Semicircular canal morphogenesis in the zebrafish inner ear requires the function of gpr126 (lauscher), an adhesion class G protein-coupled receptor gene

The three semicircular canals of the vertebrate inner ear detect angular acceleration (rotational movements) in three-dimensional space. Each canal consists of a curved fluid-filled duct, ending in a swelling or ampulla that houses a sensory organ: the crista. Formation of the semicircular canal ducts is a fundamental developmental process, involving the generation of form via movement and fusion of sheets of epithelium. In amniotes, flattened pouches form as outpocketings from the otic vesicle, and their apical sides touch and fuse to generate the semicircular canal ducts (Abraira et al., 2008; Chang et al., 2008) (reviewed by Bok et al., 2007). In a topologically equivalent process in anamniote vertebrates (zebrafish and Xenopus), the outpocketings are less pronounced, and finger-like projections of epithelium grow into the centre of the otic vesicle, where they fuse to form a pillar that becomes the hub of the canal (Waterman and Bell, 1984; Haddon and Lewis, 1991; Haddon and Lewis, 1996). Expression of versican (vcan) genes, which code for chondroitin sulphate proteoglycan core proteins, marks this process beautifully as it happens (Fig. 1).

Studies in the zebrafish, chick and mouse have begun to address the mechanisms underlying semicircular canal morphogenesis. It has been proposed that the cristae induce the ducts; signalling molecules, including Fgfs and Bmps, are likely to be involved (Cantos et al., 2000; Chang et al., 2004; Chang et al., 2008; Shawi and Serluca, 2008). Specification of canal tissue also requires the activities of various transcription factor genes, including dlx5, hmx2/3, lmo4, otx1 and sox10 (Hadrys et al., 1998; Acampora et al., 1999; Morsli et al., 1999; Fritzsch et al., 2001; Wang et al., 2001; Merlo et al., 2002; Wang et al., 2004; Lin et al., 2005; Hammond and Whitfield, 2006; Dutton et al., 2009; Deng et al., 2010).

Once canal projection or pouch tissue is specified, it must undergo morphological change to form the canal ducts. Outgrowth of projections in the Xenopus ear is driven by the production of the glycosaminoglycan hyaluronan (HA) (Haddon and Lewis, 1991). HA production in the zebrafish ear is regulated by dfna5 (orthologue of the human autosomal dominant deafness gene DFNA5) and ugdh (UDP-glucose dehydrogenase), both expressed in the projections. In dfna5 morphants and jekyll (ugdh) mutants, HA fails to be produced, and projection outgrowth is blocked (Neuhauss et al., 1996; Busch-Nentwich et al., 2004). Variable projection outgrowth defects in the zebrafish ear have also been reported after knockdown or mutation of ncs1a (Blasiole et al., 2005), atrophin2 (rerea - Zebrafish Information Network), fgf8a (Asai et al., 2006), cdh2 (Babb-Clendenon et al., 2006), atp1a1a.2 (Blasiole et al., 2006), atp2b1a (Cruz et al., 2009) and grhl2 (Han et al., 2011), and in Hedgehog pathway mutants (Hammond et al., 2010).

When the sides of the canal pouches (in amniotes) or tips of the projections (in zebrafish and Xenopus) touch each other, cells change behaviour to form a fusion plate. Establishment of the zones of fusion and non-fusion in the mouse ear involves an antagonistic interaction between Lrig3 and Netrin1 (Salminen et al., 2000; Abraira et al., 2008), and regulation by Wnt signalling (Noda et al., 2012; Rakowiecki and Epstein, 2013). Resolution at the fusion plate in mouse, chick and Xenopus involves breakdown of the basal lamina (Haddon and Lewis, 1991; Salminen et al., 2000; Abraira et al., 2008), epithelial-to-mesenchymal transition (Salminen et al., 2000; Kobayashi et al., 2008) and cell death (Haddon and Lewis, 1991; Fekete et al., 1997; Cecconi et al., 2004). In the zebrafish, however, cell death does not appear to play a major role (Waterman and Bell, 1984; Fekete et al., 1997).

The mechanisms underlying these events - in particular how the growing projections recognize each other, adhere and fuse - are not fully understood. Here, we show that several extracellular matrix (ECM) components are expressed strongly in the epithelial projections in the zebrafish ear during outgrowth, and are then rapidly downregulated after fusion. We have characterized semicircular canal defects in a zebrafish mutant, lauscher (lau) (German: ‘eavesdropper’) (Whitfield et al., 1996). The lau mutant ear displays severe abnormalities in canal development: projections overgrow and fail to fuse correctly to form pillars. There are striking concomitant defects in the expression of ECM components in the canal projections, and the ear becomes swollen.

