Hedgehog signalling is required for correct anteroposterior patterning of the zebrafish otic vesicle

The inner ears of vertebrates are responsible for the sensations of hearing and of balance. Although the two ears are symmetric about the midline of the organism, each individual organ in most vertebrate species is asymmetric about all three axes [anteroposterior/rostrocaudal (AP), dorsoventral (DV) and mediolateral (ML)]. The zebrafish otic field (defined by the expression ofpax8) is induced at the lateral edges of the neural plate, adjacent to several presumptive rhombomeres (r) of the developing hindbrain. By about 14 hours post fertilisation (hpf), condensation of cells gives rise to distinct thickenings (the otic placodes) immediately opposite r5(Kimmel et al., 1995;Pfeffer et al., 1998;Phillips et al., 2001). The expression of several genes, for example pax2a (formerlypax2.1), dlx3b (formerly dlx3) and eya1,is now detected throughout these placodes(Krauss et al., 1991;Akimenko et al., 1994;Sahly et al., 1999). The only known genes with restricted patterns of expression in the otic placode at this stage are the Delta genes. These are expressed in anterior and posterior domains, symmetrical about the AP and DV axes, but restricted to the medial side of the placode (Haddon et al.,1998). It is therefore likely that at 14 hpf only the ML axis of the otic placode has been specified.

Asymmetric gene expression patterns about both the AP and the DV axes are obvious by 18 hpf, when the placode begins to cavitate to form an otic vesicle. nkx5.1 (hmx3 — Zebrafish Information Network), which is currently the earliest known marker of an asymmetry about the AP axis, is expressed in an anterior domain from around 16 hpf;pax5 is detectable in the anterior epithelium from 17.5 hpf anddacha is detected in the dorsal otic epithelium by 17-18 hpf,suggesting that all axes of the ear have been specified by this time(Pfeffer et al., 1998;Adamska et al., 2000;Hammond et al., 2002). By 24 hpf, several further otic genes are expressed asymmetrically, and this presumably both reflects and reinforces axis specification. pax5,nkx5.1 and fgf8 are expressed anteriorly, bmp7 andfollistatin posteriorly, dlx3b and dacha dorsally,eya1 ventrally, and pax2a and dacha medially(Krauss et al., 1991;Akimenko et al., 1994;Pfeffer et al., 1998;Reifers et al., 1998;Sahly et al., 1999;Adamska et al., 2000;Mowbray et al., 2001;Hammond et al., 2002). Sensory epithelium now thickens and stratifies, and fingers of non-sensory epithelium protrude into the otic lumen and fuse to form the semicircular canal system(reviewed by Whitfield et al.,2002).

Fekete and colleagues have proposed a model in which tissues surrounding the ear provide inductive signals for both axis specification and further otic differentiation (Fekete, 1996;Brigande et al., 2000a;Brigande et al., 2000b). They propose that signals from the hindbrain have dorsalising activity, and may also be important in providing AP information and medialising signals to the otic vesicle. Several lines of evidence, ranging from early transplantation experiments carried out in Amblystoma to more recent studies of knockout and mutant mice, suggest that the hindbrain does provide signals to pattern medial and dorsal otic regions(Harrison, 1945;Deol, 1964;Mansour et al., 1993;Mark et al., 1993;McKay et al., 1996;Niederreither et al., 2000). Each hindbrain rhombomere also expresses a specific and unique group of genes,including members of the Hox gene cluster, and may thus impart AP identity to adjacent inner ear regions (Fekete,1996; Prince et al.,1998; Brigande et al.,2000a; Brigande et al.,2000b).

Fekete and colleagues also suggest that ventral midline structures (i.e. the notochord and floorplate) may specify ventral otic structures. Both the notochord and floorplate are strong sources of Hedgehog (Hh) proteins, and evidence from the chick suggests that these tissues are able to repress dorsal and lateral otic fate (Giraldez,1998). We therefore set out to test whether Hedgehog signalling from the ventral midline is required to pattern the developing ear, and in particular whether it is responsible for the specification and development of ventral and/or medial otic structures. Hh proteins are secreted peptides known to act as morphogens in the axis specification of other organs, such as the neural tube, limb bud and somites(Echelard et al., 1993;Krauss et al., 1993;Riddle et al., 1993;Roelink et al., 1995)(reviewed by Hammerschmidt et al.,1997; Ingham and McMahon,2001). In many situations, Hh is a diffusible molecule, and, in vertebrates, has been reported to act over several cell diameters (up to 300μm in chick limb bud mesenchyme, for example)(Gritli-Linde et al., 2001;Lewis et al., 2001;Zeng et al., 2001). Details of the signalling pathway have been elucidated in Drosophila. The Hh receptor, patched (Ptc), in the absence of Hh ligand, interacts with, and inhibits the action of, Smoothened (Smo). In the presence of Hh, repression of Smo via Ptc1 is lifted, and the signal is transduced through various intercellular intermediates to the transcription factor cubitus interruptus(Ci). Among the targets of the Hh signalling cascade is ptc itself,whose transcription is upregulated by active Hh signalling (for a review, seeIngham and McMahon, 2001).

