KCNK5 channels mostly expressed in cochlear outer sulcus cells are indispensable for hearing

In the cochlea, K+ is essential for mechano-electrical transduction. Here, we explore cochlear structure and function in mice lacking K+ channels of the two-pore domain family. A profound deafness associated with a decrease in endocochlear potential is found in adult Kcnk5−/− mice. Hearing occurs around postnatal day 19 (P19), and completely disappears 2 days later. At P19, Kcnk5−/− mice have a normal endolymphatic [K+] but a partly lowered endocochlear potential. Using Lac-Z as a gene reporter, KCNK5 is mainly found in outer sulcus Claudius', Boettcher's and root cells. Low levels of expression are also seen in the spiral ganglion, Reissner's membrane and stria vascularis. Essential channels (KCNJ10 and KCNQ1) contributing to K+ secretion in stria vascularis have normal expression in Kcnk5−/− mice. Thus, KCNK5 channels are indispensable for the maintenance of hearing. Among several plausible mechanisms, we emphasize their role in K+ recycling along the outer sulcus lateral route.

I n cochlear hair cells, conversion of sounds into electrical impulses relies on mechanical activation of their apical cationic channels. K þ entering across these channels depolarizes the hair cells leading to signal transduction and synaptic transmission at the afferent auditory nerve. K þ is then released into extracellular spaces through the hair cell basolateral membranes and recirculated back to the endolymph mostly by its secretion from the stria vascularis. The molecular and cellular basis for the K þ recycling mechanism remain quite hypothetical [1][2][3][4] , although it has long been established that K þ supply to the stria vascularis originates more from perilymph than blood 5 . Several lines of evidence suggest that K þ is recycled through distinct intracellular routes. Indeed, early studies pointed out the role of non-sensory cochlear cells after K þ release from outer hair cells; this lateral pathway involves supporting cells (Deiters' and Hensen's cells), outer sulcus cells (Claudius', Boettcher's and root cells), spiral ligament fibrocytes and stria vascularis cell layers 6 . Another route has also been proposed from the inner hair cells to endolymph through a medial pathway involving border and inner sulcus cells, limbal fibrocytes and interdental cells 7 . Pathways contributing to K þ cycling may not be limited to these 8 , as K þ may also be directly reabsorbed from perilymph by both limbal and spiral ligament fibrocytes, as well as from endolymph through outer sulcus cells although this path seems limited to the cochlear apical turns 6,9,10 .
Gene mutations that cause deafness highlight the importance of the sophisticated assembly of channels, transporters and gap junctions required for intracochlear K þ recycling 2,3 . The K þ channel KCNQ4 that probably forms the major pathway used by K þ to exit the hair cells is located into their basolateral membranes 11,12 . Kcnq4 deletion leads to progressive hair cell degeneration and deafness in humans lasting from early childhood to puberty 11 . Similarly, knockout mice lacking the K þ -Cl À co-transporter KCC3 (ref. 13) or KCC4 (ref. 14) suffer from hair cell degeneration and deafness. However, Kcc3 deletion triggers a progressive hearing loss spanning the animal's first year of life 13 , whereas Kcc4 deletion produces a rapid deafness starting within a couple of weeks after the hearing onset 14 . Another major player that mediates K þ transport has been characterized as the inward rectifier KCNJ10 channel expressed in root cells 4,15 and in intermediate cells of the stria vascularis, where this channel is a key element in generating the endocochlear potential (EP) 2,16 . Consistent with this, greatly compromised hearing in Kcnj10 mutant mice is associated with the absence of an EP, and reduced endolymphatic [K þ ] and volume 16 . Gene inactivation for the Na þ -K þ -2Cl À co-transporter NKCC1 expressed in outer sulcus, spiral ligament and marginal cells, induces a progressive deafness that starts in juvenile mice 17 and evolves over several dozens of weeks 18 . Finally, connexins 26 and 30 whose mutations are common causes of genetic deafness in humans are located in the gap junction networks, including the supporting cells, the spiral ligament fibrocytes, and both basal and intermediate cells of the stria vascularis 19,20 . This short review supports the concept of K þ being transported through networks of cells. However, alternative pathways other than these described above cannot be excluded, and the relative importance of any one of these pathways is still unclear.
