https://orcid.org/0000-0002-1846-8600
Figures
Abstract
Adult-onset progressive hearing loss is a common, complex disease with a strong genetic component. Although to date over 150 genes have been identified as contributing to human hearing loss, many more remain to be discovered, as does most of the underlying genetic diversity. Many different variants have been found to underlie adult-onset hearing loss, but they tend to be rare variants with a high impact upon the gene product. It is likely that combinations of more common, lower impact variants also play a role in the prevalence of the disease. Here we present our exome study of hearing loss in a cohort of 532 older adult volunteers with extensive phenotypic data, including 99 older adults with normal hearing, an important control set. Firstly, we carried out an outlier analysis to identify genes with a high variant load in older adults with hearing loss compared to those with normal hearing. Secondly, we used audiometric threshold data to identify individual variants which appear to contribute to different threshold values. We followed up these analyses in a second cohort. Using these approaches, we identified genes and variants linked to better hearing as well as those linked to worse hearing. These analyses identified some known deafness genes, demonstrating proof of principle of our approach. However, most of the candidate genes are novel associations with hearing loss. While the results support the suggestion that genes responsible for severe deafness may also be involved in milder hearing loss, they also suggest that there are many more genes involved in hearing which remain to be identified. Our candidate gene lists may provide useful starting points for improved diagnosis and drug development.
Author summary
Age-related hearing loss is a common disease which is highly heritable, and more than 150 genes are known to underlie human hearing impairment. However, most of these genes are linked to early-onset, severe forms of deafness, and involve variants which are very rare. Investigating age-related hearing loss is challenging because it is so common; there are many genes which may be involved, with multiple contributing variants which may be rare or common. Here, we have obtained audiometric and genomic data from a cohort of older volunteers (aged 55 and over), which includes a vital control group of 99 older people with good hearing. We have taken two approaches to detecting genes and variants implicated in age-related hearing loss, and followed up this work with a second, smaller, cohort. Our study provides multiple candidate genes for further investigation.
Citation: Lewis MA, Schulte J, Matthews L, Vaden KI Jr, Steves CJ, Williams FMK, et al. (2023) Accurate phenotypic classification and exome sequencing allow identification of novel genes and variants associated with adult-onset hearing loss. PLoS Genet 19(11): e1011058. https://doi.org/10.1371/journal.pgen.1011058
Editor: Thomas B. Friedman, National Institute on Deafness and Other Communication Disorders, National Institutes of Health, UNITED STATES
Received: May 17, 2023; Accepted: November 7, 2023; Published: November 27, 2023
Copyright: © 2023 Lewis et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The numerical data underlying the graphs, statistics and figures in this study are provided in the paper or as Supporting Information, where possible. Individual-level human data cannot be shared publicly because they contain potentially identifying or sensitive information. Access to the TwinsUK data used in this study may be requested through the TwinsUK registry at https://twinsuk.ac.uk/. Access to the classifications and whole exome sequencing data from the MUSC cohort may be requested via dbGaP (http://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs003327.v1.p1). Audiometry data from the MUSC cohort are available to researchers upon request and completion of institutional data use agreements, as required by the Medical University of South Carolina; please contact the Department of Otolaryngology (https://medicine.musc.edu/departments/otolaryngology). The mouse single cell RNAseq data are publicly available at https://umgear.org/ (https://identifiers.org/geo:GSE137299, https://identifiers.org/geo:GSE114157, https://identifiers.org/geo:GSE181454, https://identifiers.org/geo:GSE136196).
Funding: This work was primarily funded by the National Institutes of Health/National Institute on Deafness and Other Communication Disorders (P50 DC 000422 to JRD) awarded to the Medical University of South Carolina and by the South Carolina Clinical and Translational Research (SCTR) Institute, with an academic home at the Medical University of South Carolina, NIH/NCATS Grant number UL1 TR001450. This investigation was conducted in a facility constructed with support from Research Facilities Improvement Program Grant Number C06 RR14516 from the NIH/NCRR. TwinsUK is funded by the Wellcome Trust, Medical Research Council, Versus Arthritis, European Union Horizon 2020, Chronic Disease Research Foundation (CDRF), Zoe Ltd, the National Institute for Health and Care Research (NIHR) Clinical Research Network (CRN) and Biomedical Research Centre based at Guy’s and St Thomas’ NHS Foundation Trust in partnership with King’s College London. The immunolabelling study was supported by RNID and In2Research UK. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Hearing loss is a common, complex condition with a strong genetic component. More than 700 genes have been found to underlie Mendelian hearing loss in humans and/or mice (reviewed in [1]), but large-scale mouse studies suggest there may be as many as 1000 genes which alone can result in hearing impairment when mutated [2]. Identifying the genes and specific gene variants involved in age-related hearing loss may suggest genes or pathways that can be targeted therapeutically, as well as being useful for diagnosis.
Identifying genes and variants involved in hearing loss is challenging owing to the heterogeneity of the disease. The inner ear is a complex system with multiple molecular components that need to function and interact correctly to enable normal hearing. Family studies have led to the identification of many variants involved in adult-onset Mendelian hearing loss (for example, MIR96 [3], DMXL2 [4], reviewed in [5]), but these tend to be very rare or even private variants, and are unlikely to explain all of the hearing loss seen in humans. Some loci have been identified through genome-wide association studies (GWAS) [6–8], but very large numbers of people are needed and GWAS chips are limited by their use of common, ancient variants. Whole exome and genome sequencing offer greater scope for identifying causative variants whatever their allele frequency and, indeed, recent studies using exome sequencing [1,9] suggest that intermediate frequency variants also play a role in hearing difficulty.
Another challenge in this field is the complexity of auditory phenotypes. Hearing loss does not have a single pathogenic mechanism, but can result from multiple inner ear pathologies. At present, with few exceptions, accurate diagnosis of the underlying hearing problem is not possible. In addition, many large-scale studies make use of self-reported questionnaires to explore hearing impairment. Although self-reported hearing difficulty is fairly well correlated with overall audiometric thresholds [10–12], and hearing aid prescription is a surrogate for abnormal pure tone audiometry at least in the UK, these may be prone to subjective bias and offer no way to distinguish between different auditory phenotypes and underlying pathologies. Large cohorts with good audiometric phenotyping offer more objective classification of participants, which may allow more sensitive detection of causal genes and variants. This has been demonstrated by a recent study on such a cohort, which found an increased burden of predicted deleterious rare variants in known hearing loss genes in people with sensorineural hearing loss compared to controls with good hearing (assessed by pure-tone threshold) or no medical reports of hearing loss [13]. Furthermore, appropriate quality control is vital for genetic studies, and because adult-onset hearing loss is so common, a well-characterised age-matched group with audiometrically determined normal hearing provides a better control than volunteers reporting no hearing difficulty or younger adults with normal hearing.
Here we present our data from a cohort of older adult volunteers having extensive phenotype data, including 99 older adults with good hearing. The cohort was characterised by estimates of Metabolic and Sensory components of age-related hearing loss, calculated from audiograms [14], then participants were grouped according to their estimates. The Metabolic and Sensory components were based on the premise that underlying cochlear pathologies will have differential effects across audiometric frequencies or at higher frequencies, respectively. More specifically, Metabolic (i.e. strial) losses involve reduced cochlear lateral wall function and Sensory losses involve primary lesions within the organ of Corti or neural components of the cochlea. Different genes may contribute to each primary site of lesion. We have carried out both gene-based and variant-based tests to identify candidate genes and variants in each phenotypic group, and prioritised those candidates using a variety of methods, including repeat analyses in a second, smaller cohort.
Results
MUSC cohort classification
Our primary cohort consisted of 532 participants; 292 female and 240 male participants, with an overall average age of 72.25 years (71.96 years for women, 72.60 for men). 62 women and 187 men reported a positive noise history (Table 1, Fig 1). The difference in average thresholds between the sexes is greater than that between the groups reporting a positive or negative noise history (Fig 1C). 99 participants were classified into the Older-Normal audiogram category; 87 women and 12 men. In the Metabolic category, there were 92 women and 62 men, while in the Sensory category, there were 53 women and 101 men. PLINK v2 [15] was used to check for relatedness using common variants in linkage disequilibrium; none of the cohort were related. We used the 2504 individuals from the 1000 Genomes study [16] to plot out the ancestry of this cohort and found it to be largely non-Finnish European, most similar to the “British in England and Scotland” and “Utah residents with Northern and Western European ancestry” sub-populations (S1 Fig).
A, B: Bar charts showing the numbers of participants in each classification (not including Unclassified or Unselected cases) in the MUSC and TwinsUK cohorts (with twins removed from the latter). Black sections represent those participants reporting a positive noise history. A shows the numbers, and B shows the percentages reporting a positive noise history within each category. C shows the average audiograms of the MUSC cohort, plotted in groups separated by sex and noise history. The three people with unknown noise history have not been included. The thresholds from the left ear are shown on the left-hand plot, and those from the right ear on the right-hand plot. Numbers and average ages of each group are listed in the key. Error bars are standard deviation.
TwinsUK cohort classification
There were 159 female participants from the TwinsUK cohort meeting our requirements, including ten dizygotic twin pairs. There were no monozygotic twin pairs, and no other relatedness was reported. The overall mean age was 64.82 years. There were few positive responses to questions about noise exposure in work or leisure activities, so no participants were classified as having a positive noise history. We carried out the same ancestry analysis on these 159 participants, and found they also had a non-Finnish European ancestry, and like the MUSC cohort, they were most similar to the “British in England and Scotland” and “Utah residents with Northern and Western European ancestry” sub-populations from the 1000 Genomes (S1 Fig).
