The left-right determinant Inversin is a component of node monocilia and other 9+0 cilia

Vertebrates exhibit numerous left-right (LR) asymmetries, such as the positions of the heart and spleen on the left side of the body(Fujinaga, 1997). The process by which LR asymmetry is established can be divided into three phases: (1) the initial determination of LR polarity; (2) LR asymmetric expression of signaling molecules; and (3) LR asymmetric morphogenesis induced by these signaling molecules (Beddington and Robertson, 1999; Capdevila et al., 2000; Wright,2001; Yost, 2001;Hamada et al., 2002).

In spite of recent progress, our knowledge of LR determination remains limited. In particular, the mechanism by which symmetry is initially broken remains unknown, although several models have been proposed(Brown and Wolpert, 1990;Brown et al., 1991). Cilia and fluid flow have been implicated in LR determination(Nonaka et al., 1998;Okada et al., 1999). Monocilia on the node pit cells rotate in a clockwise direction and thereby generate a leftward flow of extra-embryonic fluid (referred to as nodal flow). We recently directly demonstrated a role for nodal flow in LR determination by examining the effects of artificial flow(Nonaka et al., 2002). An artificial rightward flow that was fast enough to reverse the endogenous leftward flow was thus able to reverse LR patterning in mouse embryos. Although it remains unclear how nodal flow contributes to LR determination, it is possible that the flow transports an unknown morphogen toward the left side of the embryo. Alternatively, mechanical stress induced by the flow might be sensed by cells in or near the node.

Various mouse mutants with defects in the initial determination of LR polarity have been identified. The absence of nodal flow, however, may account for the situs defects of most of these mutants. Mutant mice with impaired nodal flow can be classified into two groups: those lacking the node cilia and those with immotile node cilia. The first group includes mice deficient in the kinesin superfamily proteins KIF3A(Marszalek et al., 1999;Takeda et al., 1999) or KIF3B(Nonaka et al., 1998). Other mutants, such as those deficient in polaris(Murcia et al., 2000),probably also belong to this group. The second group includes theIv/Iv mutant (Supp et al.,1997; Supp et al.,1999) and, possibly, mice that lack DNA polymerase λ(Kobayashi et al., 2002) or HFH4 (Chen et al., 1998;Brody et al., 2000).

The Inv (Invs — Mouse Genome Informatics) mutation is unique in that homozygous mutant mice exhibit LR reversal instead of LR randomization (Yokoyama et al.,1993). Unexpectedly, nodal flow in the Inv mutant embryo was shown to be leftward, although it was slow and turbulent(Okada et al., 1999). It remains unclear how such a slow, turbulent flow might result in LR reversal,although plausible models have been proposed(Okada et al., 1999). TheInv/Inv mouse may thus be the only mutant whose defects in LR patterning cannot be simply explained by nodal flow. The Inv gene is expressed ubiquitously in developing mouse embryos and encodes a protein that contains ankyrin repeats and calmodulin-binding motifs(Mochizuki et al., 1998;Morgan et al., 1998). Overexpression of the mouse Inv protein in frog embryos perturbed LR decision making (Yasuhiko et al.,2001), but the precise functions of this protein remain unclear.

We have now generated transgenic mice that express a fusion construct of Inv and green fluorescent protein (Inv::GFP). Our observations of the subcellular localization of this fusion protein suggest that Inv contributes to LR determination as a component of monocilia in the node.

The coding region of the mouse Inv cDNA was linked to the 5′end of the coding sequence for enhanced GFP derived from pEGFP N3 (Clontech). The Inv::GFP fusion cDNA was subcloned into a BOS EX vector(Mizushima and Nagata, 1990)(kindly provided by S. Nagata) under the control of the elongation factor 1α (EF1α) gene promoter, to yield BOS-Inv::GFP and to ensure that the fusion gene was expressed ubiquitously and robustly. TheInvΔC::GFP construct was generated by deleting the 3′part of the coding region of the Inv cDNA at the NheI site. Transgenic mice were generated by pronuclear injection of DNA into fertilized eggs obtained from the mating of inv/+ FVB mice(Saijoh et al., 1999). Crossing of Inv/+, Inv::GFP mice with Inv/+ mice yielded Inv/Inv, Inv::GFP animals. The presence of the transgene was verified by Southern blot analysis with the GFP coding sequence as a probe. GFP fluorescence in newborn mice was observed with a stereomicroscope (Leica MZ FLIII). For Inv::GFP, ten lines of transgenic mice were initially obtained. Eight of them expressed Inv::GFP and their transgenes were able to rescue the situs defects and polycystic kidney of theInv/Invmouse. In the remaining two lines, the transgene failed to rescue these Inv/Inv defects. For InvΔC::GFP, four transgenic mice were established. In three of them, the transgene could rescue situs defects but not polycystic kidney of the Inv/Inv mouse. Among these transgenic lines produced, those expressing the transgenes at a relatively high level, B27 (Inv::GFP) and N105(InvΔC::GFP), were mainly used in this study. Founder mice for the B27 and N105 lines were Inv/+ and +/+, respectively.

