We utilized the pseudo-bulk and single-cell gene expression profiles to compare the four HC types. Since adult IHC and OHC transcriptomes have been compared extensively in the past (Li et al., 2018; Ranum et al., 2019), we focused our analyses on the differences between cochlear and vestibular HCs, as well as between type I and II HCs. Principal component analysis (PCA) was used to denote variance across the four types of HCs. The first principal component (PC1) shows that the most dramatic differences are tissue-based—between cochlear and vestibular HCs (Figure 2A). The distribution across PC2 indicates that the similarity between type I and II HCs is greater than the similarity between IHCs and OHCs, suggesting more homogeneity among vestibular HCs. Gene expression profiles of single HCs were also used to analyze similarities among the four different HC types (Figure 2B). Despite heterogeneity found among individual HCs of each type, three-dimensional PCA visualization of single cells draws similar conclusions to the pseudo-bulk PCA, suggesting higher transcriptomic similarity between vestibular HCs in contrast to cochlear HCs.

Next, we examined the number of genes that are shared and unique among the four types of HCs based on pseudo-bulk expression profiles (Figure 2C). Although approximately 71% of the detected genes are shared among all four types of HCs, the smaller proportion of unique genes in each HC population may underlie their specific biological identities. We performed pairwise differentially expressed gene (DEG) analyses between within-tissue HC types, as well as between cochlear and vestibular HCs (Figure 2D). DEGs were defined as those with an expression level above 0 and a minimum of twofold change (log₂ ≥1) between the two cell populations with statistical significance of p ≤ 0.01.

DEGs between adult IHCs and OHCs have been analyzed before using cell type-specific microarray and bulk RNA-seq techniques (Li et al., 2018; Liu et al., 2014). Here, our comparison revealed differential enrichment of 154 and 123 genes in IHCs and OHCs, respectively (Figure 2D). We show expression of previously characterized DEGs in IHCs (e.g., Otof and Slc17a8) and OHCs (e.g., Ocm, Slc26a5, and Chrna10) as well as genes whose functions have not yet been characterized, including Atp2a3, Calb2, Dnajc5b, Ripor3, and Scd1 in IHCs, and Aqp11, Dnm3, and Sh3gl3 in OHCs. Our DEG analysis comparing IHCs and OHCs is consistent with the previous studies (Li et al., 2018; Liu et al., 2014).

We next compared DEGs between type I and II HCs (Figure 2D). Although some new marker genes of adult type I and II HCs from mouse utricles have been identified (McInturff et al., 2018), differential gene expression analysis between type I and II HCs has not been conducted extensively. Our analysis revealed enrichment of 35 DEGs in type I HCs and 59 DEGs in type II HCs. Except for a few genes such as Otog and Bmp2, the roles of these DEGs in vestibular HCs have not been examined.

Grouping the tissue-specific subtypes together, we calculated DEGs between cochlear and vestibular HCs (Figure 2D). This comparison identified 674 DEGs enriched in cochlear HCs and 887 DEGs enriched in vestibular HCs. We note that many vestibular DEGs (such as Adam11, Car12, Nrxn3, Lpgat1, Spp1, Pcdh20, and Cfap126) are related to the secretion of extracellular matrix protein, cell–matrix interactions, cell adhesion, mineralized matrix, calcification, acid–base balance, and cilia. Meanwhile, many of the genes enriched in cochlear HCs (such as Slc25a5, Strip2, C1ql1, Rorb, and Ocm) are related to the extensively studied unique structure and function of cochlear HCs. It is interesting that Cib2 is differentially expressed in cochlear HCs while Cib3 is enriched in vestibular HCs. Cib2 and Cib3 act as an auxiliary subunit of the sensory mechanoelectrical transduction channel in HCs (Giese et al., 2017; Riazuddin et al., 2012; Wang et al., 2023), and mutations of these two genes are associated with deafness and Usher syndrome 1J (Riazuddin et al., 2012).

To examine the functional relevance of the calculated DEGs, we conducted overrepresentation analysis (ORA). Since the molecular properties of vestibular HCs are less known compared to cochlear HCs, we looked more closely at the biological processes enriched in vestibular HCs compared to cochlear HCs, which included gene ontology (GO) terms related to cilium organization and microtubule-based cilia motility (highlighted by red asterisks in Figure 2E). We also conducted ORA between type I and II HCs. Enriched processes in type I HCs included those related to cytoplasmic translation, p53-mediated signal transduction, actin filament depolymerization and regulation, and cilium or flagellum-dependent cell motility (Figure 2F). Terms enriched in type II HCs are related to oxidative phosphorylation, proton transmembrane transport, aerobic and cellular respiration, detection of mechanical stimulus involved in sound perception, and mechanoreceptor differentiation (Figure 2G).

