A Claudin-9–Based Ion Permeability Barrier Is Essential for Hearing

Hereditary hearing loss is one of the most common birth defects, yet the majority of genes required for audition is thought to remain unidentified. Ethylnitrosourea (ENU)–mutagenesis has been a valuable approach for generating new animal models of deafness and discovering previously unrecognized gene functions. Here we report on the characterization of a new ENU–induced mouse mutant (nmf329) that exhibits recessively inherited deafness. We found a widespread loss of sensory hair cells in the hearing organs of nmf329 mice after the second week of life. Positional cloning revealed that the nmf329 strain carries a missense mutation in the claudin-9 gene, which encodes a tight junction protein with unknown biological function. In an epithelial cell line, heterologous expression of wild-type claudin-9 reduced the paracellular permeability to Na+ and K+, and the nmf329 mutation eliminated this ion barrier function without affecting the plasma membrane localization of claudin-9. In the nmf329 mouse line, the perilymphatic K+ concentration was found to be elevated, suggesting that the cochlear tight junctions were dysfunctional. Furthermore, the hair-cell loss in the claudin-9–defective cochlea was rescued in vitro when the explanted hearing organs were cultured in a low-K+ milieu and in vivo when the endocochlear K+-driving force was diminished by deletion of the pou3f4 gene. Overall, our data indicate that claudin-9 is required for the preservation of sensory cells in the hearing organ because claudin-9–defective tight junctions fail to shield the basolateral side of hair cells from the K+-rich endolymph. In the tight-junction complexes of hair cells, claudin-9 is localized specifically to a subdomain that is underneath more apical tight-junction strands formed by other claudins. Thus, the analysis of claudin-9 mutant mice suggests that even the deeper (subapical) tight-junction strands have biologically important ion barrier function.


Introduction
The separation of compositionally distinct extracellular fluids is a vital function of the epithelial and endothelial cells that line the lumens of organs and blood vessels, respectively. Indiscriminate diffusion of solutes across epithelial and endothelial sheets is prevented by sealing of the paracellular spaces by tight junctions [1,2]. These intercellular junctions are comprised of membraneassociated and transmembrane proteins that form continuous networks of strands circumscribing the apical end of the cell. The tight-junction strands of neighboring cells are arranged in parallel fashion as they associate with each other laterally, thereby creating paracellular barriers to the diffusion of solutes [3]. In addition to forming barriers, tight junctions also function as signaling hubs, and contribute to the maintenance of cell polarity [4,5].
The ''tightness'' of tight junctions varies greatly from tissue to tissue, and depends largely on the presence of integral membrane proteins known as claudins [6]. Twenty-four claudins have been identified in humans. Each of these is a four-pass transmembrane protein with intracellular N-and C-termini and two extracellular loops. The first extracellular loop is responsible for the selective ion barrier function of the claudin [7][8][9][10], and is also critical for the polymerization of neighboring claudins into strands [11]. The second extracellular loop is thought to be important for the intercellular claudin-claudin interactions [12]. The claudin Cterminus binds to scaffolding proteins such as zonula occludens-1, -2, -3 and MUPP-1, which link the tight junction to the actin cytoskeleton [13][14][15][16][17].
The expression of individual claudins can have diverse effects on the paracellular permeability of solutes. Some claudins create paracellular ion barriers [9,[18][19][20][21][22], whereas others form ionselective paracellular channels [8,18,[23][24][25][26]. The properties of tight junctions are further diversified by the expression of multiple claudins in most epithelial cells. Moreover, some (but not all) claudins can interact with other members of the claudin family, either within or between the tight junction strands, and this affects paracellular ion permeability [27]. Thus, the ion barrier property of epithelial and endothelial cell layers depends on the type and ratio of different claudins that are present, and also on the interactions among them [28].
The hearing organ provides a unique example of the need to separate compositionally distinct extracellular fluids. The stereocilia of the cochlear sensory hair cells are surrounded by the endolymph, in which the K + concentration is high (,140 mM) and the Na + concentration is low (,1 mM). In contrast, the basolateral side of hair cells is bathed in the perilymph, which has low K + and high Na + concentrations (,2 mM and ,140 mM, respectively). These ion gradients are generated by the stria vascularis ( Figure 1A), which also produces an endocochlear potential (EP, ,80 mV) between the endolymphatic and perilymphatic spaces [29]. The EP amplifies the K + driving force. In this context, sound-induced deflections of stereocilia result in the entry of K + into the hair cell through apical ion channels. The ensuing depolarization of the cell leads to sound detection. Subsequently, the resting potential of the hair cell is restored by K + efflux from the basolateral side of the cell [30]. The K + and voltage gradients between the apical and basolateral sides of hair cells are sustained by tightly sealed epithelial cells that line the entire endolymphatic compartment and both sides of the stria vascularis ( Figure 1A) [31][32][33].
Of all the claudin family members, claudin-9 is the most highly expressed in the inner ear [34], and it is present in all of the major epithelial cell types that line the endolymphatic space [35]. However, the biological function of claudin-9 has not been established. Here we report the molecular and phenotypic characterization of a previously unrecognized claudin-9-deficient mouse strain, nmf329, which was generated in the course of an ENU-induced mutagenesis project at The Jackson Laboratory [36]. Our analysis of nmf329 mice indicates that the genetic defect in claudin-9 leads to deafness, and that a claudin-9-based ion barrier is necessary to protect the basolateral domain of the hair cell from the K + -rich endolymph.