In zebrafish, four hedgehog homologues have been reported: sonic hedgehog (shh) and tiggy-winkle hedgehog(twhh) (both orthologues of tetrapod Shh), and echidna hedgehog (ehh) and hh-a (orthologues of Indian hedgehog) (Krauss et al.,1993; Ekker et al.,1995; Currie and Ingham,1996; Zardoya et al.,1996a; Zardoya et al.,1996b). Components of the transduction cascade include two Patched genes, and at least three Gli genes, orthologues of ci(Concordet et al., 1996;Karlstrom et al., 1999;Lewis et al., 1999a;Varga et al., 2001). Importantly, only a single smoothened orthologue, smo,appears to have been retained in the zebrafish genome, and all Hh signalling is thought to require the function of this gene(Varga et al., 2001). Mutations in some components of the Hh pathway have been isolated:shh is disrupted in sonic you (syu) mutants,gli1 is disrupted in detour (dtr) mutants,gli2 is disrupted in you too (yot) mutants and thesmo gene is disrupted in mutant alleles of slow muscle omitted (smu) (Schauerte et al., 1998; Karlstrom et al.,1999; Barresi et al.,2000; Chen et al.,2001; Varga et al.,2001; Karlstrom et al.,2003). In addition, the Hh pathway is thought to be disrupted inchameleon (con), iguana (igu) andyou mutants (Schauerte et al.,1998; Lewis et al.,1999b; Odenthal et al.,2000).

Consistent with a role for Hedgehog in early medial or ventral otic patterning, we find that all essential components of the Hh signal transduction cascade are expressed in the otic vesicle, while threehh genes are expressed in adjacent midline structures (notochord and floorplate). Surprisingly, however, mutant analysis indicates that Hh signalling appears to be involved in AP patterning of the otic vesicle, rather than DV or ML patterning as predicted. Using double mutants and antisense morpholino experiments, we also show that both Shh and Twhh are involved in this AP patterning, and that either gene alone can compensate for the absence of the other.

Wild-type embryos used were WIK or AB. Mutant strains used werecon^tf18b, cyc^tf219, dtr^ts269,flh^tk241 or flh^tm229, igu^ts294,ntl^tc41, oep^t2257, smu^b641,syu^t4 and you^ty97, all recessive loss-of-function alleles, and yot^ty119, a dominant repressor of Gli-mediated Hh signalling. Phenotypically wild-type siblings were used as controls. Embryonic stages are given as hours post fertilisation(hpf) at 28.5°C, converted from somite stages in embryos younger than 24 hpf (Westerfield, 1995;Kimmel et al., 1995).

Whole-mount in situ hybridisation was carried out as described previously(Oxtoby and Jowett, 1993). Digoxigenin-labelled probes were prepared according to manufacturer's instructions (Roche). For microscopy, embryos were cleared in a glycerol/PBS series and mounted in 100% glycerol. Sense hybridisations were carried out forsmo, gli2 and ptc2; all were negative.

Embryos were whole-mount stained for actin with FITC-phalloidin as described previously (Haddon and Lewis,1996), mounted in Vectashield (Vector Laboratories) and imaged with a Leica SP confocal microscope. For dorsal views, ears were dissected.

After in situ hybridisation, embryos were fixed overnight in 4%paraformaldehyde, and cleared through a glycerol/PBS series. Sections(∼100 μm) were cut using a hypodermic needle and mounted in 100%glycerol. For thinner sections, fixed embryos were embedded in 1% low melting point agarose to facilitate correct orientation. Agarose blocks were dehydrated and cleared in an ethanol/butanol series, embedded in paraffin wax,and sectioned at 7 μm. Sections were stained with Haematoxylin and Eosin,and mounted in DePeX (Sigma).

5′methylguanosine-capped sense mRNA was produced as described previously (Krieg and Melton,1984). RNA (5 nl) was injected into one- or two-cell embryos using a Narishige microinjection rig, at 50 ng/μl to 1 μg/μl forptc1 RNA, 25 ng/μl to 100 ng/μl for shh RNA, 25 ng/μl to 400 ng/μl for dnPKA RNA(Concordet et al., 1996), and 500 ng/μl for ehh and twhh RNA. nGFP RNA (75 ng/μl)was co-injected in all experiments. GFP was visualised between shield stage and tail bud stage; embryos not expressing GFP ubiquitously were discarded.

Carboxyfluorescein-conjugated antisense morpholinos (GeneTools) were targeted to the 5′ end of the shh and twhh open reading frames (GenBank Accession Numbers, AF124382 and U30710, respectively). The sequences were: shh MO (5′ to 3′), aag ccg cat ttt gcc gca cgc tga a; and twhh MO, gct tca gat gca gcc tta cgt cca t(Lewis and Eisen, 2001). Morpholinos (MOs) were diluted to 0.5 mM or 0.25 mM using Danieau medium(Nasevicius and Ekker, 2000)and injected into one- or two-cell embryos; any embryo not showing ubiquitous fluorescence was discarded. twhh MO (0.5 mM) caused necrosis at the anterior end of the embryo, which appears to be a nonspecific effect (data not shown).

For observation, embryos were anaesthetised with tricaine (3-amino benzoic acid ethyl ester) and mounted in 3% methyl cellulose(Westerfield, 1995). Initial analysis was carried out using a Leica MZ12.5 fluorescence dissecting microscope. Detailed examination and photography was carried out using an Olympus BX51 compound microscope, Olympus Camedia (C-3030ZOOM) camera and AnalySIS software (Olympus). Images were assembled using Adobe Photoshop.

