Improving the Management of Patients with Hearing Loss by the Implementation of an NGS Panel in Clinical Practice

A cohort of 128 patients from 118 families diagnosed with non-syndromic or syndromic hearing loss (HL) underwent an exhaustive clinical evaluation. Molecular analysis was performed using targeted next-generation sequencing (NGS) with a custom panel that included 59 genes associated with non-syndromic HL or syndromic HL. Variants were prioritized according to the minimum allele frequency and classified according to the American College of Medical Genetics and Genomics guidelines. Variant(s) responsible for the disease were detected in a 40% of families including autosomal recessive (AR), autosomal dominant (AD) and X-linked patterns of inheritance. We identified pathogenic or likely pathogenic variants in 26 different genes, 15 with AR inheritance pattern, 9 with AD and 2 that are X-linked. Fourteen of the found variants are novel. This study highlights the clinical utility of targeted NGS for sensorineural hearing loss. The optimal panel for HL must be designed according to the spectrum of the most represented genes in a given population and the laboratory capabilities considering the pressure on healthcare.


Introduction
Hearing loss (HL) is the most common sensory deficit in humans [1]. According to data from the World Health Organization, it is estimated that more than 5% of the world's population suffers from this disease, that is, around 360 million people.
HL can be classified as conductive, sensorineural or mixed (a combination of both); acquired or hereditary; prelingual or postlingual; and non-syndromic (NSHL) or syndromic, as a part of a more complex phenotype, that account up to 30% of HL cases [2].
HL is one of the most common birth defects, with an incidence of 1-2 per 1000 newborns and growing as age increases, reaching more than 300 per 1000 in those over 75 years of age. This high A total of 128 patients from 118 families diagnosed with non-syndromic or syndromic HL were included in our study. Most patients were of Spanish origin, except for three patients that came from Eastern Europe, two patients that were from Maghreb, two patients that were of sub-Saharan origin and one patient that was from East Asia. Patients were recruited from September 2017 to December 2019. Most patients presented with non-syndromic hearing loss, but we also received for screening four patients with Usher syndrome (USH), two with Waardenburg syndrome (WS) and two patients with branchio-oto-renal syndrome (BOR). Patients were enrolled through the Department of Otolaryngology of the University Hospital La Fe, according to standard assistance procedures. Comprehensive clinical evaluations, imaging examination, pure-tone audiograms, auditory brainstem response and other relevant medical information were collected for the probands to characterize the type and severity of HL. All recruited patients presented sensorineural or mixed HL. Hearing loss severity was established as mild (between >25 and ≤40 dB), moderate (between >40 and ≤70 dB) or severe/profound (>70 dB).
Written informed consent was obtained from all participants or their legal guardians. This study was approved by the Hospital La Fe Ethics Committee in agreement with the Declaration of Helsinki (REV03/5/2014).
Genomic DNA (gDNA) from the patients and relatives was obtained and purified using the automated DNA extractor QIAsymphony (QIAGEN, Hombrechtikon, Switzerland). The concentration of the resulting DNA samples was determined with Nanodrop and Qubit fluorometer (Thermo Fisher Scientific, Waltham, MA, USA)

Panel Design
We designed an NGS panel for the analysis of hereditary hearing loss using the SureDesign tool (Agilent Technologies, Santa Clara, CA, USA). The genes that were included in this panel were selected according to the prevalence reported in different studies [1,[8][9][10] choosing those with the highest prevalence. Finally, the panel included the coding regions and flanking intronic regions (+/-25 bp) of 59 genes, 35 of them associated with non-syndromic HL, and 24 genes associated with syndromic HL ( Table 1). The panel also included five deep intronic regions of the USH2A gene [11][12][13].