One twin of each pair was removed from the cohort; where twins were classified into the same category, the removed twin was chosen at random (6 pairs). For 3 pairs, one twin was classified as Older-Normal or Metabolic, with the other either Unclassified or Unselected; in those cases, the Unclassified or Unselected twin was removed. The last pair consisted of one twin classified as Older-Normal and one classified as Metabolic; both were removed for the outlier analysis but for the threshold analysis, one (the Older-Normal-classified twin) was chosen at random for removal. After twin removal, there were 49 in the Older-Normal category (average age 61.37 years), 62 in the Metabolic category (average age 66.26 years) and 7 in the Sensory category (average age 68.14 years) (Table 1, Fig 1).
Selection and validation of high quality, high impact variants
After quality filtering (S1 Table), 938,008 (MUSC) and 279,434 (TwinsUK) high quality variants were obtained from the nuclear exome, and 1174 (MUSC) and 142 (TwinsUK) mitochondrial variants, most of which were homoplasmic (with a variant allele fraction > 0.95) (S2 Fig). Variants were then filtered for high predicted impact and MAF < 0.1, based on our previous work (S1 Table, [1]), resulting in 29,807 (MUSC) and 21,432 (Twins UK) high quality, high impact variants from the nuclear exome, and 226 (MUSC) and 16 (TwinsUK) high impact mitochondrial variants (S2 Fig). For the two analyses carried out, homoplasmic mitochondrial variants were treated as homozygote calls and heteroplasmic mitochondrial variants as heterozygote calls.
In order to validate the exome sequencing, Sanger sequencing was carried out on 114 variants in multiple samples from the MUSC cohort, and individual call accuracy was 94.7% (360 correct from 380 calls in total). Only two variants were not validated; the remainder of the incorrect calls were errors in zygosity (eg a heterozygote call for an individual homozygous for the alternate allele) (Table 2).
Outlier analysis
To investigate variant load in hearing loss, and in and between the specific phenotypes, the number of variants per gene in participants belonging to one group (eg Older-Normal) were compared to the number of variants in the same gene in participants belonging to another group (eg Metabolic). The participants were also compared segregated by sex, because the genetic contribution to adult-onset hearing loss differs by sex [1]; however, because there were only 12 men classified as having Older-Normal hearing, comparisons which required that group were not carried out, resulting in 9 comparisons from the MUSC cohort (Fig 2A–2I; Table 3). Two lists of outlier genes were obtained from each comparison; one with an exceptionally high variant load in the first group and one with an exceptionally high variant load in the second group (S2 Table; raw variant counts are in S3 Table). The definition of a high variant load in each gene is not based on overall variant count, but on what is expected based on the other comparison group, so a large gene with many variants (such as TTN) will not necessarily be an outlier (eg Fig 2C). At the other extreme, a gene which does not have many variants overall may still be an outlier if it has more than expected in one group, and so there are outlier genes in the bottom left quadrant of all the plots in Fig 2 (marked by colour around the plotted point).
Each point represents a gene. Outliers (S2 and S4 Tables) are marked in orange (for higher load in participants not classified as Older-Normal), blue (for higher load in participants classified as having Older-Normal hearing), purple (for higher load in participants classified as having Sensory hearing loss) or green (for higher load in participants classified as having Metabolic hearing loss). A-I show comparisons in the MUSC cohort; A,B,C,G show all participants, D,E,F,I show female participants and H shows male participants. A,D show a comparison of variant load in people in the Older-Normal category to all others in the cohort, B,E show a comparison of variant load in people in the Older-Normal category to people in the Metabolic category, C,F show a comparison of variant load in people in the Older-Normal category to people in the Sensory category, and G,H,I show a comparison of variant load in people in the Metabolic category to people in the Sensory category. J, K show comparison of variant load in the TwinsUK cohort (which is all female); J shows a comparison of variant load in people in the Older-Normal category to all others in the cohort and K shows a comparison of variant load in people in the Older-Normal category to people in the Metabolic category. Genes with a lot of variants in, at the top right of each plot, are labelled; in some cases these are highly variable genes (TTN, CCHCR1, NBPF3, TAS2R43, shown in grey). Having more variants overall does not necessarily make them of interest. What matters is whether a gene has many more or many fewer variants than expected, and that can be just as true of a gene with comparatively few variants overall (at the bottom left of the plot) as of one with many, such as TTN, which is at the top right of most plots.
Several genes with lots of variants are labelled in the panels of Fig 2, some of which are outliers with exceptionally high variant loads in one of the comparison groups. For example, DNAH3 is an outlier in metabolic hearing loss in the MUSC cohort (panels B, E, G, I) and was also previously reported as associated with hearing loss [17], but in contrast it is an outlier in normal hearing in the Twins UK cohort (panels J, K). MT-CYB is an outlier in the sensory hearing loss groups in the MUSC cohort (panels A, B, C, D, F, G, H, I) but was an outlier in the normal hearing group of the TwinsUK cohort (S4 Table), so the association was not replicated.
To investigate these gene lists, findings were compared to a list of 774 genes which are known to underlie hearing loss in humans and/or mice (S6 Table; this includes 544 human orthologues of the 519 deafness genes known only from mouse studies (Fig 3)). These are good candidates for adult-onset hearing loss, and we suggest that enrichment in these genes supports the relevance to hearing loss. Only two lists showed a significant enrichment for hearing genes; the list of genes with high variant load in Metabolic hearing loss (male and female participants together, comparing Older-Normal to Metabolic hearing loss), and the list of genes with high variant load in Metabolic hearing loss (male participants, comparing Metabolic hearing loss to Sensory hearing loss) (Table 3). The gene lists were also tested for enrichment in 1213 highly variable genes, which are genes frequently reported to carry variants in multiple exome sequencing projects [1]. A significant enrichment of highly variable genes was found in multiple gene lists (Table 3), suggesting that some of the genes included are present for reasons unrelated to hearing. The outlier lists were combined to obtain a final candidate list of 291 genes, 21 of which were known deafness genes and 37 of which were highly variable genes (S2 Table). 107 of these genes were also identified in our previous study of self-reported hearing difficulty in the UK BioBank cohort [1], 12 of which were known deafness genes (PKHD1L1, ELMO3, CDH23, UBE3B, ADGRV1, COL9A3, NAV2, DMD, AFAP1L2, MPDZ, LOXHD1, and CELSR1).
Brackets indicate orthologues (e.g. there are 544 human orthologues of the 519 mouse deafness genes).
To prioritise the list of candidate genes, a third list of outlier genes was obtained from the TwinsUK cohort. In this case there were not sufficient participants classified as having Sensory hearing loss, and so only two comparisons were carried out (Fig 2J and 2K; Table 3), resulting in a final candidate list of 435 genes, including 25 known deafness genes and 43 highly variable genes (S4 Table, raw variant counts in S5 Table).
Thirty-eight genes were common to all three analyses (Fig 4, Table 4), one of which was a known deafness gene (PKHD1L1) (Fig 4). For further prioritisation, we investigated the expression of mouse orthologues in the inner ear using publicly available single cell RNAseq data from the gEAR expression resource [18]. Thirty-two of the 38 genes had good quality mouse orthologues, and of these, eleven genes had no expression reported in the chosen ages and cell types, and a further eleven genes were expressed at low levels (up to and including the expression level of Hprt1, to which all expression was normalised). The remaining 10 genes were strongly expressed in at least one cell type and age (S3 Fig). Based on this analysis, among the most interesting novel candidate genes were FKBP2 and SYNE2, which have strong expression in multiple cochlear and lateral wall cell types, and ABCB8, which shows similar expression to the hair cell marker Myo7a (S3 Fig).
Genes are listed in S2 and S4 Tables, and the UK Biobank gene list is from our previous study [1].
In order to investigate genes associated with specific phenotypes, we also plotted the expression of genes identified only in the phenotype-specific analyses. There were 18 genes linked only to Metabolic hearing loss (including four deafness genes: DMD, DUOX2, CELSR1 and ELMO3) and 54 genes linked only to Sensory hearing loss (including four deafness genes: ARHGAP21, LMO7, UBE3B and ADGRV1) (S2 and S4 Tables). After removing genes without a good quality mouse orthologue and with low or no expression in the chosen inner ear datasets, we plotted the expression of 12 Metabolic-linked genes and 17 Sensory-linked genes (S4 Fig). The four Metabolic-linked genes most strongly expressed in the lateral wall are MT-CO1, TLN2, DPP4 and CHMP4C, and are also expressed in several organ of Corti cell types (S4 Fig). The Sensory-linked genes most strongly expressed in the organ of Corti are MADD, UBE3B and LMO7, but they have low or no expression in the lateral wall (S4 Fig). We tested the protein expression of Madd in the inner ear of wildtype mice at ages from embryonic day (E)14.5 to postnatal day (P)4, and found that there was no protein detected at E14.5, but expression in inner and outer hair cells was visible at E16.5 and increased in intensity up to P4 (S5 Fig), which correlates with the scRNAseq data from the gEAR (S4 Fig).
Threshold difference detection
We compared the thresholds of carriers of each individual variant to those of non-carriers in order to assess the contribution of each variant to threshold differences. Forty of the 29,807 high impact variants in the nuclear exome passed the filters and permutation testing. In two cases (KIRREL1 and CCDC171), both the non-segregated alternate allele group and one of the sex-segregated groups exhibited a significant difference in thresholds. In the remaining 38 cases, only one group exhibited a significant threshold difference. One mitochondrial variant (rs41518645, in MT-CYB) also was found to pass the filter and permutation tests, and was associated with better thresholds in male participants (Fig 5). There were no instances of multiple variants being identified in the same gene, and only two were in known deafness genes, S1PR2 and PIEZO1 (Table 5, Figs 5 and S6). Sixteen of the 41 variants were associated with better thresholds than the sex-matched reference group (eg TCEANC2, Fig 5), and 25 with worse thresholds than the reference group (eg CLDN3, Fig 5). Fifteen variants exhibited a significant difference in thresholds in only one sex (eg S1PR2, HADH, Fig 5), not including those instances where there were too few carriers of the opposite sex to determine if their thresholds were similarly affected, eg CAPN9 (Fig 5, Table 5; all audiograms are shown in S6 Fig).