As a first step towards identifying possible roles for Hh signalling in otic vesicle development, we analysed mRNA expression patterns of zebrafish Hh pathway components in the vicinity of the ear(Fig. 1). None of the three zebrafish hedgehog genes examined, twhh, shh andehh, are expressed in the otic vesicle itself. The closest source of Hedgehog to the ear between 16.5 hours post fertilisation (hpf) and 30 hpf is from midline tissues; shh is expressed in both the notochord and floorplate, while twhh is expressed only in the floorplate, andehh only in the notochord (Fig. 1C,E,G) (Krauss et al.,1993; Ekker et al.,1995; Currie and Ingham,1996). These structures are approximately 40 μm from the vesicle at 24 hpf, which is a feasible distance over which Hh may act(Gritli-Linde et al., 2001;Lewis et al., 2001;Zeng et al., 2001).shh is also expressed in visceral endoderm from 24 hpf(Roy et al., 2001), and in pharyngeal endoderm from 30 hpf(Piotrowski et al., 2000). At 25 hpf, endodermal expression is about 75 μm posterior to the ear, and is weaker than the floorplate expression nearer the ear (data not shown)(Roy et al., 2001).twhh expression is also detectable in the pharyngeal endoderm at 24 hpf, ∼100 μm from the otic vesicle (data not shown).

Genes encoding components for the reception and transduction of the Hh signal are expressed in the otic epithelium. ptc1 is expressed in a ventromedial domain from 16.5 to 30 hpf, initially uniformly along the AP axis of the vesicle, but becoming concentrated in the posterior by 22 hpf(Fig. 1B,I,J). This indicates active Hh signal transduction in the otic vesicle, since ptc1 is itself a transcriptional target of the Hh pathway(Concordet et al., 1996;Goodrich et al., 1996).ptc2 is expressed similarly to but more widely than ptc1, as its expression is upregulated by a lower concentration of Hh signal(Lewis et al., 1999a). By 24 hpf, ptc2 RNA is detectable throughout ventral otic epithelium of wild-type embryos, rather than being restricted to a ptc1-like ventromedial band (Fig. 1D).smo is expressed throughout the otic vesicle from 16.5 to 30 hpf(Fig. 1F). Thus, all reported essential components of the zebrafish Hh signalling pathway are expressed in locations consistent with a direct role for Hh in early ear development.gli2, however, is not highly expressed in the otic epithelium(Fig. 1H). It is possible,however, that another Gli gene is expressed here, as Gli genes are expressed differentially in other developmental contexts (reviewed byIngham and McMahon, 2001).

Providing further evidence for a direct effect of Hh signalling on otic vesicle development, otic ptc1 expression is greatly reduced or lost in two strong Hh pathway mutants, con^tf18b andsmu^b641 (Fig. 2A-C). In addition, ptc1 expression is upregulated in the ears of embryos in which shh RNA has been overexpressed(Fig. 2D). In all three of these cases, an otic phenotype is associated with the alteration inptc1 expression, as discussed below.

To investigate the effect of reduced Hedgehog signalling on otic development, we analysed the ears of all zebrafish mutants known or presumed to be defective in a component of the Hh signalling pathway(Table 1). Only three of these show gross otic patterning defects. First, con and smu ears,contrary to expectation, display AP patterning defects, as described below. Second, the ears of igu mutants lack the dorsolateral septum that divides the anterior and posterior semicircular canals, but appear normal in all other respects (data not shown); this phenotype will not be considered further in this report.

By 72 hpf, the wild-type ear is well differentiated and displays clear asymmetries about all three axes (AP, DV and ML). The most obvious structures are the otoliths, which lie over the two maculae. The smaller, anterior(utricular) otolith is situated ventral and lateral to the larger, posterior(saccular) otolith, which lies medially. In both smu and conthe two otoliths are small, ventral and lateral, resembling the anterior otolith (Fig. 3A-C). The underlying maculae can be visualised by phalloidin staining, which labels the actin-rich stereociliary bundles of hair cells. In the ears of wild-type embryos, the anterior (utricular) macula sits on the anteroventral floor, and the posterior (saccular) macula lies on the posteromedial wall(Fig. 3D,G). In smu,however, a single sensory patch covers the entire ventral floor of the vesicle, while in con the anterior macula is present but an additional ventral sensory patch develops at the posterior of the ear(Fig. 3E,F,H,I). This second patch resembles a posterior macula in shape, but is reduced in size. In neither smu nor con is there a sensory patch in the normal medial position of the posterior macula(Fig. 3E,F). Note, however,that the position of the axes of the ear with respect to the midline is altered in con and smu homozygotes(Fig. 3J-L). Midline tissue is missing, bringing the ventral surface of the ear closer to the midline than normal. Taken together, these data suggest that posterior otic regions are not specified correctly in smu and con, and may be acquiring some anterior identity.

To investigate smu and con ear patterning in detail, we used a panel of otic region-specific markers. Anterior expression domains of three genes are duplicated or expanded into posterior regions of the otic vesicle of smu and con homozygotes. Anterior domains ofotx1 expression at 48 hpf are duplicated in a mirror image manner in both con and smu (Fig. 4A-C). Similarly, wnt4 expression at 36 hpf, normally detectable at the posterior end of the anterior macula, shows duplicated expression at the anterior end of the second ventral macula in con(Fig. 4D,F). wnt4 is also expressed in the centre of the single ventral macula in smu(Fig. 4E). Third,nkx5.1, which normally marks anterior otic regions from 16 hpf, and a small posterodorsal region by 30 hpf, is expanded along the entire AP extent of the ear in both con and smu(Fig. 4G-I). These expression domains suggest a duplication of anterior otic regions. However, pax5and fgf8, both of which mark anterior otic epithelium at 24 hpf,maintain their normal expression pattern in con and smu,suggesting that this duplication is incomplete(Fig. 4J-O). Confirming the absence of posterior identity, follistatin expression, which normally marks a localised posterior epithelial region from 24 hpf, is absent or severely reduced in the ears of both con and smu(Fig. 4P-R). Taking these data together, we conclude that a partial mirror image duplication of anterior otic regions occurs at the expense of posterior identity in con andsmu homozygotes.