Previous studies have characterized some marker genes in different HC types. We assessed the expression of previously characterized and newly identified genes related to HC structure and function across all four HC types. Many genes are expressed in all four HC types (Figure 3A). Among these are well-characterized genes such as Cib2, Otof, Kcna10, Ptprq, Tmc1, and Espn, while other genes such as Tjap1, which encodes a tight junction-associated protein, are less known. We also noted genes that were enriched in a tissue-specific manner. For example, Cdh23, Rorb, and Osbp2 are more highly expressed in cochlear HCs than in vestibular HCs, while genes such as Ldhb, Fbxo32, and Gsn are enriched in vestibular HCs but only weakly expressed in cochlear HCs. Several genes expressed in vestibular HCs with no expression in cochlear HCs (such as Cfap43, Cfap126, Cib3, Cxcl14, Pcdh20, Pifo, Slc9a3r2, and Tmc2) could potentially be used as vestibular HC markers (Figure 3A). Moreover, Adam11, Cacna2d4, Car12, and Shank2 are found to be only expressed in type I HCs, while Ccer2, Cfap45, Dlk2, and Rprm are only expressed in type II HCs. Our analysis also revealed new marker genes for IHCs (Atp2a3, Rims2, and Ripor3) and OHCs (Aqp11, Mmd, and Sh3gl3). Cfap43, Cfap44, Cfap45, Cfap126, Kif3, Mlf1, and Pifo, enriched in vestibular HCs, are all associated with kinocilium structure and function. We utilized single-molecule fluorescent in situ hybridization (smFISH) and immunostaining techniques to validate the expression of 12 genes in HCs (Figure 3B, C). The expression patterns of these genes and proteins were highly consistent with our observations from scRNA-seq analysis. Overall, our results identified novel marker genes and previously unidentified expression patterns among the four HC types.

Next, we focused our analysis on evaluating the genes associated with HC function. We compared the expression of 208 genes related to HC structure and function, including stereocilia and apparatus for mechanotransduction, ion channels, and synapses (Figure 4A). Among the genes associated with stereocilia and mechanotransduction apparatus (Shin et al., 2013), Calm1, Calm2, Eps8, Espn, Fbxo2, Dynll2, Ush1c, Ywhae, Tmc1, Tmie, Cdh23, Pcdh15, Ank1, Ank3, and Lhfpl5 were highly expressed in all four HC populations. Others were highly expressed in only one or two HC populations. Atp2a3, Calb2, and Dpysl2 were highly expressed in IHCs, while Lmo7, Ocm, and Strc were highly expressed in OHCs. Moreover, expression of Dpysl2, Cdh23, and Cib2 was higher in cochlear HCs than in vestibular HCs, while Xirp2, Pls1, Slc9a3r2, and Tubb4b were more highly expressed in vestibular HCs than in cochlear HCs. Some genes were uniquely expressed in either cochlear or vestibular HCs. For example, Cib3 and Tmc2 are expressed in vestibular HCs but not in cochlear HCs.

All HCs possess ion channels. Our analysis detected several genes related to stretch-activated ion channels, such as Trpc1 and Trpm4. Genes for Cl⁻ and Na⁺ channels were expressed. For Ca²⁺ and K⁺ channels, Cacna1d and Kcna10 were expressed in all four HC types with varying levels of expression. In cochlear HCs, Cacna1d, Kcna10, Kcnab1, Kcnj16, and Kcnma1 were expressed in IHCs, whereas Cacna1d, Kcna10, Kcnk1, Kcnma1, Kcnn2, and Kcnq4 were expressed in OHCs. In the vestibular HCs, Cacna2d4, Kcna10, Kcnab1, and Kcnma1 showed relatively high expression in type I HCs, whereas type II HCs indicated a relatively high expression of Cacna2d4, Cacng5, Kcna10, Kcnb1, Kcnh2, and Kcnh7. Best1, Clic, Hcn1, and Kcnh7 were only expressed in vestibular HCs.

Next, we examined the genes related to synapses (Figure 4A). Our results indicated expression of Ctpb2, Dlg1, Homer2, Otof, Pclo, and Snap91 in all HCs at varying levels. We observed a relatively higher expression of Dnm1, Otof, Rims2, Slc17a8, Snap91, and Stx7 in IHCs, while Dlg1, Dnm3, Snap9, Chrna9, and Chrna10 showed higher expression in OHCs. Some of the highly expressed genes in type I HCs include Dnm1, Kif3a, Pclo, Shank2, Slc17a8, Syt13, Syt14, Chrna9, and Chrna10, whereas type II HCs showed relatively higher expression of Kif3a, Otof, Shank2, Stx7, and Syt13. We used smFISH to validate the expression of 8 additional genes across the four HC types. The expression patterns shown in Figure 4B are consistent with our analysis (Figure 4A).

