Vestibular Role of KCNQ4 and KCNQ5 K+ Channels Revealed by Mouse Models*

Background: KCNQ K+ channels regulate neuronal excitability, and KCNQ4 mutations cause deafness. Results: KCNQ4 and KCNQ5 expression in vestibular organ and the impact on vestibular function are investigated. Conclusion: Both channels reside in postsynaptic calyx membranes of hair cells, and loss of KCNQ4 impairs vestibular function. Significance: KCNQ4 may affect vestibular function because of a novel role in synaptic transmission. The function of sensory hair cells of the cochlea and vestibular organs depends on an influx of K+ through apical mechanosensitive ion channels and its subsequent removal over their basolateral membrane. The KCNQ4 (Kv7.4) K+ channel, which is mutated in DFNA2 human hearing loss, is expressed in the basal membrane of cochlear outer hair cells where it may mediate K+ efflux. Like the related K+ channel KCNQ5 (Kv7.5), KCNQ4 is also found at calyx terminals ensheathing type I vestibular hair cells where it may be localized pre- or postsynaptically. Making use of Kcnq4−/− mice lacking KCNQ4, as well as Kcnq4dn/dn and Kcnq5dn/dn mice expressing dominant negative channel mutants, we now show unambiguously that in adult mice both channels reside in postsynaptic calyx-forming neurons, but cannot be detected in the innervated hair cells. Accordingly, whole cell currents of vestibular hair cells did not differ between genotypes. Neither Kcnq4−/−, Kcnq5dn/dn nor Kcnq4−/−/Kcnq5dn/dn double mutant mice displayed circling behavior found with severe vestibular impairment. However, a milder form of vestibular dysfunction was apparent from altered vestibulo-ocular reflexes in Kcnq4−/−/Kcnq5dn/dn and Kcnq4−/− mice. The larger impact of KCNQ4 may result from its preferential expression in central zones of maculae and cristae, which are innervated by phasic neurons that are more sensitive than the tonic neurons present predominantly in the surrounding peripheral zones where KCNQ5 is found. The impact of postsynaptic KCNQ4 on vestibular function may be related to K+ removal and modulation of synaptic transmission.

The vestibular organ senses gravitational forces and acceleration and provides essential sensory input for the control of body equilibrium, head orientation, and eye movements. Sensory hair cells (HCs) 6 in otholithic organs (utricle and sacculus) respond to gravitational forces and linear acceleration, whereas HCs in the cristae ampullares detect rotational acceleration. Both types of sensory end organ contain different classes of HCs named type I and type II. These classes are not only distinguished by their pattern of innervation, with calyx terminals ensheathing type I cells, but also by other characteristics like the expression of distinctive plasma membrane currents (1). Vestibular sensory output depends not only on mechanosensitive ion channels located in the stereocilia of HCs, but also on ion channels present in both the basolateral membrane of HCs and in their cognate afferents. These channels may modulate receptor potentials, synaptic transmission, and neuronal excitability. At least two members of the KCNQ (K v 7) K ϩ channel family (2,3), KCNQ4 and KCNQ5, are found in the vestibular organ (4,5). KCNQ4 is important for hearing (5,6), is mutated in patients with dominant DFNA2 deafness (6), and modulates touch sensation (7). KCNQ4 displays a highly restricted expression pattern. In the cochlea, it is localized at the basal pole of outer hair cells (OHCs) that depolarize and eventually degenerate upon Kcnq4 disruption (5). KCNQ4 is also found at basal poles of vestibular type I hair cells that are ensheathed by calyx synapses (5). However, its localization to pre-or postsynaptic membranes remains controversial (4, 8 -12). Unlike cochlear OHCs, vestibular HCs do not degenerate in Kcnq4 Ϫ/Ϫ mice (5), but ϳ30% of patients with KCNQ4 mutations display mild vestibular symptoms (13).
KCNQ5 is more broadly expressed than KCNQ4 (14,15), yields similar currents, but lacks cochlear expression. In vestibular sensory epithelia (5) it partially overlaps with KCNQ4. Apart from the disappearance of an afterhyperpolarization current in hippocampal neurons (16), no phenotype has been described in Kcnq5 dn/dn mice carrying a dominant negative mutation.
