Maintenance of stereocilia and apical junctional complexes by Cdc42 in cochlear hair cells

ABSTRACT Cdc42 is a key regulator of dynamic actin organization. However, little is known about how Cdc42-dependent actin regulation influences steady-state actin structures in differentiated epithelia. We employed inner ear hair-cell-specific conditional knockout to analyze the role of Cdc42 in hair cells possessing highly elaborate stable actin protrusions (stereocilia). Hair cells of Atoh1–Cre;Cdc42flox/flox mice developed normally but progressively degenerated after maturation, resulting in progressive hearing loss particularly at high frequencies. Cochlear hair cell degeneration was more robust in inner hair cells than in outer hair cells, and began as stereocilia fusion and depletion, accompanied by a thinning and waving circumferential actin belt at apical junctional complexes (AJCs). Adenovirus-encoded GFP–Cdc42 expression in hair cells and fluorescence resonance energy transfer (FRET) imaging of hair cells from transgenic mice expressing a Cdc42-FRET biosensor indicated Cdc42 presence and activation at stereociliary membranes and AJCs in cochlear hair cells. Cdc42-knockdown in MDCK cells produced phenotypes similar to those of Cdc42-deleted hair cells, including abnormal microvilli and disrupted AJCs, and downregulated actin turnover represented by enhanced levels of phosphorylated cofilin. Thus, Cdc42 influenced the maintenance of stable actin structures through elaborate tuning of actin turnover, and maintained function and viability of cochlear hair cells.


INTRODUCTION
Dynamic actin turnover and rearrangement alter overall cell geometry and polarity, as well as local membrane topology, to form distinct cellular structures (Campellone and Welch, 2010). In some structures, such as lamellipodia and the Listeria comet tail, actin filaments are highly dynamic with a turnover rate measured in seconds (Ponti et al., 2004;Theriot et al., 1992), whereas they are relatively stable in stereocilia (Schwander et al., 2010) and apical junctional complexes (AJCs) (Ivanov et al., 2005).
Stereocilia and AJCs are present in inner ear hair cells, which are specialized sensory epithelia detecting hearing and balance. In the cochlea, hair cells are arranged in a single row of inner hair cells (IHCs) and three rows of outer hair cells (OHCs). Although IHCs and OHCs are believed to share common mechanotransduction machinery, they have distinct roles during sound detection: IHCs are true sensors, whereas OHCs function as an amplifier through an active process that involves stereociliary and somatic motility (Matsumoto et al., 2010;Schwander et al., 2010). Stereocilia, which are actin-based protrusions, are known as exquisitely organized microvilli and filopodia composed of hundreds of parallel actin filaments with the plus ends at the distal tip; however, stereocilia are organized into precise rows of graded height (Frolenkov et al., 2004). High sensitivity of hair cells depends on the coordinated movement of stereocilia upon mechanical stimulation; thus, the length and shape of stereocilia are determined precisely. Moreover, as hair cells are usually not replaced (Collado et al., 2008), turnover of stereociliary components is a lifetime requirement. Although a recent model has suggested that stereocilia maintain their steady state through treadmill-like dynamic actin turnover (Rzadzinska et al., 2004), others have reported no such finding (Zhang et al., 2012), raising a question about the actin turnover mechanism in stereocilia. AJCs are another actin structure essential for hair cell function, such as the maintenance of a proper cell arrangement and the epithelial barrier (Collado et al., 2011). However, the mode and effect of actin turnover at AJCs remain unknown (Nunes et al., 2006).
Cdc42 is a key regulator of the actin cytoskeleton. However, the function of Cdc42 in hair cells has not been studied, particularly regarding the development and maintenance of stereocilia. Cdc42, Rac1 and RhoA are the best characterized Rho-family small GTPases, and all are expressed in hair cells (Kalinec et al., 2000). Although they have common regulators and effectors, their activity results in different effects. Cdc42 is essential for filopodia formation (Chen et al., 2000;Yang et al., 2006), whereas Rac and RhoA induce lamellipodia and stress fiber formation, respectively (Heasman and Ridley, 2008). Cdc42 induces actin polymerization through its interaction with actin nucleators (Campellone and Welch, 2010), including the Arp2/3 complex component N-WASP (Takenawa and Suetsugu, 2007), and promotes filopodia induction and stabilization through the PAK/LIM kinase (LIMK) pathway and the cofilin phosphorylation cycle (Matsumoto et al., 2010;Melendez et al., 2011). Cdc42 also regulates AJC formation in vitro (Otani et al., 2006;Qin et al., 2010) and in vivo (Melendez et al., 2011).
Here, we show that Cdc42 is localized and it functions at stereociliary membranes and AJCs for their maintenance in hair cells. Cdc42 deletion resulted in abnormal stereocilia and AJC morphology, and was associated with downregulated actin turnover, leading to slowly progressive cochlear hair cell loss. These data provide new insights into the function of Cdc42 at apical protrusions and AJCs in differentiated epithelia in vivo.