Six different variants were chosen as examples from the full list of 41; see S6 Fig. Two audiograms are shown for each variant; the thresholds from the left ear are shown on the left, and those from the right ear on the right. Numbers and average ages of each group are listed on the graph. The symbols at the top of each graph mark which groups passed the criteria for each stimulus frequency compared to the relevant reference group (+ for male, = for female, and * for all participants). Carriers of the MT-CYB and TCEANC2 variants have better thresholds than non-carriers, and carriers of the MMS19, S1PR2, CLDN3 and CAPN9 variants have worse thresholds than non-carriers. The variants in S1PR2 and HADH are linked to worse thresholds only in male and female carriers respectively, and there are no female carriers of the CAPN9 variant so it is unknown whether they would be similarly affected to male carriers.
To further investigate the contribution of these 41 variants to threshold differences, carriers of each variant in the TwinsUK cohort were identified, and their audiograms plotted compared to homozygous carriers of the reference allele. Sixteen of the 41 variants were not found in any of the TwinsUK participants (Table 5), and for a further seven variants, the threshold difference was only seen in male MUSC carriers, not in female participants, so a comparison was not possible with the all-female TwinsUK cohort (Table 5). However, five carriers from the TwinsUK cohort were found to have similar audiograms to those in the MUSC cohort for the variant in HADH, and this was also found independently when the same filter and permutation testing was carried out on the TwinsUK cohort. Carriers of variants in MED12L (n = 2), and ZDHHC6 (n = 3) also had a similar average threshold shape to that seen in the MUSC cohort carriers (Table 5, S6 Fig), supporting the suggestion of a potential role for these variants in contributing to the hearing loss seen in carriers. We examined the MUSC carriers of the variants in HADH, MED12L and ZDHHC6 to check for any variants in 50 known dominant deafness genes (https://hereditaryhearingloss.org, accessed March 2023 [21]) but did not find any dominant gene consistently affected within each group.
From the TwinsUK cohort alone, only four variants passed the filters and permutation testing, one of which was the variant in HADH, also identified in the MUSC cohort. The other three genes were AKR7A3, SCN7A and ME1 (S6 Fig). There were many carriers of each of these three variants in the MUSC cohort, but for ME1 and AKR7A3, the average thresholds of carriers did not show any obvious difference to non-carriers, suggesting that if these variants do contribute to hearing loss, the impact is not reflected in audiogram shape (S6 Fig). MUSC carriers of the variant in SCN7A (n = 22; 9 female, 13 male) had, on average, slightly worse thresholds than homozygous carriers of the reference allele, resembling the thresholds of carriers in the TwinsUK cohort, but the difference was not significant (S6 Fig).
Discussion
From the outlier analysis, we identified 38 candidate genes that may contribute to overall hearing status (including one deafness gene, PKHD1L1) (Table 4), 18 genes linked to Metabolic hearing loss alone (including four deafness genes, DMD, DUOX2, CELSR1 and ELMO3) (S3 and S5 Tables), and 54 genes linked to Sensory hearing loss alone (including four deafness genes, ARHGAP21, LMO7, UBE3B and ADGRV1) (S3 Table). The threshold analysis revealed 41 candidate genes including two known deafness genes (S1PR2 and PIEZO1) (Table 5).
Known deafness genes from the candidate gene lists
Our candidate gene lists include 11 deafness genes; S1PR2, PIEZO1, PKHD1L1, DMD, DUOX2, CELSR1, ELMO3, ARHGAP21, LMO7, UBE3B and ADGRV1. Only 3 of these have been identified in humans; ADGRV1, which is an Usher syndrome type II gene [22], DMD, which has been associated with congenital hearing impairment as well as muscular dystrophy [23], and S1PR2, variants in which lead to congenital profound hearing impairment [24], although a point variant in S1pr2 in mice results in early-onset progressive hearing loss [25]. These phenotypes are more severe than the late-onset progressive hearing loss in our human subject cohorts, which supports the theory that genes responsible for severe deafness may also be involved in milder forms of hearing loss.
PIEZO1, PKHD1L1, DUOX2, CELSR1, ELMO3, ARHGAP21, LMO7 and UBE3B are all defined as deafness genes through work on the mouse orthologues. Of these, hearing loss caused by mutant alleles of Arhgap21 and Elmo3 have only been reported by the IMPC large-scale phenotyping screen at a single age of 14 weeks (www.mousephenotype.org [26, 27]); Elmo3 homozygous mutants have raised thresholds at low frequencies (https://www.mousephenotype.org/data/genes/MGI:2679007) and Arhgap21 heterozygous mutants exhibit variably raised thresholds across most frequencies tested (https://www.mousephenotype.org/data/genes/MGI:1918685). Mice with a disrupted Ube3b gene display mild hearing impairment at all frequencies at 3 months old, and this impairment was more severe when tested at 6 months old [28]. Pkhd1l1 mutants show early-onset progressive hearing loss [29], while abolishing Lmo7 expression in mice results in late-onset progressive hearing impairment [30]. All these are comparatively mild effects, but mice carrying a missense variant in Duox2 have severely raised thresholds [31], and mice carrying variants in the planar cell polarity gene Celsr1 exhibit vestibular defects and misoriented outer hair cells [32]. Finally, mice carrying a mutation in Piezo1 have recently been reported to exhibit progressive hearing loss [33].
Expression analysis of novel candidate genes
Interestingly, most of our newly associated genes have not previously been reported with hearing loss, suggesting that there are many more genes involved in hearing which remain to be identified. Our expression analysis results suggest some promising genes for further investigation, such as SYNE2, FKBP2, and ABCB8 from the main analysis (Table 4), and MADD, COG4 and CHMP4C from the phenotype-specific analysis (S3 and S5 Tables). SYNE2 forms part of the LINC (Linker of Nucleoskeleton and Cytoskeleton) Complex, which is part of the nuclear envelope and is essential for maintenance of normal hearing [34]. FKBP2 encodes FKBP13, a luminal endoplasmic reticulum (ER) protein which is upregulated in response to cellular stress such as heat shock, or the accumulation of unfolded protein precursors in the ER [35,36], and ABCB8 is a mitochondrial ABC transporter which plays a role in cellular viability and is protective against oxidative stress [37]; variants in either may contribute to the vulnerability of inner ear cells to damage and age-related deterioration. MADD, which was associated with Sensory hearing loss and has an expression pattern resembling that of Myo7a (S4 Fig and S3 Table), is an activator of the Rab3 small GTP-binding protein family, and has been shown to be critical for neurotransmitter release in neuromuscular junctions and in hippocampal neurons [38,39]; it also may play a role in inner ear synapses but that has yet to be determined. COG4, which was also associated with Sensory hearing loss (S3 Table), is strongly expressed in hair cells and Deiters cells (S4 Fig), and is known to be important for zebrafish inner ear development [40]. CHMP4C, which was associated with Metabolic hearing loss in the TwinsUK cohort (S5 Table), is expressed in the marginal and basal cells of the stria vascularis, as well as several cell types in the organ of Corti (S4 Fig). Previous whole exome sequencing and genome-wide association studies have also linked CHMP4C to hearing impairment, suggesting it is a good candidate for further study [1,6–8]. However, it should be noted that gene expression in a particular cell type is not a guarantee of a critical role in that cell type, and the absence of expression in inner ear cells at the times and stages studied does not preclude a gene from having a role in hearing. It may be needed at a later time in development, or elsewhere in the auditory pathway, or may only be needed only at very low quantities, making it difficult to detect by single cell RNAseq. Also, given the limited data available from expression studies, a role for the other candidate genes and variants in age-related hearing loss should not be discounted.
Novel candidate genes from the threshold analyses
From our threshold analyses on both cohorts, we identified a variant in the gene HADH as a candidate associated with worse hearing, and variants in the genes ZDHHC6 and MED12L were associated with better hearing (S6 Fig, Table 5). HADH (hydroxyacyl-Coenzyme A dehydrogenase) localises to the mitochondrial matrix where it plays a role in the beta-oxidation pathway, breaking down fatty acid molecules to generate acetyl-coA. Variants in other genes in the same pathway have been shown to result in mitochondrial dysfunction [41], suggesting a potential mechanism for HADH variants to affect hearing. ZDHHC6 is a palmitoyltransferase located in the ER, and defects in palmitoylation have been linked to hearing impairment [42]. MED12L is a subunit of the Mediator protein complex which is part of the basal transcriptional apparatus; post-natal deletion of the Med12 subunit of the same complex in mice results in rapid loss of basal cell organisation and disruption of the stria vascularis leading to hearing loss [43].
The genetic contribution to hearing differences between sexes
The MUSC cohort has a slight excess of female participants over male, but the difference in classification of their hearing is marked, with too few male participants classified as “Older-Normal” to carry out a robust regression analysis on men alone using that category (Table 1, Fig 1). This difference has been previously described in multiple studies [11,44–49], with hearing in women tending to be better than in men and declining later in life, generally around the onset of menopause [11,50]. However, the average age of the participants in the MUSC cohort is over 60, suggesting that there is also a genetic contribution to the difference in hearing impairment between the sexes, as observed in our previous study [1].