Expression patterns of dlx3b, a dorsal otic marker, andeya1, a ventral marker, are normal in con and smuhomozygotes (Fig. 4S-X). In addition, several of the genes discussed above with respect to AP patterning are expressed asymmetrically about the DV axis: otx1, wnt4 andnkx5.1 are all ventral markers(Fig. 4A-I). In all cases, it is only the AP patterning that is altered; the DV aspect of these expression patterns remains unaffected. In addition, a ventral neurogenic region is specified in con and smu ears (see below). We therefore conclude that the DV otic axis is patterned correctly in the absence of Hh signalling. We are unable to tell, however, whether the dorsal part of the ear is also duplicated about the AP axis in con and smu, as semicircular canal projections appear symmetrical in the wild-type ear, and no AP-restricted dorsal markers are currently available. It is possible,therefore, that Hh is involved only in patterning ventral otic structures, and that other signals pattern the AP axis of dorsal regions.

The expression of pax2a, which marks the medial side of the otic vesicle, is also unchanged in con and smu homozygotes(Fig. 4Y-Aa). Likewise, the medial position of the pax5 expression domain is normal(Fig. 4J-L). These data suggest that the ML axis of smu and con ears is also patterned correctly. We do observe a change in the medial expression of otx1(Fig. 4A-C). However, the medial expression of otx1 normally marks the posterior macula(Fig. 4A); because this structure is missing in smu and con, we conclude that this change in otx1 expression is a result of the anterior duplication,rather than a separate ML patterning defect. Note also that ectopic cristae may develop in smu and con ears, while cristae are reduced or absent in the ears of embryos in which Hh signalling is increased (see below). These data suggest that Hh may repress the development of cristae,which are lateral structures, but in the mutants, ectopic cristae may be explained by the duplication of anterior regions.

Neuroblasts that form the statoacoustic ganglion (SAG) delaminate from an anteroventral region of the otic epithelium and migrate anteriorly to form the ganglion, which is positioned anteroventral to the otic vesicle(Haddon and Lewis, 1996). As anterior regions of the otic vesicle are partially duplicated in the absence of Hh signalling, it was of interest to know whether signals from the vesicle might specify the direction of neuroblast migration. If so, we would predict that in con and smu mutants, neuroblasts would migrate both anteriorly and posteriorly, forming a second ganglion underneath posterior regions of the ear. The presumptive SAG is thought to be marked by the expression of both nkx5.1 and sna2 expression at 24 hpf(Thisse et al., 1995;Adamska et al., 2000). Incon and smu homozygotes, the expression of nkx5.1and sna2 remains detectable in the SAG in an anterior domain of variable size, but there is no evidence of a posterior duplicated region of expression of either of these genes (Fig. 4G-I and data not shown). We conclude that the specification of neuroblasts in ventral regions does occur in con and smu,but that the direction of neuroblast migration is controlled independently of AP otic vesicle patterning.

At 48 hpf, both bmp4 and msxc are expressed in three discrete ventral domains, representing the developing cristae(Ekker et al., 1997;Mowbray et al., 2001). A fourth expression domain of both of these genes is seen in 31% (5/16)con and 50% (5/10) smu ears, suggesting the presence of an ectopic crista (Fig. 4Ab-Ad and data not shown). Hair cells differentiate in these ectopic cristae(Fig. 3E). This phenotype is consistent with an anterior otic duplication in con and smu,as both the anterior and lateral cristae are located in the anterior half of the normal ear. An anterior duplication would, therefore, also lead to the presence of two cristae at the posterior of the ear. Owing to the lack of markers specific to individual cristae, however, we were unable to assign an identity to the cristae in the ears of con and smu.

To confirm that the ear phenotypes observed in con andsmu are indeed caused by decreased Hh signalling, we overexpressedptc1 RNA in wild-type embryos. This mimics a loss of function Hh pathway mutant, as an excess of Ptc1 will exert a repressive effect on Smo(Goodrich et al., 1999). We injected 5 nl of ptc1 RNA into one- or two-cell embryos at concentrations ranging from 0.05 μg/μl to 1 μg/μl. Concentrations below 0.5 μg/μl had no effect on the ear. However, at 0.5 μg/μl we phenocopied the con ear defects in approximately 21% of the ears examined. The remaining ears either had no defect or were too necrotic to classify (Fig. 5;Fig. 6E;Table 2). At 1 μg/μl,many embryos died due to nonspecific toxic effects; however, 1.5% ears now showed a slightly more severe phenotype resembling smu ears(Fig. 5;Fig. 6I;Table 2). This suggests that the smu ear phenotype is caused by a more severe reduction in Hh signalling than the con phenotype, and corroborates other studies that indicate that Hh signalling is more strongly attenuated in smuthan con (Lewis et al.,1999b; Barresi et al.,2000; Chen et al.,2001; Varga et al.,2001).

Phalloidin staining of the ears from embryos injected with 0.5 μg/μlptc1 RNA confirmed the similarity of the phenotype to that seen incon (Fig. 6F): these ears lack a posterior macula, have a second ventral patch of hair cells at the posterior, and in a number of cases possess four cristae. Expression ofmsxc confirmed the presence of four cristae in 3/20 con-like ears examined (Fig. 6G). In addition, nkx5.1 expression is expanded along the entire ventral aspect of the otic vesicle, confirming that the phenotype is identical to that seen in the loss of function Hh pathway mutants(Fig. 6H).