Non-syndromic deafness in the nmf329 mouse line
Mice of the nmf329 line were tested for their reactions to loud sound bursts generated by a click box (,90 dB sound pressure level, SPL). Whereas wild-type mice exhibited typical startle responses, nmf329 homozygotes did not react. To evaluate the hearing loss of the nmf329 mice more quantitatively, we measured the auditory brainstem response (ABR) of these animals on postnatal day 28 (P28), using broadband and pure-tone sounds. We found that the ABR threshold of nmf329/nmf329 mice was elevated by ,60 dB-SPL compared to that of wild-type littermates, whereas the ABR thresholds of heterozygous and wild-type animals were indistinguishable from each other ( Figure 1B and 1C). The hearing loss of nmf329/nmf329 mice was equally severe at high-and low-frequency, as indicated by the uniformly high ABR thresholds at 8, 16, and 32 kHz ( Figure 1C). Next, we investigated whether nmf329/nmf329 mice develop hearing for a brief period prior to P28. We found that the hearing loss of homozygous mutant mice was already severe at P16 (i.e. 3-4 days after the expected onset of hearing, Figure S1). These results indicate that the deafness caused by the nmf329 mutation has an early onset.
Given that auditory defects are often accompanied by problems with balance, we quantified the balancing abilities of nmf329 mice by measuring the time that they could remain on a slowly rotating rod. Equally good balance was demonstrated by wild-type, nmf329/+, and nmf329/nmf329 mice ( Figure S2). Also, intercrosses between heterozygotes produced deaf progeny at a frequency of 28%, suggesting that nmf329/nmf329 mice are born at the Mendelian ratio (i.e. 25%), and that the nmf329 mutation does not lead to embryonic lethality. Histopathological examination of the heart, lung, ileum, colon, liver, and kidney for other defects did not reveal abnormalities in the nmf329/nmf329 mice (data not shown). Overall, these results indicate that the nmf329 mutation leads to recessively inherited non-syndromic deafness.
Cochlear hair-cell degeneration in the nmf329 mouse line Histological analysis of cross-sectioned cochleas from nmf329/ nmf329 mice (P28) demonstrated that the overall structure of the hearing organ was maintained in the nmf329 line, and that the stria vascularis, Reissner's membrane, and spiral ganglion neurons were morphologically normal ( Figure S3). Furthermore, no signs of inflammation were detected in the hearing organ ( Figure S4). However, the organ of Corti was deformed at the basal turn of the nmf329/nmf329 cochlea; it collapsed into a compact cell layer that lacked paracellular spaces, and it contained only two rows of outer hair cells (OHCs, Figure 2B). This malformation was not present along the entire length of the hearing organ; at the apical turn, the organ of Corti appeared intact, with three rows of OHCs present ( Figure 2D). In contrast, cochlear sections from nmf329/+ mice (data not shown) and wild-type mice (Figure 2A and 2C) exhibited normal morphology both at the basal and apical turns. These data indicate that the nmf329 mutation leads to a loss of OHCs at the basal turn of the cochlea by P28.
To assess the extent of the hair-cell loss further, organ of Corti samples were dissected from wild-type, nmf329/+, and nmf329/ nmf329 mice at various stages of development, and stereociliary bundles and other actin-rich structures were visualized using fluorescently labeled phalloidin. In young nmf329/nmf329 animals (P8), the stereociliary bundles of IHCs and all three rows of OHCs were present and intact ( Figure 3B and 3C), suggesting that the initial steps of hair-cell development were unaffected by the nmf329 allele. However, by P14, widespread degeneration was observed in the organ of Corti in the homozygous mutant mice. Specifically, at the basal turn, most OHC stereociliary bundles were missing, especially from the first and second rows of OHCs ( Figure 3E; low magnification images and statistical analyses are shown in Figure  S5, S6, S7). At the apex of the cochlea, the defect was milder: approximately half of the stereociliary bundles were missing from

Author Summary
Hereditary deafness is a common birth defect in the human population, yet the majority of genes required for audition is thought to be unidentified. Genetic approaches in the mouse have greatly contributed to our understanding of the molecular mechanisms that underlie hearing. Random mutagenesis of mice, identification of deaf mutants, and subsequent analysis of the deafness-causing gene defects has led to the discovery of several previously unrecognized gene functions. Here, we report on the characterization of a new mutant mouse line (nmf329) that exhibits profound hearing loss and loss of sensory cells in the auditory organ. Genetic analysis reveals that these animals carry a mutation in the claudin-9 gene, which encodes a protein with hitherto unknown biological function. We have found that normal claudin-9-but not a mutant form-inhibits the paracellular movement of certain ions. The lack of the claudin-9 ion barrier in the inner ear leads to changes in ionic conditions that can account for the loss of sensory cells in the mutant mice. Within the cell-cell junctions, the claudin-9 layer is located basal to those of other claudins. Thus, our analysis of claudin-9-deficient animals suggests that even the deeper layers of claudins have important ion barrier function.
the first OHC row, whereas the second and third OHC rows were largely intact ( Figure 3F, Figure S5, Figure S7). Images focused at the level of cell-cell junctions showed that many of the OHCs were missing from the reticular lamina and replaced by large polygonal cells ( Figure 3E and 3F, Figure S6). The loss of OHCs from the reticular lamina was confirmed by visualizing cell-cell junctions in the organ of Corti with an anti-occludin antibody ( Figure S8). In contrast to the widespread loss of OHCs, the IHCs were preserved even at the basal turn of the cochlea in nmf329/nmf329 mice ( Figure 3H and Figure S6). Surprisingly, our analysis of older nmf329/nmf329 mice revealed that the initially rapid degeneration of the hearing organ slowed down after P14; at P80, the basal turn of cochlea still contained a few OHCs ( Figure 3H; Figure S5, S6, S7). At the cochlear apex, the loss of hair cells was particularly apparent in the first row of OHCs ( Figure 3I; Figure S5, S6, S7). In contrast to the homozygous mutants, both wild-type mice and Representative ABR waveforms for nmf329/+ mice and nmf329/nmf329 mice at P28. Broadband click stimuli were applied at the indicated intensities, in dB-SPL. Bold and underlined numbers indicate the auditory thresholds. (C) Mean ABR thresholds (6SEM) to broadband click stimuli and pure-tone sounds in wild-type (+/+), nmf329/+, and nmf329/nmf329 mice at P28. Mice with ABR thresholds higher than the maximum SPL (120 dB or 100 dB, depending on the frequency) were not included in the statistical analysis, and are listed separately (bottom row). The number of mice used for the generation of ABR data is shown in parenthesis. doi:10.1371/journal.pgen.1000610.g001 nmf329/+ mice had three intact rows of OHCs at each time point investigated ( Figure 3A, 3D, and 3G). Overall, these data demonstrate a widespread loss of OHCs in the nmf329/nmf329 mice after P8. Furthermore, the cochlear defect in the mutant mice was morphologically similar to the previously described ''phalangeal scars'', which are formed by supporting cells when OHCs are lost due to a variety of causes, such as noise trauma, ototoxic aminoglycosides, and aging [37,38].