The other clear difference in auditory phenotype between the sexes can be seen in the number of men classified as having Sensory hearing loss (101, 42% of male participants) compared to women (53, 18% of female participants) (Table 1). The proportions in the Metabolic hearing loss group are the inverse, although not so extreme (62 men, 26% of male participants, and 92 women, 32% of female participants). However, in the all-female TwinsUK cohort, there are only 7 participants classified as having Sensory hearing loss (5%, not including the twins who were removed; Table 1). Sensory hearing loss has been attributed to noise exposure, among other factors, and most of the men in the MUSC cohort had a positive noise exposure history (189, 79% of all male participants). However, the proportion of men reporting a history of noise exposure across the three classified groups was broadly similar (Older-Normal: 67%; Sensory: 79%; and Metabolic: 69%) (Table 1, Fig 1). The proportion of women in the MUSC cohort reporting a positive noise history in the different classifications was also very similar (Table 1, Fig 1). Self-reported noise history alone thus does not explain the excess of male participants classified as having Sensory hearing loss in the MUSC cohort. There may be a sex-specific genetic contribution to this observation, but more data are needed for further exploration. A more objective, quantifiable measure of noise exposure would also help in this, since noise history questionnaires can be an unreliable measure of an individual’s noise exposure.
The only regression analysis which could be performed using male participants alone from the MUSC cohort was the comparison of variant counts in men classified with Sensory hearing loss versus those classified with Metabolic hearing loss (Fig 2). Some of the genes with a high variant load in Sensory hearing loss are found in both the male and female lists, but none of the genes with a high variant load in Metabolic hearing loss are shared between the sexes (Fig 6). It is possible that there is a higher sex-specific genetic contribution to Metabolic hearing loss, but more data from larger cohorts are needed to explore this further. We also found multiple variants which appear to contribute to differences in thresholds in a sex-specific manner (Table 5), although the lack of audiograms and exome sequencing from male participants in the TwinsUK cohort means that we have not been able to follow up on those variants linked to threshold differences visible only in men from the MUSC cohort. Similarly well-characterised cohorts are necessary for further investigating the differing genetic contribution to hearing loss between the sexes.
The Venn diagram shows the overlap of genes identified as having a high variant load in Metabolic or Sensory hearing loss in all, male and female participants in the MUSC cohort. The shaded circles show the high variant load gene counts identified in all participants.
Candidate genes and variants associated with better hearing in older adults
Intriguingly, a subset of candidates from both outlier and threshold analyses are associated with better hearing, suggesting that some variants may lead to protection against age-related hearing loss and/or protection against damage from noise exposure. This is not the first report of protective variants. Examples of other protective variants include the N352S variant in B4GALT1 which is protective against cardiovascular disease [51] and protein-truncating variants in GPR75 which reduce the risk of obesity [52], as well as the A88V variant in Gjb6, which protects against hearing loss in mice [53]. Identification of genes and variants which protect against hearing loss could be a useful starting point for developing therapeutic treatments to do the same.
Methods
Ethics statement
All human subjects research was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained in this study, which was approved by the Medical University of South Carolina (MUSC) Institutional Review Board (for the MUSC cohort) and Guys & St Thomas’ Trust (GSTT) Ethics Committee (for the TwinsUK cohort).
All mouse experiments were carried out in accordance with UK Home Office regulations and the UK Animals (Scientific Procedures) Act of 1986 (ASPA) under UK Home Office licences, and the study was approved by the Wellcome Sanger Institute Ethical Review Committee.
Participants and audiometric measurements
The primary cohort consisted of 532 volunteers enrolled in an ongoing longitudinal study of age-related hearing loss at MUSC, dating from 1987, described in detail in Dubno et al, 2013 [54]. Notably, no participants exhibited any sign of conductive hearing loss or active otologic disease. The 532 individuals were aged 55 years or older. Pure tone thresholds (at 0.25, 0.5, 1.0, 2.0, 3.0, 4.0, 6.0 and 8.0 kHz) were obtained for each ear of each person, along with questionnaire responses concerning noise exposure history.
For the follow-up cohort, we selected 159 participants from the TwinsUK study based on age (55 years and older), self-reported ethnicity (“White”), and availability of both exome and pure-tone audiometry data. The pure-tone audiometry data collection has previously been described [55]; briefly, all participants underwent an otologic examination followed by an air-conduction pure-tone audiogram for each ear (0.125, 0.25, 0.5, 1.0, 2.0, 4.0, 6.0 and 8.0 kHz). Participants also answered a detailed questionnaire concerning medical history and environmental exposure to factors relevant to hearing.
Classification of audiograms
Phenotype cohorts were formed based on selection criteria to define individuals with representative metabolic or sensory hearing losses, as well as normal hearing, to enable comparisons between specific phenotypes. Audiograms were classified into one of three main categories (Older-Normal, Metabolic, and Sensory, Fig 7) based on the estimated metabolic and sensory components of the observed hearing loss [14]. The typical audiogram in metabolic cases shows mildly elevated thresholds at low frequencies sloping gently downwards towards higher frequencies, while the shape of a typical sensory pattern has normal thresholds at low frequencies and steeply downwards-sloping thresholds at high frequencies [54,56,57]. These typical profiles (obtained from 402 older adult audiograms [14]) can be used to approximate the metabolic and sensory components of the hearing loss observed in an individual ear. It is then possible to calculate the contribution of each profile to this approximation, and the quality of the approximation itself is represented by the line-fit error.
Averages of thresholds for Older-Normal (black), Metabolic hearing loss (blue) and Sensory hearing loss (red) cases in the MUSC cohort (left) and the TwinsUK cohort (right). Error bars are standard error of the mean.
To classify these cohorts, first, metabolic and sensory estimates and line-fit error were calculated for each of the right/left pairs of audiograms. Second, poorly fit audiograms were excluded from classification using the criterion of line-fit error ≥15 dB, which identifies audiograms with configurations inconsistent with age-related hearing loss (e.g., corner audiograms). These rejected audiograms are referred to as “Unselected” below. Third, a set of simple rules (below) using the metabolic and sensory estimates were applied to classify cases into the four remaining categories.
Using this approach, the Older-Normal category was defined by cases with summed metabolic + sensory estimates that were <20 dB HL, with <10 dB difference in the estimates between ears. The Metabolic category was selected from the remaining cases (i.e., not Older-Normal) with metabolic estimates ≥20 dB, ear asymmetries in the metabolic estimate ≤15 dB, sensory estimates <20 dB, and metabolic > sensory estimates. The Sensory category was selected from the remaining cases (i.e., not Older-Normal and not Metabolic) with sensory estimates ≥15 dB, ear asymmetries in the sensory estimate ≤20 dB, metabolic estimates <25 dB, and sensory > metabolic estimates. Finally, the remaining cases (i.e., not Older-Normal, not Metabolic, not Sensory) were less clearly representative of metabolic or sensory hearing loss, and were labelled Unclassified. After classifying the audiograms based on these rules, all audiograms in each category were reviewed by eye (KIV, JRD); a few anomalous cases were removed and a few cases were added based on consistency with a category. There were a total of 1–10 manual additions or removals for each category (S7 and S8 Figs).
Exome sequencing and alignment
Libraries for exome sequencing of the MUSC cohort were prepared using the Agilent SureSelect X2 Target Enrichment System (version 5) and the Agilent SureSelect Human All Exon V5 kit, which included 5’ and 3’ UTRs. DNA was sheared using the Covaris S220 focused ultrasonicator. Libraries were sequenced on the Illumina HiSeq 2500.
The exome sequencing of the Twins UK cohort has been previously described [58]. Briefly, DNA extracted from whole blood was hybridised to NimbleGen human exome arrays and sequenced using Illumina sequencing machines (NimbleGen 2.1M and the Illumina GAIIx for the first batch of sequencing, and NimbleGen EZ v2 and the HiSeq 2000 for the second).
For both cohorts, fastq files were aligned to GRCh38 using Hisat2.0 [59], following quality control steps (S7 Table). Bam files were realigned to sex-corrected genomes using XYalign [60].
Variant calling, filtering, annotation and confirmation
After read alignment, genomic variants were called using three callers; GATK HaplotypeCaller [61,62], BCFtools [63] and Freebayes [64] (S7 Table). Combining calls from multiple callers has been shown to offer more accurate variant calling [65]. HaplotypeCaller quality scores were recalibrated using the GATK Variant Quality Score Recalibrator (VQSR) tool [66], which annotates variants into tranches which represent subsequent levels of sensitivity versus specificity. Variants in the highest tranche are very high accuracy, very likely to be true but also incomplete (high specificity, lower sensitivity). The second tranche is less specific but more sensitive, and so on (https://gatk.broadinstitute.org/hc/en-us/sections/360007226651-Best-Practices-Workflows). BCFtools calls were filtered using vcfutils [63], and Freebayes calls were filtered using vcftools [67] (S7 Table).
After quality filtering each set of calls in the MUSC cohort, a selection of variants representing a range of quality scores was tested by Sanger sequencing to ascertain the best combination of filters. The resequenced variants were assessed based on whether the variant was correctly identified and also whether the sample genotypes were correctly called. We obtained 184 sequences from 66 variants, and the most accurate variant calls were those which had passed the GATK VQSR filters and had also passed at least one of the BCFtools or Freebayes filters. Those variants from the second GATK VQSR tranche which had passed the BCFtools filter were also found to be accurately called. The variant calls from all three callers were combined according to these requirements, and this combination filter was implemented for both the MUSC and the TwinsUK cohorts (S1 Table). Where there was a genotype disagreement (eg GATK and Freebayes called 0/1 and BCFtools called 1/1), the majority call was accepted. Calls like this, and calls with no disagreement, accounted for 99.8% (MUSC) and 85.5% (TwinsUK) of total calls. Where there were three different calls, one for homozygote alternate, one for homozygote reference and one for heterozygote, a heterozygous genotype was assigned (0.00025% of calls (MUSC); 0.00046% of calls (TwinsUK)). Other call combinations were considered missing (0.19% (MUSC); 14.5% (TwinsUK)). The reason for the TwinsUK sequencing having a higher call missing rate is due to the exome sequencing having been processed in two batches with different exome arrays [58], so some variants have only been called in half the participants.
941,165 (MUSC) and 281,261 (TwinsUK) variants passed these quality filters, and were then tested for excess heterozygosity using the R HardyWeinberg package [68,69] to identify and remove misaligned variants [70]. Also excluded were variants which had a high allele frequency in their cohort (defined as variants with cohort allele frequency above minor allele frequency (MAF)+0.4), which are likely to be aligner miscalls in low-complexity regions [71].