RESULTS
Inactivation of the Cdc42 gene in inner ear hair cells by use of Atoh1-Cre transgenic mice We examined and confirmed expression of Cdc42 mRNA in cochlea using a reverse transcription polymerase chain reaction (RT-PCR; data not shown) and in situ hybridization (ISH) (Fig. 1A). We used the Atoh1 promoter, a basic helix-loop-helix (bHLH) transcription factor for hair-cell-specific inactivation of Cdc42 (Chen et al., 2002;Jahan et al., 2013). The pattern of Atoh1-Cre-directed recombination was assessed in the inner ear at embryonic day (E) 13.5 and E15.5 as well as postnatal day 3 (P3) by assessing the b-galactosidase activity of rearranged CAG-flox CAT flox -LacZ reporter mice (hereafter referred to as Atoh1-Cre;LacZ). Consistent with a previous report (Chen et al., 2002), X-gal staining was first observed at E15.5 in the cochlea and vestibule of Atoh1-Cre;LacZ mice but not in control (CAG-flox CAT flox -LacZ) mice (data not shown). Cochlear X-gal staining was restricted to IHCs and OHCs at P3 (supplementary material Fig. S1A). Immunostaining against Cre also showed a specific signal in IHCs and OHCs at P1 of Atoh1-Cre +/2 mice but not in littermate Atoh1-Cre 2/2 mice (supplementary material Fig. S1B). We attempted to detect endogenous Cdc42 using two different anti-Cdc42 antibodies but could not detect a specific signal in inner ear tissues with conventional fixation. Therefore, we used trichloroacetic acid (TCA) fixation, known to be effective for several antibodies against actin-related proteins (Hayashi et al., 1999). Cdc42 immunoreactivity was detected in TCA-fixed IHCs and OHCs at P0, with particularly intense reactivity in stereocilia of Cdc42 flox/flox mice (Fig. 1B), which was consistent with a report stating that Cdc42 is a hair bundle (stereocilia) protein (Shin et al., 2013). As expected, this reactivity was absent in the hair cells of Atoh1-Cre;Cdc42 flox/flox mice (Fig. 1B).
In order to assess cochlear function, we first used auditory brainstem response (ABR), an electrophysiological hearing test that detects evoked potentials in the auditory pathway from the cochlea to the upper brainstem. We examined hearing in 2-8-week-old Atoh1-Cre;Cdc42 flox/flox mice using ABR with broadband click stimuli corresponding to the low frequencies at 2-4 kHz ( Fig. 2A). No differences were observed in hearing between age-matched Cdc42 flox/flox mice and Atoh1-Cre;Cdc42 +/+ mice at 3 weeks or heterozygous Atoh1-Cre;Cdc42 flox/+ mice at 8 weeks [n §6; 20.063.2 versus 18.362.8 dB sound pressure level (SPL); n §6; 21.764.8 versus 25.064.9 dB SPL, respectively]; Cdc42 flox/flox mice were used as controls for all subsequent studies. At 2 weeks, Atoh1-Cre;Cdc42 flox/flox mice had a slightly elevated ABR threshold compared with that in control mice ( Fig. 2A, 35.062.3 versus 23.162.2 dB SPL), progressing to 70.563.2 dB SPL by 8 weeks (Fig. 2A). An 8-32 kHz tone-burst stimulation resulted in severe hearing loss particularly at high frequencies from 2 weeks (Fig. 2B). We further confirmed hearing loss in Atoh1-Cre;Cdc42 flox/flox mice using the distortion product otoacoustic emission (DPOAE) response, which detects a low-level sound generated by the active mechanism of OHCs that is emitted to the ear canal. A significant decrease in the DPOAE level was detected at 4 weeks in response to high frequencies and progressively deteriorated to encompass all frequencies by 8 weeks (Fig. 2C). Although Atoh1 also functions in vestibular hair cells (Chen et al., 2002), Atoh1-Cre;Cdc42 flox/flox mice had no detectable balance impairment and exhibited normal gait and swimming ability throughout their lives.
No obvious changes in gross tissue morphology were detected in either the cochlea or the vestibule, including the sensory epithelia [organ of Corti, vestibular macula, and crista ampularis] of Atoh1-Cre;Cdc42 flox/flox mice at P0 and 2 weeks (supplementary material Fig. S1C). The number of IHCs and OHCs labeled with the hair-cell-specific marker myosin VIIa in Atoh1-Cre;Cdc42 flox/flox mice at 2 weeks was identical to that in control mice, suggesting that the development of IHCs and OHCs was normal (data not shown). However, losses of IHCs and OHCs were occasionally detected without any changes in the spiral ganglion or stria vascularis in Atoh1-Cre;Cdc42 flox/flox mice at 8 weeks (supplementary material Fig. S1C); therefore, we focused on hair cell viability. Phalloidin staining of the filamentous actin (F-actin) in Atoh1-Cre;Cdc42 flox/flox mice indicated loss of hair cells in the middle turn of the cochlea at 4 weeks ( Fig. 2D) and extensive loss at 8 weeks, particularly in IHCs and in the basal turn (Fig. 2D,E), which was consistent with reduced sensitivity to high frequencies. In contrast, the number of hair cells and the shapes of kinocilia and stereocilia in the vestibule were normal in Atoh1-Cre;Cdc42 flox/flox mice at 5 weeks (supplementary material Fig. S1D,E).

Localization and activation of Cdc42 at stereocilia and AJCs in cochlear hair cells
We next identified the site where Cdc42 has prominent functions in hair cells. Because precise structural features were lost in TCAfixed immunostained samples, we expressed adenovirus-encoded GFP-Cdc42 or GFP-Cdc42(T17N;4A), an inactive Cdc42 mutant lacking the membrane-targeting motif, in organotypic cochlear explants. Intense GFP-Cdc42 fluorescence was observed in confocal reconstructions at stereociliary membranes and AJCs (Fig. 3A). In contrast, GFP-Cdc42(T17N;4A) was not localized to the stereocilia or AJCs (Fig. 3B). To investigate whether Cdc42 is active and functioning at stereociliary membranes and AJCs, we examined organs of Corti harvested from P2 transgenic mice expressing a Cdc42 fluorescence resonance energy transfer (FRET) biosensor (Cdc42-FRET biosensor mice) using twophoton excitation fluorescence microscopy (Fig. 3C). The FRET: CFP ratio (hereafter FRET/CFP) was intense at stereociliary membranes ( Fig. 3D) and higher at the apical cell-cell junctions (apicolateral membranes) than that at the basolateral membranes ( Fig. 3D). Intriguingly, the FRET/CFP ratio was higher in the upper portions than in the basal portions of stereocilia ( Fig. 3D-F). These high FRET/CFP ratios decreased significantly following treatment with the selective Cdc42 inhibitor ML141 (Fig. 3G). Moreover, high FRET/CFP ratios at the stereocilia and AJCs were also observed at P9 when hair cells are functionally and structurally mature (Frolenkov et al., 2004) (data not shown).