Mapping and identification of the nmf329 allele
The Jackson Laboratory had previously mapped the mutation in the nmf329 mouse line to chromosome 17, ,9.6 cM from the centromere [36]. Since this linkage-map position implicated several hundred genes as potential candidates for nmf329, we refined the location of the nmf329 allele to a physical interval on chromosome 17. Nmf329/nmf329 mice (on the C57Bl/6 background) were outcrossed to the A/J strain, and the F1 offspring were then backcrossed to the nmf329 line. Progeny of this backcross (116 mice) were assessed for hearing by ABR measurement, and analyzed for crossovers by SNP detection over an 8-megabase genomic segment (see Figure 4A). This approach refined the position of the nmf329 locus to a 2.5-megabase region between the SNPs rs33167092 and rs32299331. The identified genomic interval contains 76 annotated genes, none of which had been associated with deafness. Therefore, we selected 3 candidate genes (claudin-9, claudin-6, and Kctd5) based on their detection in cochlear cDNA libraries (Unigene database information, NCBI). Sequence analysis of the candidate genes from the nmf329 line revealed that one of them, claudin-9, contained a T to C point mutation 102 bases downstream of the start codon ( Figure 4A). The c.102TRC alteration led to a phenylalanine (F) to leucine (L) substitution in the first predicted extracellular loop of the encoded protein ( Figure 4B). The affected amino acid (F35) is conserved among all claudin-9 orthologs that have been identified thus far ( Figure S9). Moreover, F35 is conserved amongst many members of the claudin protein family ( Figure 4C). These results suggest that the claudin-9 mutation in the nmf329 mice may have functional consequences.
Opposing effects of wild-type and mutant claudin-9 on transepithelial resistance To explore the molecular function of the wild-type and mutant claudin-9 proteins (claudin-9 wt and claudin-9 F35L , respectively), we examined the potential electrophysiological consequences of claudin-9 wt and claudin-9 F35L expression in an epithelial cell line. Since N-terminal tags do not interfere with the function of claudins but facilitate expression analysis [20], we fused the N-termini of claudin-9 wt and claudin-9 F35L to EYFP. The DNA sequences of these fusion constructs were inserted downstream of a tetracycline response element and transfected into Tet-off MDCK cells, which do not contain detectable claudin-9 mRNA (data not shown). Stable clones of transfected Tet-off MDCK cells were isolated and subsequently cultured in the presence or absence of doxycycline. Fluorescence microscopy demonstrated that doxycycline treatment ''turned off'' the expression of both EYFP-claudin-9 wt and EYFPclaudin-9 F35L , whereas the lack of doxycycline resulted in expression of transfected proteins in virtually all cells ( Figure 5A and 5B). Furthermore, both EYFP-claudin-9 wt and EYFP-claudin-9 F35L were co-localized with the tight-junction protein occludin in the plasma membrane, indicating that the F35L alteration does not prevent the localization of claudin-9 to tight junctions ( Figure 5C and 5D; the EYFP and occludin signals are shown individually in Figure S10).
To evaluate the effects of claudin-9 wt and claudin-9 F35L on transepithelial resistance, we grew the Tet-off MDCK clones on microporous filters, and turned the expression of transfected proteins ''on'' and ''off'' (using doxycycline withdrawal and supplementation, respectively). Expression of claudin-9 wt doubled the transepithelial resistance of Tet-off MDCK monolayers compared to that in control (''off'') cultures ( Figure 5E). In contrast, the expression of claudin-9 F35L lowered the transepithelial resistance of the MDCK epithelium ( Figure 5F). Thus, expression of claudin-9 wt made the MDCK monolayer electrically tighter, whereas claudin-9 F35L expression made the epithelium leakier.
Elimination of claudin-9 ion barrier function by the F35L amino acid substitution The increased transepithelial resistance of claudin-9 wt -expressing epithelia suggested that claudin-9 wt reduced the paracellular permeability for some of the ions in the assay buffer. To characterize the ion permeability profile of both wild-type and mutant claudin-9, we performed dilution-potential assays in Ussing chambers, using the stable clones of claudin-9 wt and claudin-9 F35L -transfected Tet-off MDCK cells. In these experiments, the cells were seeded on Transwell filters, and the main salt component of the assay buffer was diluted 2fold on one side of the MDCK monolayer, while osmolarity was maintained by adding mannitol. Asymmetric dilution of a salt on the apical versus basolateral side of an epithelial monolayer generates concentration gradients of the diluted anions and cations. This leads to transepithelial ion fluxes at different rates, depending on the anion and cation permeabilities of the epithelium. The unequal transepithelial fluxes of anions and cations generate a voltage termed the dilution potential. In control (i.e. claudin-9 ''off'') MDCK monolayers, permeability to Na + was greater than that to Cl 2 , as indicated by the apically negative dilution potential that was generated when both Na + and Cl 2 were diluted on the basolateral side. In MDCK monolayers expressing claudin-9 wt , basolateral dilution of NaCl led to a much smaller voltage change (compare ''on'' and ''off'' bars in Figure 6A). This indicates that the expression of claudin-9 wt either decreased the paracellular permeability to Na + or increased the permeability to Cl 2 . To distinguish between these two possibilities, we replaced Cl 2 with the poorly permeable anion aspartate (Asp). When the dilution potential assay was performed in the NaAsp buffer, claudin-9 wt expression led to a reduction in the change in transepithelial voltage ( Figure 6A). Thus, claudin-9 wt decreased the paracellular permeability for Na + . In arginine chloride buffer (ArgCl), by contrast, claudin-9 wt expression did not alter the dilution potential ( Figure 6A), indicating that claudin-9 wt did not affect the Cl 2 permeability of MDCK cultures. Since the endolymph has a uniquely high K + concentration amongst the extracellular fluids, we also tested the K + permeability of claudin-9 wt tight junctions. In KCl buffer, the dilution potential was reduced by the expression of claudin-9 wt ( Figure 6A). These results suggest that claudin-9 wt formed a paracellular barrier to K + and Na + , which is in agreement with the findings of a recent electrophysiological study in MDCK cells [39].
Next, we studied the ion permeability profile of MDCK epithelia in which claudin-9 F35L expression was turned ''on'' or ''off''. In contrast to the expression of wild-type protein, that of claudin-9 F35L led to an increase in the dilution potential measured in the NaCl, NaAsp, and KCl buffers (compare ''on'' and ''off'' bars in Figure 6B). However, claudin-9 F35L did not cause a change in the dilution potential when measured in the ArgCl buffer ( Figure 6B). Thus, the mutant protein increased the K + and Na + permeability-but not the Cl 2 permeability-of tight junctions in MDCK monolayers. Taken together, these data strongly suggest that claudin-9 wt forms paracellular ion permeability barriers to Na + and K + . Moreover, this cation barrier function is eliminated by the F35L substitution found in the claudin-9 protein of the nmf329 mouse strain.
Claudin-9 expression and increased perilymphatic K + concentration in the cochleas of nmf329 mice To study the effect of the F35L substitution on claudin-9 localization in vivo, we performed immunohistochemistry experiments with a polyclonal antibody that recognizes the unique C- terminus of claudin-9. In the organ of Corti from both nmf329/ nmf329 mice and their wild-type littermates, staining was visible at the junctional complexes of hair cells and their supporting cells ( Figure 7A). In addition to its expression in the organ of Corti, claudin-9 was also detected at the cell-cell borders in the inner sulcus, outer sulcus ( Figure 7A), stria vascularis, and Reissner's membrane (data not shown), in both homozygous mutant and wild-type mice. The specificity of the immunostaining was verified in two separate sets of negative control experiments, in which we either replaced the anti-claudin-9 antibody with normal goat IgG or pre-incubated the anti-claudin-9 antibody with a blocking peptide (data not shown). Overall, our observations indicate that the F35L amino acid substitution did not prevent claudin-9 from localizing to the plasma membrane in the hearing organ.
Next, we investigated whether claudin-9 F35L is localized to the apex of OHCs or dislocated towards the cell base. To visualize the apical surface of epithelial cells, we incubated cochlear preparations with fluorescently labeled wheat germ agglutinin (WGA) which binds to the specific carbohydrates on the surface of nonpermeabilized cells. Following WGA labeling, tissue preparations were immunostained with the anti-claudin-9 antibody. Z-stack confocal microscopy images showed that both claudin-9 wt and claudin-9 F35L were localized to the apex of OHCs ( Figure 7B; WGA and claudin-9 signals are shown individually in Figure S11). In addition, transmission electron microscopy images of freeze fracture replicas revealed that the tight junction strands in the organ of Corti of wild-type and nmf329/nmf329 mice at P5 were similar ( Figure S12). Furthermore, the expression pattern of another functionally important claudin in the cochlea-claudin-14 [40]-was not altered in the hearing organ of nmf329/nmf329 mice ( Figure S13). These results strongly suggest that the F35L alteration in claudin-9 does not affect the subcellular localization of claudin-9 and the organization of tight-junction strands in the OHCs.
To evaluate the electrophysiological consequences of the claudin-9 mutation in vivo, we measured the endolymphatic and perilymphatic K + concentrations, as well as the EP, in nmf329/ nmf329 mice and control (nmf329/+) littermates at P70-P80. These assays were performed in deeply anesthetized mice using doublebarreled microelectrodes, as we have described for other mouse models [41,42]. The EP and the endolymphatic K + concentrations were similar in the nmf329/nmf329 and nmf329/+ mice, indicating that the stria vascularis functioned normally in the homozygous mutant animals ( Figure 7C and 7D). However, the perilymphatic K + concentration was significantly higher in the nmf329/nmf329 mice than in their littermate controls ( Figure 7D). These results show that the claudin-9 defect in the nmf329 line does not abolish the endocochlear K + and voltage gradients.

