Prevalence and clinical features of hearing loss caused by EYA4 variants

Variants in the EYA4 gene are known to lead to autosomal dominant non-syndromic hereditary hearing loss, DFNA10. To date, 30 variants have been shown to be responsible for hearing loss in a diverse set of nationalities. To better understand the clinical characteristics and prevalence of DFNA10, we performed genetic screening for EYA4 mutations in a large cohort of Japanese hearing loss patients. We selected 1,336 autosomal dominant hearing loss patients among 7,408 unrelated Japanese hearing loss probands and performed targeted genome enrichment and massively parallel sequencing of 68 target genes for all patients. Clinical information of cases with mutations in EYA4 was gathered and analyzed from medical charts. Eleven novel EYA4 variants (three frameshift variants, three missense variants, two nonsense variants, one splicing variant, and two single-copy number losses) and two previously reported variants were found in 12 probands (0.90%) among the 1,336 autosomal dominant hearing loss families. The audiometric configuration of truncating variants tends to deteriorate for all frequencies, whereas that of non-truncating variants tends to show high-frequency hearing loss, suggesting a new correlation between genotype and phenotype in DFNA10. The rate of hearing loss progression caused by EYA4 variants was considered to be 0.63 dB/year, as found in this study and previous reports.

of hearing loss was divided into mild (PTA: 20-40 dB HL), moderate (41-70 dB HL), severe (71-95 dB HL), and profound (>95 dB HL). Asymmetric hearing loss was defined as a difference in PTA of over 10 dB between the right and left ears. The audiometric configurations were categorized into low-frequency, mid-frequency (U-shaped), high-frequency, flat type, and deaf as reported previously 32 .

Amplicon resequencing and variant annotation. Amplicon libraries were prepared using an Ion
AmpliSeq ™ Custom Panel for 68 genes reported to cause non-syndromic hereditary hearing loss (ThermoFisher Scientific, MA, USA), in accordance with the manufacturer's instructions. The detailed protocol has been described elsewhere 33 . MPS was performed with an Ion Torrent Personal Genome Machine (PGM) system using an Ion PGM ™ 200 Sequencing Kit and an Ion 318 ™ Chip (ThermoFisher Scientific). The sequence data were mapped against the human genome sequence (build GRCh37/hg19) with a Torrent Mapping Alignment Program. After sequence mapping, the DNA variant regions were piled up with Torrent Variant Caller plug-in software. After variant detection, their effects were analyzed using ANNOVAR software 34,35 . The missense, nonsense, insertion/deletion and splicing variants were selected from among the identified variants. Variants were further selected as less than 1% of: (1) the 1,000 genome database, (2) 6,500 exome variants, (3) the Human Genetic Variation Database (a dataset for 1,208 Japanese exome variants), and (4) 333 in-house Japanese normal hearing controls. This filtering process was performed using our original database software described elsewhere 36 . The pathogenicity of selected variants was evaluated by ACMG (American College of Medical Genetics) standards and guidelines 37 . For missense variants in particular, functional prediction software, including Sorting Intolerant from Tolerant (SIFT), Polymorphism Phenotyping (PolyPhen2), LRT, Mutation Taster, Mutation Assessor, Functional Analysis through Hidden Markov Models (FATHMM), RadialSVM, LR, and CADD, were used through the ANNOVAR software program 34,35 . Direct sequencing was utilized to confirm the selected variants.
Copy number analysis in the MPS database. We employed our recently published specialized copy number variation (CNV) detection method for Ion AmpliSeq TM sequencing that utilizes multiplex PCR-based targeted genome enrichment 38 . The depth of coverage information for each amplicon was used for copy number analysis. After normalization, the relative read depths of amplicons were visualized as described previously 38 . Variant prioritization. EYA4 was reported as a genetic cause for autosomal dominant inherited hearing loss, thus, we selected hearing loss patients from apparently autosomal dominant families. Among 1,336 autosomal dominant hearing loss families, we further selected the families with candidate EYA4 variants. The criteria for the selection process were (1) the EYA4 variant was classified into "pathogenic", "likely pathogenic" or "uncertain significance" and (2) there were no candidate variants in the other 67 genes reported to cause hearing loss. Based on the ACMG guidelines, we regarded "pathogenic" and "likely pathogenic" variants as strong candidates for EYA4-associated hearing loss. In addition, we listed the "variants of uncertain significance" identified during the filtering procedure described above in Table 1. However, we removed "variants of uncertain significance" with a CADD Phred score of less than 20, or identified in some control databases as being of "unlikely causative". The CADD Phred score threshold used in this study was <20 as all of the previously reported EYA4 pathogenic variants were predicted to have a CADD Phred score of 23.5 or more (e.g., the lowest CADD Phred score for c.978C > G is 23.5), so we employed 20 as threshold to allow a safety margin. In addition, we also removed the c.1790delT and c.1886_1899del variants as unlikely causative variants because nonsense-mediated mRNA decay was not presumed to be triggered from the location of the variants. Finally, we selected 12 variants as causative and performed a more detailed hearing loss phenotype analysis.