Progressive degeneration and loss of cochlear hair cells in Atoh1-Cre;Cdc42 flox/flox mice We analyzed the ultrastructure of organs of Corti, particularly stereocilia, in Atoh1-Cre;Cdc42 flox/flox mice using scanning electron microscopy (SEM). The regular arrangement of IHCs and OHCs was maintained in control mice at 8 weeks (Fig. 4A). Stereocilia in IHCs were arranged into a few gently curved rows with a moderately determined length in each row (Fig. 4G), whereas OHC stereocilia were arranged in three rows with a distinct W-shape alignment and precise length gradient among and within the rows (Fig. 4C).
The cellular arrangement and apical configuration of both IHCs and OHCs in Atoh1-Cre;Cdc42 flox/flox mice at P3 were identical to those in control mice (supplementary material Fig.  S1F). At P8, IHCs in Atoh1-Cre;Cdc42 flox/flox mice obtained a mature shape, in which kinocilia were ablated and stereocilia were aligned in rows in a staircase pattern, which was the same as control IHCs (Fig. 4E). The shape and array of stereocilia in OHCs were also indistinguishable from those in control mice (Fig. 4F). However, when hearing was fully mature at 2 weeks, occasional IHC degeneration with partial stereocilia fusion was observed in Atoh1-Cre;Cdc42 flox/flox mice (Fig. 4H). IHCs showed further degeneration and were randomly eliminated at 4 weeks (Fig. 4I), and IHC loss and stereocilia fusion were frequently observed in the middle turn of the cochlea at 6 weeks ( Fig. 4J). Finally, most IHCs were lost at 8 weeks, except for a small number of IHCs with fused rod-like stereocilia in the middle turn (Fig. 4B). IHCs were more frequently present in the apical turn at 8 weeks, but stereocilia were often fused ( Fig. 4K-M). In the middle turn of Atoh1-Cre;Cdc42 flox/flox mice, the percentage of IHCs with fused stereocilia was 15.2%, 31.4%, 42.8% and 88.2% at 2, 4, 6 and 8 weeks, respectively.
The degeneration and loss of OHCs were much less pronounced than that of IHCs during this period. OHCs were occasionally lost at 8 weeks in Atoh1-Cre;Cdc42 flox/flox mice (Fig. 4B), and the remaining OHCs had scattered stereocilia bundles with a disrupted W-shaped profile; however, no stereocilia fusion was observed (Fig. 4D).
Disturbed ultrastructure of stereocilia and AJCs in Atoh1-Cre;Cdc42 flox/flox mice We used transmission electron microscopy (TEM) to further examine the structural changes in stereocilia and AJCs using IHCs in the middle turn of the cochlea at 2 and 6 weeks. No apparent changes were detected in AJCs of Atoh1-Cre;Cdc42 flox/flox mice at 2 weeks (supplementary material Fig. S4A,B), whereas remarkable changes were observed in the ultrastructures of stereocilia and (A,B) Dissected organ of Corti from P1 wild-type mice cultured for 16 h and infected with the indicated adenoviruses (green). The top panels of A show membrane localization of GFP-Cdc42 covering an individual stereocilum (arrows) and at apical cell junctions in a IHC (arrowhead). Scale bar: 2 mm. The lower panels of A and B show reconstructed lateral view (in the xz axis) images of cochlear OHCs; GFP-Cdc42, but not GFP-Cdc42(17N;4A), is localized at the stereociliary membranes (arrow). Scale bars: 10 mm (C) Schematic representation of the intramolecular Cdc42 FRET biosensor: YPet and Turquoise-GL are variants of YFP and CFP, respectively. (D-G) Dissected organ of Corti from P2 Cdc42-FRET mice were observed under a two-photon excitation microscope. D, F and G are obtained from different organs of Corti. Scale bars: 10 mm. (D) The FRET/CFP ratio is highest at the stereocilia (arrowheads) and is higher at the apicolateral membranes (large arrow) than at the basolateral membranes (small arrow). Note that the image plane is oblique to the apical surface to show cross-sections of various depths, from the level of stereocilia to that of the basolateral membrane (see the illustration above it). A 3D movie is available in supplementary material AJCs at 6 weeks. In contrast to the regular rows of IHCs and supporting cells (where there is a row of supporting cells adjacent to each row of IHCs) in control mice (Fig. 5A), IHCs in Atoh1-Cre;Cdc42 flox/flox mice were often absent and replaced by supporting cells (Fig. 5B). IHCs in control mice were bordered by the arcuate-shaped apical junctional membrane and the underlying thick perijunctional density of the circumferential actin belt (Fig. 5A). Stereocilia in control IHCs were apically located and each had a single rootlet inserted into the cuticular plate (Fig. 5A). In contrast, the membranes at the base of stereocilia in most IHCs of Atoh1-Cre;Cdc42 flox/flox mice were elevated and contained some actin cores with rootlets ( Fig. 5B), which penetrated a visibly normal cuticular plate and appeared to be normal in length, indicating that these stereocilia were fused at the base. The vertical shape of AJCs was ruffled in Atoh1-Cre;Cdc42 flox/flox mice compared with that in control mice ( Fig. 5C). At high magnification, it could be seen that the circumferential actin belt in IHCs was thinner in Atoh1-Cre;Cdc42 flox/flox mice than in control mice (Fig. 5D).