A 0.6 kb PvuII fragment of the mouse Inv cDNA corresponding to the region of Inv located downstream of the ankyrin repeats was subcloned into the pGEX-4T vector (Pharmacia). The encoded glutathione S-transferase (GST)-Inv fusion protein was expressed in Escherichia coli strain AD202 (Akiyama and Ito,1990), purified by chromatography on glutathione-Sepharose 4B(Pharmacia), and injected into rabbits. Polyclonal antibodies specific for Inv were isolated by preabsorption of rabbit serum with GST followed by affinity purification.

Primary fibroblasts expressing the Inv::GFP fusion protein were established from the skin of newborn transgenic animals. The cells were cultured in plastic dishes with Dulbecco's modified Eagle's medium supplemented with 10%fetal bovine serum and antibiotics. For immunofluorescence staining, the cells were cultured on cover slips for several days, fixed with 2% paraformaldehyde and stained with specific antibodies as described below.

Primary fibroblast cells were lysed in cell lysis buffer containing 20 mM Tris Hcl, 1% TritonX-100, 0.25% Deoxicorate, 250 mM NaCl, 5 mM EDTA, 1 mM PMSF. The lysates containing 20 μg protein were loaded on 10%SDS-polyacrylamide gels. An affinity-purified polyclonal anti-Inv antibody and a horseradish peroxidase-conjugated goat anti-rabbit antibody (Jackson) were used as the primary and secondary antibodies, respectively. Immune complexes were detected with ECL Western blot Detection Systems (Amersham).

Freshly isolated tissues were frozen in OCT compound (Tissue-Tec) and then cryosectioned (at 8 μm). After fixation with 2% paraformaldehyde, the sections were incubated with rabbit polyclonal antibodies to Inv, to GFP (MBL)or to calmodulin (Zymed), or with mouse monoclonal antibodies to acetylated tubulin (Sigma) or to γ-tubulin (Sigma). Antibodies were diluted in phosphate-buffered saline containing 5% skim milk and 0.05% Tween 20. All antibodies were diluted 1/1000 for staining sections, or 1/5000 for whole-mount staining. Immune complexes were detected with Alexa 488-conjugated goat antibodies to rabbit immunoglobulin G (Molecular Probes) or Alexa 568-conjugated goat antibodies to mouse immunoglobulin G (Molecular Probes). For whole-mount immunofluorescence analysis, freshly isolated embryos were fixed with ice-cold acetone and then incubated consecutively with primary and secondary antibodies at 4°C for 48 and 24 hours, respectively, in phosphate-buffered saline containing 5% skim milk and 1% Triton X-100. Confocal images were obtained with a Carl Zeiss confocal microscope (LSM 510)and three-dimensional images were reconstructed with the LSM 510 software(version 2.5).

Kidneys were fixed in Bouin's solution, dehydrated and embedded in paraffin wax. Serial sections (5 μm) were prepared and stained with Hematoxylin and Eosin according to standard procedures.

To characterize the subcellular localization of the Inv protein, we generated transgenic mice that express an Inv::GFP fusion protein. The transgene, BOS-Inv::GFP, was designed to express a fusion protein in which the GFP sequence is fused in frame to the C terminus of Inv(Fig. 1A). TheInv::GFP construct was placed under the control of the promoter of the Ef1a (Eif1a — Mouse Genome Informatics) gene(Mizushima and Nagata, 1990),which confers ubiquitous and high-level expression. The BOS-Inv::GFPtransgene was microinjected into mouse fertilized eggs (+/+ or+/Inv), resulting in the generation of several transgenic lines expressing Inv::GFP. Unless indicated otherwise, the B27 line, which expresses Inv::GFP at a relatively high level, was used in the present study. Mice harboring the Inv::GFP transgene (+/+ or +/Inv, Inv::GFP)were normal and fertile. Transgenic embryos expressed the fusion protein in a ubiquitous manner (Fig. 1B;data not shown), and newborn transgenic mice were readily recognized by the presence of fluorescence in their limbs(Fig. 1C).