Mitochondrial variants were called using GATK Mutect2 [72] and filtered using GATK FilterMutectCalls. Although none of the kits used (Agilent SureSelect All Exon v5, NimbleGen 2.1M and NimbleGen EZv2) include the mitochondrial chromosome, off-target reads have been found to map correctly [73]. Griffin et al [74] tested this using three different exome kits (including the Agilent SureSelect Human All Exon 50Mb kit and the NimbleGen SeqCap EZ Exome Library v2.0) and conventional mitochondrial DNA sequence, and found that if the coverage was high enough (>30x), heteroplasmy over 40% could be reliably detected. The mitochondrial calls were therefore further filtered by read depth and variant allele fraction (S1 Table). For the two analyses carried out (described below), homoplasmic variants were treated as homozygote calls and heteroplasmic variants as heterozygote calls.
Genomic and mitochondrial variants were annotated using the Ensembl Variant Effect Predictor (VEP) v100 [75]. Annotation sources included 5’UTR variant prediction (Sutr, [76]), splice site variant prediction (SpliceAI, [77]), pathogenicity prediction (CADD, [78]) and minor allele frequency (gnomAD, TOPMED, ESP6500 and 1000Genomes [16,19,79,80]). The final filter, for high impact variants with MAF < 0.1, was based on our previous work (S2 Table).[1]
Chosen variants from the MUSC cohort were resequenced using Sanger sequencing (carried out by Eurofins Genomics LLC, Kentucky, USA). Primers for Sanger sequencing were designed using primer3 [81], and sequence traces were checked using Gap4 [82].
Regression analysis of number of variants per gene
Four comparisons were carried out: Older-Normal hearing to all hearing loss (including Unselected and Unclassified participants); Older-Normal hearing to Metabolic hearing loss; Older-Normal hearing to Sensory hearing loss; and Metabolic hearing loss to Sensory hearing loss. 12176 genes (including mitochondrial genes) had at least one variant called in one sample, and were assessed in each analysis. For each comparison, a linear regression was carried out on the total number of variants per gene per group. In the first three comparisons, the number of variants in the Older-Normal hearing group was used to predict the expected number of variants in the hearing loss group, and in the fourth comparison, the number of variants in the Metabolic hearing loss group was used to predict the expected number of variants in the Sensory hearing loss group. The residuals (the difference between the observed and predicted variant load for each gene) were used to determine the outlier genes. Briefly, the first (Q1) and third (Q3) quartile and the interquartile distance D (Q3-Q1) were calculated for each regression’s residuals, and outlier genes were defined as those with residuals > Q3 + 6D and those with residuals < Q1 – 6D [83]. Hypergeometric tests for enrichment were carried out using R.
Compilation of the list of known deafness genes.
The list of deafness genes consists of those genes known to underlie hearing impairment in humans or in mice, and was manually compiled and curated from the literature. It includes all the genes listed in the Hereditary Hearing Loss Homepage (hereditaryhearingloss.org/) and genes which, when mutated, result in altered hearing thresholds in mutant mice, as reported by the International Mouse Phenotyping Consortium (www.mousephenotype.org) (average thresholds were individually checked for shifts > 10dB and low variance between individuals). This list is an update of that reported in our previous study [1]; it consists of 519 genes linked to hearing impairment in mice, 102 genes linked to hearing impairment in humans, and 128 genes linked to hearing impairment in both mice and humans (Fig 3, S6 Table).
Expression analysis of outlier genes
Gene expression in the mouse inner ear was assessed using single cell RNAseq data obtained from the gEAR portal (https://umgear.org [18]). Datasets were chosen to include multiple ages (embryonic day (E)16, postnatal day (P)1, P7 [84], P15 [85], P20 [86] and P30 [87]), and expression was normalised within each dataset and cell type to Hprt expression. Where a dataset had more than one set of measurements for a cell type (eg the E16 dataset has “OHC_1” and “OHC_2”, both representing outer hair cells), expression levels were averaged. The expression of each gene was plotted in 12 different cell types, as defined by the original experiments. Eleven marker genes were plotted for comparison (hair cells: Myo7a; inner hair cells: Fgf8; outer hair cells: Slc26a5; non-sensory cells: Sox2; inner pillar cells: S100b; Deiters’ cells: Hes5; marginal cells: Kcne1; intermediate cells: Met; basal cells: Cldn11; spindle and root cells: Slc26a4; fibrocytes: Gm525). These are known marker genes for their cell types, with the exception of Gm525, which was chosen based on its fibrocyte-specific expression at P30 [87].
Immunohistochemistry
Wildtype mice from the C57BL/6N strain were used for the Madd expression study. Animals were collected, fixed, embedded in paraffin wax and sectioned as described in [88], with the only difference being the postnatal ages chosen (P0 and P4). Sections were dewaxed and rehydrated, then subjected to heated antigen retrieval in Citric acid after endogenous peroxidase blocking. Sections were then incubated with anti-Madd (1:200, AbCam, cat. no. ab134117) for 16 hours at room temperature. The secondary antibody used was Goat anti-rabbit IgG antibody (H+L) (1:5000, Vector Laboratories, BA-1000). Sections were stained using the Vectastain Elite ABC kit (Vector Laboratories, cat. no. PK-6100), counterstained with haemalum, dehydrated, and mounted in DPX. Images were taken using a Zeiss Axioskop 2 microscope with a Zeiss Axiocam camera, and processed using Adobe Photoshop CS6; minimal adjustments were made, including rotating and resizing.
Threshold difference detection
To assess each individual variant, audiograms were plotted with participants separated into groups by genotype and sex. Variants with fewer than 5 people/group in all alternate allele groups were excluded. For each group, the average thresholds at each stimulus frequency were compared to the reference group, testing for a difference of 20dB or more. This is a greater average threshold difference than that seen between those reporting a positive noise history and those reporting a negative noise history (Fig 1C). A maximum limit was imposed on standard deviation in the alternate allele group which differed by stimulus frequency (15 dB for 0.125–0.5 kHz, 20 dB for 1–2 kHz, 25 dB for 3–4 kHz, 30 dB for over 4 kHz) to prioritise variants associated with consistent threshold patterns. All variants where at least two stimulus frequencies in each ear passed this filter were put through to permutation testing.
We carried out 20,000 permutations for each variant to assess the likelihood that those stimulus frequencies which passed the filter were observed by chance. For each permutation, individuals from the cohort were assigned randomly to groups of the same number and sex as the original alternate allele and reference groups, and the relevant stimulus frequencies tested using the same settings as before. If more than 1000 random shufflings produced a similar result (i.e. p > 0.05), the variant was rejected. This was carried out automatically, and the scripts can be found at github.com/moraglewis/ThreADD.
The final p-values, derived from the permutation testing, represent the likelihood that the differences between the alternate and reference allele groups arose by chance. For the MUSC cohort, which had 41 variants passing the permutation test, we corrected for multiple testing using the Benjamini-Hochberg correction.
Supporting information
S1 Table. Tools and settings used for filtering variants by quality and impact.
https://doi.org/10.1371/journal.pgen.1011058.s001
(XLSX)
S2 Table. Outlier genes from the MUSC cohort analyses.
https://doi.org/10.1371/journal.pgen.1011058.s002
(XLSX)
S3 Table. Variant counts per gene from the MUSC cohort.
https://doi.org/10.1371/journal.pgen.1011058.s003
(XLSX)
S4 Table. Outlier genes from the TwinsUK cohort analyses.
https://doi.org/10.1371/journal.pgen.1011058.s004
(XLSX)
S5 Table. Variant counts per gene from the TwinsUK cohort.
https://doi.org/10.1371/journal.pgen.1011058.s005
(XLSX)
S6 Table. Genes known to underlie hearing loss in humans and/or in mice.
https://doi.org/10.1371/journal.pgen.1011058.s006
(XLSX)
S7 Table. Details of exome sequence data processing.
https://doi.org/10.1371/journal.pgen.1011058.s007
(XLSX)
S1 Fig. Both cohorts are predominantly of Non-Finnish European ancestry.
Population principal component analysis for the MUSC (orange, A) and TwinsUK (red, B) cohorts, along with data from participants from different genetic ancestries in the 1000 Genomes project [16]. The global population distribution is shown on the left and the distribution of European subpopulations on the right.
https://doi.org/10.1371/journal.pgen.1011058.s008
(PDF)
S2 Fig. Schematic of the pipeline used for variant calling and filtering.
See S1 Table for details of the filter settings.
https://doi.org/10.1371/journal.pgen.1011058.s009
(PDF)
S3 Fig. Expression levels at different developmental stages of the mouse orthologues of ten genes of interest from the outlier analysis (Syne2, Atp2c2, Dhrs4 (DHRS4L2), Rasal1, Patj (INADL), Tacc2, Abcb8, Itsn2, Pkhd1l1, and Fkbp2).
Single cell RNAseq data from the gEAR (http://umgear.org) was plotted for each of the ten genes. Expression was normalised to Hprt (represented by a horizontal line at y = 1 on each plot). Marker genes included for comparison are Myo7a (hair cells), Fgf8 (inner hair cells), Slc26a5 (outer hair cells), Sox2 (non-sensory cells), S100b (inner pillar cells), Hes5 (Deiters’ cells), comparison (Kcne1 (marginal cells), Met(intermediate cells), Cldn11 (basal cells), Slc26a4 (spindle and root cells) and Gm525 (fibrocytes). Two sets of plots are presented; the first set show expression in organ of Corti cell types and the second show expression in lateral wall cell types.
https://doi.org/10.1371/journal.pgen.1011058.s010
(PDF)
S4 Fig. Expression levels at different developmental stages of genes linked to specific subtypes of hearing loss.