Two-pore domain potassium (K2P) channels produce background or leak K þ currents over a large range of membrane potential 21,22 . In the brain and the kidney, they play major roles in setting membrane potential, pH homeostasis and/or reabsorption of bicarbonate [23][24][25][26] . Interestingly, significant levels for KCNK1, 2, 5, 6 and 17 have been detected from a survey of various inner-ear complementary DNA collections 27 . Indeed, in the inner ear, several members of the K2P channel family have been identified in both vestibular (dark cells and afferent nerve fibres) and cochlear (stria vascularis, spiral ganglion cells and Reissner's membrane) components 28,29 . Changes in profiles of K2P channel expression across development suggested various roles for K2P channels in postnatal maturation of the cochlea, K þ recycling and/or fluid homeostasis 30 , but their potential role in auditory function has not been examined. Therefore, we analysed the implication of several K2P channels (KCNK3, KCNK5 and KCNK9) in hearing. We found that only Kcnk5 knockout mice had severe auditory impairment, whereas Kcnk3 and/or Kcnk9 gene deletion produced no obvious hearing deficit. Deafness in null Kcnk5 mice appeared extremely early, within 2 days after the hearing onset. The organ of Corti showed progressive morphological alterations, with early damage of the sensory hair cells and supporting cells in young mice, followed by complete disappearance of epithelial outer sulcus cells above the basilar membrane in the adults, that is Claudius' and Boettcher's cells. Adult, but not juvenile, Kcnk5 À / À mice also presented a late degeneration of spiral ganglion neurons. These structural changes were associated with an age-dependent decrease in EP. However, the expression of KCNJ10, KCNQ1 and pendrin was not altered. Overall, our results demonstrate an essential role of KCNK5 in the maintenance of hearing. Our data support the hypothesis that KCNK5 in outer sulcus cells play an important role in K þ recycling. Other mechanisms mediated by KCNK5 channels including endolymphatic pH-buffering, K þ secretion and/or leakage and reabsorption may also contribute to the present findings.

Results
Deletion of Kcnk5 but not Kcnk3 and/or 9 alters hearing. The role of K2P channels in hearing was first tested in adult mice (P30-P90) by recording far-field auditory brainstem-evoked potentials. Similar patterns of auditory brainstem response (ABR) waves were recorded from all Kcnk3 À / À (n ¼ 9), Kcnk9 À / À (n ¼ 4), Kcnk3-9 À / À (n ¼ 12), Kcnk5 þ / þ (n ¼ 6) and Kcnk5 þ / À (n ¼ 19) mice, whereas no signal was detected from the Kcnk5 À / À (n ¼ 18) mice. Figure 1a presents ABR waves for the three Kcnk5 mice types. Audiograms (Fig. 1b), displaying the mean acoustic pressure values of the ABR thresholds for a wide range of frequencies, showed a similar acoustic sensitivity (no statistical difference) in all tested mice but the Kcnk5 À / À animals, which had no detectable response at any frequency, even for the highest sound pressure level (SPL) tested (110 dB SPL). Furthermore, recordings from the round window performed on three additional Kcnk5 À / À adult mice confirmed the lack of evoked potential in the 8 th nerve for all frequencies and SPLs tested, thus verifying the loss of auditory function. Monitoring the postnatal development through ABR waves was performed between P14 and P21 (Fig. 1c). In both Kcnk5 þ / þ (n ¼ 9) and Kcnk5 þ / À (n ¼ 8) mice, ABR waves appeared by P15-P16 and were similar between the two groups. A slightly delayed onset of hearing was observed in Kcnk5 À / À (n ¼ 6) mice, their ABR waves appearing at P17 (n ¼ 1) and at P19 (n ¼ 5). At P19, similar thresholds were measured in Kcnk5 þ / þ , Kcnk5 þ / À and Kcnk5 À / À mice. However, ABR waves had disappeared by P21 in all the six tested Kcnk5 À / À mice.

KCNK5 does not play an obvious role in vestibular function.
Monitoring spontaneous motor behaviours (for example, righting and postural adjustments) did not provide any evidence for a vestibular dysfunction in any of the Kcnk5 À / À mice tested. In addition, a swim test was conducted on Kcnk5 þ / þ (n ¼ 3), Kcnk5 þ / À (n ¼ 5) and Kcnk5 À / À (n ¼ 7) mice of different ages (between P20 and P340) taken from several litters (n ¼ 8). Not a single sinking episode during the full length of the tests was ever displayed by any one of the tested mice, even by the two oldest (Kcnk5 À / À of P262 and P340), suggesting no impairment and no delayed degradation of the vestibular function.