Lack of OHC degeneration in the claudin-9-defective cochleas under organ culture conditions
We speculated that the small increase in the perilymphatic K + concentration of nmf329/nmf329 mice could reflect a spatially restricted defect of tight junctions. Also, the local K + level close to the ''leaky'' epithelium may be higher than the K + concentration in the bulk of the perilymph. To test whether extracellular conditions play a role in the hair-cell defect of nmf329/nmf329 mice, we examined the rates of OHC degeneration in organ culture (5.4 mM K + concentration). Organ of Corti samples from nmf329/nm329 mice and control (nmf329/+) littermates were explanted on P5 ( Figure 8A and 8D). The explanted tissue was maintained under previously described culture conditions that do not delay the differentiation of OHCs [43]. After 9 days of culturing, stereocilia were visualized by staining for F-actin. For comparison, organ of Corti samples were removed from 14-day old nmf329/nmf329 mice, and were stained for F-actin. In contrast to the in vivo loss of OHCs, all three rows of OHCs survived when the organ of Corti samples from nmf329/nmf329 mice were cultured ex vivo (compare Figure 8B and 8C). The OHCs of control (nmf329/+) littermates did not exhibit signs of degeneration either in organ culture or in vivo ( Figure 8E and 8F). Thus, although the organs of Corti of nmf329/nmf329 mice and littermate controls show different morphology in vivo, they are indistinguishable when maintained ex vivo (see OHC counts in Figure S14). These data are consistent with the notion that the degeneration of OHCs in the nmf329 mouse line is due to unfavorable extracellular conditions.