Library Preparation and Sequencing
The library preparation was carried out according to the Bravo NGS SureSelectQXT Automated Target Enrichment protocol (Agilent Technologies, Santa Clara, CA, USA) for Illumina Multiplexed Sequencing. Sequencing analysis was performed sequentially in batches of 16 patients. The libraries were sequenced on a MiSeq instrument with a MiSeq v2 300 cycle reagent kit (Illumina, San Diego, CA, USA).
To predict the potential effect of the variants on the splicing, we used the bioinformatic tools MaxEnt and Splice AI.
Sanger sequencing (BigDye Terminator kit v1.1, Applied Biosystems, Carlsbad, CA, USA) was carried out to validate the pathogenic and likely pathogenic point variants and to perform segregation analysis when patients' relatives were available.
To detect copy number variations (CNVs), we carried out an analysis using the DECoN v1.0.2 program [14], which is a tool that detects variants in copy number from aligned sequences based on the number of reads for each position. The CNVs obtained by this program were checked using the multiplex ligation-dependent probe amplification technique (MLPA): OTOA + STRC (P461 salsa) (MRC-Holland, Amsterdam, The Netherlands). Deletions previously described to affect the DFNB1 locus were confirmed by multiplex PCR [15]. These MLPA reagents were also performed in patients with only one pathogenic variant detected in a gene with (autosomal recessive) AR inheritance.

Results
We aimed to obtain a median read depth greater than 100×. Coverages obtained were around 150×-200×, and 98% of analyzable target regions were covered by at least 20 reads. However, some regions of 3 genes with homologous pseudogenes (ESPN, OTOA and especially STRC) were not well covered. These regions are detailed in Table S1.
We detected the variant(s) responsible for the disease in 47 out of 118 families (40%), 27 with an AR inheritance pattern, 18 with AD and 2 with an X-linked pattern ( Table 2). Detailed clinical data from the diagnosed patients are shown in Table 2.

Autosomal Recessive HL
Twenty-nine cases belonging to 27 families carried biallelic pathogenic or likely pathogenic variants associated with an autosomal recessive pattern of inheritance (Table 2A).
Twenty-six cases presented with NSHL. These were linked to GJB2/GJB6 (DFNB1) (nine cases), STRC (three cases), OTOF (three cases belonging to two families), LOXHD1 (two cases), OTOA (two cases), TMPRSS3 (two cases belonging to one family) and one case in the MYO15A, SLC26A4, OTOG, TECTA and MYO7A genes (Figure 1). Family trees for families with more than one affected patient are displayed in Figure 2. The remaining three solved cases suffered from Usher syndrome due to putative pathogenic variants in ADGRV1, CDH23 and USH2A, one family for each gene.
The most prevalent variants found were c.35del (GJB2) and del (GJB6-D13S1830), both affecting DFNB1 locus, followed by the complete deletion of the STRC gene. In all cases, the deletion of STRC was associated with mild to moderate postlingual hearing loss.
Five of the detected pathogenic variants were novel. Four of them produced a premature stop codon: three frameshift variants (c.

Autosomal Dominant HL
We identified variants responsible for the disease associated with an autosomal dominant pattern of inheritance in 25 patients belonging to 18 families (Table 2B).
Twenty-four of these patients had been referred as non-syndromic HL. Nine patients belonging to six families presented variants in MYO6, four patients from three families in TECTA, four patients from two families in COL11A2, two patients from two families in WFS1 and two patients from the same family in KCNQ4; pathogenic variants in ACTG1 and EYA4 were detected in one patient each (Table 2B and Figure 2).
One of the families linked to COL11A2 (family 38) was found to present the pathogenic variant c.4392 + 1G > A, previously described by Brunner et al. (1994) [48] as associated with Stickler syndrome without eye affectation. This family was clinically re-evaluated and re-classified as Stickler syndrome.
was associated with mild to moderate postlingual hearing loss.
Five of the detected pathogenic variants were novel. Four of them produced a premature stop codon: three frameshift variants (c.