(Po0.001 for all tests) compared with the other groups of mice (n ¼ 9, 8.74±0.51, n ¼ 9, 9.21±0.27 and n ¼ 9, 8.15±0.41 cells per surface unit in adult Kcnk5 þ / þ , adult Kcnk5 þ / À and P23 Kcnk5 À / À respectively). No significant difference was found between the three other groups (adult Kcnk5 þ / þ , adult Kcnk5 þ / À and P23 Kcnk5 À / À ). These observations are in line with the usual time course of spiral ganglion neuron loss occurring after hair cell disappearance 31 . In contrast in all animals, the spiral ligament, the stria vascularis and the Reissner's membrane had normal structure and thickness.
KCNK5 is mainly expressed in cochlear outer sulcus. The use of b-galactosidase (b-gal) as a surrogate of KCNK5 expression allowed specific identification of cochlear cells that normally express this K þ channel (Fig. 3). In adult mice, X-gal staining of in toto cochleas (Fig. 3a) mainly appeared along the cochlear duct as two stripes (arrowhead) below the stria vascularis (asterisk). In cryostat sections of cochleas (Fig. 3b-e), the highest X-gal staining was seen in the outer sulcus epithelial cells and in the root cells under the spiral prominence. Detailed examination of cells located above the basilar membrane shows that both Claudius' and Boettcher's cells express KCNK5. Indeed, the X-gal staining overlaps a small group of cells underlying the Claudius' cells ( Fig. 3e) whose location and shape are very similar to those reported by others as Boettcher's cells (see Fig. 4d in ref. 4). A lighter b-gal expression was identified in spiral ganglion neurons, Reissner's membrane and as small spots in few other compartments, including the stria vascularis. Of note, b-gal expression appeared less pronounced under spiral prominence in Kcnk5 þ / À (n ¼ 4, Fig. 3c) than in Kcnk5 À / À (n ¼ 4, Fig. 3d) cochleas.
Since both the outer sulcus root cells and the spiral prominence epithelial cells express pendrin, an anion transporter protein essential for hearing [32][33][34] , the above results suggested that KCNK5 might be present in these pendrin-expressing cells. Thus, immunohistochemistry was used to detect either pendrin alone or pendrin and b-gal in adult (P45-P60) cochleas of Kcnk5 þ / þ (n ¼ 3), Kcnk5 þ / À (n ¼ 7) and Kcnk5 À / À (n ¼ 7) mice. In whole sections, pendrin was expressed in spiral prominence and in root cells with a similar pattern in all tested mice, including Kcnk5 þ / À and Kcnk5 À / À animals (Fig. 4a,b). As illustrated in a Kcnk5 þ / À mouse (Fig. 4c-k), confocal imaging confirmed that pendrin is located in the root cells ( Fig. 4c-h) and that the highest b-gal staining is found in Claudius' cells ( Fig. 4c,d,i,j), some of which being adjacent to the pendrin-positive root cells. A lower level of KCNK5 expression was also observed in the root cells as small puncta or spots of b-gal staining (Fig. 4e,k). Merged images indicated a colocalization of pendrin and KCNK5 in the root cells (Fig. 4e).

Expression of known K þ channels in root cells and stria.
Additional immunohistochemical experiments were performed to analyse the effect of Kcnk5 deletion on the expression of KCNJ10 (Kir4.1) and KCNQ1 (Kv7.1), two K þ channels playing key roles in K þ recycling along the root cells, and EP generation and/or regulation via intermediate and marginal cells of the stria vascularis [35][36][37][38][39][40] . Qualitative analyses in adult (P46-47) Kcnk5 þ / À (n ¼ 3) and Kcnk5 À / À (n ¼ 3) mice revealed similar patterns of expression of the two K þ channels in both groups of mice, with KCNJ10 localized in intermediate cells (Fig. 5a,b) and KCNQ1 expressed in marginal cells (Fig. 5c,d) of the stria vascularis. In the root cell areas of Kcnk5 þ / À (n ¼ 2, Fig. 5e,g,h) and Kcnk5 À / À (n ¼ 2, Fig. 5f,i,j) mice, no obvious difference was found in levels of expression of KCNJ10 and tubulin, a component of the cytoskeleton of the root cell bodies and their processes 15 . Thus, the similar patterns of localization of KCNJ10, KCNQ1 and tubulin suggest no major alteration in their expressions in both stria vascularis and root cells in mice lacking KCNK5.