Single cell RNAseq data from the gEAR (http://umgear.org) was plotted for each of the 29 genes associated with Metabolic (M) or Sensory (S) hearing loss. Expression was normalised to Hprt (represented by a horizontal line at y = 1 on each plot). Marker genes included for comparison are Myo7a (hair cells), Fgf8 (inner hair cells), Slc26a5 (outer hair cells), Sox2 (non-sensory cells), S100b (inner pillar cells), Hes5 (Deiters’ cells), comparison (Kcne1 (marginal cells), Met(intermediate cells), Cldn11 (basal cells), Slc26a4 (spindle and root cells) and Gm525 (fibrocytes).
https://doi.org/10.1371/journal.pgen.1011058.s011
(PDF)
S5 Fig. Madd is expressed in the hair cells of the mouse inner ear.
Images of the cochlear duct (basal turn) at ages from E14.5 (where no staining is visible) to P4. At least three mice were examined at each age. Brown shows where Madd is present (visible from E16.5 and older); hair cells are marked with arrowheads. Scale bar = 20μm.
https://doi.org/10.1371/journal.pgen.1011058.s012
(PDF)
S6 Fig. All audiograms from the MUSC and TwinsUK cohorts plotted in groups by sex and genotype.
For the 41 variants identified in the MUSC cohort, audiograms of TwinsUK carriers are shown on the right where available. The variant in HADH was identified in both cohorts. For the remaining 3 variants identified in the TwinsUK cohort, the audiograms of MUSC carriers are shown on the left. Two audiograms are shown for each variant in each cohort; the thresholds from the left ear are shown on the left, and those from the right ear on the right. Numbers and average ages of each group are listed on the graph. The symbols at the top of each graph mark which groups passed the criteria for each stimulus frequency compared to the relevant reference group (+ for male, = for female, and * for all participants). Error bars are standard deviation.
https://doi.org/10.1371/journal.pgen.1011058.s013
(PDF)
S7 Fig. MUSC cohort classification.
A) Schematic showing the classification process and numbers at each stage. B) Plot of the sensory estimate against the metabolic estimate for each well-fit case, with the blue/red shading indicating the magnitude of each estimate. The small dots are the Unclassified cases. C) shows the mean audiograms for the cases assigned to each category (error bars are standard error of the mean).
https://doi.org/10.1371/journal.pgen.1011058.s014
(PDF)
S8 Fig. TwinsUK cohort classification.
A) Schematic showing the classification process and numbers at each stage. B) Plot of the sensory estimate against the metabolic estimate for each well-fit case, with the blue/red shading indicating the magnitude of each estimate. The small dots are the Unclassified cases. C) shows the mean audiograms for the cases assigned to each category (error bars are standard error of the mean).
https://doi.org/10.1371/journal.pgen.1011058.s015
(PDF)
S2 Data. Data for audiograms in Figs 7, S7 and S8.
https://doi.org/10.1371/journal.pgen.1011058.s017
(XLSX)
Acknowledgments
We thank Safa Iqbal and Carl Hobbs for their work on the Madd expression study, and Elysia James for help with annotation. We are grateful to the participants in each cohort for contributing their data.
References
- 1. Lewis MA, Schulte BA, Dubno JR, Steel KP. Investigating the characteristics of genes and variants associated with self-reported hearing difficulty in older adults in the UK Biobank. BMC Biol. 2022;20(1):150. Epub 20220627. pmid:35761239; PubMed Central PMCID: PMC9238072.
- 2. Ingham NJ, Pearson SA, Vancollie VE, Rook V, Lewis MA, Chen J, et al. Mouse screen reveals multiple new genes underlying mouse and human hearing loss. PLoS Biol. 2019;17(4):e3000194. Epub 2019/04/12. pmid:30973865; PubMed Central PMCID: PMC6459510 following competing interests: ZP is employed by Ambry Genetics, and exome sequencing is among the commercially available tests. The other authors have declared that no competing interests exist.
- 3. Mencia A, Modamio-Hoybjor S, Redshaw N, Morin M, Mayo-Merino F, Olavarrieta L, et al. Mutations in the seed region of human miR-96 are responsible for nonsyndromic progressive hearing loss. Nat Genet. 2009;41(5):609–13. Epub 2009/04/14. pmid:19363479.
- 4. Chen DY, Liu XF, Lin XJ, Zhang D, Chai YC, Yu DH, et al. A dominant variant in DMXL2 is linked to nonsyndromic hearing loss. Genet Med. 2017;19(5):553–8. Epub 20160922. pmid:27657680.
- 5. Ahmadmehrabi S, Brant J, Epstein DJ, Ruckenstein MJ, Rader DJ. Genetics of Postlingual Sensorineural Hearing Loss. Laryngoscope. 2021;131(2):401–9. Epub 20200403. pmid:32243624.
- 6. Ivarsdottir EV, Holm H, Benonisdottir S, Olafsdottir T, Sveinbjornsson G, Thorleifsson G, et al. The genetic architecture of age-related hearing impairment revealed by genome-wide association analysis. Commun Biol. 2021;4(1):706. Epub 2021/06/11. pmid:34108613; PubMed Central PMCID: PMC8190123.
- 7. Kalra G, Milon B, Casella AM, Herb BR, Humphries E, Song Y, et al. Biological insights from multi-omic analysis of 31 genomic risk loci for adult hearing difficulty. PLoS Genet. 2020;16(9):e1009025. Epub 2020/09/29. pmid:32986727; PubMed Central PMCID: PMC7544108.
- 8. Wells HRR, Freidin MB, Zainul Abidin FN, Payton A, Dawes P, Munro KJ, et al. GWAS Identifies 44 Independent Associated Genomic Loci for Self-Reported Adult Hearing Difficulty in UK Biobank. Am J Hum Genet. 2019;105(4):788–802. Epub 2019/10/01. pmid:31564434; PubMed Central PMCID: PMC6817556.
- 9. Praveen K, Dobbyn L, Gurski L, Ayer AH, Staples J, Mishra S, et al. Population-scale analysis of common and rare genetic variation associated with hearing loss in adults. Commun Biol. 2022;5(1):540. Epub 20220603. pmid:35661827; PubMed Central PMCID: PMC9166757.
- 10. Cherny SS, Livshits G, Wells HRR, Freidin MB, Malkin I, Dawson SJ, et al. Self-reported hearing loss questions provide a good measure for genetic studies: a polygenic risk score analysis from UK Biobank. Eur J Hum Genet. 2020;28(8):1056–65. Epub 2020/03/24. pmid:32203203; PubMed Central PMCID: PMC7382483.
- 11.
Davis A, Research MRCIoH. Hearing in adults: the prevalence and distribution of hearing impairment and reported hearing disability in the MRC Institute of Hearing Research’s National Study of Hearing. London: Whurr Publishers; 1995. xv, 1011 pages: illustrations p.
- 12. Nondahl DM, Cruickshanks KJ, Wiley TL, Tweed TS, Klein R, Klein BE. Accuracy of self-reported hearing loss. Audiology. 1998;37(5):295–301. pmid:9776206.
- 13. Ahmadmehrabi S, Li B, Hui D, Park J, Ritchie M, Rader DJ, et al. A Genome-First Approach to Rare Variants in Dominant Postlingual Hearing Loss Genes in a Large Adult Population. Otolaryngol Head Neck Surg. 2022;166(4):746–52. Epub 20210720. pmid:34281439.
- 14. Vaden KI, Eckert MA, Matthews LJ, Schmiedt RA, Dubno JR. Metabolic and Sensory Components of Age-Related Hearing Loss. Journal of the Association for Research in Otolaryngology. 2022. pmid:35064426
- 15. Chang CC, Chow CC, Tellier LC, Vattikuti S, Purcell SM, Lee JJ. Second-generation PLINK: rising to the challenge of larger and richer datasets. Gigascience. 2015;4:7. Epub 2015/02/28. pmid:25722852; PubMed Central PMCID: PMC4342193.
- 16. Genomes Project C, Auton A, Brooks LD, Durbin RM, Garrison EP, Kang HM, et al. A global reference for human genetic variation. Nature. 2015;526(7571):68–74. Epub 2015/10/04. pmid:26432245; PubMed Central PMCID: PMC4750478.
- 17. Lewis MA, Nolan LS, Cadge BA, Matthews LJ, Schulte BA, Dubno JR, et al. Whole exome sequencing in adult-onset hearing loss reveals a high load of predicted pathogenic variants in known deafness-associated genes and identifies new candidate genes. BMC Med Genomics. 2018;11(1):77. Epub 20180904. pmid:30180840; PubMed Central PMCID: PMC6123954.
- 18. Orvis J, Gottfried B, Kancherla J, Adkins RS, Song Y, Dror AA, et al. gEAR: Gene Expression Analysis Resource portal for community-driven, multi-omic data exploration. Nat Methods. 2021;18(8):843–4. Epub 2021/06/27. pmid:34172972.
- 19. Karczewski KJ, Francioli LC, Tiao G, Cummings BB, Alfoldi J, Wang Q, et al. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature. 2020;581(7809):434–43. Epub 2020/05/29. pmid:32461654; PubMed Central PMCID: PMC7334197.
- 20. Chen S, Francioli LC, Goodrich JK, Collins RL, Kanai M, Wang Q, et al. A genome-wide mutational constraint map quantified from variation in 76,156 human genomes. bioRxiv. 2022:2022.03.20.485034.
- 21. Van Camp G, Smith RJH. Hereditary Hearing Loss Homepage [cited 2021 December]. Available from: https://hereditaryhearingloss.org/.
- 22. Weston MD, Luijendijk MW, Humphrey KD, Moller C, Kimberling WJ. Mutations in the VLGR1 gene implicate G-protein signaling in the pathogenesis of Usher syndrome type II. Am J Hum Genet. 2004;74(2):357–66. Epub 20040120. pmid:14740321; PubMed Central PMCID: PMC1181933.