Cdc42-KD in MDCK cells as a model of Cdc42-deleted cochlear hair cells
To further confirm the effect of Cdc42 deletion and to investigate the role of Cdc42 in stereocilia and AJCs, we established an in vitro model using MDCK cells, which possess microvilli (structures analogous to primordial stereocilia) and have been used as a model for AJCs (Ben-Yosef et al., 2003;Nakano et al., 2009). First, we examined the subcellular localization of Cdc42 in MDCK cells stably expressing GFP-Cdc42 (MDCK GFP-Cdc42 ) plated on Matrigel to produce cysts. GFP-Cdc42 fluorescence was intense at the apical membrane and weakly high at the lateral membrane of MDCK GFP-Cdc42 cells in these cysts ( Fig. 6A), consistent with a previous report (Qin et al., 2010). Next, we established MDCK cells with stable knockdown of Cdc42 (MDCK Cdc42-KD ) using the most effective shRNA plasmid sh197 (Fig. 6B,C). Using SEM, we found that the number of microvilli was significantly reduced and the cell border was often dentated at the cell junctions to partially overlie the surface of the neighboring cells in two MDCK Cdc42KD clones (MDCK Cdc42-KDa and MDCK Cdc42-KDg ) in comparison with the control MDCK cells (MDCK cont ) (Fig. 6D). MDCK Cdc42-KDg was used for all further studies. The reduced number of microvilli and the dentated cell border in MDCK Cdc42-KDg cells were almost completely rescued by adenoviral-mediated expression of an shRNA-resistant form of GFP-Cdc42 but not an shRNA-resistant form of the inactive GFP-Cdc42(T17N;4A) ( Fig. 6E,F). Highresolution morphological examination of MDCK Cdc42-KDg cells by scanning helium-ion microscopy (SHIM) showed that microvilli were scattered and had abnormally ragged, fused, short or elongated morphologies (Fig. 6G).
Finally, we examined the formation of tight junctions (TJs) in MDCK Cdc42-KDg cells. Although the TJ marker ZO1 was localized correctly in MDCK cont cells (Fig. 7A), ZO1 was not targeted at TJs in MDCK Cdc42-KDg cells ( Fig. 7B; similar results were obtained using MDCK Cdc42-KDa cells; data not shown). The localization of ZO1 at the cell junctions was recovered by  The disturbed ultrastructures in the stereocilia and AJCs of Atoh1-Cre;Cdc42 flox/flox mice, together with the abnormalities in microvilli and TJs of MDCK Cdc42-KD cells, suggested that Cdc42 deletion compromises actin dynamics in stereocilia, microvilli and AJCs. To understand Cdc42 signaling in these stable actin structures, we examined several molecules associated with actin turnover in MDCK Cdc42-KD cells.
N-WASP, which is a downstream target of Cdc42, has a closed conformation when in the inactive state. N-WASP is activated by binding of Cdc42 to its Cdc42/Rac interactive binding (CRIB) region, together with phosphorylation at Tyr 256 (Suetsugu et al., 2002). The phospho-N-WASP was localized at the apical surface and apicolateral membranes of MDCK cont , but not MDCK Cdc42-KD ,  Fig. S2C,D). Furthermore, the 'wavy' staining of ZO1 in MDCK NWASP-KD cells indicated that TJs were abnormal (supplementary material Fig. S2E). Thus, MDCK NWASP-KD cells showed similar, but milder, phenotypes than did MDCK Cdc42-KD cells.
p21-activated kinases (PAKs) are well-known downstream targets of Cdc42, and the levels of phospho-PAK was reduced in MDCK Cdc42-KD cells (Fig. 8A), whereas phosphorylation of LIMK (Fig. 8A) and cofilin (Fig. 8B), which are downstream targets of PAKs, was unexpectedly increased in both clones of MDCK Cdc42-KD cells, consistent with a previous report (Garvalov et al., 2007). Given that LIMK is a substrate of Rho-associated protein kinases (ROCKs) (Amano et al., 2010), we examined phosphorylation of the ROCK substrate myosin phosphatase targeting subunit 1 (MYPT1) and found that it was enhanced in MDCK Cdc42-KD cells (Fig. 8C). After treatment with the ROCK inhibitor Y27632, the enhanced phosphorylation of cofilin was partially normalized and that of MYPT1 was almost fully normalized (Fig. 8D). The activation of the RhoA-ROCK pathway in MDCK Cdc42-KD cells was confirmed using an antibody to active RhoA and an antibody to phospho-myosin light chain 2 (MLC2) (supplementary material Fig. S3A,B).
The dislocalized phospho-N-WASP and enhanced levels of phospho-cofilin suggested that Cdc42 deletion leads to reduced actin turnover in MDCK Cdc42-KD cells. To test whether the actin turnover was also reduced in hair cells, we evaluated the actin depolymerization rate in the hair cells of Atoh1-Cre;Cdc42 flox/flox mice. We added an actin polymerization inhibitor cytochalasin D, which arrests actin polymerization and causes shortening of stereocilia by continuous depolymerization at the minus ends of F-actin, to the explant culture of organ of Corti, as previously described (Rzadzinska et al., 2004). The length of stereocilia in the areas of interest (supplementary material Fig. S3D,E) was comparable between control mice (1.4760.03-1.8560.04 mm) and Atoh1-Cre;Cdc42 flox/flox mice (1.4660.06-1.7860.05 mm) before treatment with cytochalasin D, but was significantly reduced by treatment with cytochalasin D (supplementary material Fig. S3F). As expected, stereocilia after treatment was slightly but significantly longer in Atoh1-Cre;Cdc42 flox/flox mice (1.1160.02-1.2060.04 mm) than in control mice (0.9960.02-1.0460.02 mm) (supplementary material Fig. S3F), i.e. the rate of stereociliary shortening was less in Atoh1-Cre;Cdc42 flox/flox mice (0.35-0.58 mm/32 h) than in control mice (0.48-0.81 mm/32 h) in all the three areas observed (supplementary material Fig. S3G).