Discussion
This study shows that Kcnk5 knockout mice develop an early and profound deafness within 2 days after hearing onset (that is, P19) associated with an early alteration of sensory hair cells and supporting cells (from P21) followed by a late degeneration of outer sulcus (Os) epithelial cells and of spiral ganglion neurons in adults. In contrast, major K þ channels such as KCNJ10 and KCNQ1 involved in K þ secretion or recycling have a normal expression in Kcnk5 À / À mice. These results markedly differ from that obtained after deletion of other genes thought to play a role in K þ recycling along the lateral route (Kcnq4, Nkcc1, Kcc3 and Kcc4), all resulting in progressive deafness over weeks or months 11,13,14,18,41 . Thus, our data demonstrate that KCNK5 channels are indispensable for the maintenance of hearing. Since high expression level of KCNK5 was revealed in the Os cells, notably in Claudius' cells, but also in Boettcher's and root cells, the foremost hypothesis is that KCNK5 channels contribute to K þ recycling through the lateral epithelial gap junction network 4,15,42 . Also, our results further suggest that other K þ recycling pathways or known K þ channels that are expressed in the cochlea seem unable to compensate for KCNK5 loss. In the upper cochlear turn, however, K þ recycling may be unaltered since Claudius' cells do not extend over the lower part of the lateral wall under the spiral prominence and since a specific K þ reabsorption current has been found there 6,9,10 . Several preserved mechanisms of K þ secretion and/or production of the endolymph, such as those mediated by functional KCNQ1 or KCNE1 subunit channels 43 , are likely responsible for the lack of Reissner's membrane collapse, normal cochlear maturation, adequate endolymphatic K þ concentration and at least the partial build-up of the EP that are necessary for hearing onset as observed in P17-P19 Kcnk5 null mice. In contrast, a few days later, juvenile mice lacking KCNK5 suffered from an early and severe auditory impairment. This is consistent with an indispensable role of KCNK5 channels in sustaining the EP when an increased metabolic demand for K þ recycling arises. The lack of significant decrease in [K þ ] measured in Kcnk5 null mice does not necessarily contradict our hypothesis, since a normal endolymphatic [K þ ] (1) may be achieved via blood supply through the stria vascularis before hearing onset (see introduction), (2) fits with a normal mechano-transduction in very young Kcnk5 À / À mice and (3) was assessed at the very beginning of audition (at P19) when the EP was high enough to sustain hearing. In Kcnk5 À / À mice older than P19, we postulate that the lack of proper K þ recycling through the Os induces a toxic K þ accumulation around sensory hair cells, leading to an early damage of the organ of Corti and to the loss of hearing. The losses of Corti's organ, Os epithelial cells and spiral ganglion neurons observed only in adults were likely secondary to the loss of hearing. It is, however, worthy to note that the normal vestibular function in these mice might be related to a lower requirement for K þ recycling to set the vestibular endolymph potential to a few millivolts 8 . Nevertheless, relations between K þ concentration in endolymph and EP are complex, as illustrated by a high positive EP associated with a low endolymphatic K þ concentration observed in the mouse model of Bartter syndrome 44 . Kcnk5 knockout mice may represent another interesting model to further investigate these relations. Finally, no detectable cochlear defect or impaired hearing was found in mice lacking KCNK3 and/or KCNK9. Although it has been suggested that KCNK3 could be important in the postnatal development of the cochlea 30 , our data do not support this assumption for any of the K2P channels investigated herein. The slight delay in hearing onset observed in Kcnk5 À / À mice may suggest that KCNK5 exerts a small effect on cochlear maturation. However, this delay could also be a consequence of the general hypotrophy already shown in these animals 23 .