Autosomal Dominant HL
We identified variants responsible for the disease associated with an autosomal dominant pattern of inheritance in 25 patients belonging to 18 families (Table 2B).
Twenty-four of these patients had been referred as non-syndromic HL. Nine patients belonging to six families presented variants in MYO6, four patients from three families in TECTA, four patients from two families in COL11A2, two patients from two families in WFS1 and two patients from the same family in KCNQ4; pathogenic variants in ACTG1 and EYA4 were detected in one patient each (Table 2B and Figure 2).
One of the families linked to COL11A2 (family 38) was found to present the pathogenic variant Additionally, we also detected pathogenic variants in two families with syndromic hearing loss. We found the variants responsible for the disease in one patient diagnosed with Waardenburg syndrome, presenting the variant responsible for the disease in MITF, and one patient diagnosed with BOR syndrome was found to present with the pathogenic variant in EYA1.
No prevalent pathogenic variants associated with an autosomal dominant (AD) pattern of inheritance was detected. Seven of the AD pathogenic variants identified in the present study were novel. One novel stop codon (c.1666C > T/p.(Arg556 *)) was detected in MYO6. Two splicing variants, none previously described, were detected; one of them was located at a canonical site (c.1674 + 1G > A in MYO6), and the other was located at c.1224-9del in MYO6. Furthermore, an in-frame novel duplication was found in WFS1, c.1463_1474dup/p.(Val491_Pro492insLeuIleThrVal) and two missenses variants in COL11A2 (c.1748G > A/p.(Gly583Asp)) and MYO6 (c.494T > G/p.(Leu165Arg)) ( Table 3).
The audiogram of patient 40431, harboring the c.1463_1474dup/p.(Val491_Pro492insLeuIleThrVal) variant, showed a characteristic profile with severe threshold increases for low-frequency tones ( Figure 3).

X-linked HL
Variants responsible for the disease associated with an X-linked pattern of inheritance were found in three cases belonging to two families (Table 2C). One case presented a novel missense variant in POU3F4 (recessive X-linked) and the other two cases were a boy and his mother, both carrying a novel frameshift variant in SMPX (dominant X-linked) (Figure 2 and Table 3).

Partially Diagnosed Patients
In 11 patients we detected one or several pathogenic variants in the heterozygous state in genes with an AR inheritance pattern. In seven cases we identified a pathogenic variant in only one gene: USH2A (2), GJB2 (2), STRC (1), OTOF (1) and CDH23 (1). In four patients we detected pathogenic variants in several different genes (Table 4)

X-Linked HL
Variants responsible for the disease associated with an X-linked pattern of inheritance were found in three cases belonging to two families (Table 2C). One case presented a novel missense variant in POU3F4 (recessive X-linked) and the other two cases were a boy and his mother, both carrying a novel frameshift variant in SMPX (dominant X-linked) (Figure 2 and Table 3).

Partially Diagnosed Patients
In 11 patients we detected one or several pathogenic variants in the heterozygous state in genes with an AR inheritance pattern. In seven cases we identified a pathogenic variant in only one gene: USH2A (2), GJB2 (2), STRC (1), OTOF (1) and CDH23 (1). In four patients we detected pathogenic variants in several different genes (Table 4)  [29] NSHL: non-syndromic hearing loss, USH: Usher syndrome. Novel variants are marked in bold.