- 23. Pfister MH, Apaydin F, Turan O, Bereketoglu M, Bylgen V, Braendle U, et al. A second family with nonsyndromic sensorineural hearing loss linked to Xp21.2: refinement of the DFN4 locus within DMD. Genomics. 1998;53(3):377–82. Epub 1998/11/04. pmid:9799605.
- 24. Santos-Cortez RL, Faridi R, Rehman AU, Lee K, Ansar M, Wang X, et al. Autosomal-Recessive Hearing Impairment Due to Rare Missense Variants within S1PR2. Am J Hum Genet. 2016;98(2):331–8. Epub 20160121. pmid:26805784; PubMed Central PMCID: PMC4746333.
- 25. Ingham NJ, Carlisle F, Pearson S, Lewis MA, Buniello A, Chen J, et al. S1PR2 variants associated with auditory function in humans and endocochlear potential decline in mouse. Sci Rep. 2016;6:28964. Epub 20160707. pmid:27383011; PubMed Central PMCID: PMC4935955.
- 26. Dickinson ME, Flenniken AM, Ji X, Teboul L, Wong MD, White JK, et al. High-throughput discovery of novel developmental phenotypes. Nature. 2016;537(7621):508–14. Epub 2016/09/15. pmid:27626380; PubMed Central PMCID: PMC5295821.
- 27. Groza T, Gomez FL, Mashhadi HH, Munoz-Fuentes V, Gunes O, Wilson R, et al. The International Mouse Phenotyping Consortium: comprehensive knockout phenotyping underpinning the study of human disease. Nucleic Acids Res. 2023;51(D1):D1038–D45. pmid:36305825; PubMed Central PMCID: PMC9825559.
- 28. Basel-Vanagaite L, Dallapiccola B, Ramirez-Solis R, Segref A, Thiele H, Edwards A, et al. Deficiency for the ubiquitin ligase UBE3B in a blepharophimosis-ptosis-intellectual-disability syndrome. Am J Hum Genet. 2012;91(6):998–1010. Epub 20121129. pmid:23200864; PubMed Central PMCID: PMC3516591.
- 29. Wu X, Ivanchenko MV, Al Jandal H, Cicconet M, Indzhykulian AA, Corey DP. PKHD1L1 is a coat protein of hair-cell stereocilia and is required for normal hearing. Nat Commun. 2019;10(1):3801. Epub 2019/08/25. pmid:31444330; PubMed Central PMCID: PMC6707252.
- 30. Du TT, Dewey JB, Wagner EL, Cui R, Heo J, Park JJ, et al. LMO7 deficiency reveals the significance of the cuticular plate for hearing function. Nat Commun. 2019;10(1):1117. Epub 20190308. pmid:30850599; PubMed Central PMCID: PMC6408450.
- 31. Johnson KR, Marden CC, Ward-Bailey P, Gagnon LH, Bronson RT, Donahue LR. Congenital hypothyroidism, dwarfism, and hearing impairment caused by a missense mutation in the mouse dual oxidase 2 gene, Duox2. Mol Endocrinol. 2007;21(7):1593–602. Epub 20070417. pmid:17440044.
- 32. Curtin JA, Quint E, Tsipouri V, Arkell RM, Cattanach B, Copp AJ, et al. Mutation of Celsr1 disrupts planar polarity of inner ear hair cells and causes severe neural tube defects in the mouse. Curr Biol. 2003;13(13):1129–33. pmid:12842012.
- 33. Lee JH, Perez-Flores C, Park S, Kim HJ, Chen Y, Kang M, et al. The Piezo channel is central to the mechano-sensitive channel complex in the mammalian inner ear. Res Sq. 2023. Epub 20230712. pmid:37502846; PubMed Central PMCID: PMC10371147.
- 34. Horn HF, Brownstein Z, Lenz DR, Shivatzki S, Dror AA, Dagan-Rosenfeld O, et al. The LINC complex is essential for hearing. J Clin Invest. 2013;123(2):740–50. Epub 20130125. pmid:23348741; PubMed Central PMCID: PMC3561815.
- 35. Nigam SK, Jin YJ, Jin MJ, Bush KT, Bierer BE, Burakoff SJ. Localization of the FK506-binding protein, FKBP 13, to the lumen of the endoplasmic reticulum. Biochem J. 1993;294 (Pt 2)(Pt 2):511–5. pmid:8373365; PubMed Central PMCID: PMC1134484.
- 36. Partaledis JA, Berlin V. The FKB2 gene of Saccharomyces cerevisiae, encoding the immunosuppressant-binding protein FKBP-13, is regulated in response to accumulation of unfolded proteins in the endoplasmic reticulum. Proc Natl Acad Sci U S A. 1993;90(12):5450–4. pmid:7685904; PubMed Central PMCID: PMC46738.
- 37. Ardehali H, O’Rourke B, Marban E. Cardioprotective role of the mitochondrial ATP-binding cassette protein 1. Circ Res. 2005;97(8):740–2. Epub 20050915. pmid:16166555; PubMed Central PMCID: PMC2756018.
- 38. Tanaka M, Miyoshi J, Ishizaki H, Togawa A, Ohnishi K, Endo K, et al. Role of Rab3 GDP/GTP exchange protein in synaptic vesicle trafficking at the mouse neuromuscular junction. Mol Biol Cell. 2001;12(5):1421–30. pmid:11359932; PubMed Central PMCID: PMC34594.
- 39. Yamaguchi K, Tanaka M, Mizoguchi A, Hirata Y, Ishizaki H, Kaneko K, et al. A GDP/GTP exchange protein for the Rab3 small G protein family up-regulates a postdocking step of synaptic exocytosis in central synapses. Proc Natl Acad Sci U S A. 2002;99(22):14536–41. Epub 20021018. pmid:12388783; PubMed Central PMCID: PMC137918.
- 40. Clement A, Blanco-Sanchez B, Peirce JL, Westerfield M. Cog4 is required for protrusion and extension of the epithelium in the developing semicircular canals. Mech Dev. 2019;155:1–7. Epub 20181001. pmid:30287385; PubMed Central PMCID: PMC6226009.
- 41. Foomani FH, Jarzembowski JA, Mostaghimi S, Mehrvar S, Kumar SN, Ranji M. Optical Metabolic Imaging of Mitochondrial Dysfunction on HADH Mutant Newborn Rat Hearts. IEEE J Transl Eng Health Med. 2021;9:1800407. Epub 20210816. pmid:34462673; PubMed Central PMCID: PMC8396955.
- 42. Steinke KV, Gorinski N, Wojciechowski D, Todorov V, Guseva D, Ponimaskin E, et al. Human CLC-K Channels Require Palmitoylation of Their Accessory Subunit Barttin to Be Functional. J Biol Chem. 2015;290(28):17390–400. Epub 20150526. pmid:26013830; PubMed Central PMCID: PMC4498076.
- 43. Huang TW, Iyer AA, Manalo JM, Woo J, Bosquez Huerta NA, McGovern MM, et al. Glial-Specific Deletion of Med12 Results in Rapid Hearing Loss via Degradation of the Stria Vascularis. J Neurosci. 2021;41(34):7171–81. Epub 20210712. pmid:34253626; PubMed Central PMCID: PMC8387121.
- 44. Cruickshanks KJ, Wiley TL, Tweed TS, Klein BE, Klein R, Mares-Perlman JA, et al. Prevalence of hearing loss in older adults in Beaver Dam, Wisconsin. The Epidemiology of Hearing Loss Study. Am J Epidemiol. 1998;148(9):879–86. Epub 1998/11/04. pmid:9801018.
- 45. Dubno JR, Lee FS, Matthews LJ, Ahlstrom JB, Horwitz AR, Mills JH. Longitudinal changes in speech recognition in older persons. J Acoust Soc Am. 2008;123(1):462–75. Epub 2008/01/08. pmid:18177174.
- 46. Dubno JR, Lee FS, Matthews LJ, Mills JH. Age-related and gender-related changes in monaural speech recognition. J Speech Lang Hear Res. 1997;40(2):444–52. Epub 1997/04/01. pmid:9130212.
- 47. Helzner EP, Cauley JA, Pratt SR, Wisniewski SR, Zmuda JM, Talbott EO, et al. Race and sex differences in age-related hearing loss: the Health, Aging and Body Composition Study. J Am Geriatr Soc. 2005;53(12):2119–27. Epub 2006/01/10. pmid:16398896.
- 48. Lee FS, Matthews LJ, Dubno JR, Mills JH. Longitudinal study of pure-tone thresholds in older persons. Ear Hear. 2005;26(1):1–11. Epub 2005/02/05. pmid:15692300.
- 49. Pearson JD, Morrell CH, Gordon-Salant S, Brant LJ, Metter EJ, Klein LL, et al. Gender differences in a longitudinal study of age-associated hearing loss. J Acoust Soc Am. 1995;97(2):1196–205. Epub 1995/02/01. pmid:7876442.
- 50. Hederstierna C, Hultcrantz M, Collins A, Rosenhall U. The menopause triggers hearing decline in healthy women. Hear Res. 2010;259(1–2):31–5. Epub 2009/09/29. pmid:19781610.
- 51. Montasser ME, Van Hout CV, Miloscio L, Howard AD, Rosenberg A, Callaway M, et al. Genetic and functional evidence links a missense variant in B4GALT1 to lower LDL and fibrinogen. Science. 2021;374(6572):1221–7. Epub 2021/12/03. pmid:34855475.
- 52. Akbari P, Gilani A, Sosina O, Kosmicki JA, Khrimian L, Fang YY, et al. Sequencing of 640,000 exomes identifies GPR75 variants associated with protection from obesity. Science. 2021;373(6550). Epub 2021/07/03. pmid:34210852.