Results
Identified variants and the frequency of EYA4-associated hearing loss. Among the 1,336 probands with ADNSHL, we identified 12 (0.90%) who carried a possible EYA4 pathogenic variant (Table 1, Fig. 1). These 12 probands did not show any pathogenic variants or candidate variants in the 67 previously reported deafness genes apart from EYA4. Among the 12 candidate variants, eleven were novel, and one was previously reported. Three of them were missense variants, three were frameshift insertion/deletion variants, three were nonsense variants, one was a splicing variant, and two were copy number losses. Six of them were located in the eyaVR (amino acids 0-369) and four were located in eyaHR (amino acids 370-639). The one previously reported variant was classified as "Pathogenic". Eight variants were classified as "Likely pathogenic" according to the ACMG guidelines, whereas three remained as "variants of uncertain significance (VUS)". Further, none of the 12 variants was found in the Japanese 333 in-house controls (666 control alleles), and none of the three VUS variants was observed in the ExAC03 database.
In addition to the above 12 causative EYA4 variants, we also identified 6 variants in the EYA4 gene from our cohort (Table 1), but we regarded these 6 variants as unlikely to be causative. Three missense variants, c.887C > T, c.936G > T, and c.995C > T were identified in the ExAC03 database over 0.0001, suggesting these three variants were not causative variants. In addition, the c.278T > C variant was predicted to be an unlikely causative variant from its low CADD score (CADD Phred score 14.76). Furthermore, the c.1790delT and c.1886_1899del variants were regarded as unlikely causative variants as nonsense-mediated mRNA decay was not presumed to be triggered from the location of the variants (these variants were located in the final exon or one exon before the final exon). From these results, the pathogenicity of these six variants was unclear. Thus, we performed further detailed clinical characteristic analysis for 12 patients with causative EYA4 variants.
Clinical characteristics of the EYA4-associated hearing loss patients identified in this study. www.nature.com/scientificreports www.nature.com/scientificreports/ hearing loss that deteriorated in all frequencies, whereas the patients with non-truncating variants showed high-frequency hearing loss. We also analyzed the rate of hearing deterioration by using the patients in this study and previously reported case results (Fig. 3) and found that the average rate of progression in PTA was 0.63 dB/ year (95%CI: 0.41-0.85 dB/year).