We show that lau mutations disrupt gpr126, an adhesion class G protein-coupled receptor (GPCR) gene. Adhesion GPCRs are one of five classes of GPCR, with dual roles in cell adhesion and signalling (Yona et al., 2008). These are exactly the functions that might be predicted to trigger changes in cell behaviour at the fusion plate. We show that the lau mutant ear phenotype can be ameliorated by treatment with cAMP agonists, indicating that Gpr126 is likely to signal through G protein-mediated activation of adenylyl cyclase and production of cAMP. The expression pattern of gpr126 is very similar to that of sox10, and is partially dependent on Sox10 function. These data identify a new signalling mechanism involved in semicircular canal formation in the zebrafish, implicating Gpr126 in the control of projection outgrowth, contact recognition and fusion in the developing ear.

Standard zebrafish husbandry methods were employed (Westerfield, 2000). Wild-type strains used were AB (ZDB-GENO-960809-7), WIK (ZDB-GENO-010531-2) and EKW (ZDB-GENO-990520-2); mutant alleles were lau^tk256a (ZDB-GENE-070117-2161), lau^tb233c (formerly bge^tb233c) (Whitfield et al., 1996), lau^fr24 [isolated in the Hammerschmidt lab (Carney et al., 2010)] and lau^vu39 (isolated in the Topczewski lab). Embryos of the nac (mitfa^w2/w2) strain (ZDB-GENO-990423-18), which lack melanophores (Lister et al., 1999), were used as wild types for some experiments. Embryos were raised in E3 (Westerfield, 2000) at 28.5°C. Work in Sheffield conformed to UK Home Office regulations.

lau^tb233c homozygous adults (-/-) were crossed to the polymorphic WIK strain. Offspring of the heterozygous progeny (tb233c/WIK) were used for bulked segregant analysis. Six hundred and ninety-six mutant embryos from tb233c/WIK incrosses were used for fine mapping with SSLP and SNP markers. Candidate gene cDNA was generated using a SuperScript III First Strand Kit (Invitrogen). Sequencing was carried out using an ABI3730 capillary sequencer. Sequence data were analysed using FinchTV (www.geospiza.com); primers were designed using FastPCR (www.biocenter.helsinki.fi). Sequence numbering is given according to the NCBI reference sequence NM_001163291.1 (Monk et al., 2009).

Genomic DNA from mutant and wild-type embryos was amplified by PCR (see supplementary material Table S1 for primer sequences). The resulting tb233c, tk256a and fr24 products were digested with the restriction enzymes SfaNI, BsmFI and BfaI (New England Biolabs), respectively, for 5 hours at 37°C and separated on a 2% TBE agarose gel.

A morpholino oligonucleotide (Gene Tools) was designed to target the splice donor site of exon 19 of gpr126; a 5-base mismatch morpholino was used as a control (see supplementary material Table S1 for sequences). Morpholinos were resuspended in water, and injected into one-cell stage embryos with 0.05% Phenol Red as described previously (Hammond and Whitfield, 2006).

Wild-type embryos at 48-50 hours post-fertilization (hpf) were anaesthetized with tricaine (Sigma) and immobilized in 3% methylcellulose. Streptomyces hyaluronidase (Sigma) was dissolved at 0.5-1 mg/ml in PBS/0.5% Phenol Red; 1 nl was injected into the lateral projection of one ear, as described for Xenopus (Haddon and Lewis, 1991). Control injections were performed by injecting PBS/0.5% Phenol Red without hyaluronidase.

In situ hybridization was performed as described previously (Nüsslein-Volhard and Dahm, 2002). Probes used are listed in supplementary material Table S1. Embryos were bleached post-in situ where necessary in 0.5×SSC, 10% H₂O₂ and 5% formamide.

Embryos were fixed and dehydrated as for in situ hybridization. Following rehydration and washes in 1×PBS/0.1% Tween-20 (PBTw), samples were permeabilized with 20 μg/ml proteinase K (1 hour, room temperature) followed by PBTw washes. After blocking in 10% bovine serum, embryos were incubated overnight at 4°C in anti-collagen II antibody (1:500; II-II6B3; DSHB). After further washes and overnight incubation in anti-mouse IgG-HRP (Sigma), staining was visualized using the Vectastain DAB kit (Vector Labs).

After fixation, embryos were rinsed in PBS and permeabilized in PBS/2% Triton-X100 (for 4 hours at room temperature). After PBS rinses, embryos were incubated overnight in 2.5 μg/ml FITC-phalloidin (Sigma) followed by PBS washes and mounting in Vectashield medium.

Wild-type and homozygous tb233c mutant embryos were treated between 60 hpf (high pec) and 90 hpf with forskolin (Sigma; 2×6 hour pulses, 60-66 hpf and 84-90 hpf) at 25-100 μM or IBMX (3-isobutyl-1-methylxanthine) (Sigma; continuous treatment, 60-90 hpf) at 10-100 μM in E3. Control embryos were treated with equivalent amounts of DMSO (Sigma). To facilitate treatment times, embryos were incubated at 24°C from 8 hpf prior to drug treatment and at 28.5°C during treatments.