Discussion
The genetic diagnosis of hereditary hearing loss is highly difficult due to its enormous underlying genetic heterogeneity (more than 120 genes described up to date), which is a reflection of the high complexity of the ear structure and organization.
In the last 10 years (from 2006 to 2016), the genetic analysis of patients with hearing loss in our tertiary hospital was restricted to detect the most frequent pathogenic variants responsible for hereditary sensorineural hearing loss in Spain, specifically the complete coding sequence of the GJB2 gene, the deletions D13S1830 and delD13S1854 in the GJB6 gene and the OTOF p.Q829X variants. The implementation of our custom NGS panel containing 59 HL genes improved the management of our patients, as it has allowed us to detect putative pathogenic variants in 26 different genes. Furthermore, we have been able to genetically diagnose syndromic cases suffering from Deafness Infertility syndrome, Usher syndrome, Stickler syndrome, Waardenburg syndrome and BOR syndrome.
However, pathogenic variants in a few genes still explain a great number of hearing loss cases. The main example is GJB2, encoding connexin 26. Pathogenic variants in this gene are the most common cause of hereditary hearing loss in many populations [63]. In the present work, biallelic variants in GJB2, together with GJB6 (DFNB1 locus), were responsible for the disease in nine families with AR inheritance, followed by pathogenic variants in STRC (three AR families). Regarding AD inheritance families, heterozygous pathogenic variants in MYO6 were found in six families, followed by pathogenic variants in TECTA (four AD families). An additional patient presented a homozygous AR pathogenic variant in TECTA. All inheritance patterns have been described for HL: recessive, dominant, X-linked and mitochondrial. In some genes (like MYO6, TECTA or ESPN), a group of variants follow a dominant inheritance pattern, whereas others follow a recessive inheritance pattern, complicating the interpretation of genetic analysis [64]. Another feature that complicates the genetic studies of HL is the existence of some pseudogenes with high homology to some prevalent genes (like STRC, OTOA or ESPN). In the panel design, we tried to include some extra probes for the regions of these genes showing high homology with their pseudogenes, in addition to the default probes generated by SureDesign. However, low coverage was still obtained, and those point variants suspected to be pathogenic had to be confirmed by Sanger sequencing using primers specifically designed to hybridize only with the gene, not the pseudogene [65,66].
When a CNV affecting STRC or OTOA was suspected after DECoN v1.0.2 analysis, its presence was confirmed by MLPA using SALSA P461 (MRC Holland).
Several pathogenic variants identified in this study are reported in a large number of studies, suggesting a high prevalence. The pathogenic variant in OTOF c.2485C > T/p.(Gln829 *) is the third most frequent in the Spanish population that causes prelingual hearing loss [67], and STRC deletions are the second most frequent cause of mild-to-moderate hearing loss after the DFNB1 locus [68]. The variant c.1540C > A/p.(Gln514Lys) is the most frequent variant in SLC26A4 in the Spanish population, described in more than 36 Spanish families to date [69]. Furthermore, the pathogenic change c.9799T > C/p.(Cys3267Arg) in USH2A is one of the most frequent variants in the Spanish population, specifically the third most common cause of Usher syndrome [70,71]. Finally, the pathogenic variant c.5668C T/p.(Arg1890Cys) that affects the TECTA gene has been described in some families from Spain, America and The Netherlands. In the most unrelated families, patients present the same haplotype, which suggests that the variant is derived from a common ancestor (founder effect) [46]. Nowadays, all known HL genes can be simultaneously analyzed thanks to the technological development of NGS. Even so, the rate of genetic diagnosis using NGS in patients with hearing loss varies around 40-60% [8,29,64,[72][73][74][75][76] depending on many factors: the degree of HL (profound, severe, moderate), age of HL onset, the existence of family history, the ethnic origin or the number of genes contained in the NGS panel. The highest rates have usually been obtained for patients with a positive family history or when the HL was congenital and symmetric [8]. In the present work, the global diagnostic yield was 40%. This is a satisfactory yield, since our custom NGS panel included a limited number of genes (59), and the exclusion criteria for the genetic testing was very lax. Thus, the analyzed patient sample was very heterogeneous, including all types of sensorineural/mixed hearing loss (congenital, prelingual and postlingual; mild, moderate, severe and profound; and stable and progressive) with ages ranging from 0 to 61 years.