Deletion of the pou3f4 gene prevents the loss of OHCs in nmf329 mice
The pou3f4 transcription factor is required for the generation of EP, but it is not necessary for the development of hair cells [44]. In the nmf329 line, the loss of OHCs occurs after the endocochlear K + concentration gradient is established (P2-P8 [45]) but coincides with the development of EP (P8-P16 [46]). Therefore, we tested whether the EP-dependent increase in the K + -driving force contributes to hair-cell loss in the claudin-9 mutant mice in vivo. EP in the claudin-9-defective cochleas was reduced by breeding nmf329 homozygotes with a mouse line lacking the pou3f4 gene. Animals in the second backcross generation were examined for potential alterations in the morphology of OHC at P14-15. Figure 6. Claudin-9 wt but not claudin-9 F35L reduces the paracellular permeability for K + and Na + . Dilution potential across (A) claudin-9 wt -and (B) claudin-9 F35L -transfected Tet-off MDCK cell clones, as assessed using Ussing chambers. Expression of the two claudin-9 constructs was turned ''off'' and ''on'' using doxycycline treatment as indicated. For baseline recordings, the basal and apical chambers were filled with the indicated buffers. For measurement of the dilution potential, the buffer on the basal side was replaced with a ''K buffer'' that contained half the concentration of the main salt component indicated below each pair of columns (the osmolarity was maintained by substituting mannitol). Expression of claudin-9 wt decreased the dilution potential in the NaCl, NaAsp, and KCl buffers, which suggests that this protein has a K + and Na + barrier function (mean6SEM, n = 3 in each of the 3 tested clones, paired t test, ***p,0.0001). Expression of claudin-9 F35L did not affect the dilution potential in the ArgCl buffers, but increased the dilution potential in the NaCl, NaAsp, and KCl buffers (mean6SEM, n = 3 in each of the 3 tested clones, paired t test, *p,0.05). doi:10.1371/journal.pgen.1000610.g006 Actin staining of organ of Corti samples indicated that OHCs were intact in the nmf329/+:pou3f4 Y/2 animals ( Figure 9A), which is consistent with previous histological analyses of pou3f4 Y/2 mice [44,47]. Nmf329/nmf329 mice with an unaltered pou3f4 gene (nmf329/nmf329:pou3f4 Y/+ ) showed widespread loss of OHCs at P14 ( Figure 9B). In contrast, nmf329/nmf329 littermates in which the pou3f4 gene was deleted (nmf329/nmf329:pou3f4 Y/2 ) exhibited a morphologically intact organ of Corti with three rows of OHCs ( Figure 9C and 9D). Thus, deletion of the pou3f4 gene in the nmf329 line prevented the loss of OHCs.