To determine whether the Inv::GFP protein is functional, we transferred the transgene to the Inv/Inv background. Homozygous Inv mutant mice exhibit situs inversus and die within a few weeks after birth as a result of polycystic kidney disease (Yokoyama et al., 1993). The laterality defects of Inv/Inv newborn mice are evident from the position of the stomach, which is located on the right side (Fig. 1D). The lung lobation pattern is also reversed, with one lobe on the right and four lobes on the left (Fig. 1E). The defects caused by the Inv mutation were rescued in Inv/Invmice by expression of the Inv::GFP transgene. Thus, in Inv/Inv,Inv::GFP mice, the stomach was located on the left(Fig. 1D) and lung lobation was normal (one lobe on the left, four lobes on the right)(Fig. 1E). Moreover,Inv/Inv, Inv::GFP mice grew normally and were fertile, and they did not develop the polycystic kidney disease and jaundice characteristic ofInv/Inv animals (Fig. 1F,G). These observations thus indicated that the Inv::GFP protein is fully functional.

We examined the subcellular localization of Inv::GFP in primary fibroblasts established from newborn Inv/Inv, Inv::GFP mice. In nonfixed cells,GFP fluorescence was detected in rod-like structures protruding from the cell body (Fig. 2A-C). To determine whether these structures were primary cilia, we subjected fixed fibroblasts to immunofluorescence staining for acetylated tubulin, a marker of primary cilia and centrioles. The rod-like structures that were positive for GFP were indeed detected by antibodies to acetylated tubulin(Fig. 2D-I). These results thus indicated that, in primary fibroblasts, Inv::GFP is preferentially localized to primary cilia.

We also examined the localization of the endogenous Inv protein in primary fibroblasts derived from newborn wild-type (non-transgenic) mice, with an anti-Inv antibody. In Western blot analysis(Fig. 3A), the anti-Inv antibody detected an 116 kDa protein in +/+, and Inv/+ fibroblasts but not in Inv/Inv fibroblasts. The same antibody detected the Inv::GFP fusion protein of 140 kDa in Inv/Inv, Inv::GFP fibroblasts. In immunohistological staining, the anti-Inv antibody detected primary cilia in +/+ and Inv/Inv, Inv::GFP fibroblasts but not those inInv/Inv fibroblasts (Fig. 2J-L, Fig. 3B),confirming the specificity of this antibody. These results indicate that the native Inv protein is also localized to primary cilia. The subcellular localization of Inv::GFP thus appeared to be indistinguishable from that of the native Inv protein.

Double staining of subconfluent cultures of primary fibroblasts derived from Inv/Inv, Inv::GFP mice with antibodies to GFP and antibodies toγ-tubulin revealed several distinct patterns of centriolar staining(Fig. 4A).

1. Neither of the pair of centrioles was associated with a GFP signal (data not shown).

2. One of the pair of centrioles was associated with a spot-like GFP signal(Fig. 4B).

3. One of the pair of centrioles was associated with a rod-like GFP signal(Fig. 4C).

4. One of the pair of centrioles was associated with an extended rod-like GFP signal (Fig. 4D).

Time-lapse observations revealed that these different staining patterns represent a temporal sequence from (1) to (4) during the formation of primary cilia (see Movie 1). Thus, the Inv::GFP signal first appeared associated with a mother centriole as it began to form a primary cilium; the fusion protein signal then elongated as the cilium extended from the older centriole, which becomes the basal body(extension of the Inv::GFP signal over a distance of 5 μm required ∼ 50 minutes).