To confirm that the difference between the con and smuear phenotypes is due to differences in the level of residual Hh activity, we injected 0.5 μg/μl ptc1 RNA into embryos from a con/+× con/+ mating. This should reduce Hh signalling further in thecon homozygotes, but circumvents the use of toxic levels ofptc1 mRNA. In 14.8% of ears from injected embryos, we observe asmu-like otic phenotype; these embryos presumably represent the homozygous con mutants. A further 16.1% of ears show acon-like phenotype; these embryos may include homozygous,heterozygous or wild-type siblings (Fig. 5; Table 2). The level of death due to nonspecific toxic effects of RNA injection is low (5.2%)and comparable with that seen in the wild-type injection experiments(Table 2). Phalloidin staining confirmed the similarity of the phenotype to that seen in smu ears:the posterior macula was absent, a single ventral macula was observed and four cristae were seen in some ears (Fig. 6J). No in situ markers were used, as we have found none that distinguishes between the con and smu ear phenotypes. Theptc1 injection data therefore confirm that the difference between thecon and smu ear phenotypes is due to a difference in Hh activity levels in the two mutants.

At least three Hedgehog genes are expressed in the vicinity of the ear(Fig. 1) but a mutant is available in only one of these, syu, which removes the function of Shh. There are as yet no reported zebrafish twhh or ehhmutants. To investigate which of these three zebrafish Hedgehog genes have a role in AP otic patterning, we therefore made use of the midline mutantsno tail (ntl) and cyclops (cyc) in addition to syu. ntl mutants carry a mutation in the zebrafish homologue of the Brachyury gene and lack a differentiated notochord(Schulte-Merker et al., 1994).ntl embryos therefore never express ehh because ehhis expressed only in the notochord (Currie and Ingham, 1996). cyc, a nodal-related mutant,lacks the floorplate, and hence twhh expression, as twhh is only found in the floorplate (Rebagliati et al., 1998; Sampath et al.,1998).

In all three single mutants (syu, cyc or ntl) the ears are largely phenotypically normal, and show none of the defects found incon or smu embryos (data not shown). Although cychomozygotes have slightly abnormally shaped ears, phalloidin staining reveals that the sensory patches are fully formed and present in the correct relative positions. syu mutant embryos, including the developing ear, are developmentally retarded. Otherwise, the ear appears normal, although the posterior macula may occasionally be positioned slightly too far towards the anterior. ntl ears appear normal in all respects. In all three single mutants, nkx5.1 (an anterior marker), follistatin (a posterior marker) and bmp4 (a crista marker) are expressed normally(data not shown). The loss of function of each individual Hedgehog protein is therefore not sufficient to cause gross AP patterning defects in the developing ear.

To examine the ears of fish lacking function of two of the three Hedgehog proteins, we used double mutants and morpholino knockdowns(Nasevicius and Ekker, 2000;Odenthal et al., 2000;Lewis and Eisen, 2001) (Tables3 and4). To remove functionalehh and shh, we crossed fish heterozygous for bothntl and syu. The ears of 137 live embryos from four clutches appeared morphologically normal, although some were developmentally retarded. As ntl;syu double homozygous embryos are difficult to distinguish from ntl homozygotes, all embryos displaying a ntl phenotype from three clutches (lacking a tail; n=19) were examined by phalloidin staining. In every case the sensory patches were well formed and positioned correctly, although in five fish, the presumed ntl;syudouble mutants, the development of the sensory patches was retarded. These data indicate that removal of functional Ehh in addition to Shh is not sufficient to give a con or smu-like ear phenotype(Table 3 and data not shown).

By contrast, cyc;syu double homozygotes (identifiable because they have both U-shaped somites and cyclopia) show an anteriorised ear phenotype similar to, but perhaps not quite as strong as, the smu phenotype. The posterior otolith is small and too lateral, resembling the anterior otolith (Fig. 7A,B); the posterior macula is missing, and either a single macula covering the ventral surface of the ear (as in smu) or two separate ventral maculae (as incon) are seen (Fig. 7D,E). Additionally, anterior nkx5.1 expression is expanded posteriorly and posterior follistatin expression is absent(Fig. 7G,H,J,K). bmp4expression indicated that four cristae instead of the usual three were present in 7/18 (38%) ears examined (data not shown). Removal of both twhhand shh function from the embryo is thus sufficient to cause anteriorisation of the otic vesicle, although removal of the function of either alone is not.

We confirmed this result by the use of antisense morpholinos to knock down Shh and Twhh function. Carboxyfluoresceintagged antisense morpholinos were designed against shh and twhh, and injected into wild-type or syu mutant embryos. Injection of 0.25 mM twhh morpholino to wild-type (AB) embryos caused circulation defects and very slight somite defects. Although necrosis was observed in the head in some cases, the ears of all injected embryos developed normally(Table 4). To deplete bothtwhh and shh function, we injected 0.25 mM twhh MO into a clutch of embryos from a mating between syu/+ parents. We observed a phenotype similar to that seen in twhh MO-injected wild-type embryos in 82% of cases. In the remaining 18% (the presumedsyu homozygotes), the eyes were cyclopic and the ears resembledcon and smu anteriorised ears. The otoliths both resembled the anterior otolith, a ventral sensory patch could be seen at the posterior of the ear and the posterior macula was absent(Fig. 7C,F). In addition,posterior otic follistatin expression was lost and anterior oticnkx5.1 expression extended posteriorly, although not to the extent seen in cyc;syu double mutants(Fig. 7I,L).

