Identification of alternative splicing in WFS1 associated with low-frequency hearing loss in the common marmoset

Summary

Approximately 200 genes have been identified as causative in hereditary hearing loss. Genetic testing is increasingly important, not only for accurate diagnosis but also for predicting audiometric profiles, prognoses, and potential syndromic features. Hereditary hearing loss can be syndromic or nonsyndromic, with nonsyndromic forms further classified by inheritance: autosomal-dominant or autosomal-recessive. In autosomal-dominant cases, three pathological mechanisms—haploinsufficiency, dominant-negative effects, and gain of function—are often implicated. Moreover, specific genes correlate with distinct audiometric patterns: WFS1 variants typically cause low-frequency hearing loss, whereas KCNQ4 and POU4F3 variants are linked to high-frequency loss. To investigate the underlying mechanisms of these frequency-dependent patterns, gene expression across cochlear turns was compared in mice, but interpretations of the results were limited because of inherent structural differences between rodent and primate cochleae. Therefore, the common marmoset (Callithrix jacchus), which offers closer anatomical and functional similarity to human cochleae, was utilized herein as an improved model. Using RNA sequencing (RNA-seq) across cochlear turns of common marmosets, the present study aimed to uncover gene expression and alternative splicing patterns that may explain tonotopic manifestations in hereditary hearing loss, including those caused by WFS1 variants, the present study being one such using common marmoset cochlear RNA-seq data, and these findings are highly valuable for genetic diagnosis and the development of gene therapies.

Frequency-specific hearing loss in autosomal-dominant deafness is linked to distinct genes, such as WFS1 for low-frequency loss. RNA sequencing of cochlear turns in the common marmoset reveals alternative WFS1 splicing patterns that may underlie tonotopic vulnerability.

Introduction

According to the Hereditary Hearing Loss homepage (https://hereditaryhearingloss.org/), approximately 200 causative genes have been identified for hereditary hearing loss as of August 2025. Genetic testing has emerged as increasingly important clinically, not only for diagnosis but also for the prediction of audiometric configurations, prognosis, and potential syndromic features. Hereditary hearing loss can be classified as either syndromic or nonsyndromic, depending on the causative gene. Nonsyndromic hearing loss is further categorized by inheritance patterns into autosomal-dominant or autosomal-recessive forms. Among these, three principal mechanisms have been proposed for the pathogenesis of autosomal-dominant nonsyndromic hearing loss (ADNSHL): haploinsufficiency, dominant-negative effects, and gain of function. In ADNSHL, the causative genes are often associated with characteristic audiometric profiles. For instance, pathogenic variants in WFS1 typically result in low-frequency hearing loss,¹ whereas those of KCNQ4² and POU4F3³ are generally associated with high-frequency hearing loss. Despite these well-known associations, the mechanisms underlying these frequency-specific phenotypes remain unclear.

We previously attempted to elucidate the mechanisms underlying the characteristic audiometric patterns observed in various forms of hereditary hearing loss by comparing gene-expression levels across cochlear turns in mice.⁴ However, these analyses failed to fully explain the basis of frequency-specific hearing-loss patterns. One possible reason for this is the morphological differences in cochlear structures between rodents and primates.

The common marmoset (Callithrix jacchus) has recently attracted attention as a small nonhuman primate model that can bridge the translational gap between mice and humans. In auditory research, marmosets have been used widely for objective auditory assessments⁵ and analyses of protein-expression patterns in the cochlea,⁶ highlighting notable similarities with human inner ear structures⁷ and functions.⁸ While the tonotopic organization of the cochlea in the common marmoset has not yet been fully established, its auditory frequency range (approximately 125 Hz–36 kHz) is more comparable to that of humans (20 Hz–20 kHz) than that of rodents (1 kHz–100 kHz).⁹ Furthermore, studies using fetal common marmosets have demonstrated their applicability as developmental models of the human inner ear.^10,11

We hypothesized that differential gene expression and/or alternative splicing across cochlear turns contribute to the pathogenesis of frequency-dependent hearing loss in ADNSHL. Given the anatomical and morphological similarities between human and common marmoset cochlea,^7,8 we employed a marmoset model to investigate the spatial expression of known deafness-related genes.