A key morphological feature in the hair bundle of vestibular HCs is the presence of kinocilium, which has been regarded as a type of specialized primary cilia. Since our analysis revealed an enrichment of cilium-related GO terms (Figure 2F) and axonemal genes such as Cfap43, Cfap44, Cfap45, Cfap126, Kif3, Plf1, and Tubb4b (Figure 3A) in vestibular HCs, we sought to investigate the composition and molecular nature of the kinocilium (Figure 5A). Proteomics-based approaches have contributed to the development of cilia-associated protein databases. We utilized several well-established databases, including CiliaCarta (van Dam et al., 2019), the SYSCILIA gold standard (SCGSv2) (Vasquez et al., 2021), and CilioGenics (Pir et al., 2024) to compile a list of ~1000 cilia-related genes. We noted a significant overlap of these genes with our HC transcriptomic profiles (Figure 5B). GO analysis of the overlapping genes revealed enrichment of cellular component and biological process terms primarily related to cilia organization, assembly, maintenance, intracellular transport, and microtubule dynamics, particularly associated with motile cilia (Figure 5—figure supplement 1). Additionally, molecular function analysis highlights associations with motor activity, dynein chain binding, and BBSome binding (Figure 5—figure supplement 1). BBSome (Bardet–Biedl syndrome) is an octameric protein complex crucial for regulating transport in primary cilia (Tian et al., 2023).

The primary cilium is a sensory organelle that responds to and transmits external signals to the interior of the cell. Structurally, primary cilia are characterized by the presence of nine microtubule doublets encircling the shaft, which transition into a disorganized structure distally (Figure 5C). We first examined the expression of primary cilia-related genes in vestibular and cochlear HCs. Among approximately 420 primary cilia-related genes/proteins (van Dam et al., 2019; Pir et al., 2024; Vasquez et al., 2021), we detected the expression of 410 genes in at least one type of HC. Figure 5D shows the top 50 abundantly expressed primary cilia-related genes in type II HCs compared to the other three HC types. Most of these genes were detected in all four HC types except a few genes, such as Mlf1, Ttc21a, and Tmem218, which were weakly or not expressed in cochlear HCs.

Primary cilia are enriched in receptors and effectors for key pathways, including GPCR, cAMP, Ca²⁺, RTK, TGF-β, MAPK, TOR, BMP, Wnt, Notch, and Rho signaling (Anvarian et al., 2019), localized to the ciliary shaft, transition zone, and BBSome (Hansen et al., 2025). We analyzed the expression of these genes in HCs and found higher expression levels of these pathway-related genes in vestibular HCs than in cochlear HCs (Figure 5—figure supplement 2).

Intraflagellar transport (IFT) involves anterograde and retrograde transport of molecules along the axoneme of cilia, facilitating the transport of components between the ciliary base and tip (Ma et al., 2023). IFT is essential for the proper assembly and maintenance of both primary and motile cilia. Thus, we assessed the expression of IFT-associated genes in HCs. Most of the genes are enriched in vestibular HCs compared to cochlear HCs (Figure 5D). Immunostaining confirmed the expression of IFT172 and CLUAP1 in the kinocilia of vestibular HCs (Figure 5E).

Curiously, our GO analysis returned many terms related to cilium motility. Motile cilia are highly conserved organelles across different organisms and tissues, although they exhibit organism- and tissue-specific adaptations (Leung et al., 2025). Recent advances using proteomics of isolated motile cilia from various ciliated tissues have enabled the profiling of genes associated with motile cilia. To explore whether the kinocilium possesses a molecular composition characteristic of motile cilia, we compared our HC transcriptomes with multi-tissue proteomics datasets derived from different organisms, including human, bovine, porcine, and murine, as well as diverse motile ciliary tissues such as sperm, oviduct, ventricle, and trachea (Leung et al., 2025; Figure 5—figure supplement 3). Our analysis revealed a significant overlap, particularly among axonemal components of motile cilia, with strong enrichment in vestibular HCs compared to cochlear HCs. The axoneme of motile cilia and flagella is a cylindrical structure harboring a canonical ‘9 + 2’ arrangement, where nine doublet microtubules (DMTs) surround two microtubule singlets in the center (Figure 5C). Altogether, the axoneme machinery consists of nine DMTs, two rows of inner and outer dynein arms (IDAs and ODAs), nexin–dynein regulatory complex (N-DRC), two singlet central pair complexes (CPCs), three radial spokes (RSs), microtubule inner proteins (MIPs), and external coiled-coil regions. The motility unit is arranged in a 96-nm repeat module along the CPC. Recent cryo-electron microscopy (cryo-EM) and cryo-electron tomography (cryo-ET) studies have provided a more comprehensive identification of this module repeat (Chen et al., 2023; Leung et al., 2025; Walton et al., 2023). To understand the molecular composition and function of kinocilia, we focused our analysis on the expression of genes associated with the structure of motile cilia. First, we assessed the expression of gene sets related to the motile cilium and each of its structural components. We noted a robust expression of motile cilia-related gene signatures in vestibular HCs, while cochlear HCs expressed little to none (Figure 5F). Next, we further assessed the expression of the key genes related to motile cilia machinery (Figure 5G), including the 96 nm module and CPC, based on the current known localization from biochemistry and proteomics.