Using KCNQ mutant mouse models we now resolve the controversy surrounding the contribution of KCNQ4 and KCNQ5 to vestibular HC currents and determine their physiological relevance for the vestibular system. We found that neither channel is significantly expressed in adult vestibular HCs, but rather postsynaptically in calyx terminals where their expression partially overlaps. Accordingly, currents of vestibular HCs were unchanged upon gene disruption. Vestibulo-ocular reflexes were reduced both in Kcnq4 Ϫ/Ϫ and Kcnq4 Ϫ/Ϫ / Kcnq5 dn/dn double mutant mice, whereas only marginal effects were observed in Kcnq5 dn/dn mice. The larger impact of Kcnq4 disruption on vestibular function may be related to its preferential expression in the central zones of vestibular sensory epithelia.

EXPERIMENTAL PROCEDURES
Mouse Models and Genotyping-The generation of Kcnq4 Ϫ/Ϫ , Kcnq4 dn/dn , and Kcnq5 dn/dn mice and their genotyping has been described previously (5,16). KCNQ4 mouse models were initially kept in C3H/HeJ and Kcnq5 dn/dn mice in C57BL/6 background. Double mutant mice were generated by crossing Kcnq4 Ϫ/Ϫ and Kcnq5 dn/dn mice. They were viable, fertile, and had no immediately apparent phenotype. They were kept and investigated in the mixed background. Subsequently, all mutant genotypes were obtained by breeding mice heterozygous for both genes. Mice of either sex were used for experiments.
Electrophysiology-Whole cell recordings were done on HCs from acutely dissected utricles that were perfused at 0.5 ml/min with oxygenated solution containing 137 mM NaCl, 5 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , 10 mM HEPES, 11 mM glucose, pH 7.4. The epithelium was cut transversally and anchored to the chamber floor with two pieces of glass, exposing the basolateral membrane of hair cells. Central and peripheral hair cells could not be easily distinguished in this semi-intact preparation. However, patches were mostly obtained in the central and adjacent areas of the exposed tissue which may span the whole striola and juxtastriola and a minor portion of extrastriola. Recordings were done with 3-4-megohm electrodes filled with 140 mM KCl, 5 mM NaCl, 10 mM HEPES, 10 mM glucose, 5 mM EGTA, 3 mM MgATP, 1 mM NaGTP, pH 7.3. Type I and type II HCs were characterized (20) by the presence or absence, respectively, of the g K,L conductance (1). Microscopic inspection further helped to identify HC classes.
Whole cell recordings were done using a Multiclamp 700B amplifier (Molecular Devices). After seal formation (Ͼ10 G⍀) onto the basolateral membrane of HCs and membrane disruption, we estimated membrane capacitance (C m ) and series resistance (R s ) from the decay of capacitive transients induced by a Ϯ10-mV pulse from a holding potential of Ϫ80 mV. After cancellation of capacitive transients, R s was compensated up to 85%. The resting potential was measured as the zero current voltage in the current clamp mode. Data analysis used pClamp 10 software (Molecular Devices) and Origin 7.5 (OriginLab). Data were sampled at 10 kHz and filtered at 2 kHz. The voltage clamp protocol started from a holding potential of Ϫ70 (for type II HCs) or Ϫ90 mV (for type I HCs), followed by a 50-ms pulse to Ϫ110 mV for type I HCs and Ϫ100 mV for type II HCs followed by test pulses of 200 ms between Ϫ110 and ϩ40 mV for type I HC and Ϫ100 to ϩ50 mV in type II HC in steps of 10 mV, and a constant step to Ϫ30 mV for tail currents. In statistical analysis, the number of cells is given by n, and number of animals by N.
Expression in Xenopus Oocytes and Two-electrode Voltage Clamping-Xenopus oocytes were obtained and injected as described (15). A final amount of 20 ng of cRNA/oocyte was injected for each condition. Oocytes were incubated 3-4 days at 17°C and then examined by two-electrode voltage clamping using a Turbo Tec03 (NPI Electronics) amplifier and pClamp 8.0 software (Molecular Devices). The protocol consisted of 5-s steps from Ϫ110 to ϩ50 mV in 10-mV increments from a holding potential of Ϫ80 mV.