By generating detailed profiles of gene expression and alternative splicing across cochlear turns, we aimed to identify the patterns potentially underlying the tonotopic manifestation of hereditary hearing loss.

In the present study, we elucidate one of the mechanisms underlying characteristic audiometric patterns in ADNSHL by analyzing gene expression and alternative splicing using RNA sequencing (RNA-seq) of the common marmoset cochlea.

Material and methods

Sample collection

Three adult common marmosets (C. jacchus) were obtained from the Central Institute for Experimental Medicine and Life Science (Kawasaki, Kanagawa, Japan). Subject 1 was a 12.7-year-old female, subject 2 was a 5.6-year-old male, and subject 3 was a 4.5-year-old female. None of the animals had any known genetic modifications related to deafness. Pre-anesthesia was administered via an intramuscular injection of ketamine (50 mg/kg) and xylazine (4 mg/kg), followed by deep anesthesia using isoflurane inhalation. Exsanguination was performed via the abdominal aorta, and death was confirmed by loss of palpebral reflex and respiratory arrest. Decapitation was then performed, and the tympanic membrane was perforated with forceps, followed by dislocation of the ossicles to circulate RNAlater in the cochlea. The heads were immediately immersed in RNAlater solution (Ambion, Austin, TX) and stored at −80°C. The left cochlea was dissected from the temporal bone in RNAlater solution, and the bony cochlear wall was removed to extract the membranous labyrinth. The cochlear membranous labyrinth was divided into apical, middle, and basal turns according to angular distance from the apex: 0°–360° (apex), 360°–720° (middle), and >720° (base). Each region was treated as a separate sample under a microscope (Figure 1).

Figure 1.

Sample collection procedure

(A) The malleus (m) and stapes are visible after removal of the tympanic membrane.

(B) The stapes (s), facial nerve (fn), and cochlea (c) are visible after removal of the malleus and incus.

(C) The membranous labyrinth is exposed after removal of the cochlear bony wall.

(D) The membranous labyrinth is dissected and separated into apical, middle, and basal turns.

The animal study protocol was approved by the institutional review board of the Central Institute for Experimental Medicine and Life Science (approval no. 20061A).

mRNA extraction and cDNA synthesis

Total RNA was extracted using an RNeasy Mini Kit (QIAGEN, Hilden, Germany), according to the manufacturer’s protocol. β-Mercaptoethanol was added to Buffer RLT before use. Cells were lysed using 0.5-mm zirconia beads with a bead crusher (Micro Smash MS-100; Tomy Digital Biology, Tokyo, Japan) at 3,000 rpm for 30 s. RNA quantification and integrity assessments were performed using an Agilent 2100 Bioanalyzer and RNA Pico 6000 Kit (Agilent Technologies, Waldbronn, Germany). cDNA synthesis and library preparation were conducted using the SMART-Seq HT PLUS Kit (Takara Bio, Shiga, Japan) following the manufacturer’s protocol. Reverse transcription and second-strand synthesis were performed with 12–13 cycles to ensure sufficient yield. The resulting cDNA was subjected to quality control using the Agilent 2100 Bioanalyzer.

RNA-seq

Long-read sequencing

Isothermal single-primer polymerase chain reaction (PCR) was used to amplify long reads. Specific barcode adapters from the Oxford Nanopore Native Barcoding Kit 96 V14 (SQK-NBD114.96) were used for the 15-cycle PCR. A one-dimensional sequencing adapter was ligated to the cDNA, which was then loaded onto a FLOPRO002 R10.4.1 flow cell (Oxford Nanopore Technologies, Oxford, UK) using a PromethION sequencer. Data were acquired using MinKNOW software and exported in FASTQ format. Reads were mapped to the Ensembl 113 marmoset reference genome (version 113.1) using Minimap2, and the data were used for the analysis of alternative splicing events. The long-read RNA-seq data have been lodged in the DNA Data Bank of Japan (DDBJ) Sequence Read Archive as run accession numbers DRR905161–DRR905169.