We examined genes related to the 96 nm axonemal repeat of mammalian epithelial cilia. This structural unit contains proteins encoded by 128 genes (Gui et al., 2021). We found that 112 of these genes were expressed in adult vestibular HCs (Supplementary file 2). Genes encoding axonemal dynein (IDAs and ODAs), such as Dnah5 and Dnah6, as well as RS components (Wdr66, Cfap206, Cfap61, and Iqub), N-DRC components (Drc1 and Iqca), and MIPs (Cfap126 and Wdr63) were predominantly expressed in vestibular HCs with little to no expression in cochlear HCs (Figure 5G). Axonemal CCDC39 and CCDC40, which form external coiled-coil regions, are the molecular rulers that organize the axonemal structure in the 96 nm repeating interactome and are required for the assembly of IDAs and N-DRC for ciliary motility (Becker-Heck et al., 2011; Brody et al., 2025; Merveille et al., 2011; Oda et al., 2014). Our results indicate a high expression of Ccdc39 and Ccdc40 in vestibular HCs, whereas little to no expression was observed in cochlear HCs. We should point out that unlike axonemal dynein proteins, which are uniquely required for cilia motility, the encoded proteins of several genes in Figure 5G (marked by red asterisks) are also expressed in cytoplasm and/or are multifunctional.

Next, we examined the expression of genes encoding transcription factors that are known key regulators of ciliome gene activation, including RFX and FOXJ1 transcription factor families. RFX controls genes in both motile and non-motile cilia, while FOXJ1 specifically governs motile cilia formation (Choksi et al., 2014). Our data show moderate expression of Foxj1 in vestibular HCs and weak expression in cochlear HCs (Figure 5G). Other genes that regulate motile cilia formation, including Lrrc6, were also expressed at high to moderate levels in vestibular HCs compared to cochlear HCs. We also found a strong enrichment of transcriptional targets associated with vestibular HCs, particularly those involved in motile cilia programming and maintenance. The expression of these transcription factors may reflect their importance in the maintenance of kinocilia in adult vestibular HCs. Furthermore, analysis of a published ATAC-seq dataset (Jen et al., 2019) from adult mouse vestibular tissue revealed increased chromatin accessibility in the promoter regions of genes associated with motile cilia machinery (Figure 5—figure supplement 4), suggesting elevated transcriptional activity in vestibular HCs. These findings are consistent with our observations from scRNA-seq datasets.

To examine the conservation of expression of these genes in vestibular HCs across species, we obtained the orthologs of axoneme-related genes in adult zebrafish inner ear HCs and human vestibular HCs using published datasets (Barta et al., 2018; Wang et al., 2024). Figure 6A shows the expression of these genes in adult mouse, zebrafish, and human vestibular HCs. While the expression levels vary, most of these genes are expressed across species. The exceptions are Dnah3 and Dnah12, which are expressed in zebrafish HCs but not in mammalian vestibular HCs.

We conducted immunostaining and high-resolution confocal imaging to validate the expression of key motile cilia markers in the kinocilia. FOXJ1 is expressed in the nucleus of vestibular HCs, while CCDC39, CCDC40, TEKT1, DNAH5, and DNAH6 are expressed in kinocilia (Figure 6B). Collectively, our findings provide evidence corroborating the presence of motile cilium machinery in the kinocilia of vestibular HCs.

Since nascent cochlear HCs possess kinocilia (Figure 6C), we used published P2 cochlear and vestibular HC transcriptomes (Burns et al., 2015; McInturff et al., 2018) to investigate whether neonatal cochlear HCs express motile cilium-related genes. Assessment of expression of genes related to motile cilia machinery revealed a less drastic difference between neonatal cochlear and vestibular HCs than that between adult cochlear and vestibular HCs (Figure 6D). We note that proteins of some shared genes are expressed in both cilia and cytosol (marked by red asterisks in Figure 6D). However, some key motility-associated genes such as Dnah6 and Dnah5 (marked by red arrows in Figure 6D) were not detected in the P2 cochlear HCs. These axonemal dynein heavy chains are ATPase-driven force-generating motors that produce the ciliary power stroke in concert with other axonemal components. Immunostaining confirmed the lack of expression of CCDC39, CCDC40, and DNAH6 in cochlear HCs at P2 (Figure 6E). In contrast, these key proteins were expressed in kinocilia of vestibular HCs (Figure 6E). Lack of expression of Dnah5 and Dnah6 and the molecular rulers CCDC39 and CCDC40 suggests that the kinocilium of neonatal cochlear HCs does not possess signatures of motile cilia. Therefore, our analysis indicates that the molecular composition of kinocilia is different between neonatal cochlear and vestibular HCs.