During the experiment the mouse was placed head-fixed in a holder tube on a vestibular motion platform (R2000 Rotopod; Parallel Robotic Systems Corporation). Left eye orientation was measured using video pupil tracking with a table fixed CCD camera (Pulnix TM-6710CL, 120 frames/s) and IR illumination (850-nm LED, 6.5-cm distance from the eye). Pilocarpine (2%) eye drops were applied before the experiment to limit pupil dilatation in darkness. Online image analysis was performed to extract the location of pupil edges and corneal light reflections using custom built software for Labview (National Instruments, Austin, TX). Angular eye velocity was computed offline using custom software written for Matlab (The Mathworks Inc., Natick, MA) using the algorithm outlined elsewhere (22,23). Saccadic eye movements and nystagmus fast phases were removed using a 50°/s velocity threshold and 200-ms margins at each threshold crossing. Each mouse was accustomed to the setup and experimental paradigm in a period of up to 3 training days before the experimental data were collected.
The horizontal VOR was characterized in darkness and in the light using sinusoidal rotation about the yaw axis, using frequencies ranging between 1/16 and 4 Hz, presented in a sequence of increasing order. Peak velocity was held constant at 25°/s. The number of cycles ranged between 5 (at 1/16 Hz) and 60 at 4 Hz. Eye velocity amplitude and phase shift relative to head movement were calculated using multiple linear regression of eye velocity to in-phase and quadrature components of the stimulus velocity. Gain of the eye movement response was defined as the ratio between the eye velocity amplitude and the stimulus velocity amplitude. Phase shift is expressed in degrees; positive phase shifts indicate phase lead. Error bars represent the S.E. Circular statistics were used to compare phase values. Transfer function profiles were statistically compared using MANOVA for repeated measures (SPSS Statistics 17.0.).

Differential Distribution of KCNQ4 and KCNQ5 in Vestibular End
Organs-Previous studies examining the distribution of KCNQ4 and KCNQ5 in vestibular end organs were performed only with WT animals (4, 8, 11). We now used three genetic mouse models to control immunohistochemical labeling and to explore potential cross-talk between the expression of both proteins. Kcnq4 Ϫ/Ϫ mice lack the KCNQ4 protein, whereas Kcnq4 dn/dn mice express a dominant negative KCNQ4 variant carrying a point mutation (G286S) in the pore-forming P-loop (5). It corresponds to a mutation of a DFNA2 patient with dominantly inherited hearing loss (6). Kcnq5 dn/dn mice carry an equivalent mutation (G278S) in KCNQ5 (16). KCNQ5(G278S) does not yield currents. It exerts strong dominant negative effects on co-expressed KCNQ3 and KCNQ5 WT subunits with which it can form heterotetramers (6,15,16).
Labeling vestibular end organs with our antibodies revealed a differential distribution of KCNQ4 and KCNQ5 in both the otholithic organs ( Fig. 1) and cristae (supplemental Fig. S1). As displayed for utricles in Fig. 1A, KCNQ4 was most highly expressed in the central striolar and the adjacent juxtastriolar regions. It extended at weaker expression levels to the adjacent extrastriola ( Fig. 1, A and J). By contrast, KCNQ5 was found mainly in the extrastriola and extended at lower expression levels into the juxtastriola (Fig. 1, B and K). As described previously (4, 5), both KCNQ4 and KCNQ5 were found in calyx-like structures that ensheathe type I vestibular hair cells (Fig. 1, J and K). Calyces were co-labeled for KCNQ4 and KCNQ5 in the juxtastriolar and extrastriolar regions (Fig. 1L). Likewise, KCNQ4 was found in central, intermediate, and peripheral zones of the cristae ampullares, whereas KCNQ5 was expressed in intermediate and peripheral zones but was almost undetectable in central zones (supplemental Fig. S1, A-C, J-L).