Short-read sequencing

Libraries were constructed using the Smart-seq HT Plus Kit and Unique Dual Index Kit, followed by quality control using the Agilent 2100 Bioanalyzer with a DNA High Sensitivity Kit (Agilent Technologies). Sequencing was outsourced to Novogene (Beijing, China) and performed on an Illumina NovaSeq platform with 2 × 150-bp paired-end (PE150) chemistry, yielding 200 Gbp/8 samples. The resulting reads were mapped using STAR Aligner (Dobin lab, Cold Spring Harbor Laboratory, Cold Spring Harbor, NY). Next, transcript abundance was estimated using StringTie (Pertea lab, Johns Hopkins University, Baltimore, MD) and expressed as transcripts per million. The short-read RNA-seq data have been lodged in the DDBJ Sequence Read Archive as run accession numbers DRR905170–DRR905177.

Gene-expression analysis

Short-read RNA-seq data were obtained for each cochlear turn (apex, middle, and base) from three animals (nine samples in total). Because of the low concentration in one of the basal turn samples, sequencing could not be performed. A gene was considered differentially expressed if there was a greater than 2-fold difference between the apical and basal turns across all samples. A 2-fold-change threshold was adopted as a biological filtering criterion to identify candidate genes exhibiting robust tonotopic gradients, rather than as a measure of statistical significance. Similar fold change cutoffs have been used widely in exploratory transcriptomic studies aimed at detecting spatially regulated gene expression patterns.¹²

Alternative splicing analysis

Long- and short-read RNA-seq data were visualized and analyzed using Sashimi plots in Integrative Genomics Viewer (Broad Institute, Massachusetts Institute of Technology and Harvard University, Boston, MA). Alternative splicing was considered significant if a splice junction had a read coverage of at least 5% of the most abundant junction for each gene. Previously unreported isoforms that differ from the known Ensembl references have also been reported. Exon-level comparisons were made with known human variants using the University of California, Santa Cruz Genome Browser.

Quantitative PCR

To validate the alternative splicing events of WFS1 identified by RNA-seq, quantitative PCR (qPCR) was performed using Power SYBR Green PCR Master Mix (Thermo Fisher Scientific, Waltham, MA), according to the manufacturer’s instructions. GAPDH and ACTB were used as an internal control. The following primers were used: GAPDH forward, 5′-GCACCGTCAAGGCTGAGAAC-3′; GAPDH reverse, 5′-TGGTGAAGACGCCAGTGGA-3′; ACTB forward, 5′-GATGGTGGGCATGGGTCAGAA-3′; ACTB reverse, 5′-AGCCACACGCAGCTCGTTGT-3′.

For WFS1, the known (WFS1_Known) and previously unreported (WFS1_Identified) splice variants differed in exon 8. Primers were designed to span the exon 7 to exon 8 splice junctions using Primer3 Plus (https://www.primer3plus.com). The primer sequences are listed in Table S1. Samples were analyzed per cochlear turn. Relative expression was calculated using the ΔΔCt method, where the Ct value of each splicing variant was first normalized to GAPDH (ΔCt) and then compared to the ΔCt of the apical-turn sample (sample 2) to derive ΔΔCt values.

Results

Gene expression in the common marmoset cochlea

Scatterplot analyses were conducted for apical, middle, and basal cochlear turns to assess the technical reliability of cochlear dissection and RNA extraction, and to estimate global changes in gene expression. Gene-expression patterns were highly similar among the three regions, with most genes showing less than a 2-fold difference in expression (Figure 2). Short-read RNA-seq was primarily used to compare the expression levels between the apical and basal turns. Of the 199 genes listed as causative for hereditary hearing loss on the Hereditary Hearing Loss website, three genes showed greater than 2-fold expression differences between the apical and basal turns; we defined this threshold as noteworthy. The differentially expressed genes identified through short-read sequencing were COCH (DFNA9/DFNB110), ESRRB (DFNB35), and S1PR2 (DFNB68) (Table 1).

Figure 2.

Comparison of gene-expression patterns between cochlear turns

The horizontal and vertical axes represent gene expression levels in different cochlear turns. Most genes showed no significant differences in expression between turns.

(A) Horizontal axis: apical turn; vertical axis: middle turn.

(B) Horizontal axis: middle turn; vertical axis: basal turn.

(C) Horizontal axis: apical turn; vertical axis: basal turn.

Table 1.