Interestingly, several membrane ion co-transporters and exchangers may share similar physiological functional profile with KCNK5 channels. Homeostasis of ionic concentrations in the endolymph is achieved by a balance between secretion and reabsorption mechanisms and, both Reissner's membrane and outer sulcus epithelial cells contribute to Na þ and K þ reabsorption 34 . Since KCNK5 is expressed in these structures, the ionic concentrations in the endolymph may be altered in Kcnk5 À / À mice. It cannot be ruled out that the lack of KCNK5 alters the integrity of the Reissner's membrane, and thus of the endolymph-perilymph barrier. However the fact that Reissner's membrane presents a normal structure and thickness even in old animals does not favour this hypothesis. In addition to the present demonstration of a crucial role of KCNK5 in hearing, this K þ channel contributes to bicarbonate reabsorption by the kidney, and thus participates to the acid-base homeostasis 22 . Mice lacking the K þ -Cl À co-transporter KCC4 (ref. 14) or the a subunit of the H þ -ATPase 45 suffer from renal tubular acidosis associated with hearing loss. Further, lack of pendrin, a Cl À /HCO 3 À exchanger involved in endolymphatic pH homeostasis 33 , results in progressive deafness in both humans and mice. Since normal endolymph presents an unusual HCO 3 À concentration, thought to act as a pH buffer 34  its colocalization in pendrin-positive root cells suggest that KCNK5 also contributes to pH-buffering of the endolymph. Therefore, ionic mechanisms involving KCNK5 may be another example of striking similarities occurring between epithelial transport in the inner ear and kidney 35,45 . Interestingly, K þ currents that are very similar to KCNK5 currents 46 have been recorded from root cells 15 , but the molecular nature of the K þ channel producing these currents remains hypothetical. It must be noted that KCNK5 channels have different primary structures, tissue localization and pH sensitivity when compared with KCNK3 or KCNK9 channels 21,25,47 . Indeed, KCNK3 and KCNK9 belong to the TASK subgroup being very sensitive to variations of extracellular pH within the physiological range, while KCNK5 belongs to the TALK subgroup of alkalineactivated K2P channels. KCNK5 expression is highly restricted in both the cochlea (present results) and the brainstem 25 , whereas KCNK3/KCNK9 expression display a more ubiquitous pattern in these organs 30,47 . Furthermore, respiratory adaptation to changes in central pH levels critically depends on KCNK5 channels 25,26 , but not on KCNK3/KCNK9 ones 24 . In central auditory structures, KCNK3 and KCNK9 are expressed in cochlear spiral ganglion 28 , KCNK3 expression decreases in rat inferior colliculus after cochlear ablation 48 and KCNK5 was observed in superior olive and inferior colliculus 25 . However, no data are available on their functional roles in these structures. Although our results clearly show that neither KCNK3 nor KCNK9 plays a role in auditory sensitivity, the lack of these channels could be compensated by closely related members of the K2P family, as observed in other physiological functions 21 .
There is presently no human pathology known to be directly related to KCNK5 mutation even though disturbances in cochlear K þ recycling are clearly involved in auditory pathology 2,3 . There are indications of individual susceptibility to noise-induced hearing loss being linked with gene mutations affecting cochlear K þ circulation 43,49 . In addition, it has been reported that mice lacking the acid-sensing ion channel 2 were more resistant to noise-induced temporary hearing loss than wild-type (WT) mice 50 . KCNK5 specifically found in outer sulcus epithelial cells in connection with root cells might also be involved in presbycusis, since CBA/CaJ mice, which are affected by early age-related hearing loss, show a concomitant loss of cochlear root cells 51 .

Methods
Animals. Juvenile (P14-P23) and adult (from P30 to P340) mice of either sex were used in this study. All experiments were performed on C57BL6/J mouse littermates, bearing or not a mutation for the gene coding for KCNK5, KCNK3, KCNK9 or KCNK3 and KCNK9 (double knockout). Genotyping was performed by a PCR technique on tail DNA using a three primers in the same amplification sample: a forward primer (5 0 -AGACCAGCCACAACCATACAAT-3 0 ) was common to the WT and knockout products and two reverse primers, one in intron 1 specific for the WT band of 1.2 kb (5 0 -TAGCAAGGAAGTAGCCAGAGGT-3 0 ) and the other in the introduced exontrap sequence 52 , specific for the knockout band of 0.9 kb (5 0 -CACTCCAACCTCCGCAAACT-3 0 ). Amplification conditions were 95°C per 3 min; (95°C per 20 s; 65°C per 20 s ( À 1°C per cycle); 72°C per 1 min) 6 Â ; (95°C per 20 s; 60°C per 20 s; 72°C per 1 min) 30 Â . The animals had unlimited access to food and water, and were exposed to 12-h light-12-h dark cycles. The experimental procedures were carried out in accordance with French national legislation (JO 87-848, European Communities Council Directive (2010/63/EU,74) and local ethics committee 'Direction Départementale de la Protection des Populations', with permit numbers 13-47, 13-06 and 13-227 delivered to Y.C., M.B. and C.G., respectively.