As a control for nonspecific effects of twhh MO injection, we injected 0.5 mM shh MO into a clutch of embryos from a syu/+× syu/+ mating. In no case was an anteriorised ear phenotype observed, although all embryos now resembled syu homozygotes in possessing circulation defects and U-shaped somites, confirming that ourshh MO knocks down Shh function. These data indicate that the anteriorised ear phenotype seen in twhh MO-injected syuhomozygotes is not due to nonspecific effects of morpholino injection (seeTable 4). It therefore appears that both Twhh and Shh function to specify the posterior part of the ear, but that either can compensate for the absence of the other. We have not tested the role of Ehh with morpholinos, but suggest that it is unlikely to play a major role, given that the ears of syu;ntl double homozygotes, which lack both functional shh and ehh, are patterned normally.

As a loss of Hedgehog signalling leads to a loss of posterior character and a concomitant gain of anterior character at the posterior of the ear, we predicted that an increase in Hedgehog signalling should lead to a gain of posterior character by the anterior part of the ear. We tested this hypothesis by overexpression of shh RNA or a dominant negative (dn) PKARNA in wild-type embryos. We injected shh RNA into wild-type zebrafish embryos at concentrations of 25 ng/μl to 100 ng/μl. Approximately 65% of the ears of these embryos had a very small otic vesicle containing either a single central otolith or a fused dumb-bell shaped central otolith (Table 5;Fig. 8A-C). Semicircular canal projections were very reduced or absent in these ears(Fig. 8A-C). A few more weakly affected embryos displayed a variable semicircular canal phenotype where one or more of the canal projections was reduced or absent (data not shown). Phalloidin staining in those ears with a single or fused central otolith revealed the absence of an anterior macula and the presence of a single medial macula. This had a characteristic `bow-tie' or `butterfly' shape, and is presumed, based on its shape and position, to represent a twinned, double posterior macula (Fig. 8D-F). Increasing the concentration of RNA injected affects this phenotype. At 25 ng/μl, out of 15 posteriorised ears, ten showed the `bow-tie' shape shown in Fig. 8F and five showed the`butterfly' shape shown in Fig. 8E. At 100 ng/μl, out of six posteriorised ears, five showed the `butterfly' phenotype, while one showed the `bow-tie'.

Phalloidin staining and in situ hybridisation with msxc also indicated that the cristae were variably reduced in the posteriorised ears. In most cases, all cristae were absent, but in a number of cases one or more were observed (data not shown). As before, we were unable to assign an identity to the cristae present. Anterior markers (nkx5.1, pax5 andfgf8) were absent or severely reduced in 60% or more of cases(Fig. 8J-O), which is consistent with the number of ears with twinned posterior maculae inshh-injected embryos. follistatin, a posterior marker, was duplicated at the anterior in 5/22 ears examined or expanded anteriorly along the medial wall of the otic vesicle in 5/22(Fig. 8G-I). These data suggest that the otic phenotype of shh-injected fish is indeed a posteriorisation of anterior otic regions. Injections of either twhhRNA or ehh RNA, at concentrations up to 500 ng/μl, had no effect on the otic vesicle.

A dominant-negative form of PKA (dnPKA) was also used to repress Hedgehog signalling activity. PKA acts downstream of Smo to repress Hh signalling and therefore dnPKA causes constitutive activation of the pathway(Concordet et al., 1996). dnPKA RNA (400 ng/μl) was injected into wild-type embryos,resulting in a phenotype identical to that caused by shh injections in 14% embryos (Fig. 8C,F). Lower concentrations had no effect on the otic vesicle. These data confirm that ectopic Hh activity can lead to a duplication of posterior structures at the expense of anterior domains.

In this study, we have examined the role of signalling molecules of the Hedgehog (Hh) family in patterning the axes of the zebrafish otic vesicle. The closest sources of Hh protein to the developing ear are ventral midline structures (notochord and floorplate). As these are situated ventral and medial to the otic vesicles, we predicted that Hh would have a mediolateral(ML) or dorsoventral (DV) patterning role in the developing ear. However, we found that both the ML and DV axes of the ear appeared to be patterned correctly in the absence of Hh signalling. Instead, our data indicate that both Shh and Twhh have a role in specifying posterior otic structures. Severely reduced or absent Hh signalling leads to a loss of posterior otic structures and a concomitant partial mirror image duplication of anterior regions. Increased Hh signalling throughout the embryo leads to the reverse phenotype: a loss of anterior otic structures and a mirror image duplication of posterior regions.

Based on the expression of ptc1 and ptc2 in the ears of wild-type embryos, con and smu mutants, and embryos overexpressing shh RNA, we argue that the effect of Hh signalling on the ear is likely to be direct. However, we cannot rule out the possibility that Hh acts in a permissive manner on surrounding tissues to potentiate or block the production or action of a localised factor, which then acts secondarily to pattern the ear. We find no evidence, however, that AP pattern in the hindbrain is altered in embryos with attenuated Hh signalling: AP expression of krx20 (egr2 — Zebrafish Information Network) hoxb4a and val/mafb, for example, is normal in the rhombomeres of con mutants (data not shown).