Novel VUS/Likely Pathogenic Variants
The development of NGS has revolutionized the field of genetic diagnosis, especially in extremely genetically heterogeneous diseases, such as hereditary HL. However, an elevated number of genetic variants of uncertain clinical significance (VUS) has been detected using this technology [77]. Variants predicted to generate direct stop codons or changes in the reading frame of the proteins and variants located at canonical splice sites (+/-1 and +/-2 positions of introns) are usually classified as pathological for proteins for which loss of function is reported as cause of the disease. However, the interpretation of missense, isocoding and intronic variants located out of canonical splice sites is more complex, and many times these variants remain classified as VUS. In these cases, bioinformatics predictions, segregation analyses or functional studies are required to infer the pathological character of these variants. In our study, a lot of a priori VUS variants were detected, and only seven of them were classified as likely pathogenic based upon bioinformatics predictions and/or segregation analyses: four missense, one intronic variant and one in-frame duplication.
Missense variants: The c.235T > C/p.(Cys79Arg) change in TMPRSS3 was not found in gnomAD exomes/genomes, and 11 computational programs predicted it as pathogenic in Varsome. Furthermore, it was found in trans with other previously described pathogenic variants in the TMPRSS3 (see Table 2 and Figure 2). The MYO6 (c.494T > G/p.(Leu165Arg)) variant was found in patient 41268. He was referred to as AD non-syndromic hearing loss, being her mother, her sister and her sister´s son were also affected. Although this variant was classified as VUS following the ACMG guidelines, we should not rule it out since it was not found in healthy control databases, had a high conservation score and showed a pathogenic computational verdict based on 13 pathogenic predictions. The COL11A2 (c.1748G > A/p.(Gly583Asp)) change was found in a patient and his affected father (family 37). This variant was not present in healthy population databases, and it showed pathogenic predictions in the Alissa Interpret program based on MutationTaster, MutationAssessor, LRT, PolyPhen2 and PROVEAN. Finally, the c.977T > C/p.(Phe326Ser) (POU3F4) variant was found in a boy with mixed hearing loss and cochlear malformations (bilateral corkscrew cochlea, incomplete splitting of turns, absence of meatus and stapes fixation); clinical characteristics of hearing loss are linked to this gene. Furthermore, this change was absent in healthy controls databases, and it showed a pathogenic computational verdict based on 10 pathogenic predictions in Varsome.
Intronic variant: The c.1224-9del variant in MYO6 was found in a patient with an AD pattern of inheritance in her family, given that her mother was also affected. Unfortunately, the patient´s mother refused to collaborate in the genetic study. This variant was not found in healthy control population databases, and the MaxEnt bioinformatic tool predicted the loss of the wild-type acceptor site. This variant was classified as VUS following the AMCG, but we consider that MYO6 c.1224-9del could be a good candidate, and functional studies at the RNA level would be necessary to definitively confirm or discard the pathologic effect of this novel variant.
In frame duplication: The WFS1 in-frame duplication (c.1463_1474dup/p. (Val491_Pro492insLeuIleThrVal)) was detected in a patient presenting HL also in a cousin and her son, but they did not collaborate in the study. This change was classified as VUS according to the ACMG. However, we consider that it is necessary to take this variant into account since it is not described in the population databases, has an acceptable value of conservation and, following the criteria of the ACMG, if it had been possible to show that the variant segregates correctly within the family, the WFS1 c.1463_1474dup/p.(Val491_Pro492insLeuIleThrVal) variant would be directly classified as likely pathogenic. Additionally, the clinical phenotype of this patient is similar to other patients with pathogenic variants in WFS1, showing a characteristic audiogram with low frequencies more affected (Figure 3).

Patients with Pathogenic Variants in Two Different Genes
NGS panels allow the simultaneous analysis of a great number of genes, and, sometimes, pathogenic variants in different genes are found in the same patient.
In the present work, the 37439 patient presented the AR c.101T > C/p. The 36163 patient was found to carry two different heterozygous novel pathogenic variants in two different genes: c.1674 + 1G > A in MYO6 and c.2467C > T/p.(Gln823 *) in ESPN. Segregation analysis in this family showed that the affected father also carried the variant in MYO6, whereas the healthy mother carried the variant in ESPN. From these results it can be deduced that the variant in MYO6 is responsible for AD hearing loss, whereas the ESPN variant presents an AR inheritance pattern ( Figure 2).
The 29272 and the 41950 patients from the same family carried two different previously described AD pathogenic variants in two different genes: c.2751dup/p.(Gln918Thrfs *24) in MYO6 and c.2230G > A/p.(Asp744Asn) in ESPN. These two patients belong to a large family with more affected members, but these were geographically dispersed, and it was not possible to segregate these two variants with all family members in order to definitely elucidate the genetic basis and the inheritance pattern of HL in this case ( Figure 2).
Finally, the likely pathogenic novel MYO6 (c.494T > G/p.(Leu165Arg)) variant was found in patient 41268, referred to as AD non-syndromic hearing loss. Furthermore, two previously described AR pathogenic variants in MYO7A (c.1997G > A/p.(Arg666Gln) and c.3527G > A/p.(Ser1176Asn)) were found in this patient. Segregation analysis would be necessary to definitely elucidate the genetic basis and the inheritance pattern of HL in this family and to offer accurate genetic reproductive genetic counseling.