Images were taken using an Olympus BX51 microscope, C3030ZOOM camera and CELLB software, and assembled with Adobe Photoshop. All panels are lateral views with anterior towards the left and dorsal towards the top, unless otherwise indicated.

The zebrafish lauscher (lau) mutant is characterized by its swollen ears at 5 days post-fertilization (dpf) (Fig. 2). We have analysed four alleles: two of these, lau^tk256a and bge^tb233c, were originally identified as separate loci (Whitfield et al., 1996). However, we have since shown that bge^tb233c is non-complementing and therefore allelic to lau. A third allele, fr24, isolated in a separate screen (Carney et al., 2010), was also allelic to lau by complementation testing (data not shown). A fourth allele, vu39, was also identified separately on the basis of its swollen ears.

At early stages of ear development in the lau mutant, otic patterning was normal (supplementary material Fig. S1), and sensory patches and otoliths developed correctly (Whitfield et al., 1996) (and data not shown). Initiation of semicircular canal formation was also normal: at 45 hpf, the anterior, posterior and lateral projections began to protrude into the otic vesicle. However, the projections subsequently failed to touch and fuse correctly (Fig. 2). The acellular core of the lateral projection often became enlarged, and did not form clear anterior, posterior and ventral bulges. In some individuals, the anterior and posterior projections continued to elongate, whereas in others, they ceased. The ventral projection also initially appeared to grow normally, but later also failed to fuse with the ventral bulge of the lateral projection. In some cases, the projections grew past one another (Fig. 2I); in the tb233c allele, projections occasionally touched and fused but did not resolve to form a normal pillar (Fig. 2J). By the end of 3 dpf, the ears became swollen, most strongly in the fr24 allele (Fig. 2E). Swelling in the tb233c allele was variable (see also Fig. 8).

Endolymphatic hydrops of the mammalian inner ear is characteristic of a number of inner ear disorders (Everett et al., 2001; Hulander et al., 2003; Megerian et al., 2008). Initially, we hypothesized that the ear swelling in lau mutants might reflect this condition. However, the endolymphatic duct, a structure involved in the homeostatic regulation of endolymph, appeared to develop normally in the lau ear, as assayed by bmp4 and foxi1 expression (Fig. 3A-D). In addition, expression of atp1a1a.4, kcnq1 and nkcc1 (slc12a2) was normal or reduced (Fig. 3E-J). As these gene products have endolymph-generating functions (Lang et al., 2007; Abbas and Whitfield, 2009), a reduction in their levels is unlikely to explain the ear swelling. In addition, measurements of the perimeter and cross-sectional area of the ear suggested that the swelling was due to a shape change rather than indicating a significant increase in endolymph volume (supplementary material Fig. S2).

To test whether the loss of pillar formation might secondarily cause the swelling, we disrupted outgrowth of the projections with hyaluronidase, which is known to collapse the canal projections in the Xenopus ear (Haddon and Lewis, 1991). We injected hyaluronidase into the acellular space inside the lateral projection of one ear in wild-type embryos at 50 hpf, before fusion occurred. As in Xenopus, this blocked further projection outgrowth and fusion to form pillars, and at 5 dpf, 60% injected ears (18/30) were swollen (Fig. 2K-N). This was comparable with the degree of swelling in tb233c (Fig. 2M), but not as severe as in the other alleles. Control ears injected with a similar volume of PBS into the lateral projection did not swell (23/27 normal; four mildly swollen) (Fig. 2L). Taken together, these data suggest that the ear swelling in lau mutants is partly a secondary problem caused by a lack of projection fusion. However, we cannot rule out an additional effect on endolymph production or homeostasis (see Discussion).

To characterize the semicircular canal defects further, we examined gene expression in the developing projections and pillars in the ear (Fig. 4). Most striking were changes in the expression of genes coding for extracellular matrix (ECM) proteins or ECM-modifying enzymes. Expression of the hyaluronan and proteoglycan link protein gene hapln1a persisted abnormally in the disorganized canal projections in lau mutants (Fig. 4A-F). Conversely, expression of hapln3, normally initiated as projections fused in wild-type embryos, was greatly reduced in lau mutants (Fig. 4G-L). The chondroitin sulphate proteoglycan core protein genes versican a (vcana) and versican b (vcanb) were expressed at high levels in the mutant unfused canal tissue, failing to be restricted to the dorsolateral septum as in the wild-type (Fig. 4M-X). There was precocious and ectopic accumulation of Type II Collagen protein in lau mutant canal tissue (Fig. 4Y-DD). