first time point when hearing has been measured (Fig. 1B). However, by the 6 month time point, 30–43% of mice were categorized as being deaf at different frequencies (Fig. 1B), demonstrating the presence of alleles that cause relatively early onset of deafness. Previous mouse studies identified multiple loci related to hearing and hearing loss. The current study replicates some of the genetic loci found in previous studies. The lack of replication for other loci may simply reflect the fact that CFW mice do not segregate the same variants as other populations or could be due to insufficient sample size or type I and type II errors in the current or prior studies.

Previous studies of the genetics of hearing loss in mice were focused mostly on early-onset hearing loss. Many mouse strains are homozygous for an ahl locus, corresponding to the Cdh23^{753 A} variant (SNP rs257098870) which has been reported to cause early hearing loss [88, 89]. The position of this variant corresponds to the last nucleotide of exon 7 in GenBank sequence AF308939 and to the last nucleotide of exon 9 in Ref Seq NM_023370 and NM_001252635. This SNP is predicted to alter splicing: a G at this position results in normal exon splicing, whereas an A disrupts the donor splice site sequence and causes in-frame exon skipping [90]. The Cdh23^753G allele is associated with resistance to ARHL and is dominant to the recessive Cdh23^{753 A} allele, which is associated with ARHL susceptibility. CFW mice carry the Cdh23^{753 A} variant at ~ 0.63 allele frequency (estimated by using a subset of 86 mice that had at least three sequence reads spanning rs257098870, enabling the genotype for rs257098870 to be approximately called by bcftools without imputation). Using these data, we confirmed that homozygosity for Cdh23^{753 A} significantly increases susceptibility to age-related hearing loss in our cohort (Supplemental Fig. 2). However, our imputation-based genotyping strategy was not able to reliably genotype the region around rs257098870. In CFW mice, this region appears to have structural variation that does not align with the reference genome (Supplemental Fig. 3). However, this SNP is clearly not the only cause of age-related hearing loss, and our other results remain valid, despite our inability to accurately genotype rs257098870.

Our genetic analysis in CFW mice did not find several other QTLs that have been previously reported in other mouse strains. The regions corresponding to ahl3 have not been narrowed down to a specific gene or a variant. ahl3 (chromosome 17, 67.2 Mb) was discovered in consomic C57BL/6 J and MSM mice [91]. There is no variability in this region in the eight CFW mice that were sequenced with 30 × coverage, suggesting that this region is not variable in the CFW population, and therefore does not contain the variant detected by [91].

The regions corresponding to ahl6 also have not been narrowed down to a specific gene or a variant. ahl6 (chromosome 18, 44.2 cM) was discovered in outbred Black Swiss mice [92]. This region on chromosome 18 does contain some genetic variability in the eight CFW mice that were sequenced with 30 × coverage, and the variants were successfully called by the imputation-based method in the 946 test subjects. The fact that we did not detect a QTL at this locus suggests that CFW mice do not carry the causal variant that is responsible for ahl6. A similar result was obtained for the chromosomal regions corresponding to other previously reported loci that were both polymorphic and successfully genotyped in CFW mice. We can confidently report that we did not detect hearing loss in CFW mice that is associated with the following loci: ahl2 (chromosome 5, 79.6 Mb) discovered in the C57BL/6 J × NOD/LtJ cross [93, 94], ahl5 (chromosome 10, 81.1 Mb), corresponding to gene Gipc3 and discovered in outbred Black Swiss mice [92], ahl8 (chromosome 11, 120 Mb), corresponding to gene Fscn2 and discovered in BXD mice [95], M5 Ahl8 (chromosome 5, approximately 78–118 Mb) discovered in BXD mice [96], and ahp (chromosome 16) discovered in BXD mice [87]. The most parsimonious explanation for our failure to detect these QTLs found in other populations is that the causal alleles are not polymorphic in CFW mice. It is also possible that the original findings were false positives or that our results represent a false negative, for example, because of insufficient sample size.

This study is not without limitations. The number of discoveries in a GWAS is dictated by sample size. In this study, the largest sample sizes were obtained at the 6-month time point. We had originally planned to use a 14-month time point, but found that a significant fraction of mice did not live that long. Thus, the 10-month time point was introduced mid-way through the study, and so, fewer mice were phenotyped at the 10-month time point. In addition, because a significant number of mice were deaf by the 6- and 10-month time points, they could not be used for the analysis of ABR threshold. Another limitation of this study is that we did not explore the onset of deafness, which occurred in 30–43% of mice (varies for different frequencies) sometime between the 1- and 6-month time points. Future studies that examine this process could yield additional insights. A limitation that is common for this type of study is a seeming disconnect between heritability estimates and GWAS results, when traits with low heritability produce significant QTLs. For example, the trait “deaf at 06 mo 08 kHz” has h² = 0.054 ± 0.052, p-value 0.157 (see Table 2), and produces QTL on chromosome 12 with the − log₁₀(p) for the top SNP at chr12:117,951,339 equal to 5.99. This seeming discrepancy arises because the two analyses (SNP heritability estimate and GWAS) address different aspects of genetic architecture. The SNP heritability is estimated as the average contribution of all measured SNPs to the trait’s variance. If most SNPs have very small or no effects, the overall estimate might be low (or statistically non-significant) even if one or a few loci have a relatively large impact. GWAS tests each SNP individually and can identify specific QTLs with significant effects. Another limitation is that our analysis of genes within implicated regions accounted for coding polymorphisms, but did not examine eQTLs because no eQTL data for the cochlea of CFW mice are available. We plan to develop such data in the future. Finally, the limitation of using a mouse model for Prkag2 ablation was derived in 2008 by excising exon 3 of the isoform NM_145401.2. The current genome build annotates at least three protein-coding transcripts: NM_145401.2, NM_001310480.1, and NM_001170555.1. These transcripts encode proteins with different N-termini and contain different exons 1–3. The isoform NM_145401.2 encodes a protein that contain N-terminal nuclear localization signal. Our data highlight the importance of nuclear localization of Prkag2 for development of ARHL.

In conclusion, we performed a GWAS for ARHL traits using 946-CFW outbred mice. We identified ten QTLs that offer new insights into the genetic underpinning of ARHL, identifying novel candidate genes, including Dnah11, Rapgef5, Cpne4, Prkag2, and Nek11. Using a constitutive knockout mouse model, we confirmed that Prkag2 plays a role in age-related hearing loss. Other candidate genes identified in this and future studies can be manipulated to explore their role in hearing loss. Another important future direction will be to explore the expression of the candidate genes in a spatial manner to better identify the cells and structures that are affected.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contribution

All authors contributed to the study conception and design. The study was conceptualized by Abraham A. Palmer and Rick Friedman. Material preparation and data collection were performed by Ely Boussaty, Olivia Lamonte, Thomas Zhou, Eric Du, Khai-Minh Nguyen, and Mika Okamoto. Analyses were performed by Oksana Polesskaya, Thiago Missfeldt Sanches, and Riyan Cheng. The first draft of the manuscript was written by Oksana Polesskaya, and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding

This work was funded by the National Institutes of Health grant R01DC018566 to Rick Friedman.

National Institutes of Health,R01 DC018566,Rick A. Friedman

Data Availability

1. Genotype data for 946 CFW (stock Crl:CFW(SW)-US_P08) is available from UC Digital Collection: https://library.ucsd.edu/dc/object/bb5507359c

2. Sequencing data for low-coverage whole genome for CFW mice: ttps://www.ncbi.nlm.nih.gov/bioproject/PRJNA1230210

3. Sequencing data for 8 deeply sequenced CFW mice: https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1229615

4. The full Genetic Analysis Report:

https://palmerlab.s3.sdsc.edu/tsanches_dash_genotypes/gwas_results/r01_friedman_arhl_2020/results/gwas_report.html

Code Availability

Not applicable.

Declarations

Conflict of interest

The authors declare no competing interests.