The inner ear phenotype of mice homozygous for a mutation in theshh gene has recently been reported(Liu et al., 2002;Riccomagno et al., 2002). The defects do not appear to mimic those we see in smu or con,but they primarily affect structures that have no direct counterpart in the fish ear. In particular, the cochlear duct and cochleovestibular ganglia, all ventral otic derivatives, are rudimentary or absent, and Pax2expression is diminished in the vesicle; by contrast, in smu andcon mutants, specification of otic neuroblasts does occur, and oticpax2a expression is retained (Fig. 4; data not shown). Note that in the mouse, inactivation ofShh gives rise to much more severe overall head defects than in any of the zebrafish Hh pathway mutants, characterised by holoprosencephaly and cyclopia (Chiang et al.,1996). In vitro and in vivo evidence also suggests a later role for Shh in chondrogenesis of the murine otic capsule(Liu et al., 2002;Riccomagno et al., 2002).

Although our experiments do not address the timing of Hh action rigorously,we suggest that the zebrafish ear is likely to respond to Hh signalling between 19 and 24 hpf. Exogenous RNA injected at the one- to two-cell stage and protein translated from it are likely to degrade before 24 hpf(Hammerschmidt et al., 1999)(K. L. H., unpublished), but injection of ptc1 RNA is sufficient to repress endogenous Hh signalling and phenocopy the defects seen insmu and con. Hh may not, however, exert its posterior inductive abilities as early as 16-17.5 hpf, when the earliest AP molecular asymmetries appear (nkx5.1 and pax5 expression)(Pfeffer et al., 1998;Adamska et al., 2000). This is because Hh signalling activity, as indicated by the expression of its target gene ptc1, only becomes concentrated in posterior otic epithelium between 19 hpf and 22 hpf. Before this, ptc1 expression (and hence Hh activity) is detectable in a ventromedial domain along the entire AP length of the otic vesicle.

If Hh does act prior to 19 hpf, it is possible that ventromedial cells specified by Hh later move to occupy more posterior positions, thus transforming a DV or ML signal into an AP pattern. Although a fate map of the zebrafish otic vesicle exists (Haddon,1997), it is not sufficiently detailed to tell whether such movements do generally occur. Grafting experiments in salamander embryos are suggestive of an anterior to posterior movement of ventromedial otic cells(Kaan, 1926), but species-specific differences are likely: in the chick otic cup, ventral cells tend to move in an anterodorsal direction(Brigande et al., 2000a). In addition, amphibian ears may show a high degree of cell mixing(Kil and Collazo, 2001), but this appears to be more limited in the zebrafish and chick otic vesicle(Haddon, 1997;Brigande et al., 2000a).

Alternatively, Hh may act after 19 hpf to reinforce and maintain AP polarity in the ear rather than establish it, in a similar fashion to the role of Hh in the vertebrate limb bud. Here, Shh expression in the zone of polarising activity (ZPA) is established by a prepattern involving mutual antagonism between the transcription factors GLI3 and dHAND(te Welscher et al., 2002). In the zebrafish fin bud, for example, a transient AP polarity is established(but not maintained) in the fin buds of syu (shh) mutants(Neumann et al., 1999), but no AP patterning is ever apparent in the fin buds of hands off(hand2) mutants (Yelon et al.,2000).

Only a low level of Hh signalling is required for correct patterning of the zebrafish otic vesicle; defects are evident only in those mutants with the strongest phenotypes in other tissues (smu and con), or when the activity of both Shh and Twhh are removed. The ears are also patterned correctly in ntl, flh and oep embryos, in which the development of subsets of tissues expressing Hh is compromised (data not shown; see below). Despite this, we observe phenotypes that appear to differ according to the level of Hh activity. The anterior duplication is incomplete in both smu and con ears, but the smu phenotype appears to be stronger; we see a single fused ventral macula rather than the two separate ventral maculae found in con. This correlates with the fact that smu homozygotes show a more complete loss of Hh signalling than con homozygotes (Lewis et al., 1999b; Barresi et al.,2000; Varga et al.,2001). Moreover, the different otic defects can be phenocopied by different levels of Hh inhibition via ptc1 injection. We also observe a dose-dependent effect of Shh or dnPKA injection; low doses more frequently result in a `bow-tie'-shaped posterior macula, whereas higher doses (100 ng)more frequently result in the `butterfly' phenotype.

The ears described in this work are enantiomorphic twins: they consist of two mirror image halves. Mirror image duplications of various tissues have been observed in several other contexts, and are frequently associated with alterations of Hh signalling. Examples include the development of adult abdominal segments and the wing in Drosophila(Basler and Struhl, 1994;Capdevila and Guerrero, 1994;Kojima et al., 1994;Kopp et al., 1997;Struhl et al., 1997a;Struhl et al., 1997b;Lawrence et al., 2002), and development of the limb bud in vertebrates(Riddle et al., 1993;Yang et al., 1997). Mutants for hh, ci and ptc in Drosophila are themselves members of the segment polarity class, in which a proportion of each embryonic abdominal segment is deleted and the remainder present as a mirror image duplication (Nüsslein-Volhard and Wieschaus, 1980).

Mirror image duplications of the inner ear have also been documented previously. Harrison transplanted ear rudiments in salamander(Amblystoma) embryos such that their AP axis was reversed with respect to that of the host (Harrison,1936; Harrison,1945). Ear rudiments transplanted early (at the neural plate stage) developed with an AP axis corresponding to that of the host, while ear rudiments transplanted later (after closure of the neural folds) developed according to the donor AP axis. However, in transplantations performed at intermediate stages (during neural tube closure; stage 19-21), up to 27% of transplanted ear rudiments developed as mirror image twins. These ears consisted of two posterior halves, two anterior halves or incomplete duplications, and show a remarkable similarity to the zebrafish phenotypes we describe. In the double anterior ears, four cristae and two utricular maculae and otoliths were observed; in the double posterior ears, utricular maculae were missing, and cristae were reduced(Harrison, 1936;Harrison, 1945). Similar duplications have been observed in Xenopus embryos after ablation of either the anterior or posterior half of the otic placode. Regeneration after anterior ablations results in mirror image double posterior ears, while posterior ablations can cause the reverse phenotype(Waldman et al., 2001).