The differentially expressed genes identified through short-read sequencing

Gene symbol	Gene name	Deafness-causing locus	TPM (Transcripts Per Million) apex	Middle	Base	Apex/base
COCH	cochlin	DFNA9/DNB110	19,746.186	13,068.797	8,497.298	2.324
ESRRB	estrogen-related receptor β	DFNB35	12.737	14.052	25.671	0.496
S1PR2	sphingosine-1-phosphate receptor 2	DFNB68	0.061	0.147	0.210	0.289

Alternative splicing in the common marmoset cochlea

Alternative splicing was analyzed using both long- and short-read sequencing data. Among the known hearing loss-associated genes, 56 of 199 showed multiple alternative splicing isoforms. Of these, 23 out of 78 ADNSHL-related genes exhibited multiple isoforms. These were KCNQ4 (DFNA2), CEACAM16 (DFNA4B), GSDME (DFNA5), WFS1 (DFNA6), EYA4 (DFNA10), MYO7A (DFNA11), MYO6 (DFNA22), REST (DFNA27), NLRP3 (DFNA34), TMC1 (DFNA36), TNC (DFNA56), DIABLO (DFNA64), TBC1D24 (DFNA65), CD164 (DFNA66), DMXL2 (DFNA71), SLC44A4 (DFNA72), PDE1C (DFNA74), PLS1 (DFNA76), ELMOD3 (DFNA81), MAP1B (DFNA83), USP48 (DFNA85), THOC1 (DFNA86), and PI4KB (DFNA87) (Figure 3).

Figure 3.

Schematic of the splicing isoforms identified among causative genes for ADNSHL

The red squares indicate exons that have not been previously reported. The gray dashed squares represent known exons that were absent in the isoforms identified from the cochlea in the present study. The double slashes denote omitted exons that are shared by both previously described isoforms and the isoforms identified in this study. For SLC44A4, four alternative patterns were observed between exons 1–5, and five patterns between exons 8–11, suggesting the potential presence of up to 4 × 5 = 20 distinct isoforms. Similarly, for THOC1, three alternative patterns were found between exons 1 and 2, three between exons 4–6, and two between exons 15 and 16, indicating up to 3 × 3 × 2 = 18 possible isoforms. Because of the large number of potential isoforms for SLC44A4 and THOC1, their exon structures are presented in a simplified format. ADNSHL, autosomal-dominant nonsyndromic hearing loss.

Interestingly, we identified WFS1 (DFNA6) with alternative splicing for exons in which pathogenic variants have been reported in humans. Two alternative splice variants of WFS1 were identified herein. One was previously identified in the common marmoset (WFS1_Known), whereas the other represented an isoform identified in this study (WFS1_Identified) with skipping exon 8. WFS1_Identified was absent in the apical turn but was present in the middle and basal turns. According to long-read sequencing data, the proportion of WFS1_Identified among the total WFS1 isoforms was 8.3% and 11.6% in the middle and basal turns, respectively (Figure 4A). These findings were also observed in the analysis using short-read sequencing (Figure 4B).

Figure 4.

Analysis of WFS1

(A) Sashimi plot analysis of WFS1 using long-read sequencing. Only the known isoform (WFS1_Known) was detected in the apical turn. Both WFS1_Known and WFS1_Identified were observed in the middle and basal turns. The ratios of WFS1_Identified to WFS1_Known were 16/192 (8.3%) and 17/146 (11.6%) in the middle and basal turns, respectively.

(B) Sashimi plot analysis of WFS1 using short-read sequencing. In the short-read sequencing analysis as well, previously unreported isoforms were detected predominantly in the middle and basal cochlear turns, similar to the findings with long-read sequencing.

(C) Quantitative PCR analysis of WFS1_Known and WFS1_Identified. Relative expression was calculated using the ΔΔCt method, where the Ct value of each splicing variant was first normalized to GAPDH (ΔCt) and then compared to the ΔCt of the apical-turn sample (sample 2) to derive ΔΔCt values. For each sample, ΔΔCt values were converted to 2^−ΔΔCt values, and the mean 2^−ΔΔCt value for each cochlear turn was calculated. For both WFS1_Known and WFS1_Identified, the mean 2^−ΔΔCt value of the apical turn was used as the reference (1.0). The vertical axis indicates how many times greater the mean 2^−ΔΔCt values of the middle and basal turns are relative to those of the apical turn. The ΔΔCt method revealed no significant difference in the expression of WFS1_Known across cochlear turns. In contrast, WFS1_Identified expression was more than 8-fold higher in the middle and basal turns than in the apical turn.