Given that the monocilia on the node pit cells are implicated in LR determination (Nonaka et al.,1998), we next examined the localization of Inv::GFP in the node. Observation of the node of Inv/Inv, Inv::GFP embryos from the ventral side without fixation (Fig. 5A)revealed a dot-like pattern of GFP fluorescence signals within the node(Fig. 5B). Lateral observation of the node revealed rod-shaped fluorescence signals protruding from the ventral node cells into the node cavity(Fig. 5C). Examination of the cells on the ventral side of the node at high magnification showed that Inv::GFP was present within the monocilia of these cells(Fig. 5D). The fusion protein was detected uniformly along the entire length of all the monocilia. Immunohistological analysis confirmed the preferential localization of Inv::GFP to monocilia of the ventral node cells. Monocilia that were detected with antibodies to acetylated tubulin (Fig. 5E) were thus also stained with antibodies to GFP(Fig. 5F). Together, these results demonstrated the preferential localization of Inv protein in node monocilia.

The cilia on node pit cells are monocilia; that is, they exhibit the 9+0 organization of microtubules. The preferential localization of Inv::GFP in the node monocilia prompted us to examine other 9+0 cilia as well as 9+2 cilia inInv/Inv, Inv::GFP mice. Epithelial cells of the kidney collecting tubules possess 9+0 cilia. Examination of the kidneys of Inv/Inv,Inv::GFP mice revealed the presence of Inv::GFP in most of the cilia on the epithelial cells of the renal tubules, which were also positive for acetylated tubulin (Fig. 6A-C). Inv::GFP was also detected in the 9+0 cilia of the pituitary gland(Fig. 6D-F) and retina(Fig. 6G,H).

By contrast, Inv::GFP was not preferentially detected in 9+2 cilia, which are abundant in the epithelial cells of the respiratory system, oviduct and ependyma. Thus, in the oviduct (Fig. 6J-L) and ependyma (data not shown) of Inv/Inv, Inv::GFPembryos, Inv::GFP was not preferentially expressed in the cilia, revealed by their content of acetylated tubulin, but was instead present predominantly in the cytoplasm. In the trachea, Inv::GFP was detected in basal bodies stained by antibodies to γ-tubulin but not in the cilia(Fig. 6M-R). These observations thus revealed that, in cells with 9+2 cilia, Inv::GFP localized either to the cytoplasm or to the basal body, but did not accumulate preferentially in the ciliary body.

Inv is expressed ubiquitously in developing mouse embryos(Mochizuki et al., 1998). Inv::GFP was also expressed widely in transgenic mice because the transgene was placed under the control of the Ef1a gene promoter. In those cells without cilia, Inv::GFP was detected predominantly in the cytoplasm(Fig. 6I), but was occasionally detected in nuclei of some cell types such as skeletal muscle cells (data not shown). At adult stage, the transgene expression was detected in many organs,including the brain, muscle and testis (data not shown).

The mouse Inv protein contains two potential calmodulin binding (IQ) motifs(Mochizuki et al., 1998;Morgan et al., 1998) and interacts with calmodulin in vitro(Yasuhiko et al., 2001;Morgan et al., 2002). Expression of the wild-type Inv protein in frog embryos perturbed LR determination, whereas expression of mutant Inv proteins that lack either of the two calmodulin binding motifs did not(Yasuhiko et al., 2001). These observations suggested that Inv may function through direct interaction with calmodulin. To test this hypothesis, we examined whether calmodulin colocalizes with Inv in cilia. Immunostaining of wild-type mouse embryo tissues with antibodies to calmodulin revealed the presence of this protein in 9+2 cilia, including those in the trachea(Fig. 7A-C), oviduct and ependyma (data not shown). By contrast, calmodulin was not detected in any of the 9+0 cilia examined, including those in the kidney(Fig. 7D-F), node(Fig. 7G-I), and pituitary gland (data not shown). Thus, whereas Inv is localized to 9+0 cilia,calmodulin is localized to 9+2 cilia.

The Inv protein also contains an ankyrin repeat domain, a motif that mediates protein-protein interaction. To investigate whether Inv might function by interacting with other proteins through its ankyrin repeats, we examined whether the N-terminal region of Inv including the ankyrin repeats was able to rescue the defects of Inv/Inv mice. We thus established permanent transgenic lines that express InvΔC::GFP, a chimeric protein in which the N-terminal region of Inv is fused to GFP(Fig. 8A). One of these lines(N105) that expressed InvΔC::GFP at a relatively high level was used for the studies described below. InvΔC::GFP protein appears to be unstable since the level of InvΔC::GFP protein detected in fibroblasts fromInv/Inv; InvΔC::GFP mice was always much lower than that of Inv::GFP in Inv/Inv; Inv::GFP fibloblasts(Fig. 3A).