Importantly, labeling for KCNQ4 was abolished in tissue from Kcnq4 Ϫ/Ϫ mice ( Fig. 1D and supplemental Fig. S1D), validating the specificity of our staining. Surprisingly, labeling for KCNQ5 was virtually abolished in utricles from Kcnq5 dn/dn mice that express a full-length KCNQ5 protein carrying a single point mutation (Fig. 1H). A similar pattern was observed in cristae ampullares (supplemental Fig. S1, H and N). Besides validating our KCNQ5 antibody staining, this finding suggested that the mutant protein was unstable or incorrectly targeted.
The partially overlapping expression patterns of KCNQ4 and KCNQ5 (Fig. 1, C and L, and supplemental Fig. S1, C and L) suggested that they may form heteromeric channels in vivo, as has been reported for co-transfected Chinese hamster ovary cells (24). Using dominant negative mutants, including the one expressed in Kcnq5 dn/dn mice, we confirmed these findings in the Xenopus oocyte system that allows a more quantitative control of co-expression levels (supplemental Fig. S2, A and B). Importantly, KCNQ4/5 heterotetramers did not yield higher currents than the respective homotetramers, contrasting with findings for KCNQ3-KCNQ5 (15,16) and KCNQ2-KCNQ3 channels (25,26). Hence, the disruption of Kcnq4 should not decrease K ϩ currents more than expected from a loss of KCNQ4 currents alone. On the other hand, instead of a compensatory up-regulation, we observed a mild, but consistent decrease of KCNQ5 labeling in Kcnq4 Ϫ/Ϫ vestibular organs (Fig. 1E). This finding hints at a role of KCNQ4-KCNQ5 complexes in trafficking, anchoring, or stabilizing KCNQ5 at its target membrane. The loss of KCNQ5 current in Kcnq4 Ϫ/Ϫ mice, however, is expected to be mild. This contrasts with the strong dominant negative effect of the KCNQ5 dn mutant which will suppress currents from all KCNQ4 subunits present in KCNQ4/5 heteromers. Of note, no dominant negative effect is expected in striolae (or central zones of cristae) as they lack appreciable KCNQ5 expression ( Fig. 1, B, E, and K, and supplemental Fig. S1, B and K) (11). Hence, the output from the striola, which harbors the more sensitive, phasic HC-neurite complexes (27), will not be affected in Kcnq5 dn/dn mice.
KCNQ4 and KCNQ5 Are Not Expressed in Adult Vestibular Hair Cells, but in Afferent Neurons-The lack of KCNQ-mediated currents in Ͼ3-week-old type I HCs suggested that KCNQ4 and -5 are not expressed in those sensory cells, but rather reside in postsynaptic calyx membranes that ensheathe type I HCs. Confocal light microscopy cannot resolve the postsynaptic inner calyx membrane from the closely apposed basal, presynaptic HC membrane. A postsynaptic localization of KCNQ4, however, is strongly suggested by the extension of weaker KCNQ4 immunoreactivity along the outer calyx membrane into neurites (4, 11), a finding we now confirmed by Kcnq4 Ϫ/Ϫ -controlled immunostaining of striolar calyces (Fig.  3A). By contrast, KCNQ5 labeling (controlled by Kcnq5 dn/dn HCs) was detected neither in outer calyceal membranes, nor in neurites (Fig. 3B). We correlated these findings with immunodetection of either channel protein in neuronal cell bodies of the vestibular ganglion (Fig. 4). Whereas somata of a neuronal subpopulation showed plasma membrane and intracellular labeling for KCNQ4 (Fig. 4A), KCNQ5 was not, or only barely, detectable (Fig. 4D). Hence, KCNQ5 may be more stringently targeted to the inner calyx membrane than KCNQ4. In Kcnq5 dn/dn mice, however, KCNQ5 was detected in a punctate pattern in neuronal cell bodies (Fig. 4F), while disappearing from vestibular end organs ( Fig. 1H and supplemental Fig. S1, H and N). Such a trafficking defect might also be expected