- 53. Kelly JJ, Abitbol JM, Hulme S, Press ER, Laird DW, Allman BL. The connexin 30 A88V mutant reduces cochlear gap junction expression and confers long-term protection against hearing loss. J Cell Sci. 2019;132(2). Epub 2018/12/19. pmid:30559251.
- 54. Dubno JR, Eckert MA, Lee FS, Matthews LJ, Schmiedt RA. Classifying human audiometric phenotypes of age-related hearing loss from animal models. J Assoc Res Otolaryngol. 2013;14(5):687–701. Epub 2013/06/07. pmid:23740184; PubMed Central PMCID: PMC3767874.
- 55. Wolber LE, Steves CJ, Spector TD, Williams FM. Hearing ability with age in northern European women: a new web-based approach to genetic studies. PLoS One. 2012;7(4):e35500. Epub 20120430. pmid:22558162; PubMed Central PMCID: PMC3340381.
- 56. Schmiedt RA, editor The Physiology of Cochlear Presbycusis2010.
- 57. Vaden KI Jr, Matthews LJ, Eckert MA, Dubno JR. Longitudinal Changes in Audiometric Phenotypes of Age-Related Hearing Loss. J Assoc Res Otolaryngol. 2017;18(2):371–85. Epub 2016/11/11. pmid:27830350; PubMed Central PMCID: PMC5352606.
- 58. Williams FM, Scollen S, Cao D, Memari Y, Hyde CL, Zhang B, et al. Genes contributing to pain sensitivity in the normal population: an exome sequencing study. PLoS Genet. 2012;8(12):e1003095. Epub 20121220. pmid:23284290; PubMed Central PMCID: PMC3527205.
- 59. Kim D, Langmead B, Salzberg SL. HISAT: a fast spliced aligner with low memory requirements. Nat Methods. 2015;12(4):357–60. Epub 2015/03/10. pmid:25751142; PubMed Central PMCID: PMC4655817.
- 60. Webster TH, Couse M, Grande BM, Karlins E, Phung TN, Richmond PA, et al. Identifying, understanding, and correcting technical artifacts on the sex chromosomes in next-generation sequencing data. Gigascience. 2019;8(7). pmid:31289836; PubMed Central PMCID: PMC6615978.
- 61. McKenna A, Hanna M, Banks E, Sivachenko A, Cibulskis K, Kernytsky A, et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res. 2010;20(9):1297–303. Epub 2010/07/21. pmid:20644199; PubMed Central PMCID: PMC2928508.
- 62. Poplin R, Ruano-Rubio V, DePristo MA, Fennell TJ, Carneiro MO, Van der Auwera GA, et al. Scaling accurate genetic variant discovery to tens of thousands of samples. bioRxiv. 2018:201178.
- 63. Danecek P, Bonfield JK, Liddle J, Marshall J, Ohan V, Pollard MO, et al. Twelve years of SAMtools and BCFtools. Gigascience. 2021;10(2). pmid:33590861; PubMed Central PMCID: PMC7931819.
- 64. Garrison EM, Gabor . Haplotype-based variant detection from short-read sequencing. aRxiv. 2012:1207.3907.
- 65. Bao R, Huang L, Andrade J, Tan W, Kibbe WA, Jiang H, et al. Review of current methods, applications, and data management for the bioinformatics analysis of whole exome sequencing. Cancer Inform. 2014;13(Suppl 2):67–82. Epub 20140921. pmid:25288881; PubMed Central PMCID: PMC4179624.
- 66. Van der Auwera GA, Carneiro MO, Hartl C, Poplin R, Del Angel G, Levy-Moonshine A, et al. From FastQ data to high confidence variant calls: the Genome Analysis Toolkit best practices pipeline. Curr Protoc Bioinformatics. 2013;43(1110):11 0 1–0 33. pmid:25431634; PubMed Central PMCID: PMC4243306.
- 67. Danecek P, Auton A, Abecasis G, Albers CA, Banks E, DePristo MA, et al. The variant call format and VCFtools. Bioinformatics. 2011;27(15):2156–8. Epub 20110607. pmid:21653522; PubMed Central PMCID: PMC3137218.
- 68. Graffelman J. Exploring Diallelic Genetic Markers: The HardyWeinberg Package. Journal of Statistical Software. 2015;64(3):1–23.
- 69. Graffelman J, Camarena JM. Graphical tests for Hardy-Weinberg equilibrium based on the ternary plot. Hum Hered. 2008;65(2):77–84. Epub 20070926. pmid:17898538.
- 70. Fuentes Fajardo KV, Adams D, Program NCS, Mason CE, Sincan M, Tifft C, et al. Detecting false-positive signals in exome sequencing. Hum Mutat. 2012;33(4):609–13. Epub 2012/02/02. pmid:22294350.
- 71. Maffucci P, Bigio B, Rapaport F, Cobat A, Borghesi A, Lopez M, et al. Blacklisting variants common in private cohorts but not in public databases optimizes human exome analysis. Proc Natl Acad Sci U S A. 2019;116(3):950–9. Epub 2018/12/29. pmid:30591557; PubMed Central PMCID: PMC6338851.
- 72. Benjamin D, Sato T, Cibulskis K, Getz G, Stewart C, Lichtenstein L. Calling Somatic SNVs and Indels with Mutect2. bioRxiv. 2019:861054.
- 73. Picardi E, Pesole G. Mitochondrial genomes gleaned from human whole-exome sequencing. Nat Methods. 2012;9(6):523–4. Epub 20120530. pmid:22669646.
- 74. Griffin HR, Pyle A, Blakely EL, Alston CL, Duff J, Hudson G, et al. Accurate mitochondrial DNA sequencing using off-target reads provides a single test to identify pathogenic point mutations. Genet Med. 2014;16(12):962–71. Epub 20140605. pmid:24901348; PubMed Central PMCID: PMC4272251.
- 75. McLaren W, Gil L, Hunt SE, Riat HS, Ritchie GR, Thormann A, et al. The Ensembl Variant Effect Predictor. Genome Biol. 2016;17(1):122. Epub 2016/06/09. pmid:27268795; PubMed Central PMCID: PMC4893825.
- 76. Pajusalu S. 5utr [cited 2021]. Available from: https://github.com/leklab/5utr.
- 77. Jaganathan K, Kyriazopoulou Panagiotopoulou S, McRae JF, Darbandi SF, Knowles D, Li YI, et al. Predicting Splicing from Primary Sequence with Deep Learning. Cell. 2019;176(3):535–48 e24. Epub 2019/01/22. pmid:30661751.
- 78. Rentzsch P, Witten D, Cooper GM, Shendure J, Kircher M. CADD: predicting the deleteriousness of variants throughout the human genome. Nucleic Acids Res. 2019;47(D1):D886–D94. Epub 2018/10/30. pmid:30371827; PubMed Central PMCID: PMC6323892.
- 79. Fu W, O’Connor TD, Jun G, Kang HM, Abecasis G, Leal SM, et al. Analysis of 6,515 exomes reveals the recent origin of most human protein-coding variants. Nature. 2013;493(7431):216–20. Epub 2012/12/04. pmid:23201682.
- 80. Taliun D, Harris DN, Kessler MD, Carlson J, Szpiech ZA, Torres R, et al. Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program. Nature. 2021;590(7845):290–9. Epub 2021/02/12. pmid:33568819; PubMed Central PMCID: PMC7875770.
- 81. Untergasser A, Cutcutache I, Koressaar T, Ye J, Faircloth BC, Remm M, et al. Primer3—new capabilities and interfaces. Nucleic Acids Res. 2012;40(15):e115. Epub 2012/06/26. pmid:22730293; PubMed Central PMCID: PMC3424584.
- 82. Bonfield JK, Smith K, Staden R. A new DNA sequence assembly program. Nucleic Acids Res. 1995;23(24):4992–9. Epub 1995/12/25. pmid:8559656; PubMed Central PMCID: PMC307504.
- 83. Vuckovic D, Mezzavilla M, Cocca M, Morgan A, Brumat M, Catamo E, et al. Whole-genome sequencing reveals new insights into age-related hearing loss: cumulative effects, pleiotropy and the role of selection. Eur J Hum Genet. 2018;26(8):1167–79. Epub 2018/05/05. pmid:29725052; PubMed Central PMCID: PMC6057993.
- 84. Kolla L, Kelly MC, Mann ZF, Anaya-Rocha A, Ellis K, Lemons A, et al. Characterization of the development of the mouse cochlear epithelium at the single cell level. Nat Commun. 2020;11(1):2389. Epub 2020/05/15. pmid:32404924; PubMed Central PMCID: PMC7221106.
- 85. Ranum PT, Goodwin AT, Yoshimura H, Kolbe DL, Walls WD, Koh JY, et al. Insights into the Biology of Hearing and Deafness Revealed by Single-Cell RNA Sequencing. Cell Rep. 2019;26(11):3160–71 e3. Epub 2019/03/14. pmid:30865901; PubMed Central PMCID: PMC6424336.
- 86. Xue N, Song L, Song Q, Santos-Sacchi J, Wu H, Navaratnam D. Genes related to SNPs identified by Genome-wide association studies of age-related hearing loss show restriction to specific cell types in the adult mouse cochlea. Hear Res. 2021;410:108347. Epub 2021/09/19. pmid:34536825.
- 87. Korrapati S, Taukulis I, Olszewski R, Pyle M, Gu S, Singh R, et al. Single Cell and Single Nucleus RNA-Seq Reveal Cellular Heterogeneity and Homeostatic Regulatory Networks in Adult Mouse Stria Vascularis. Front Mol Neurosci. 2019;12:316. Epub 2020/01/11. pmid:31920542; PubMed Central PMCID: PMC6933021.
- 88. Carlisle FA, Steel KP, Lewis MA. Specific expression of Kcna10, Pxn and Odf2 in the organ of Corti. Gene Expr Patterns. 2012;12(5–6):172–9. Epub 20120316. pmid:22446089; PubMed Central PMCID: PMC3368262.