Auditory physiology. Mice were anesthetized with a mixture of ketamine chloride (50 mg kg À 1 ) and medetomidin chloride (500 mg kg À 1 ) through intraperitoneal injection, and at the end of a recording session, animals were awakened by injection of atipamezol (500 mg kg À 1 , intraperitoneally). All recordings were made in a soundproof chamber.
Auditory brainstem-evoked potentials. For auditory brainstem-evoked potentials, mice were placed in front of a loudspeaker (FF1-TDT) and tone pips (with 2 ms linear rise/fall time and no plateau) were delivered at octave frequencies from 2 to 32 kHz at a rhythm of 10 s À 1 . Acoustic levels were measured in dB of SPL with a free-field microphone (Bruel and Kjaer microphone 4191) positioned at the animal's head. ABR waves were recorded via small needle electrodes introduced under the skin at the skull vertex (active electrode), behind one mastoid (reference electrode) and in a neck muscle (ground electrode). Signals were amplified 10 4 times, filtered between 300 and 3,000 Hz (Grass ICP 511 amplifiers), digitally converted and averaged (10 3 sweeps) with a Micro1401 Plus system (Cambridge Electronic Devices, UK). Recordings of ABR waves started from an acoustic level of 110 dB SPL and proceeded down in steps of 10 dB. The ABR thresholds for various acoustic frequencies, defined as the lowest intensity (in dB) that generates a clearly identifiable and reproducible response wave, are represented as audiograms.
Round window recordings. Round window recordings allow the measure of compound action potentials from the eighth cranial nerve with much larger signalto-noise ratio than far-field brainstem recordings. They were performed after piercing a small retro-auricular hole in the bulla to gently place a platinum ball electrode into the round window niche.
EP recordings. The head of the animal was fixed at frontal bones level to a micromanipulator with screws and acrylic cement. The head was tilted sideways, the pinna was cut, the tympanum and middle ear bones resected and the posterior ridge of the ear canal was gently broken into small pieces with fine tweezers to visualize and access the cochlear first turn. A small hole was pierced with a fine needle through the bone at the level of the stria vascularis pigmented area above the stapedial artery. Then a glass microelectrode (B10 MO) filled with 150 mM KCl solution was slowly inserted through the hole with a micromanipulator while the electric potential was monitored (NeuroLog DC amplifier) and recorded (Micro1401Plus) at least for 2 min.
K þ concentration measurements. Potassium concentration [K þ ] was measured by means of ion-sensitive microelectrodes 53 . The reference channel was back-filled with 154 mM NaCl, while the ion-sensitive channel was salinized with methyltrichlorosilane in dichloromethane (Fluka/Sigma-Aldrich), tip-filled with the K þ ionophore I-cocktail A (Fluka) and back-filled with 100 mM KCl. A differential amplifier subtracted the reference potential from that of the ionsensitive channel to obtain the pure signal related to changes in the ion concentration. These changes were then calculated using the Nernst equation. The K þ -sensitive microelectrodes were tested and calibrated using known [K þ ] solutions.
Vestibular function: swim test. Since auditory and vestibular systems are closely related, we evaluated the role of KCNK5 in vestibular function using a swim test 54,55 , on control and Kcnk5 À / À mice. Mice of either sex were individually lowered in a 20-by-40-cm clear plexiglas aquarium filled with warm water (25-27°C) to a depth of 20 cm. Control mice displayed a normal swimming behaviour characterized as follows: maintenance of a horizontal body position, nose and tail kept above the water, and swimming towards the side of the cage. The swimming ability was assessed over periods of 60 s by scoring the total duration of sinking episodes, that is, a vertical body position and the nose below the water level.