for KCNQ4-KCNQ5 heteromers containing KCNQ5(G78S) subunits. Indeed, juxtastriolar and extrastriolar KCNQ4 labeling was mildly reduced in maculae and cristae of Kcnq5 dn/dn mice (Fig.  1, A and G, and supplementary Fig. S1M). This reduction was more evident in the extrastriola where the KCNQ5/KCNQ4 ratio is larger (insets in Fig. 1, A and G). A, potassium currents from type I (left) and type II HCs (right) obtained from semi-intact preparations of mouse utricle from wildtype and double-mutant mice Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn . Insets show voltage protocol used to obtain currents in type I and type II HCs, respectively. Cells were clamped before the voltage step to a Ϫ110 mV (type I HC) or Ϫ100 mV (type II HC) for 50 ms and after it to Ϫ30 mV. B, current-voltage relationship of potassium currents from type I (left) and type II (right) HCs for wild-type (black dots), Kcnq4 Ϫ/Ϫ (blue triangle), Kcnq5 dn/dn (green triangle), Kcnq4 Ϫ/Ϫ / Kcnq5 dn/dn (red star) mice. Currents were normalized to cell capacitance. Number of measured cells from WT, Kcnq4 Ϫ/Ϫ , Kcnq5 dn/dn , and Kcnq4 Ϫ/Ϫ / Kcnq5 dn/dn mice were 7, 5, 4, and 4 for type I HCs, and 26, 13, 4, and 5 for type II HCs, respectively. C, utricular type I (upper panels) and type II (lower panels) HC currents in the absence (black traces) and presence (red traces) of 10 M XE991 perfused for 5 min after control trace. Currents were activated at ϩ20 mV and obtained from wild-type (left panel) and Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn (middle panel) mice. Right diagrams, averaged currents (at ϩ 20 mV, and normalized to capacitance) for both genotypes. Numbers of animals were: type I HCs: n ϭ 5 and 4 for wild-type and Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn mice, respectively; and type II HCs: n ϭ 6 for either wild-type or Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn mice.

FIGURE 3. Expression of KCNQ4 and KCNQ5 in type I hair cell calyces.
A, immunohistochemistry of striolar type I HC for KCNQ4 (green), counterstained for the nuclear marker DAPI (blue), and in right panels for ␤-III-tubulin (red), a cytosolic marker for calyces. Top, results from wild-type mice; bottom panels, from Kcnq4 Ϫ/Ϫ mice. Note that weak KCNQ4 staining extends beyond the basal pole of the HC toward the neck of the calyx, where inner and outer membranes are labeled (arrows), and to the neurite. This is better visible in insets which show the areas enclosed by dotted lines with increased brightness. Kcnq4 Ϫ/Ϫ sections are not labeled by the KCNQ4 antibody, demonstrating the specificity of staining. B, similar labeling for KCNQ5 of extrastriolar HCs. Note that KCNQ5 immunoreactivity does not extend beyond the basal part of the calyx and that labeling is abolished in Kcnq5 dn/dn section. Scale bars, 5 m.
Similar mutations in KCNQ4 are known to interfere with plasma membrane expression (7,33,34). Indeed, neuronal somata in the vestibular ganglion of Kcnq4 dn/dn mice (5), which express the trafficking-deficient G286S mutant (7), showed intense cytoplasmic KCNQ4 labeling (Fig. 4B) compared with WT somata (Fig. 4A). KCNQ4 labeling of vestibular end organs disappeared in parallel, as shown for Kcnq4 dn/dn crista ampullaris in Fig. 5, A and B. It is instructive to compare these findings with those for the cochlea where KCNQ4 is expressed in outer hair cells (4 -6). Unlike vestibular HCs, OHCs displayed detectable levels of the mutant KCNQ4 protein (Fig. 5D). Again consistent with impaired trafficking and ER retention, the mutant protein was no longer detected at the basal OHC plasma membrane (Fig. 5C), but in a punctate intracellular pattern (Fig. 5D).