The InvΔC::GFP transgene derived from N105 mice was transferred to the Inv/Inv background. Although Inv/Inv,InvΔC::GFP mice exhibited normal LR patterning of visceral organs,including the stomach and the lung, they developed polycystic kidney disease a few weeks after birth (Fig. 8B). The InvΔC::GFP protein was detected in the primary cilia of cultured fibroblasts derived from Inv/Inv, InvΔC::GFPmice as well as in the monocilia of the node(Fig. 8C). It was not apparent,however, in the 9+0 cilia of the kidney(Fig. 8C), consistent with the failure of the transgene to rescue the kidney defect of Inv/Inv mice. These results thus suggest that InvΔC is able to localize to 9+0 cilia and to function as an LR determinant. The absence of InvΔC::GFP in the kidney is most probably due to the unstable nature of this protein.

The localization of the Inv::GFP protein in 9+0 cilia prompted us to examine the node monocilia of Inv/Inv embryos. Scanning electron microscopy revealed that the cilia of the Inv/Inv embryos were indistinguishable from those of wild-type embryos in their length, thickness and shape (data not shown). Video microscopy also revealed that the node cilia of Inv/Inv embryos rotate at a speed of 600 rpm in a clockwise direction, just like those of wild-type embryos. Thus, we were not able to detect any abnormalities of the node monocilia in Inv/Inv mice.

We recently developed an experimental system for the culture of mouse embryos under conditions of artificial flow of the culture medium(Nonaka et al., 2002). With this system, we showed that artificial nodal flow is able to determine LR patterning of wild-type embryos in a manner dependent on the direction and speed of flow and on the developmental stage of the embryo(Nonaka et al., 2002). We therefore tested whether artificial nodal flow was able to rescue the laterality defects of Inv/Inv mice. Exposure of Inv/Invembryos to a fast leftward flow resulted in normal heart looping and normal embryonic turning (Fig. 9). This rescue of the LR patterning defects of Inv/Inv mice by artificial leftward flow suggested that nodal flow is impaired inInv/Inv embryos. This conclusion is consistent with previous observations that nodal flow is slow and turbulent in Inv/Inv embryos(Okada et al., 1999), and it supports the idea that Inv plays a role in LR decision making as a component of the node monocilia.

Our data indicate that Inv is preferentially localized to various 9+0 cilia(monocilia) but not to 9+2 cilia. A recent study(Nurnberger et al., 2002)reported the presence of Inv immunoreactivity in nuclei and at the cell membrane, but not in the primary cilia, of cultured Madin-Darby canine kidney(MDCK) cells, which resemble kidney epithelial cells. However, transfection of MDCK cells with pBOS-Inv::GFP revealed that the Inv::GFP protein was localized to the monocilia (data not shown). The reason for this apparent discrepancy is not clear. We also observed the nuclear localization of Inv::GFP in certain cell types, such as skeletal muscle cells, of transgenic mice, suggesting that the subcellular distribution of Inv may be dependent on cell type.

In mammals, primary cilia are present in a variety of organs at embryonic and adult stages. With a few exceptions, such as the role of node monocilia in LR patterning, the precise functions of these cilia remain unknown. Although node monocilia are motile (Sulik et al.,1994; Nonaka et al.,1998), most primary cilia are thought to be immotile. For example,monocilia present in the proximal tubule of the kidney are relatively long compared with the diameter of the collecting tube and coexist with numerous microvilli, making it unlikely that they are able to move actively. Such immotile primary cilia have been proposed to function as signal sensors, in which case Inv may be required not only for the movement of primary cilia (as in the node cilia) but also for the reception of external signals by these structures.

Substantial evidence implicates the node monocilia and nodal flow in LR determination. The localization of Inv to 9+0 monocilia further supports a role for node cilia in LR decision making. However, the mechanism by which nodal flow directs LR patterning is not known. As previously suggested(Nonaka et al., 1998), nodal flow may transport a molecule that acts as a LR determinant toward the left side. Alternatively, the direction of or mechanical stress generated by the flow may be sensed differentially by cells on the two sides of the embryo. The Inv protein may thus be required for correct movement of the node cilia or for receiving signals generated by the flow. Our observations favor the former possibility but do not exclude the latter.