Discussion
In this study, we show that claudin-9 deficiency is associated with deafness. We also show that wild-type claudin-9 is a paracellular ion permeability barrier for Na + and K + , and that the claudin-9 mutation of deaf nmf329 mice eliminates the ion barrier function of the encoded protein without preventing its plasma membrane localization. In addition to being expressed in the cochlea, claudin-9 has been detected in the vestibular system [35], liver [48], and developing kidney [49], yet claudin-9 mutant mice exhibited no signs of vestibular, hepatic, or renal defects. Thus, the ion barrier function of claudin-9 is essential in the cochlea, but appears to be dispensable in other organs.
Sound detection requires uniquely large K + and voltage gradients across the sensory hair cells of the hearing organ. During sound stimulation, hair cells are depolarized by an apical influx of K + ions. Subsequently, hair cells repolarize as K + ions are exported into the basolateral fluid. An increase in the K + concentration in the basolateral fluid is known to inhibit immunostaining (green) of WGA-labeled (red) organ of Corti samples from wild-type and nmf329/nmf329 mice (P5). Horizontal blue lines indicate the position of Z stack slices shown in the upper panels. OHC rows are indicated with arrows. Scale bars: 10 mm. (C) Endocochlear potential (EP) was measured in the cochleas of nmf329/nmf329 mice and nmf329/+ littermates in vivo (P70-P80). The EPs were not significantly different in the two groups (mean6SEM, n = 5 mice per group, unpaired t test). (D) In vivo measurement of K + concentration in the endolymph and perilymph of nmf329/ nmf329 mice and nmf329/+ littermates (P70-P80). The perilymphatic K + concentration was higher in the nmf329/nmf329 cochleas, whereas the endolymphatic K + concentration was similar in cochleas of both genotypes (mean6SEM, n = 5 mice per group, unpaired t test, *p = 0.016). doi:10.1371/journal.pgen.1000610.g007 repolarization and to be toxic to hair cells [30,32]. Therefore, the basolateral fluid of hair cells must be insulated from the K + -rich apical milieu. This insulation is likely to be impaired in the nmf329 mice because: 1) the nmf329 mutation eliminated the K + barrier function of claudin-9, 2) the timing of hair-cell degeneration in the nmf329 mice coincided with the development of EP, 3) the OHCs of claudin-9 mutant mice did not degenerate ex vivo under normal culture conditions, and 4) the OHCs of claudin-9 mutant mice were rescued from degeneration by a second gene defect that reduces the K + -driving force in the cochlea. Since K + can freely diffuse from the basolateral fluid of the hair cells to the perilymph in the scala tympani, it may seem surprising that the K + concentration was only marginally elevated in the scala tympani of claudin-9 mutant mice. However, the K + concentration in the bulk of the perilymph may be significantly lower than the local K + level in the basolateral fluid, because K + is likely to become proportionally diluted as the distance from the ''leaky'' epithelium increases. In addition, simple diffusion is insufficient to protect the hair cells from K + intoxication. Indeed, even during the normal hearing process, K + is thought to be removed from the basolateral fluid by the K-Cl co-transporters of the supporting cells known as Kcc3 and Kcc4, rather than by diffusion, and genetic deletion of either Kcc3 or Kcc4 causes hair-cell degeneration [50,51]. Thus, we suggest that the claudin-9 defect of nmf329 mice leads to toxic levels of K + in the basolateral fluid of hair cells, but that K + ions are diluted to a nearly physiological concentration as they diffuse across the basilar membrane into the scala tympani.
The extent of hair-cell degeneration in the nmf329 cochlea was not uniform throughout the organ. More OHCs were missing at the basal turn than at the apex, and the OHCs were more affected than the IHCs. A similar pattern of hair-cell loss is often observed in damaged cochleas. However, in nmf329 mice, the pattern of OHC loss was not mirrored by the frequency distribution of the ABR thresholds: we detected uniformly elevated auditory thresholds whether the sound stimulated the base or the apex of the cochlea ( Figure 1C). These data suggests that although the majority of OHCs at the apex of the claudin-9-deficient cochlea survive, they do not function properly. Thus, a ''sublethal'' functional defect of OHCs may be ubiquitous in the claudin-9deficient cochlea. This functional defect may culminate in the loss of OHCs, depending on the susceptibility of individual OHCs to the local basolateral K + concentration.
IHCs are the primary sensory cells of the auditory system, whereas the OHCs are eletromechanical amplifiers that improve hearing sensitivity by more than 40 dB. A previous study showed that the ABR threshold to pure tone sounds was increased by 40 dB when the OHCs (but not the IHCs) in parts of the cochlea were destroyed by kanamycin treatment [52]. Furthermore, the hearing threshold was elevated by 40-60 dB in mice deficient of prestin, a molecule key to eletromechanical amplification by OHCs [53]. In the nmf329 mouse line, the auditory threshold was increased by ,60 dB at P28 (Figure 1C), and the IHCs were morphologically intact ( Figure 3). Thus, the phenotype of nmf329 mice is compatible with a predominantly OHC defect, and the IHCs seem to retain at least some sensory function in the mutant mice up to P28.
Claudin-9 is widely expressed in the inner ear [35], so it is surprising that the claudin-9 mutation does not lead to reductions in either the endolymphatic K + concentration or the EP. However, 10 other claudins have been detected in the inner ear [34,35]. Some of these proteins are present at low levels or can be detected only in certain areas of the cochlea. Nevertheless, the collective expression pattern of multiple claudins may restrict the paracellular leak of claudin-9-deficient tight junctions to a limited area such as the organ of Corti. A localized K + leak may be compensated by increased K + secretion from the stria vascularis, and this could stabilize both the EP and the endolymphatic K + concentration.
Claudin-9 F35L was localized to tight junctions in transfected MDCK cells and in cochlear epithelial cells. Furthermore, heterologous expression of claudin-9 F35L reduced the tightness of tight junctions in MDCK cell cultures and increased the paracellular permeability to K + and Na + . Based on these data, we propose that claudin-9 F35L is incorporated into tight-junction strands, resulting in the disruption of the paracellular cation barrier. Thus, a phenotype might be expected not only in the nmf329/nmf329 mice but also in the nmf329/+ animals. However, the nmf329/+ mice exhibited normal auditory thresholds and intact OHCs at each time point investigated. This suggests that tight-junction strands containing both claudin-9 wt and claudin-9 F35L are sufficiently tight to prevent cochlear degeneration. Alternatively, claudin-9 wt and claudin-9 F35L may be sorted into separate tight-junction strands, and the number of claudin-9 wtcontaining strands may be sufficient to maintain the paracellular cation barrier. An additional possibility is that claudin-9 wt is expressed at a higher level than claudin-9 F35L in the heterozygous mice.
The claudin-9 sequence is highly conserved between mouse and human ( Figure S9), which raises the possibility that a claudin-9 deficiency may cause hearing loss not only in mice but also in humans. Although no deafness gene has been mapped to human chromosome 16p13 where claudin-9 is located, a large number of deafness loci likely await recognition and mapping [54]. Interestingly, a DNA donor has already been identified with a nonsynonymous SNP (id#: rs34769999) in claudin-9 that leads to a non-conservative amino acid substitution in the encoded protein (R116C). Although the DNA donor carrying this nonsynonymous SNP was heterozygous for the R116C alteration, future studies may identify deaf patients homozygous for genetic alterations in claudin-9.
In addition to claudin-9, several other claudins have been detected in the junctional complexes of the organ of Corti [34]. Most notably, claudin-14 has been localized to the junctional complexes of hair cells and supporting cells [22,35]; mutations in claudin-14 have been shown to cause extensive hair-cell loss and deafness [22,40]. Thus, the phenotypes of claudin-9 mutant mice and claudin-14 ''knock-out'' animals are similar. In the junctional complexes of OHCs, claudin-14 and claudin-9 are sorted into two separate subdomains; claudin-14 is found only in the most apical tight-junction strands, whereas claudin-9 is detected solely in the deeper (subapical) strands [34]. In addition, both claudin-9 and claudin-14 form paracellular ion permeability barriers for K + (Figure 6 and ref. [22]). Therefore, in the organ of Corti, claudin-14 would be expected to render claudin-9 redundant. In contrast to this expectation, analysis of the nmf329 mice suggests that not only the apical tight-junction strands, but also those that are subapical, can contribute significantly to the ion barrier capacity of junctional complexes. In summary, our characterization of the nmf329 mouse line not only reveals the biological significance of claudin-9, but also provides insight into the functional architecture of tight-junction complexes in the hearing organ.

Mouse lines and genotyping
Heterozygous nmf329 mice (C57BL/6J-nmf329/J) and hemizygous pou3f4 mutant mice (C3HeB/FeJ-Pou3f4del-J/J) were obtained from The Jackson Laboratory. Experimental procedures were approved by the Animal Care and Use Committees of the University of Iowa and Kansas State University. For genotyping, the one-nucleotide difference between the wild-type and mutant claudin-9 alleles was detected by PCR using HotStart Taq DNA polymerase (Qiagen), tail DNA extracts, a forward primer that anneals to a sequence identical in the wild-type and mutant genes (59-TGGTTCATGGCAGATCTGGAGG-39), and HPLC-purified reverse primers whose 39 ends anneal to the affected nucleotide (59-ACGATGCTGTTGCCGATG/A-39 for the wildtype and nmf329 alleles, respectively). The presence of the pou3f4 allele was detected by PCR using the following primers: 59-CACTCTGATGAAGAGACTCCAAC-39 and 59-CACCGT-GTGCGAATAAACCTC-39.