(D) Schematic representation of WFS1 splicing isoforms. Known hearing loss-associated variants in humans cluster within exon 8. As the WFS1 isoform identified in this study evades mutations in exon 8, its transcripts may preserve hearing in the middle and basal turns of the cochlea.

qPCR

qPCR was performed to validate the expression levels of the two WFS1 isoforms (WFS1_Known and WFS1_Identified). As determined using the ΔΔCt method, WFS1_Known showed no significant differences in expression among the apical, middle, and basal turns. In contrast, WFS1_Identified exhibited a higher expression in the middle and basal turns than in the apical turn (Figure 4C).

Discussion

In the present study, we aimed to elucidate the mechanisms underlying the onset of hearing loss associated with characteristic audiometric patterns caused by specific genetic variants. To address this, we performed RNA-seq analysis using a common marmoset (C. jacchus), focusing on approximately 200 genes that were previously reported to cause hereditary hearing loss. Three main mechanisms explain the pathology of ADNSHL: haploinsufficiency, dominant-negative effects, and gain of function. Haploinsufficiency refers to a condition in which the loss or dysfunction of one allele results in insufficient protein production for maintaining normal function. Disorders caused by haploinsufficiency arise when a loss-of-function variant in one allele downregulates expression below the threshold required to maintain function.¹³ In contrast, dominant-negative effects occur when a mutant protein produced from the variant allele interferes with the function of the wild-type protein, regardless of its expression level. Gain-of-function effects confer a distinct biological activity or enhance an existing function of a gene or its product. These effects are not necessarily associated with expression levels. For ADNSHL genes exhibiting haploinsufficiency, we hypothesized that a gradient in gene expression between cochlear turns means that regions with lower expression may fall below the functional threshold upon variation, resulting in region-specific hearing loss. In the absence of an expression gradient, the presence or absence of alternative splicing isoforms may contribute to the preservation or impairment of function. Alternative splicing has been implicated in the pathogenesis of human diseases such as cancer and cardiovascular disorders.^14,15

In our analysis of ADNSHL-related genes, only COCH showed a significantly higher expression in the apical turn than in the basal turn. However, because COCH-related hearing loss is attributed to dominant-negative effects or gain-of-function mechanisms,^16,17,18 this difference in expression is unlikely to be related to its pathogenesis.

Alternative splicing analysis using Integrated Genome Viewer revealed that the exonic locations of the splicing events matched previously reported pathogenic human variants of WFS1. For WFS1, which is associated with a distinctive low-frequency hearing loss pattern, we identified an alternative splice variant (WFS1_Identified) present in the middle and basal turns but absent from the apical turn. Both long- and short-read sequencing revealed that the proportion of this isoform at the basal turn was higher than that in the middle turn (Figures 4A and 4B). qPCR analysis confirmed that WFS1_Identified expression was significantly higher in the middle and basal turns than in the apical turn (Figure 4C), whereas WFS1_Known expression showed no significant variation across turns. The results of the long-read sequencing and qPCR were logically consistent. Even in the presence of a variant affecting exon 8, the transcript produced by the previously unreported isoform in the middle and basal turns could maintain auditory function in the mid- to high-frequency range (Figure 4D). Additionally, auditory function might not be preserved in the low-frequency region lacking this isoform. This finding supports the clinical observation that WFS1 variants are associated with low-frequency hearing-loss patterns.¹ Therefore, the characteristic audiometric profile of WFS1 could be governed by transcript diversity generated through alternative splicing.

The complete marmoset genome was published in 2014. Using this as a reference, our mapping analysis revealed differences from the canonical reference genome for many deafness-related genes. This suggests the possibility of using organ-specific reference sequences and underscores the utility of our dataset as a cochlear reference.