Our immunohistochemical analysis of mutant mice strongly suggests that KCNQ4 and -5 are expressed in postsynaptic, but not presynaptic membranes of calyx terminals. In situ hybridization further strengthened this conclusion. Both Kcnq4 (Fig. 6, A and B) and Kcnq5 (Fig. 6, C and D) mRNAs were detected in ganglionar cell bodies, but not in vestibular sensory epithelia as would have been expected with an expression in hair cells. Hence, all of our experiments point to a postsynaptic rather than vestibular hair cell expression of KCNQ4 in mice older than 3 weeks.
Consequences of the Loss of KCNQ4 and KCNQ5 for Vestibular Function-Kcnq4 Ϫ/Ϫ mice, Kcnq5 dn/dn mice, and even mice homozygous for both mutations (Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn ) lacked shaker/waltzer behavior that is indicative of a strong vestibular deficit phenotype as found, e.g. in mice lacking the NaK2Cl co-transporter Nkcc1 (35) or the K ϩ ␤-subunit gene Kcne1 (36). Both gene products are involved in generating the high endocochlear K ϩ concentration that is required for mechanotransduction currents in cochlear and vestibular hair cells.
To find out whether more subtle vestibular deficits can result from a loss of KCNQ4 and/or KCNQ5 function we measured the VOR in awake, head-restrained Kcnq4 Ϫ/Ϫ , Kcnq5 dn/dn , and  OHCs of wild-type mice (C) is changed to a weak punctate cytoplasmic labeling in Kcnq4 dn/dn mice (D). Inset, area enclosed by the dotted line is at higher magnification and increased laser intensity. Note the morphological abnormalities of prestin-positive OHCs of Kcnq4 dn/dn mice due to their incipient degeneration at this age (3 months). Nuclei were labeled with DAPI. Scale bars, 20 m.
Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn mice and WT littermates. In this assay, mice are subjected to controlled angular acceleration on a turntable while measuring evoked eye movements with a video camera (Fig. 7, A and B). These experiments were performed at different rotation frequencies both in the dark and in the light. Stimulus peak velocity was held constant at 25°/s. In the dark, VOR of all four experimental groups (WT, Kcnq4 Ϫ/Ϫ , Kcnq5 dn/dn , and Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn ) displayed the typical high pass filter behavior, with low gain and a substantial phase lead at low frequencies and high gain and a small phase lag at the high end of the frequency range (Fig. 7, C and D). Importantly, the VOR gain values of Kcnq4 Ϫ/Ϫ and those of Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn mice were significantly lower compared with those of WT over the entire frequency range (Fig. 7C) (Kcnq4 Ϫ/Ϫ : n ϭ 13, p Ͻ 0.001; Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn : n ϭ 13, p Ͻ 0.001). There was no significant difference in gain values among Kcnq4 Ϫ/Ϫ and Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn mice (p ϭ 0.531). The VOR transfer function profile of Kcnq5 dn/dn mice showed a moderately increased gain in the stimulus frequency range from 0.5 to 2 Hz, but was only marginally different from WT controls (p ϭ 0.043).
In the light, visual feedback elicits an optokinetic reflex that, in addition to the VOR, enhances compensatory eye movement (therefore also referred to as visually enhanced VOR or VVOR). As a consequence, the performance at low frequencies improved to similar levels as observed at high frequencies (Fig.  7, E and F). The VVOR gain of Kcnq4 Ϫ/Ϫ mice was not significantly different from that in WTs (p ϭ 0.161), whereas that of Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn mice was significantly lower than that in WTs (p ϭ 0.014). The largest differences in gain and phase of Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn mice were present at the high end of the frequency range, where the contribution of the optokinetic reflex to the total oculomotor performance is the smallest (Fig.  7, E and F). Rather surprisingly, the Kcnq5 dn/dn group showed a higher gain over the entire frequency range (about 0.05 gain increase, p ϭ 0.003) compared with WT (Fig. 7E).

DISCUSSION
Whereas KCNQ4 K ϩ channels are expressed in basal membranes of mouse cochlear OHCs throughout postnatal life (4,5), in adult vestibular end organs both KCNQ4 and KCNQ5 were only detectable in calyx terminals that innervate type I hair cells. Changed VORs in mice deficient for these channels suggest that these K ϩ channels play significant roles in synaptic transmission of vestibular hair cells and/or in the excitability of their afferent neurons. Our findings are relevant for patients with KCNQ4-related dominant hearing loss. Some of them present mild vestibular symptoms (13,37), which, however, cannot be directly compared with the present mouse phenotype because of methodological differences.