TEM studies have characterized the ultrastructure of kinocilia across species, revealing evidence of complex and regionally specialized organization (Kikuchi et al., 1989; Nagel et al., 2014; O’Donnell and Zheng, 2022). We used TEM to examine the ultrastructure of kinocilia from bullfrog crista HCs. TEM images, including longitudinal sections of frog vestibular hair bundles (Figure 7A), highlight a distinct zonal architecture along the kinocilium axis. At the distal tip, a prominent kinociliary bulb is observed, while the base anchors the axoneme within the cuticular plate. Two pairs of microtubule doublets span the full length of the kinocilium in the longitudinal section. The central pair of singlet microtubules, typical of motile cilia, is maintained throughout most of the shaft but disappears in the distal and transitional zones. This configuration results in a dynamic shift from a canonical 9 + 2 arrangement centrally to a 9 + 0 pattern at both the base and tip. These observations underscore the heterogeneous and hybrid nature of vestibular kinocilia, integrating structural features of both primary and motile cilia.

Vestibular kinocilia are traditionally regarded as non-motile, lacking the rhythmic beating characteristic of respiratory cilia. Kinocilia are not only tightly connected to stereocilia but are also embedded in the overlaying gelatinous membrane (Eatock and Songer, 2011; Leibovici et al., 2005; Li et al., 2008), an extracellular matrix required for physiological mechanotransduction. Using acute preparations of bullfrog semicircular canal sensory epithelia (cristae), we observed robust spontaneous kinociliary motility in some HCs (Figure 7B; Videos 1 and 2). This motility exhibited the characteristic beating pattern of flagella and cilia and occurred at a frequency of approximately 5–10 Hz at room temperature. The observed displacements were sufficiently large to induce deflection of the entire hair bundle, indicating it can generate forces to influence hair bundle dynamics. Interestingly, this phenomenon was detected in only ~1–5% of crista HCs, likely due to variable preservation of HC and kinociliary integrity in vitro in acutely dissected tissue. These findings suggest that vestibular kinocilia are capable of active movement and challenge the strict classification of these structures as non-motile even though motility is not experimentally observed in most HCs in the excised vestibular sensory epithelium.

Next, we explored whether the kinocilia of vestibular HCs from adult mice are motile by utilizing the photodiode technique to detect bundle motion (Jia and He, 2005). This technique can detect motion in the 10 nm range for synchronized signals after averaging. Since spontaneous motions are not synchronized and cannot be averaged to improve the signal-to-noise ratio, we captured the responses in the time domain and averaged them in the frequency domain after a fast Fourier transform. This allows us to detect unsynchronized cilia motion even if the signal is close to the noise level at the time domain. We first measured spontaneous movement of cilia from the epithelial lining of the Eustachian tube, which is an extension of airway epithelia, as cilia in the respiratory tract are a typical example of motile cilia (Li et al., 2014). Figure 7C shows two representative waveforms of airway cilia beat with a magnitude between 700 and 1500 nm. The two responses shown in Figure 7C have main frequency components at 6–9 Hz at room temperature, consistent with a previous study (Li et al., 2014). Next, we measured the movement of the top segment of the hair bundle from mouse crista ampullaris. Since the kinocilium is tightly attached to the stereocilia bundle, we measured the motion of the whole bundle due to the difficulty of taking measurements from a single kinocilium. The waveform (black trace in Figure 7D) was obtained from crista HCs bathed in perilymph-like solution (L-15 medium) with 2 mM of Ca²⁺. Like the beat frequency of airway cilia, kinocilia also moved at the frequency of ~7 Hz (Figure 7D). To rule out the possibility that the bundle motion is driven by the mechanotransduction-related activity (Benser et al., 1996; Martin et al., 2003), we treated the crista ampullaris in Ca²⁺-free medium with EGTA for 2 min to break the tip-link (Assad et al., 1991; Jia et al., 2007; Kachar et al., 2000; Ricci et al., 2003). Spontaneous motion was still detected (two red traces in Figure 7D), suggesting that the motion is independent of the transduction channel activity. We measured spontaneous bundle motions from 52 crista HCs from six mice. Spontaneous motion was only detected in eight HCs. An example of a lack of response is shown in Figure 7D (blue trace). Although we were unable to determine which of the HC subtypes were exhibiting kinocilia motility, it is conceivable that both type I and II HCs have this capability since they both express the genes related to motile cilia. We note the kinocilia motion of mouse crista HCs was substantially smaller than that of airway cilia (Figure 7D) and bullfrog crista HCs.