Among the autosomal-recessive nonsyndromic hearing loss (ARNSHL) genes, ESRRB and S1PR2 genes showed significant differences in expression between cochlear turns; however, distinct audiometric patterns specific to ESRRB- or S1PR2-related hearing loss have not been previously reported. The expression levels of ARNSHL genes are generally not linked to pathogenesis, but an exception has been reported for the SLC26A4 variant c.2029C>T (p.Trp677Arg). This variant reduces mRNA and protein levels, probably because of aberrant splicing or protein instability caused by its proximity to the splice site.¹⁹ These data suggest that regions of the cochlea with lower expression may fail to maintain function in such variants, consequently contributing to the observed audiometric pattern and explaining the unique profile associated with ARNSHL.

We also compared our marmoset RNA-seq data with those from previous mouse studies conducted at our institution.⁴ In mice, four ADNSHL genes (Pou4f3, Slc17a8, Tmc1, and Crym) and nine ARNSHL genes (Otof, Strc, Ush1c, Pcdh15, Grxcr1, Dfnb59, Slc26a5, Lhfpl5, and Ptprq) exhibited greater than 2-fold differences in gene expression between cochlear turns. These genes showed higher expression levels in the apical turns than in the basal turns. In contrast, no gradient in gene expression across cochlear turns was observed for any of the genes in common marmosets. In terms of alternative splicing, the only gene with an identical splicing event between mice and marmosets was Map1b. This may highlight interspecies differences between rodents and primates and suggest the limitations of mouse models in accurately recapitulating hereditary hearing loss in humans.

In the current marmoset dataset, WFS1 was the only ADNSHL gene, the characteristic audiometric phenotype of which could be linked to alternative splicing, whereas no such evidence was observed for other genes. This may also indicate the limitations of using marmosets as an experimental model for elucidating the mechanisms underlying all forms of hereditary hearing loss. As a limitation of this study, the sample size in this study was relatively small, with variation in age and sex among individuals. The potential effects of these factors on cochlear gene expression and alternative splicing cannot be excluded completely. It should also be noted that transcript-level findings do not exclude the involvement of post-transcriptional regulatory mechanisms such as regional differences in translation efficiency or post-translational modifications. Furthermore, the functional relevance of WFS1_Identified at the protein level remains to be clarified, as isoform-specific antibodies or proteomic tools are needed to distinguish the encoded products. Future studies integrating spatial proteomics and functional assays are essential for fully elucidating the physiological significance of this isoform.

In conclusion, we successfully collected foundational data to investigate the mechanisms underlying audiometric phenotypes of hereditary hearing loss. Our findings help explain the basis of low-frequency hearing loss patterns associated with WFS1 variants. Understanding the mechanisms underlying the characteristic audiometric patterns of hereditary hearing loss is highly valuable not only for genetic diagnosis but also for the development of future gene therapies. Future studies should use these insights to explore the mechanisms underlying audiometric diversity in hereditary hearing loss.

Data and code availability

The RNA-seq datasets generated in this study have been deposited in the DDBJ Sequence Read Archive under the following run accession numbers: DRR905161–DRR905177. Sequencing reads were aligned to the reference genome using STAR, and transcript abundance was quantified using StringTie, as described in material and methods.

Acknowledgments

This research was funded by a Grant in Aid for Scientific Research (C) from the Japan Society for the Promotion of Science (24K12644 to H.Y.). We are profoundly grateful to Erika Sasaki and Keisuke Mukasa, from the Department of Biological Functions and Regulation, Central Institute for Experimental Medicine and Life Sciences, who graciously provided the common marmosets for this study. We thank Editage (www.editage.com) for English-language editing.

Author contributions

S.Y., formal analysis, investigation, writing – original draft, and writing – review and editing; H.Y., conceptualization, investigation, methodology, and writing – review and editing; S.-y.N., data curation, formal analysis, investigation, methodology, and writing – review & editing; E.S., sample preparation and writing – review & editing; K.M., sample preparation and writing – review & editing; S.-i.U., investigation and writing – review & editing; Y.T., conceptualization, investigation, and writing – review & editing.

Declaration of interests

The authors declare no competing interests.

Declaration of generative AI and AI-assisted technologies in the writing process

The authors declare that no generative AI or AI-assisted technologies were used in the writing of this manuscript.