DISCUSSION
Here, we demonstrated for the first time the localization and activation of Cdc42 at the stereociliary membranes and AJCs in cochlear hair cells by the following two methods: exogenous expression of adenovirus-encoded GFP-Cdc42 in hair cells and FRET imaging of hair cells in an explant organ of Corti from Cdc42-FRET biosensor transgenic mice. This result was supported by a study that showed that the highest FRET/CFP ratio of Cdc42 was at the apical surface of 3D-cultured MDCK cells (Yagi et al., 2012), thereby suggesting that Cdc42 plays a role in the microvilli. Consistent with the localization and activation profiles of Cdc42, prominent structural disturbance was observed in stereocilia and AJCs of Atoh1-Cre;Cdc42 flox/flox cochlear hair cells. The phenotype observed in cochlear hair cells of Atoh1-Cre;Cdc42 flox/flox mice might be a consequence of the direct influence of Cdc42 deletion during actin turnover in stereocilia and AJCs. This speculation was further supported by the enhanced phospho-cofilin levels in MDCK Cdc42-KD cells and a reduced actin depolymerization rate in the stereocilia of Atoh1-Cre;Cdc42 flox/flox cochlear hair cells.
The mature stereocilia in Atoh1-Cre;Cdc42 flox/flox mice showed a variety of degenerative changes, such as loss, shortening, fusion, and elongation. Stereocilia loss and shortening are also caused by the deletion of actin regulatory molecules located at the tip of stereocilia, e.g. espins, Eps8, myosin XVa and whirlin (Manor et al., 2011;Sekerková et al., 2011;Zampini et al., 2011). Eps8 is known to promote filopodial growth by forming the Eps8-IRSp53-Cdc42 complex (Disanza et al., 2006), and myosin XVa is the carrier protein of whirlin and Eps8 to form the myosin XVa-whirlin-Eps8 complex at the tip of stereocilia (Manor et al., 2011). The reduction of stereocilia might be partly explained by the alteration of actin polymerization and stabilization at the tip of stereocilia due to Cdc42 deficiency. Intense Cdc42 activity, particularly at the upper portions of stereocilia in Cdc42-FRET biosensor mice, also supports this possibility. However, we cannot exclude the possible involvement of a hitherto undefined or unconventional Cdc42-dependent pathway in stereocilia. Stereocilia fusion is another specific change seen in Atoh1-Cre;Cdc42 flox/flox mice that seemed to initiate at the base of stereocilia. Stereocilia fusion might be a consequence of the downregulation of actin depolymerization, which prevents shaping of the tapered actin core at the base and results in the 'zipping up' of the interstereociliary membrane towards the tip (Sakaguchi et al., 2008;Self et al., 1999). Alternatively, stereocilia fusion might be due to a dysfunction in the tethering of the plasma membrane to the actin cytoskeleton at the base of stereocilia. The overlapping phenotype is found in several other knockout or mutant mouse lines considered to have disturbed membrane tethering, such as those of radixin (Kitajiri et al., 2004), protein tyrosine phosphatase receptor Q (PTPRQ) (Goodyear et al., 2003) and myosin VI (Self et al., 1999). PTPRQ and myosinVI are known to cooperate at the base of stereocilia (Sakaguchi et al., 2008). However, Atoh1-Cre;Cdc42 flox/flox mice exhibited maintenance of specific distribution of PTPRQ and phospho-ERM proteins at the base of stereocilia (data not shown). TRIOBP-knockout mice, which maintain the stereocilia rootlet, also show fused stereocilia (Kitajiri et al., 2010); however, the rootlet structure of Atoh1-Cre;Cdc42 flox/flox stereocilia was unchanged (Fig. 5B). Finally, the presence of an unidentified Cdc42 function in shaping of stereocilia bases cannot be excluded.
Although Cdc42 activity in stereocilia was confirmed in both developing (P2) and mature (P9; data not shown) stages in Cdc42-FRET biosensor mice, the late-onset stereocilia phenotype observed in Atoh1-Cre;Cdc42 flox/flox mice suggests that the dominant involvement of Cdc42 is in the steady-state actin turnover in mature stereocilia. Late-onset progressive hearing loss associated with impaired actin turnover also occurs in Atoh1-Cre-mediated bor c-actin single knockout hair cells at 6 weeks of age, whereas deletion of both band c-actin leads to stereocilia loss by P5 (Perrin et al., 2010). Thus, the common phenotype of late-onset progressive stereocilia disruption observed in b-actin or c-actin single knockout and Atoh1-Cre;Cdc42 flox/flox mice suggests that a partial reduction of actin polymerization affects only the maintenance of stable actin protrusions and not their development. Alternatively, the late-onset phenotype might be explained by the compensation of Cdc42 in developing stereocilia by other actin polymerization factors, such as RhoQ, which is a small GTPase highly homologous to Cdc42 (Heasman and Ridley, 2008), and ELMOD1, which is a GTPase-activating protein of small GTPases that is temporarily expressed in hair cells during development and whose mutation results in hair cell phenotypes similar to those of Atoh1-Cre;Cdc42 flox/flox mice (Johnson et al., 2012).