Since kinocilia motility in mouse vestibular HCs is substantially smaller than the motility of airway cilia, we investigated the structural basis underlying comparatively reduced motility of kinocilia. This diminished motility may result from differences in the molecular composition and organization of the axonemal machinery, particularly the 96 nm modular repeat that houses key dynein motors and regulatory complexes. Recent advances in structure prediction powered by artificial intelligence and cryo-EM have facilitated the generation of highly conserved atomic models of the 96 nm axonemal repeat from human respiratory cilia and bovine sperm flagella (Chen et al., 2023; Leung et al., 2025; Walton et al., 2023). We applied our axonemal gene dataset to these atomic models to predict the molecular architecture of the 96 nm repeat in vestibular kinocilia. By mapping the expression of known axonemal components onto the structural frameworks derived from human respiratory (PDB: 8J07) (Walton et al., 2023) and bovine sperm (PDB: 9FQR) (Leung et al., 2025) axonemes, we generated two composite models that reflect the unique molecular composition of the vestibular kinocilium. These predicted structures are shown in Figure 8A, B. 3D videos of the predicted models are provided in Videos 3 and 4.

We chose the human respiratory 96 nm axonemal structure as our reference because it best reflects vestibular kinocilia and provides a more comprehensive representation of axonemal components than the sperm model, which shows more missing elements (highlighted in gold in Figure 8A, B). Using a curated reference gene list derived from the human respiratory 96 nm axonemal structure, we mapped vestibular HC gene expression and identified transcripts for all 18 ODA genes and their docking complex components, all 11 N-DRC genes, and all 7 MAPs, along with 36 of 37 RS genes and 19 of 23 IDA-related genes. Additionally, 20 of the 31 genes encoding MIPs, which are known to stabilize the microtubule doublets, were also expressed (Figure 5G). In contrast, vestibular HCs lacked the expression of several sperm-specific genes from distinct compartments of 96 nm repeat (Leung et al., 2025), including TRiC chaperonin subunits (Brown et al., 2025; Meng et al., 2026), Camk4, Efacb5, Lrrd1, Stkld1, Ccdc63, Wdr64, and several MIPs (Supplementary file 3).

Based on our models, the absence of Dnah3 and Dnah12, along with Acta2, is predicted to result in the loss of two of the six single-headed IDA components (highlighted in gold). Cfap100, which forms the modifier of inner arms (MIA) complex with Cfap73 and contributes to tethering the double-headed IDAf (inner dynein arm f), is missing in the model, whereas Cfap73 and the remaining IDAf components are present. Notably, IDAf has multiple attachment points, and MIA is not essential for docking IDAf to the DMTs (Yamamoto et al., 2013). Overall, 4 of the 23 IDA-related genes were not expressed in vestibular HCs; nonetheless, we predict that the IDA structure is reduced but not entirely absent, as DNAH6 remains localized in vestibular kinocilia (Figure 5B). In addition to the missing IDA components, we identified 11 unexpressed genes associated with MIPs, whose absence is predicted to result in reduced MIP density in the models (highlighted in orchid and gold in cross-sectional views in Figure 8A, B). Unlike other axonemal structures, MIPs exhibit greater variability across species, which may account for their lineage-specific absence in vestibular kinocilia (Figure 8; Andersen et al., 2024; Tai et al., 2023; Xia et al., 2025). The missing genes in our HC datasets and their roles in the axonemal complex, cilia motility, and ciliopathy are listed in Figure 8C and Supplementary file 3. Based on our predicted models, we speculate that the absence of Dnah3 and Dnah12 plays a major role in limiting kinocilia motility in mouse vestibular HCs, contributing to the smaller movements compared to respiratory motile cilia.

Male and female CBA/J mice aged 10 weeks were used for scRNA-seq. Cochlear and vestibular end organs (utricle, saccule, and crista) were dissected from the inner ear and placed in Petri dishes containing cold L-15 medium (Gibco; #11320033). After the cochlear and vestibular sensory epithelia and neurons were dissected out, they were transferred into two individual 1.5 ml tubes for enzymatic digestion (Collagenase IV from Sigma, concentration: 1 mg/ml collagenase) in L-15 medium. After 10 min of incubation at room temperature, the enzymatic solution was removed and replaced with 400 µl L-15 media containing 10% fetal bovine serum. The tissues in two tubes were mechanically triturated by 200 µl Eppendorf pipette tips. After that, the suspension containing cochlear and vestibular cells was then passed through 40 µm strainers for filtration and pelleted at 300 × g for 5 min. After removing extra media, cells were then reconstituted in the 50 µl L-15 with 10% fetal bovine serum media and used for cDNA library preparation. Seven mice were used for each biological replicate. Six biological replicates for cochlear sensory epithelium and four biological replicates for vestibular sensory epithelia were prepared for scRNA-seq.

The emulsion droplets were constructed using a 10x Genomics Controller device following the manufacturer’s instruction manual. cDNA libraries were constructed using the 10x Genomics Chromium Single Cell 3′ Reagent Kits V3.1. High Sensitivity DNA Kits (Agilent Technologies) were used to perform quality control for each library in an Agilent 2100 Bioanalyzer. cDNA libraries were sequenced in an Illumina NextSeq 6000 sequencer aiming for 240 billion 150 bp long paired-end reads.