Histology of mouse cochlea. Cochlear samples. The animals were killed by respiratory arrest under isoflurane overdose. The tympanic bulla was exposed and the otic capsule opened. After removal of the stapes and opening the round window, a small hole was made in the apex of the cochlea, over the helicotrema. Whole cochleas were perfused in situ through the round window with either 1% (X-gal staining) or 4% (immunohistological staining) paraformaldehyde in PBS (0.1 M phosphate with 0.15 M NaCl buffered at pH 7.4) or in 2.5% (resin embedding) glutaraldehyde in PBS with 5 mM EGTA, 2 mM MgCl 2 and 58 mM sucrose. The cochleas were severed from the temporal bone and post-fixed in the same fixative.
Cochlear structure. In toto cochleas were decalcified in EGTA (500 mM, pH 8) for 48 h, osmicated (2% OsO 4 in H 2 O), dehydrated in a grade series of alcohols (70, 80, 95 and two 100% baths), cleared in propylene oxide, kept in half propylene oxide and half soft Durcupan resin overnight and then transferred to pure resin. The resin was cured at 60°C for 48 h. After cooling, each resin block was trimmed so that the embedded cochlea was transversally sectioned in midmodiolar planes. Sections of 10 mm thickness were obtained with a microtome (Reichert-Jung Supercut 2050), serially collected onto glass slides and air dried. They were then covered with pure resin, coverslipped and cured.
X-gal staining. The Kcnk5 mutant mice were generated using an exon-trapping approach, in which the targeting vector contained the Lac-Z gene coding for b-galactosidase 55 . Thus KCNK5 expression was localized by staining for b-galactosidase enzyme activity using the X-gal substrate (5-bromo-4-chloro-3indolyl-b-d-galactopyranoside). In toto, 1% paraformaldehyde-fixed whole cochleas were washed in PBS, immersed in staining buffer (PBS, 5 mM EGTA, 2 mM MgCl 2 , 0.01% Na-desoxycholate and 0.02% Nonidet P-40) for 1 h at 37°C and left overnight at 37°C after the addition of 0.245 mM X-gal solution (from a stock solution of 5% X-gal in N,N dimethylformamide), 0.1 mM potassium ferricyanide and 0.1 mM potassium ferrocyanide. After decalcification, cochleas were immersed and positioned in Tissue-Tek OCT compound before freezing ( À 20°C). Cryostat sections of 40 mm thickness made in the modiolar cochlear plane were serially collected onto gelatin-chromalun-coated slides. Then, the sections were air dried, hydrated in PBS and counterstained with haematoxylin. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9780 ARTICLE Whole cochleas and sections were dehydrated in a grade series of alcohols, cleared in xylene, mounted in DPX mountant for histology (Fluka) and photographed.
Analysis of spiral ganglion cell loss. High-magnification microphotographs were obtained from resin-embedded cochleas' sections and used for semi-automatic quantification of spiral ganglionic cells. First, all spiral ganglionic cells with clearly visible cell somata were manually plotted, and the spiral ganglionic boundaries were also delimited (Photoshop). Then, the number of cells per section, the surface of tissue (in a.u.) and the relative packing density of cells per section, estimated as the ratio between cell number and surface, were calculated (ImageJ, Wayne Rasband, National Institutes of Health, USA), using three sections from each animal. Mean numbers were then determined in adult Kcnk5 þ / þ (n ¼ 3), adult Kcnk5 þ / À (n ¼ 3), P23 Kcnk5 À / À (n ¼ 3) and adult Kcnk5 À / À (n ¼ 3) groups of mice.

Images and figures.
Observations and photographs were made with either an Olympus BX50 or a Zeiss ApoTome (1D-SIM) AxioObserver ZI microscope equipped with both epifluorescence and a high-resolution digital camera either an Olympus DP50 (Olympus Optical, Germany) or a Zeiss AxioCam MRm (Carl Zeiss MicroImaging, Germany), or with a Laser Scanning Confocal Microscope (LSM510; Carl Zeiss MicroImaging). The figures were prepared with Photoshop software (Adobe Systems Inc.) and both brightness and contrast of the images were adjusted.
Statistics. Data were expressed as mean values ± s.e.m. from n observations and compared using either a paired t-test (two groups) or an analysis of variance (one way) followed by Dunnett's test correction for multiple comparisons. A linear regression analysis was used to test the age effect on EP values recorded in Kcnk5 À / À mice. Significance level was taken as Pr0.05. All computations were performed using SigmaStat software.