Mapping of the nmf329 locus
The nmf329 mice were crossed to the A/J strain. Female F1 mice from the intraspecific cross were backcrossed to nmf329/ nmf329 animals. F2 progeny were generated and analyzed for both hearing threshold (ABR measurement) and crossovers (SNP detection). Scanning of the Mouse Genomic Informatics database revealed 9 suitable SNPs over an 8-megabase genomic region, in the following genes: Vmn2r105 (rs29506797), Zfp160 (rs33247972), 4930432O21Rik (rs32299331), Zfp758 (rs32470819), Kctd5 (rs33185969), Gbl (rs33167092), Tsc2 (rs33261332), Lmf1 (rs29499590), and Uhrf1bp1 (rs33716796). Short DNA sequences (,500 bp) encompassing the SNPs were PCR amplified, and the PCR products were denatured by heat and re-annealed by slow cooling. Thus, if two different alleles were amplified, they formed heterodimers with a single nucleotide mismatch. For the detection of DNA heterodimers, the re-annealed PCR products were treated with a plant enzyme (CEL1) that cleaves DNA at points of mismatch [55]. DNA cleavage was analyzed by agarose gel electrophoresis.

Cloning of claudin-9 and expression in Tet-off MDCK cells
Total RNA was isolated from the inner ear of wild-type and nmf329/nmf329 mice and reverse transcribed. The coding regions of the claudin-9 wt and claudin-9 F35L cDNAs were PCR amplified and subcloned into the EYFP-C1 vector (Clontech). The EYFPclaudin-9 wt and EYFP-claudin-9 F35L fragments were further subcloned into the pUHD10-3 vector. Linearized plasmids were transfected into Tet-off MDCK II cells (Clontech) using the Effectene transfection reagent (Qiagen), and stable clones were isolated following hygromycin (300 mg/ml) selection. Expression of endogenous claudin-9 in MDCK cells were tested using RT-PCR, DNAse-treated MDCK cell RNA, and the following primers: 59-CTTCAACAGCCCTTGAACTC-39, and 59-TAGTCCCTC-TTGTCCAGCC-39.

Measurement of TER
EYFP-claudin-9 wt -and EYFP-claudin-9 F35L -transfected stable clones of Tet-off MDCK II cells were seeded onto Transwell inserts (9610 4 cells/cm 2 ) and incubated in the presence or absence of doxycycline (100 ng/ml) for 8 days. At the end of the incubation period, the electrical resistance across the cultures was measured using a Millicell-ERS (Millipore) ohmmeter in the cell culture medium (DMEM). Electrical resistance was also measured across blank inserts. TER was calculated by subtracting the resistance measured across the blank inserts from the resistance measured across the inserts with the MDCK monolayers.

Measurement of dilution potential
EYFP-claudin-9 wt -and EYFP-claudin-9 F35L -transfected stable clones of Tet-off MDCK II cells were seeded onto Transwell inserts (9610 4 cells/cm 2 ) and incubated in medium with or without doxycycline (100 ng/ml) for 8 days. At the end of the incubation period, Transwells were mounted into a custom-made Ussing chamber system (Jim's Instruments, Iowa City, Iowa). The apical and basal chambers were filled with one of the following solutions: NaCl buffer, NaAsp buffer, ArgCl buffer, or KCl buffer. These four buffers differed only in their main salt component, as indicated by their names (140 mM NaCl, 140 mM NaAsp, 140 mM ArgCl, and 140 mM KCl); the other components were identical in every buffer (2 mM CaCl 2 , 1 mM MgCl 2 , 10 mM glucose, 5 mM Tris, pH was adjusted to 7.3 with HCl).
The Ussing chambers were connected to Ag/AgCl electrodes via 3 M KCl agar bridges, and the electrodes were connected to a Quick Data Acquisition DT9800 USB board (Data Translation, Inc, Marlborough, MA) to record the voltage between the apical and basal chambers. The asymmetry of voltage-sensing Ag/AgCl electrodes and the liquid junction potentials were corrected using an offset-removal circuit. During the experiment, buffers were maintained at 37uC and bubbled constantly with air. Dilution potentials were measured by replacing the buffer on the basal side with a ''K buffer''. In select experiments, the assay buffer was diluted on the apical side of MDCK cultures to verify that the absolute value of the dilution potential did not depend on the direction of the ion gradients. The K NaCl, K NaAsp, K ArgCl, and K KCl buffers contained half the concentration (70 mM) of the salt indicated by their names, and mannitol at a concentration of 140 mM (to maintain osmolarity); concentrations of the other buffer components (i.e. Ca 2+ , Mg 2+ , glucose, Tris, H + ) were not altered. Buffer replacement was considered complete after 3 quick, successive washes with any given buffer.

ABR measurements
ABR measurements were performed as previous described [56]. In brief, mice were anesthetized with ketamine and xylazine, and were placed on a heating pad in a sound-attenuating chamber. Needle electrodes were introduced just under the skin, with the active electrode placed between the ears above the vertex of the skull, the ground electrode between the eyes, and the reference electrode underneath the left ear. Mice were presented with 1024 pure-tone or click stimuli at a rate of 21.1/s (Intelligent Hearing Systems, Miami, FL). Responses were bandpass filtered (100-1500 Hz) and recorded for 12 ms. Thresholds were determined by increasing the sound intensity in 10 dB increments, followed by 5 dB increases and decreases to determine the lowest level at which a distinct ABR wave pattern could be recognized.

Rotarod test
The balance and motor coordination of WT, nmf329/+, and nmf329/nmf329 mice were studied using rotarod tests as previously described [57]. Mice were placed onto a cylinder rotating at fixed speed (10 rpm), and latency to fall was measured for a maximum of 180 s. Each mouse underwent 4 trials. The latency to fall was measured in the 2 nd , 3 rd , and 4 th trials.