Discussion
In this report, we analyzed 1,334 ADNSHL patients and identified 12 candidate variants for EYA4-associated hearing loss. This is the largest population studied for EYA4-associated hearing loss to date. The prevalence of EYA4-associated hearing loss in ADNSHL was 0.90% (12/1,334 cases) in the Japanese population. This prevalence is slightly less than those of other ADNSHL genes such as KCNQ4, TECTA, POU4F3, and WFS1. KCNQ4 is one of the most frequently observed responsible genes for ADNSHL in the Japanese population, and its prevalence is 6.6% 39 . Likewise, the prevalence of ADNSHL caused by TECTA variants is 2.9%, 2.7% for POU4F3 variants, and 2.5% for WFS1 variants [40][41][42] .
The responsible genes for ADNSHL differ among ethnic groups. For example, KCNQ4 is the most frequent causative gene for ADNSHL in the Japanese population, whereas TECTA is the most frequent causative gene in the American population 43 . One plausible reason of this difference among populations is the effect of founder  www.nature.com/scientificreports www.nature.com/scientificreports/ or recurrent mutations. Indeed, the variant of KCNQ4; c.211delC, which is commonly observed in Japanese ADNSHL patients, was reported to be caused by a founder effect 39 . Most of the EYA4 variants found in this report were novel (we summarized the clinical features and identified variants of previous reports in Table 3) and the identified variants differed among patients. Only one variant (p.Q393X) was identified in a Korean patient 9 . According to this result, most of the EYA4 variants are not recurrent. From these results, it appears difficult to find EYA4 variants among autosomal dominant hereditary hearing loss patients by various genotyping analysis methods such as Invader assay or microarray, thus MPS is useful for identifying rare causative variants such as those in the EYA4 gene in ADNSHL patients.
In previous reports, the audiometric configuration for EYA4-associated hearing loss was a gradual high-frequency hearing loss or a flat-type hearing loss 44 . Further, no genotype-phenotype correlation was identified in previous reports. Kim et al. reported that no genotype-phenotype correlation existed for EYA4-associated hearing loss 9 . In their report, they analyzed only 87 ADNSHL patients, and identified only two patients carrying EYA4 variants. In this study, we analyzed 1,334 ADNSHL patients, and identified 12 candidate EYA4 variants. We also analyzed the detailed audiometric configurations of 12 patients identified in this study and previously reported cases and identified a genotype-phenotype correlation. High-frequency hearing loss was observed in patients with non-truncating EYA4 variants, whereas flat-type hearing loss was observed in patients with truncating EYA4 variants. In contrast, there were no significant differences in the severity of hearing loss among the different types of variants and/or variant locations (domain).
We also analyzed the rate of hearing deterioration in EYA4-associated hearing loss patients identified in this study and previously reported cases. The rate of progression of hearing loss caused by EYA4 was considered to be 0.63 dB/year (95%CI: 0.41-0.85 dB/year). In previous reports on ADSNHL hearing loss, the progression rate for the POU4F3 gene was 0.5-0.9 dB/year 41 , that for MYO6 was 2.0 dB/year 45 , and that for ACTG1 was 2.0-6.0 dB/ year 46 , and the results in this study suggests that the rate of hearing loss progression caused by EYA4 may be relatively mild.
In this study, we identified nine truncating variants including two EYA4 copy number loss cases. Thus, we speculated that the mechanism of EYA4-associated hearing loss was haploinsufficiency. In the gnomAD database (https://gnomad.broadinstitute.org/gene/ENSG00000112319), a non-negligible number of truncating variants were identified in large control populations. The probability of a loss of function intolerant score (pLI score) was 0.05. This low score may mean the loss of function in this gene is tolerant and without pathogenicity. However, most of the loss of function variants were located in specific exons that only included some splicing variants and were seldom observed in other exons (Fig. 4). From these observations, we hypothesized that these specific isoforms may not be expressed in the inner ear or may not play an important role in hearing function. It is unknown which isoforms are expressed in the human inner ear. As another hypothesized mechanism, loss of function variants in the gnomAD database were accumulated in the second to last exon, and these variants might not trigger nonsense-mediated mRNA decay. Thus, these loss of function variants may not cause hearing loss. The identified truncating variants, except for c.1790delT, were located in the exons which were included in all isoforms carried. The prevalence of loss of function variants in the EYA4 gene was 19 among about 250,000 alleles in the gnomAD database, but 9 among 2,672 alleles in this study. The summarized odds ratio between our hearing loss cohort vs. gnomAD was ). This result also supports haploinsufficiency as the mechanism underlying EYA4-associated hearing loss. The patient who carried c.1790delT, located in specific exons (truncating variants accumulated in the exon in gnomAD) suffers from an enlarged vestibular aqueduct, and this phenotype was not matched with hearing loss caused by EYA4 mutations. For these reasons, we classified this variant (c.1790delT) as "unlikely causative".
In conclusion, we performed MPS analysis of large cohort of 1,334 ADNSHL patients and successfully identified 12 novel and promising pathogenic variants. Based on this, we estimated the incidence of EYA4-associated hearing loss was 0.90% in Japanese families with autosomal dominant hearing loss. The audiometric configuration of truncating variants tended to exhibit flat-type, whereas that of non-truncating variants tended to be high-frequency hearing loss, suggesting a novel genotype-phenotype correlation in DFNA10.