For the cochlea, in situ hybridization (6), immunohistochemistry (4), and the disappearance of XE991-sensitive K ϩ currents in Kcnq4 Ϫ/Ϫ mice (5) provide irrefutable evidence for KCNQ4 being expressed in OHCs. As KCNQ4 was previously immunolocalized to basal poles of cochlear OHCs and type I vestibular HCs, it was assumed that KCNQ4 was expressed in both types of hair cells (4). This notion was strengthened by immunogold electron microscopy (4,12), which suggested additional postsynaptic expression at calyx terminals. Furthermore, overexpression of a dominant negative KCNQ4 mutant suppressed g K,L K ϩ currents in type I HCs at P6-P10 (9). This effect likely resulted from a suppression of currents carried by KCNQ4 or other KCNQ subunits with which KCNQ4 can associate.
Later work suggested that KCNQ4 (and possibly KCNQ5) are expressed in type I vestibular HCs only during an early postnatal period (8). Single-cell PCR detected Kcnq4 and -5 mRNAs in type I HCs at P1, but their levels declined in parallel to XE991-sensitive HC currents at P14 (8). Membranes of calyx terminals concomitantly started to express XE991-sensitive currents and KCNQ4 protein. The detection by RT-PCR of Kcnq4 mRNA in P35 maculae (10), however, seems to contra-dict these findings, but might be caused by small amounts of Kcnq4 mRNA in dendrites of vestibular neurons.
Our data now show convincingly that in mice older than P28 KCNQ4 and KCNQ5 are expressed in postsynaptic calyx membranes but are not detectable in HCs. Contrasting with the large effects of Kcnq4 disruption on OHC currents (4), neither loss of Kcnq4 nor expression of the dominant negative KCNQ5 mutant changed vestibular HC currents of mice older than P28. KCNQ4 and KCNQ5 pore mutant proteins were retained in neuronal somata rather than being trafficked to calyces. In situ hybridization performed at that age found mRNAs for both Kcnq4 and -5 in vestibular ganglion neurons, but not in vestib- ular HCs. The apparent discrepancy with earlier work that showed KCNQ4 expression in type I vestibular HCs might be explained by a difference in age and species, with HC expression of KCNQ4 being lost during the 1st week of life in mice, but about 1 week later in rats (1,8,38).
Thus the changes of VORs in Kcnq4 Ϫ/Ϫ mice are not caused by a direct, cell-intrinsic effect on sensory HCs. VOR measurements depend on cristae ampullares that detect rotational acceleration, whereas our patch clamp analysis was performed in otholithic organs. However, the very similar expression patterns of KCNQ4 and KCNQ5 in both organs strongly suggest similar channel functions in either end organ. In view of the sparse Kcnq4 expression in brain (4) and its robust postsynaptic expression in calyx synapses, VOR impairment likely originates from altered synaptic transmission and spike initiation in afferent neurons. The presence of a large postsynaptic K ϩ conductance that is already active at resting potentials may shunt postsynaptic currents and thereby reduce excitatory postsynaptic potentials (EPSPs). In this scenario, the loss of postsynaptic KCNQ should increase, rather than decrease, the efficiency of synaptic transmission. A decreased electric shunt in the spike initiation zone of afferent neurons will further contribute to a more efficient coupling of HC depolarization to action potential output. Both KCNQ4 and KCNQ5 are partially open at resting voltages and can be further slowly activated by depolarization (5,6,14,15). They may thus contribute to an adaptation of afferent neurons that results in phasic responses (2). Reduced postsynaptic shunting may also decrease the time constant of EPSP decay. The resulting EPSP broadening may affect circuits that compare exactly timed inputs as in the central auditory pathway (39), but is less likely to interfere with the less timecritical signals of vestibular organs.
Synaptic clefts of calyx synapses have extraordinary large areas. Ion concentrations in those clefts must be regulated by transmembrane transport rather than by passive diffusion through the open end of the calyx at the neck of HCs. Both vestibular and cochlear HCs have to cope with apical K ϩ influx through mechanosensitive cation channels that occurs at rest and is modulated by mechanical forces. Apical K ϩ influx, which depends on the high endolymphatic K ϩ concentration, must be balanced by basolateral K ϩ efflux. In cochlear OHCs, KCNQ4 provides a major pathway for the basal efflux of K ϩ (4 -6) which is then removed by the KCC4 K-Cl co-transporter of closely apposed Deiters' cells (40). KCC4 apparently has no role in removing K ϩ from the calyx cleft as our unpublished immunohistochemistry did not detect KCC4 in calyx synapses while robustly labeling Deiters' cells. Whereas ion channel(s) mediating K ϩ efflux from adult type I vestibular HCs remain to be identified (with a likely contribution of erg channels (8)), we propose that K ϩ is removed from the cleft through postsynaptic KCNQ4, KCNQ5, and possibly Kir4.1 K ϩ channels, which were also found in type I HC calyces (41). Cellular K ϩ uptake through ion channels requires unusual electrochemical K ϩ gradients. For example, glial Kir4.1 K ϩ channels can take up extracellular K ϩ when its concentration rises during neuronal activity (42,43). Model calculations (44) and experiments (45,46) suggest a significant activity-dependent rise in the K ϩ concentration in clefts of calyx synapses ([K ϩ ] cleft ). If the postsynaptic mem-brane is held at sufficiently negative voltages by K ϩ channels at the outer calyceal face, postsynaptic K ϩ uptake through KCNQ channels appears feasible. Assuming a calyx membrane potential similar to that of cultured vestibular neurons (Ϫ70 to Ϫ60 mV; (47)) and [K ϩ ] i ϭ 140 mM, [K ϩ ] cleft would have to rise to Ͼ10 -15 mM for entering the calyx through K ϩ channels. This seems realistic as recent measurements suggested an activitydependent increase of [K ϩ ] cleft to Ͼ50 mM (45).
Hence, loss of KCNQ4 or KCNQ5 may increase [K ϩ ] cleft and thereby depolarize type I hair cells, which may affect their transduction currents (46) or synaptic vesicle exocytosis. However, unlike cochlear OHCs of Kcnq4 Ϫ/Ϫ (5) or Kcc4 Ϫ/Ϫ (40) mice, vestibular HCs showed no significant degeneration at least up to 10 months of age. Constitutively open K ϩ channels in pre-and postsynaptic membranes may also lead to nonquantal neurotransmission between type I hair cells and their cognate afferents (5,44,46,48). Disruption of Kcnq4 and Kcnq5 might interfere with this nonconventional type of synaptic communication.
Vestibulo-ocular reflexes were impaired in Kcnq4 Ϫ/Ϫ and Kcnq4 Ϫ/Ϫ /Kcnq5 dn/dn mice, but not in Kcnq5 dn/dn mice. The most conspicuous effect was an overall lower VOR gain that was not significantly increased by the additional loss of KCNQ5 function. Interestingly, Kcnq5 dn/dn mice showed a slight increase in VOR gain in the dark and light. It remains to be shown whether this effect results from the expression of KCNQ5 in the retina (49).
As both KCNQ4 and KCNQ5 channels have roughly similar biophysical properties, the marked difference in their impact on VOR responses in the dark might be caused by differences in global expression levels (that are difficult to determine) or by their differential distribution in vestibular organs. Irregularly firing, phasic afferents innervate type I HCs in the central and striolar regions of cristae and maculae, respectively (27). Exactly these neurons rely predominantly on KCNQ4. By contrast, regularly firing, tonic afferents rather contact peripheral and extrastriolar hair cells (27,50) which rely more on KCNQ5. Hence, the dependence on KCNQ4 rather than KCNQ5 suggests that vestibular-ocular reflexes depend more on hair cells eliciting phasic rather than tonic responses, which may be related to the fact that these adapting, irregularly firing neurons, together with their upstream hair cells, are more sensitive to mechanical stimulation than the tonic ones.