The mouse Inv protein possesses two calmodulin binding (IQ) motifs, both of which indeed bind calmodulin in vitro(Yasuhiko et al., 2001;Morgan et al., 2002). Whereas mouse Inv perturbed LR determination when ectopically expressed inXenopus embryos, mutant Inv proteins lacking either of the two calmodulin binding motifs exhibited no such activity(Yasuhiko et al., 2001). These observations were thus suggestive of a functional interaction between Inv and the Ca²⁺-binding protein calmodulin. However, our data do not support a role for such an interaction in LR patterning. First, we showed that these two proteins are not colocalized in the mouse embryo; rather, they show reciprocal expression patterns in that calmodulin is present in 9+2 cilia,whereas Inv is localized to 9+0 cilia. Second, in contrast to the observations made with frog embryos (Yasuhiko et al.,2001), the InvΔC::GFP protein, which lacks the C-terminal region (including one of the calmodulin binding motifs) of Inv, was able to rescue the LR defects of Inv/Inv mice. Nonetheless, it is formally possible that Inv interacts with low levels of calmodulin because the InvΔC::GFP protein retains the N-terminal calmodulin binding motif. Our data do not exclude a role for Ca²⁺ in Inv function, in which regard mice lacking polycystin 2, a putative Ca²⁺ channel, have been shown to exhibit LR defects(Pennekamp et al., 2002).

Artificial nodal flow was able to rescue the LR patterning defects ofInv/Inv mice. This result is consistent with previous observations that nodal flow of Inv/Inv embryos is leftward but slow and turbulent(Okada et al., 1999). Inv may thus be required for the correct movement of node cilia. Our examination of the node monocilia of Inv/Inv mice, however, failed to reveal any apparent anomalies in their morphology or motility. Further investigations of the fine structure and movement of the node monocilia in these mutant animals are thus required to detect possible abnormalities. The development of an imaging system that allows observation of the cilia of live mouse embryos at high magnification would facilitate this goal.

Node monocilia may also function as signal sensors. Our data with artificial nodal flow may not favor this notion, but they do not exclude the possibility that Inv plays dual roles in generating correct nodal flow and in transducing signals generated by such flow. Numerous examples of immotile cilia that function in sensory perception, including chemoreception,photoreception and mechanoreception, have been described in nonvertebrates. InCaenorhabditis elegans, cilia are present in 60 out of the 302 neurons, in which they function as sensory organelles(Bargmann and Horvitz, 1991;Dwyer et al., 1998). InChlamydomonas, flagella transduce signals during gamete adhesion in mating (Solter and Gibor,1977; Pan and Snell,2002). Furthermore, in the mouse, the polaris protein is required for formation both of 9+0 cilia such as node monocilia(Murcia et al., 2000) and of 9+2 cilia such as those in brain ependymal cells. A polaris homolog in C. elegans, OSM-5, localizes to the basal body and cilia and functions in intraflagellar transport; in osm-5 mutants, sensory neurons lack cilia that function as sensors of osmotic pressure(Haycraft et al., 2001;Taulman et al., 2001). Whether or not node cilia (and Inv) function in sensory perception remains to be determined.

Homozygous Inv mutant mice develop polycystic kidney disease a few weeks after birth. Furthermore, many of the genes required for LR determination, including Inv, polaris and polycystin 2, are also implicated in polycystic kidney disease in humans. In general, the histological features of polycystic kidneys are first apparent in the proximal collecting tubules. Our examination of eight independent Inv/Inv,Inv::GFP transgenic lines revealed that the abundance of Inv::GFP protein in monocilia of the proximal collecting tubules in newborns correlated with the severity of polycystic kidney disease at later stages (data not shown). This correlation, together with the localization of Inv::GFP in the monocilia of the kidney epithelium, suggests that Inv functions in these cilia. However,examination of the kidney monocilia of Inv/Inv newborn mice (before polycystic kidney disease becomes severe) revealed no apparent anomalies with respect to the number or morphology of the cilia (data not shown). Although the precise mechanism responsible for the development of polycystic kidney disease in Inv/Inv mice is unknown, the kidney monocilia may function as organelles that sense external signals, such as ion concentration, osmotic pressure or mechanical flow, and Inv may be required for transduction of such signals.

We thank Dr Tadanobu Ban for introducing us electron microscopy; and Kyoko Mochida and Sachiko Ohishi for technical assistance. This work was supported by CREST (Core Research for Evolutional Science and Technology) of the Japan Science and Technology Corporation.