We found increased levels of phosphorylated (inactivated) cofilin in MDCK Cdc42-KD cells, consistent with a study that reported that Cdc42-deficient neurons display increased inactivation of cofilin and arrested filopodial dynamics (Garvalov et al., 2007). Moreover, neurons lacking both the actin depolymerization factor (ADF) and cofilin (Flynn et al., 2012) exhibit reduced filopodia. These data indicate that the formation and/or maintenance of actin protrusions requires the coordinated action of actin polymerization factors (e.g. Cdc42 and N-WASP) and actin depolymerization factors (e.g. ADF and cofilin), both of which are detected as stereociliary proteins (Shin et al., 2013). The balance between actin polymerization and depolymerization should be elaborately tuned in stable actin protrusions; thus, deletion of actin polymerization factors or actin depolymerization factors or both might cause downregulation of net actin turnover, which leads to structural disturbances. In fact, we confirmed the presence of a significantly, but not drastically, reduced actin depolymerization rate in the stereocilia of Atoh1-Cre;Cdc42 flox/flox mice. The absence of a drastic effect might be attributed to the slow actin turnover rate in stereocilia, with the exception of the distal portion. The reduced length detected in cytochalasin-D-treated stereocilia (0.48-0.81 mm in control mice and 0.35-0.58 mm in Atoh1-Cre;Cdc42 flox/flox mice) was within the range of the reported length (0.3-0.5 mm) of the distal portion of stereocilia, in which rapid actin turnover occurs (Zhang et al., 2012). Alternatively, the mild effect might be due to the limited time scale (32 h) or explant culture at P5, a time at which the morphological abnormality of hair cells had not yet appeared in Atoh1-Cre;Cdc42 flox/flox mice.
Modulation of cofilin phosphorylation is complex and tightly regulated by several competitive pathways, including PAKs, LIMKs, myotonic dystrophy kinase-related Cdc42-binding protein kinase (MRCK) and ROCKs. Using MDCK Cdc42-KD cells, we found that Cdc42 regulates cofilin activity by antagonizing the RhoA-ROCK pathway, based on the following observations: (1) an increase of active RhoA levels (supplementary material Fig. S3A) and (2) the moderate recovery of enhanced phospho-cofilin upon treatment with Y27632 (Fig. 8D). In accordance with these results, Cdc42 suppresses RhoA signaling upstream of ROCK in brain-derived neurotrophic factor (BDNF)stimulated neurons (Chen et al., 2006). A feedback mechanism from F-actin to RhoA for sensing the actin polymerization status might be present to control the Cdc42-mediated antagonism of RhoA signaling (supplementary material Fig. S3C). In addition, although MDCK Cdc42-KD cells exhibited phenotypes in microvilli similar to those of Atoh1-Cre;Cdc42 flox/flox stereocilia, MDCK NWASP-KD cells had a less severe phenotype than those of MDCK Cdc42-KD cells, suggesting that there are other Cdc42-mediated actin polymerization signaling mechanisms (supplementary material Fig. S3C).
Persistent disruption of stereociliary structure and function might ultimately lead to hair cell death. For instance, deficiency of molecules with specific functions in stereocilia, e.g. PTPRQ, myosin XVa and whirlin, are known to show extensive loss of both IHCs and OHCs by the age of 12 weeks (Gong et al., 2006;Goodyear et al., 2003;Mburu et al., 2003). Stereocilia degeneration in Atoh1-Cre;Cdc42 flox/flox mice began at 2 weeks when hair cell loss and AJC disruption were not observed (supplementary material Fig. S4A,B), further supporting this possibility. Cdc42 knockdown impaired AJCs in MDCK cells, as indicated by disturbed ZO1 localization in MDCK Cdc42-KD cells, which is compatible with the results of previous reports (Otani et al., 2006;Popoff and Geny, 2009). The ultrastructural disturbances of AJCs in Atoh1-Cre;Cdc42 flox/flox mice were observed at 6 weeks of age ( Fig. 5D) but not at 2 weeks (supplementary material Fig. S4B), suggesting that Cdc42 is also involved in the long-term maintenance of the circumferential actin belt of AJCs in cochlear hair cells. The disturbance of AJCs in Atoh1-Cre;Cdc42 flox/flox mice was more mild than that observed in MDCK Cdc42-KD cells, probably because of the heterologous (only hair-cell-supporting-cell but not hair-cellhair-cell) cellular connection at the level of AJCs in the organ of Corti (Fig. 3E). The complex between Cdc42, PAR proteins and atypical PKC (aPKC) is a well-known polarity regulator at the apical domain of cells (Ishiuchi and Takeichi, 2011;Suzuki and Ohno, 2006). Our immunohistological analysis showed impaired apical localization of aPKC in both IHCs (supplementary material Fig. S4C-F) and OHCs (data not shown) of Atoh1-Cre;Cdc42 flox/flox mice, suggesting that the Cdc42-PARs-aPKC complex might also be involved in the disturbance of apical domains in hair cells, including stereocilia roots and bases, and AJCs. Dysfunction of the epithelial barrier might affect hair cell morphology and viability in Atoh1-Cre;Cdc42 flox/flox mice, similar to that previously reported in mice with mutations in the TJ proteins claudin 14 and claudin 9. However, because claudin-mutant mouse lines are severely deaf as early as P15-16 and hair cell loss is almost exclusively limited to OHCs (Ben-Yosef et al., 2003;Nakano et al., 2009), it is unlikely that the Cdc42-deficient phenotypes are primarily or solely caused by dysfunction of the epithelial barrier. Finally, IHCdominant loss is unique to Atoh1-Cre;Cdc42 flox/flox mice. In general, IHCs are less vulnerable than OHCs to various etiologies such as aging, noise damage, ototoxic agents and genetic disorders (Schacht et al., 2012). A simple explanation for the predominance of the IHC phenotype is compensation by actin-regulatory molecules specific to OHCs, such as erythrocyte protein p55 (Mburu et al., 2006) and gelsolin (Mburu et al., 2010), which might be required for the maintenance of a more rigorously ordered alignment and length of stereocilia in OHCs than that in IHCs.
In summary, Cdc42 deletion caused the downregulation of actin turnover in cochlear hair cells, which resulted in morphological abnormalities in stereocilia, with additional disturbances in AJCs after maturation, ultimately leading to slowly progressive cochlear hair cell loss. Further research is required to explore further the Cdc42 signaling(s) associated with maintenance of stereocilia (supplementary material Fig. S3C) and to verify whether abnormal ciliation is directly related to hair cell viability.