In the above examples involving Hh, a localised source of Hh provides the necessary information to generate AP polarity. In most cases this is either a point source (as in the vertebrate limb bud) or a linear boundary (as in the fly wing disc). However, in the fish ear, the strongest and closest source of Hh appears to be constant along the AP axis. It is possible that endodermal expression of Shh influences the ear, but we think this unlikely, given its late onset. Unless Hh-receiving cells move to the posterior of the ear (as discussed above), a mechanism must exist to restrict the effects of Hh activity to the posterior of the ear.

One possibility is that posterior otic regions receive more Hh than anterior domains because of positioning of the otic vesicle relative to the midline and the notochord. At 22 hpf, when active Hh signalling is first concentrated in posterior otic epithelium, anterior otic regions are a little further from the midline than posterior regions (seeFig. 1J). Although slight, this difference may play some part in the concentration of higher level Hh activity in posterior otic regions. In addition, the anterior limit of the notochord coincides roughly with the anterior limit of the otic vesicle(Fig. 9), and thus notochord-derived Hh may be reduced at the anterior of the vesicle. We have found, however, that either the notochord or the floorplate alone suffices to pattern the ear correctly. The ears of ntl mutants, which lack a notochord (Schulte-Merker et al.,1994), flh mutants, which lack chordamesoderm(Halpern et al., 1995), andcyc and oep mutants, both of which lack a medial floorplate(Schier et al., 1997;Rebagliati et al., 1998), all show correct AP patterning (data not shown). Assuming that post-transcriptional and post-translational processing, release and diffusion of Hh are equivalent at different AP levels in the otic region, it appears that a constant source of Hh from midline tissues, encoded by twhh orshh, is able to pattern posterior regions of the ear.

If the source of Hh is constant, it is likely that other factors,originating either from within or outside the otic vesicle, synergise with Hh in posterior regions or antagonise it at the anterior(Fig. 9). Members of the bone morphogenetic protein (BMP), BMP antagonist, Wnt and fibroblast growth factor(FGF) families are good candidates for such factors, as they are known to potentiate or antagonise Hh in many developmental contexts (seeMarcelle et al., 1997;Meyers and Martin, 1999;Patten and Placzek, 2002)(reviewed by Cohn and Tickle,1996; McMahon et al.,2003). Members of all four families are expressed in the developing zebrafish ear (Blader et al.,1996; Reifers et al.,1998; Mowbray et al.,2001).

Among the best candidates for antagonists of Hh activity in anterior otic epithelium are Fgf3 and Fgf8. Both have an early role in otic placode induction: disruption of the function of either Fgf results in a small otic placode, and if both are disrupted the otic placode is severely reduced or fails to form entirely (Phillips et al.,2001; Raible and Brand,2001; Maroon et al.,2002). Later, at the stages when Hh is likely to be active,fgf3 continues to be expressed in r4 (now positioned adjacent to the anterior part of the otic vesicle), while fgf8 is expressed in the anterior otic epithelium (Fig. 4M) (Reifers et al.,1998; Phillips et al.,2001; Maroon et al.,2002). Both factors appear to have anterior otic inducing ability. In valentino (mafb) embryos, hindbrain fgf3expression is expanded posteriorly, resulting in the expansion of anterior-specific gene expression in the ear. Conversely, in embryos where Fgf3 function is depleted by morpholino injection, expression of some anterior-specific otic genes is reduced or missing(Kwak et al., 2002). In theacerebellar (fgf8) mutant, nkx5.1 expression is reduced, suggesting that some loss of anterior character has occurred(Adamska et al., 2000). Thus Fgf3 from the hindbrain and Fgf8 in the otic epithelium are excellent candidates for antagonists of Hh activity in the anterior otic vesicle(Fig. 9).

In all likelihood, more than one mechanism operates to concentrate Hh activity in posterior regions of the otic vesicle. Whichever mechanism is responsible, it is clear that Hh is essential for the specification of posterior otic identity in the zebrafish. We still do not understand, however,how this is effected. The mirror image duplications observed appear to reveal an underlying prepattern, where the otic vesicle is an equipotential system in which the two ends (or the centre) are specified, but an A or P identity has not been assigned to either. This is similar to the `global mirror-symmetric system' proposed for the Drosophila adult abdominal segment by Kopp and Duncan (Kopp and Duncan,1997). We note that the early expression of genes marking the positions of the presumptive maculae at the two ends of the otic vesicle, such as the Delta genes, is initially mirror symmetric(Haddon et al., 1998). A symmetric prepattern would then be acted on by Hh, Fgf and other signals, from surrounding tissues and within the ear, to establish and maintain AP polarity.

We thank J. Lewis for inspiration, A. Jamieson for help with a pilot study(funded by a Nuffield Foundation vacation studentship), P. Ingham, S. Roy, C. Wolff and J. Begbie for helpful discussion, S. Roy and C. Wolff for help with establishment of stocks, and F. Wilson and L. Gleadall for care of the fish facility. We are greatful to R. Karlstrom and B. Riley for communicating data before publication, and many members of the zebrafish community for supplying cDNAs. This work was supported by a Wellcome Trust project grant (057101). Support for confocal microscopy was provided by Yorkshire Cancer Research.