Raw transcriptomic datasets of adult cochlear and vestibular HCs from scRNA-seq have been deposited to GEO (GSE283534). The FASTQ files were mapped to mm10 reference genome to generate the single-cell expression matrices following the CellRanger count pipeline (version 6.1.2). The Cellranger output data was then processed with the Seurat package (version 4.3.0) in R (version 4.1.3).

Genes expressed in at least ten cells were included in the analysis. Cells with numbers of expressed genes <200 or >3000 and cells with numbers of unique molecular identifiers >15,000 were filtered out. Cells with >20% mitochondrial genes were also excluded from the analysis.

The gene expression data from individual samples were converted into a natural logarithm and normalized under the same condition. Data from six cochlear and four vestibular replicates were integrated separately based on the anchors identified from the top 2000 highly variable genes of individual normalized expression matrices. The Shared Nearest Neighbor graph method can calculate the neighborhood overlap (Jaccard index) between every cell and its nearest neighbors, which was used for cluster determination at a resolution of 0.6 on PCA-reduced expression data for the top 30 principal components.

Clustering results for cochlear and vestibular datasets were visualized separately using t-SNE. Cluster annotations were initially produced using SingleR and then corrected where appropriate based on well-known cellular markers for cochlear and vestibular cells as described before (Xu et al., 2022).

Entrez Gene, HGNC, OMIM, and Ensembl database were used for verification, reference, and analyses. Online Databases of Ciliogenics (https://ciliogenics.com/), CiliaCarta (https://ngdc.cncb.ac.cn/databasecommons/database/id/6383), and Primary Cilium Proteome (https://esbl.nhlbi.nih.gov/Databases/CiliumProteome/) were also used for reference.

GO analysis in Figure 5—figure supplement 1 was performed using ShinyGO 0.82 (Ge et al., 2020). The Venn diagrams in Figure 5B, Figure 5—figure supplement 3 were generated using Venny 2.1 (https://bioinfogp.cnb.csic.es/tools/venny/), except for the sperm Venn diagram in Figure 5—figure supplement 3E, which was plotted using nVenn (https://degradome.uniovi.es/cgi-bin/nVenn/nvenn.cgi). Figure 5C was created with BioRender.com and further modified using Adobe Photoshop.

Inner ears were fixed overnight with 4% paraformaldehyde at 4°C. Cochlear and vestibular sensory epithelia were dissected out. After several washes with PBS, the tissue was blocked for 1 hr with 0.25% normal goat serum in PBS containing Triton X-100 (0.01%) and goat serum (10%). Primary antibodies against DNM1 (NBP2-48950, Novus Biologicals), SLC7A14 (HPA045929, Sigma), TJAP1 (NBP1-80902, Novus), FOXJ1 (14-9965-82, Thermo Fisher), CCDC39 (HPA035364, Sigma), CCDC40 (PA5-54653, Thermo Fisher), DNAH5 (31079-1-AP, Thermo Fisher), DNAH6 (HPA036391, Sigma), TEKT1 (HPA044444, Millipore Sigma), CLUAP1 (PA5-83710, Thermo Fisher), IFT172 (28441-1-AP), and acetylated tubulin (T6793, Sigma) were incubated with the tissues for 12 hr at 4°C. After washes with PBS, secondary antibody (1:500) (Alexa Fluor Molecular Probe 488 or 555; Invitrogen) was added and incubated for 2 hr at room temperature. Alexa Fluor 405 or 488 phalloidin (A30104 or A12379, Invitrogen) was used to label stereocilia bundles. Tissues were washed with PBS and mounted on glass microscopy slides with antifade solution (5 ml PBS, 5 ml glycerol, 0.1 g n-Propyl gallate). Images were captured using a Nikon TI-2 Spinning Disk or Zeiss LSM 980 Inverted confocal microscope. Immunostaining of each type of antibody was repeated in four mice.

Single-molecule fluorescence in situ hybridization was used to validate the expression of 15 genes in 10-µm-thin sections prepared from three mice. Samples were prepared in formalin-fixed paraffin-embedded tissue. Probes for 18 genes were purchased from ACD. These genes include Adam11 (Cat#: 580971), Aqp11 (Cat#: 803751), C1ql1 (Cat#: 465081), Cdh23 (Cat#: 567261-C2), Chrna10 (Cat#: 818521), Cib2 (Cat#: 846681), Cib3 (Cat#: 1105771), Dnm3 (Cat#: 451841), Ikzf2 (Cat#: 500001), Kcnq4 (Cat#: 707481), Otof (Cat#: 485678), Pcdh20 (Cat#: 467491), Slc7a14 (Cat#: 544781), Tmc1 (Cat#: 520911-C2), and Tbx2 (Cat#: 448991-C2). Methods for the RNAscope 2.5 HD Duplex Assay from Advanced Cell Diagnostics were followed.