Syndromic Cases
Most patients included in this study suffered from non-syndromic hearing loss, but eight cases were referred as syndromic: four patients with Usher syndrome (USH), two Waardenburg syndrome (WS) patients and two branchio-oto-renal syndrome (BOR) patients. We could find the variants responsible for the disease in five of them (Table 2A,B).
The patient 35238 and his mother (42783) were referred as NSHL, but they were found to carry a pathogenic variant in COL11A2: c.4392 + 1G > A. This variant had been previously reported by Brunner et al. (1994) [48] associated with Stickler syndrome without eye affectation. These patients were clinically re-evaluated, and both presented with osteoarticular problems and flattened facial profiles (Table 2B). Thus, this family was re-classified as Stickler syndrome.
Three unrelated cases with bilateral, symmetrical, postlingual, moderate and stable HL (33416, 37112 and 37439) presented biallelic contiguous-gene deletions at chromosome 15q15.3 that included both CATSPER2 and STRC. This deletion causes deafness-infertility syndrome (DIS) in males due to CATSPER haploinsufficiency results in sperm abnormalities [78]. The patient 37112 was a male of 4 years old. Thus, the patient´s parents were informed that their son will be infertile in adulthood.
In another case (12228), a complex rearrangement involving STRC and CATSPER2 was detected (Table 3 and Figure 4). DECoN analysis using NGS data showed a partial deletion involving exons 1-15 of STRC. However, based on coverage data from NGS, we could not differentiate between STRC and pSTRC. Thus, an MLPA analysis was performed using P-461 SALSA (MRC Holland). This SALSA includes specific probes only for exons 19, 24-25 of STRC and also some specific probes for some exons of CATSPER2. MLPA results showed a partial deletion affecting exons 23, 24 and 25 of STRC (chr15:41680256-41682666), whereas STRC exon 19 showed a normal dosage (chr15:41684606-41684940). However, chromosome coordinates chr15:41711482-41728076 corresponding to CATSPER2 showed again a ratio of 0.5. Segregation analysis would be helpful in this case to find out if this complex rearrangement is carried in the same chromosome or if there are two different deletions affecting the 15q15.3 locus, located in different alleles.

Conclusions
A large number of genes has been associated with HL, but still many cases remain unexplained. Novel HL genes are expected to be discovered and also genetic variants affecting regulatory regions of the genome, which are currently not screened in diagnosis genetic testing. Furthermore, the possibility of multigenic inheritance patterns must be explored in the near future [79].
Nowadays, a huge number of DNA variants are being detected in countless genetic diagnostic laboratories around the world, and a non-negligible number of them are possibly being misinterpreted. It is necessary to share this information with the scientific community and to establish close collaborations to interpret the functional implications of DNA variability. Working altogether, we will be able to decipher the secrets that we still ignore about the human genome.
Supplementary Materials: The following are available online at www.mdpi.com/xxx/s1, Table S1: Regions of the panel design with a poor coverage.

Conclusions
A large number of genes has been associated with HL, but still many cases remain unexplained. Novel HL genes are expected to be discovered and also genetic variants affecting regulatory regions of the genome, which are currently not screened in diagnosis genetic testing. Furthermore, the possibility of multigenic inheritance patterns must be explored in the near future [79].
Nowadays, a huge number of DNA variants are being detected in countless genetic diagnostic laboratories around the world, and a non-negligible number of them are possibly being misinterpreted. It is necessary to share this information with the scientific community and to establish close collaborations to interpret the functional implications of DNA variability. Working altogether, we will be able to decipher the secrets that we still ignore about the human genome.

Conflicts of Interest:
The authors declare no conflict of interest.