Preparation of adenovirus solutions for expression of GFP, GFP-Cdc42 and GFP-N-WASP GFP, GFP-Cdc42 and GFP-Cdc42(T17N;4A), resistant to sh197, were amplified by PCR and cloned into the pDONR221 entry vector (Invitrogen). GFP, GFP-Cdc42 and GFP-Cdc42(T17N;4A) in the pAd/ CMV/V5-DEST destination vector (Invitrogen) were obtained using the pDONR221 constructs and Gateway technology (Invitrogen). The initial adenovirus solution was made by transfection of HEK293-FT cells (Invitrogen) with PacI-digested destination plasmid using FuGENE HD (Promega). The final high-titer virus solution was obtained by CsCl buoyant density centrifugation followed by dialysis. The final titer of the virus solution was ,10 11 PFU/ml.
X-gal and H&E stainings X-gal staining was used to detect b-galactosidase expression in whole embryos and dissected inner ears, as previously described (Kassai et al., 2008). Samples for histology were embedded in paraffin wax and 5-mmsections were collected on glass slides, de-paraffinized, and stained using Myer hematoxylin and Eosin (H&E) solution (Muto Pure Chemicals). Xgal and H&E stainings were photographed by using a light microscope (Axioplan II; Carl Zeiss) with a DP26 camera (Olympus).

ABR and DPOAE measurements
To assess hearing, Atoh1-Cre;Cdc42 flox/flox and littermate control mice at the age of 2, 3, 4, 5, 6 and 8 weeks were tested by auditory brainstem response (ABR) measurement either unilaterally or in both ears separately. At least four animals were tested in each group, and the number of tested ears was 13, 16, 22, 6, 14, 28 for control mice and 16, 12, 31, 8, 28, 40 for Atoh1-Cre;Cdc42 flox/flox mice at 2, 3, 4, 5, 6 and 8 weeks, respectively. Mice were anesthetized with 50 mg/kg of body weight pentobarbital by intraperitoneal injection and placed on a heating pad. Reference, ground and earth needle electrodes were placed subcutaneously just posterior to the subjected ear, just anterior to the other ear and at the vertex, respectively. Click or tone burst stimuli at 8, 16, 24, or 32 kHz were generated using SigGenRp software through an EC1 condenser speaker and conducted to the testing ear canal with a plastic acoustic tube. ABR recording was performed using BioSigRP software together with TDT System 3 Real-time Signal Processing Systems (Tucker-Davis Technologies, FL, USA). ABR waveforms were recorded for 12.8 ms at 40,000 Hz by using 50-5000 Hz band-pass filter settings, and ABR waveforms from 500 stimuli were averaged. Hearing thresholds (dB SPL) were defined by decreasing the sound intensity by 10 dB steps and recording the lowest sound intensity level resulting in a recognizable ABR wave pattern.
For DPOAE, measurements were tested bilaterally. The number of ears tested was 10, 8 and 10 for control, and 12, 14 and 10 for Atoh1-Cre;Cdc42 flox/flox mice at the age of 2, 4 and 8 weeks, respectively. DPOAEs were measured by commercial instrumentation HearID TM Auditory Diagnostic System (Mimosa Acoustics; IL, USA) combined with CUBeDIS II v2.40 (Etymotic Research; IL, USA) software. DPOAE at frequency of 2f1-f2 were elicited using two primary tone stimuli, f1 and f2, with sound pressure levels of 65 and 55 dB SPL respectively, with f2/f151.20. A custom plastic ear tip (diameter of 3 mm) attached to an ER-10C (Mimosa Acoustics; IL, USA) probe was inserted into the ear canal and DPOAE amplitude was measured at f2 frequencies of 8, 12, 16 and 20 kHz and plotted after subtraction of noise floor amplitude.

Immunohistochemistry
Dissected tissues were fixed by 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer (pH 7.4) or 10% TCA solution (Hayashi et al., 1999). After permeabilization with PBS containing 0.3% Triton X-100 (PBS-0.3T), fixed tissues were incubated with primary antibody for 2 h at 25˚C in PBS-0.03T and 0.5% fat-free BSA, followed by Alexa-Fluor-488 (or 568)-conjugated secondary antibody or phalloidin for 1 h at 25˚C, and mounted in Prolong anti-fade (Invitrogen) with a coverslip. MDCK cells on the glass-bottomed dish, slide glass with Matrigel (BD Biosciences) or 0.45-mm polyester filter insert were fixed by 4% PFA in 0.1 M phosphate buffer. After permeabilization with PBS-0.3T, fixed cells were incubated with primary antibody in PBS-0.03T with 0.5% fat-free BSA, followed by Alexa-Fluorconjugated secondary antibody, and counterstained with Alexa-Fluorconjugated phalloidin and DAPI (Invitrogen). Immunostainings were observed under a LSM700 confocal microscope (Carl Zeiss).

Counting of IHCs and OHCs, and evaluation of vestibular hair cells
For cell counting of IHCs and OHCs from three Cdc42 flox/flox and Atoh1-Cre;Cdc42 flox/flox mice at 2 weeks, fixed organs of Corti were stained for myosin VIIa and with phalloidin. Three visual fields of interests were selected from the apical, middle and basal turns respectively, and the number of myosin-VIIa-positive hair cells per unit area was counted. The unit area was defined as a field including ten pillar cells, to normalize for shrinking or skewing of the sample. For evaluation of vestibular hair cells, fixed utricular maculae were stained with myosin VIIa antibody and propyl iodide. Five confocal images per hair cell layer, which was identified as the area where myosin-VIIa-positive cells mostly occupied, of each macula obtained from three Cdc42 flox/flox and Atoh1-Cre;Cdc42 flox/flox mice were taken, and the number of nuclei in myosin-VIIa-positive cells per image was analyzed.