Histology and immunofluorescence
For routine histological examination, the temporal bones were fixed in 4% PFA for 24 hours (4uC) and decalcified in Immunocal solution (Decal Chemical Corporation). The heart, lung, ileum, colon, liver, and kidney were fixed in Prefer solution (Anatech Ltd). Samples were embedded in paraffin, sectioned, deparaffinized, and stained with hematoxylin and eosin. For F-actin labeling, WGA labeling and immunostaining, the organ of Corti was dissected from the cochlea and fixed in one of the following solutions: Prefer solution (WGA labeling, claudin-9 immunostaining and claudin-14 immunostaining), 4% PFA (F-actin labeling and CD68 immunostaining), 10% TCA (occludin immunostaining). MDCK cells were fixed in ethanol (30 min, 4uC) and acetone (3 min, 220uC). When visualization of the cell surface was required, WGA-Alexa Fluor 594 (10 mg/ml) (Invitrogen Corp.) was added to the specimen before permeabilization. Tissue samples were permeabilized with 0.5% Triton X-100 in PBS for 20 min, and blocked with BSA. The following reagents and antibodies were used: phalloidin-Alexa Fluor 488 (Invitrogen Corp.), goat anti-claudin-9 antibody (C-20, Santa Cruz Biotech Corp.), rabbit anti-claudin-14 antibody (Invitrogen Corp.), monoclonal rat anti-CD68-Alexa Fluor 488 antibody (AbD Serotec, Raleigh, NC, USA), monoclonal anti-occludin antibody labeled with either Alexa Fluor 488 or Alexa Fluor 594 (Invitrogen Corp.). The sensitivity of the anti-CD68 antibody was verified using activated peritoneal macrophages as previously described [58]. The specificity of the anti-claudin-9 antibody was verified by immunofluorescence in claudin-9-and vector-transfected HEK293 cells. The specificity of the immunostaining in the organ of Corti was tested by pre-incubating the anti-claudin-9 antibody with a blocking peptide (10 mg/ml, Santa Cruz Biotech Corp.) for 1 h before adding it to the tissue samples. In additional control experiments, anti-claudin-9 and anti-claudin-14 antibodies were replaced with normal IgG from goat and rabbit (Calbiochem), respectively. Secondary anti-goat and anti-rabbit antibodies were labeled with Alexa Fluor 488 (Invitrogen Corp.). Images were obtained using an LSM-510 confocal microscope (Carl Zeiss Inc.).

Freeze-fracture analysis
Organ of Corti samples were fixed in 2.5% glutaraldehyde solution supplemented with 3 mM CaCl 2 and buffered with cacodylate (0.1M, pH 7.2) for 2 hours at room temperature. Tissue samples were then cryoprotected in 30% glycerol, mounted on gold stubs with central wells, and frozen by plunging into liquid propane. After freezing, the samples were fractured in a Balzers 301 unit using a fracture knife, followed by platinum/carbon replication. Replicas were washed in household chlorine bleach, rinsed in distilled water, and retrieved on 200-mesh formvar coated copper grids. The images of the replicas were recorded using a JEOL JEM-1230 TEM microscope equipped with a Gatan Ultrascan 1000 camera.

Measurements of EP and the K + concentration in vivo
Adult mice (P70-P80) were anesthetized with 4% tribromoethanol (13 ml/g body weight i.p.), and the K + concentration and the EP were measured using double-barreled microelectrodes, as described previously [41,42]. The EP electrode was filled with 500 mM NaCl; the K + -selective barrel was silanized and the tip filled with a liquid ion exchanger (Fluka 60398, K + ionophore I-Cocktail B) that was backfilled with 150 mM KCl. Measurements in the basal turn of the cochlea were made by a round-window approach, through the basilar membrane of the first turn. The K +selective electrode was calibrated in solutions with known cation (K + and Na + ) concentrations in situ at 37uC.

Ex vivo culture of the organ of Corti
Organ of Corti samples were dissected from the cochleas of nmf329/nmf329 and nmf329/+ mice on P5. The tissue pieces were mounted on poly-L-lysine-coated glass coverslips, and cultured for 9 days in Neurobasal-A medium containing N2 supplement, B-27 supplement, BDNF (50 ng/ml), and L-glutamine (0.5 mM) as previously described [43].

Statistical analysis
Results are presented as the mean6SEM. The results were statistically evaluated using the paired t test, unpaired t test, or oneway ANOVA followed by a post hoc Dunnett's test, as indicated in the figure legends. Figure S1 Early-onset hearing loss in the nmf329 strain. ABR thresholds (dB-SPL) to broadband click stimuli in wild-type (+/+), nmf329/+, and nmf329/nmf329 mice at P16. Data are mean6SEM (n = 5; one-way ANOVA, p,0.0001; post hoc Dunnett's test, control group is +/+, ** p,0.01).  . At P14, the organ of Corti is undamaged in the +/+ mouse (C), whereas several stereociliary bundles of OHCs are missing from the nmf329 cochlea, especially from the first row (D). At P80, the control cochlea is intact (E), but many OHC bundles are missing from the nmf329/nmf329 cochlea (F). Arrowheads indicate points where two images of the same cochlea were joined digitally to provide in-focus pictures for the entire tissue preparation. Scale bars: 100 mm. The actin content of stereociliary bundles was visualized in the organ of Corti preparations from +/+ and nmf329/nmf329 mice, using phalloidin-Alexa Fluor 488. At P8, the stereociliary bundles are present in all three rows of OHCs in the cochlea of +/+ (A) and nmf329/nmf329 mice (B, C). At P14, the organ of Corti is intact in the +/+ mouse (D), whereas the basal (E) and apical turns (F) in the nmf329 cochlea lack numerous stereociliary bundles. At P80, the control cochlea is undamaged (G), but the nmf329/nmf329 cochlea contains only a few OHCs at the basal turn (H, arrowheads), and most OHCs are missing from the first row at the apex (I). Panels E and F are epifluorescence images; all other panels are confocal microscopy images. Scale bars: 10 mm.