The mouse inner ears were fixed with 4% paraformaldehyde and 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.4) with 2 mM CaCl₂. The bullfrog crista ampullaris was fixed with 3% paraformaldehyde, 2% glutaraldehyde, 2% tannic acid, and 0.5% calcium chloride in 0.1 M sodium cacodylate buffer (pH 6.8) with 0.1 mM CaCl₂ (Kachar et al., 1990). The tissues were then post-fixed for 1 hr with 1% OsO₄ in 0.1 M sodium cacodylate buffer and washed. For SEM, cochleae and vestibular tissues were dehydrated via an ethanol series, critical point dried from CO₂, and sputter-coated with platinum. The morphology of the HC stereocilia bundle was examined in a FEI Quanta 200 scanning electron microscope and photographed. For TEM, the bullfrog crista ampullaris was embedded in plastic (Epon 812 Epoxy Resin) after dehydration via an ethanol series. 70 nm thin sections were cut with a diamond knife and collected on 300-mesh grids. The thin section was post-stained with 3% uranyl acetate for 15 min and 1.5% lead citrate for 3 min. The preparations were examined in an electron microscope (JEOL 100CX) and photographed. Three animals were used for EM studies.

Bullfrogs were anesthetized with 20 µg/g of 3-aminobenzoic acid ethyl ester and decapitated. The saccular macula was carefully dissected out under cooled frog’s Ringer solution. The otoliths were gently removed with forceps in fresh Ringer solution (Jaeger et al., 1994). The saccular macula was placed under an Olympus upright microscope and kinocilia motility was visualized using DIC and a 60x objective and captured with a camera.

The Eustachian tube was dissected out from CBA/J mice and sectioned along its longitudinal length. The preparation was bathed in L-15 medium (Invitrogen), containing 136 mM NaCl, 5.8 mM NaH₂PO₄, 5.4 mM KCl, 1.4 mM CaCl₂, 0.9 mM MgCl₂, 0.4 mM MgSO₄, and 10 mM HEPES-NaOH (pH 7.4, 300 mmol/l) in an experimental chamber mounted on the stage of a Leica upright microscope. Crista ampullaris was also dissected from 10-week-old CBA/J mice and bathed in L-15 medium. The tissue was attached to the bottom of the chamber by the weight of two thin platinum rods (0.5 mm in diameter). The tissue was mounted with the cilia or hair bundles facing upward toward the water-immersion objective. The cilia and hair bundles were imaged using a 63x water immersion objective (Leica) and magnified by an additional 10x relay lens. Ciliary motion was measured and calibrated by a photodiode-based measurement system mounted on the Leica upright microscope (Jia and He, 2005). The magnified image of the hair bundle was projected onto a photodiode through a rectangular slit. The image was positioned to one side of the slit with 50% of the rectangular slit being covered by the magnified image of the bundle. Cilia motion modulated the light influx to the photodiode. The photocurrent response was calibrated to displacement units by moving the slit a fixed distance (0.5 μm) with the image of the cell in front of the photodiode. After amplification, the photocurrent signal was low-pass filtered by an antialiasing filter before being digitized by a 16-bit A/D board (Digidata 1550A; Molecular Devices). The motile responses were low-pass filtered at 250 Hz and digitized at 1 kHz. Ciliary motion was acquired in a 2-s window for each trial, and 20 trials were captured for each cell in one recording. The power spectrum of the response was averaged and analyzed in the frequency domain using Clampfit software (version 10, Molecular Devices). The experiments were performed at room temperature (22 ± 2°C).

The 96 nm repeat structures of the human respiratory axoneme (PDB: 8J07) and bovine sperm flagellum (PDB: 9FQR) were used as structural references to model the vestibular kinocilia axoneme. Candidate vestibular HC genes identified through transcriptomic profiling were annotated and mapped to their respective axonemal compartments based on known or predicted protein localizations within these reference structures. The 8J07 model lacked atomic modeling for the full-length RS3 (Zhao et al., 2025), despite its presence in the corresponding cryo-EM density map (EMD-35888), due to the uncharacterized proteome of RS3 in the human respiratory system. To address this, RS3 components from the sperm axoneme structure (PDB: 9FQR), excluding sperm-specific proteins, were extracted, fitted into the EMD-35888 density using ChimeraX’s ‘fit-to-map’ tool, and overlaid onto the 8J07 model to complete the RS3 architecture. All structural visualization and model integration were performed using UCSF ChimeraX v1.6. (Pettersen et al., 2021). In the models, DNAH5 and DNAH9 (present in our HC data) occupy the ODA region in human respiratory cilia, while DNAH8 and DNAH17 occupy the same region in sperm (but are not expressed in our dataset). Since they correspond structurally, we did not mark them as missing in the sperm-based kinocilia model.

Publicly available software, standard packages, and algorithms were used for the analysis, including Cell Ranger (v6.1.2) (RRID:SCR_017344), Seurat and built-in tools (v4.3.0) (RRID:SCR_016341), and UCSF ChimeraX (RRID:SCR_015872, v1.6). No custom code or algorithms were used or generated.