SEM and TEM
Freshly dissected inner ear tissues were fixed in 2% PFA, 2.5% glutaraldehyde (GA) in 0.1 M PB. Organ of Corti epithelia were dissected in the same buffer and postfixed with 1% OsO 4 in H 2 O for 2 h. For SEM, tissues were dehydrated in an ethanol series, followed by isoamyl acetate, and dried in a HCP-2 critical-point dryer (Hitachi Koki, Japan). Dried samples were mounted on stubs and examined on a Hitachi S-2380 scanning electron microscope at 10 kV. Precise visualization of MDCK microvilli was performed using an SMT ORION PLUS scanning heliumion microscope (SHIM; Carl Zeiss), which has much higher resolution than conventional SEM (Notte et al., 2010). For TEM analysis, samples were embedded in Epon 812 resin after post-fixation (Okenshoji, Japan), polymerized at 60˚C overnight and ultra-thin sections (thickness ,70 nm) were cut on an ultramicrotome (EM-UC7; Leica Microsystems, Germany), placed on copper grids and examined on Hitachi H-7100 electron microscope at 80 kV.
Organotypic explant culture of cochlea and adenoviral infection P2 organotypic organs of Corti explant cultures were prepared as previously described (Sakaguchi et al., 2008). For adenoviral infection, 2 ml of the high-titer adenoviral solution was added to the culture medium (ex vivo day 1) for 2 h. The explants were fixed 4-10 days after the infection with 4% PFA in 0.1 M phosphate buffer, counterstained with Alexa-Fluor-568-conjugated phalloidin, and observed under a confocal microscope.
Cytochalasin D experiments using organotypic explant culture of organs of Corti The explants of P5 organs of Corti were cultured for 6 h, and then the medium was replaced with medium containing 0.02 mM cytochalasin D.
The tissues were fixed after 32 h, stained with Alexa-Fluor-488conjugated phalloidin, and observed under a confocal microscope. IHCs, which have longer stereocilia than OHCs, in the apical turns at 0-45˚, 45-90˚and 90-135˚from the apex were used in this experiment, as hair cells in the middle or basal turns were often severely damaged. Coverslips were slightly compressed when the tissues were mounted, and only IHCs with stereocilia appropriately spread on a plane were analyzed. The length of the highest stereocilia was measured using NIH ImageJ software. In total, analysis was performed using 120-151 IHCs in ten organs of Corti (Cdc42 flox/flox ) and 60-77 IHCs in five to six organs of Corti (Atoh1-Cre;Cdc42 flox/flox ) for cytochalasin D treatment, and 32-33 IHCs in two organs of Corti (Cdc42 flox/flox ) and 21-29 IHCs in two organs of Corti (Atoh1-Cre;Cdc42 flox/flox ) without treatment.

FRET imaging
Organs of Corti from P2 Cdc42-FRET biosensor mice were dissected in Leibovitz's L-15 medium (Invitrogen), then attached on 3.5-mm Cell-Tak coated dishes (150 mg/ml; BD Biosciences), and maintained in DMEM/F-12 supplemented with 10% FBS. FRET imaging under twophoton excitation microscopy was performed as described previously (Goto et al., 2013). Samples were maintained in an incubation chamber (Tokai Hit, Japan) and imaged using a BX61WI/FV1000 upright microscope equipped with 606 water-immersion objective (LUMPlanFLN; Olympus, Japan) connected to a Mai Tai DeepSee HP Ti:sapphire Laser (Spectra Physics, Mountain View, CA). FRET/CFP images were acquired and analyzed with MetaMorph software (Universal Imaging, West Chester, PA) and Imaris Software (Bitplane AG, Switzerland), and represented by the intensity-modulated display (IMD) mode, in which eight colors from red to blue are used to represent the FRET/CFP ratio.

Three-dimensional culture of MDCK cells
To produce cysts, trypsinized MDCK cells were suspended (1610 4 /ml) in culture medium containing 2% Matrigel. A suspension of 250 ml was placed on an eight-well glass slide (Lab-Tek II; NUNC) coated with 40 ml of polymerized Matrigel, and incubated for 7 days (Yagi et al., 2012). Fixed cells counterstained with Alexa-Fluor-568-conjugated phalloidin were observed under a confocal laser microscope.
To grow microvilli on the apical surface, 1610 4 MDCK cells were grown on a 0.45-mm polyester filter insert (12-mm diameter Transwell; Coning Inc.) for 24 h. Then, 2 ml of adenovirus solution was added to the culture medium for 2 h, and cells were cultured for an additional 4 days. After fixation, samples were observed by SEM. The percentage of MDCK cells with a reduced number of microvilli was calculated using 200 cells from four independent samples.

Immunoblotting and pull-down assay
Plasmids were transfected into MDCK cells using NEPA21 electroporator. For immunoblotting, cells were lysed in homogenizing buffer (Ueyama et al., 2006) by sonication in the presence of protease inhibitor cocktail, protein phosphatase inhibitor cocktail (Nakalai tesque), and 1% Triton X-100. Total cell lysates were centrifuged at 12,000 g for 20 min at 4˚C, and the supernatants were subjected to SDS-PAGE followed by immunoblotting for 2 h at 25˚C using primary antibody diluted in PBS-0.03T containing 0.5% fat-free BSA. Active RhoA was detected with the RhoA Activation Assay Kit (NewEast Biosciences) based on pull-down assays using the monoclonal antibody specifically recognizing active RhoA. The bound primary antibodies were detected with secondary antibody conjugated to horseradish peroxidase (HRP) using the ECL detection system (GE Healthcare).

Statistical analysis
All data are presented as the mean6s.e.m. Two groups were compared using unpaired two-tailed Student's t-test. For comparisons of more than two groups, one-way ANOVA or two-way ANOVA was performed and followed by Bonferroni's post hoc test of pairwise group differences. Statistical analyses were performed using Prism 5.0 software (GraphPad); P,0.05 was considered statistically significant.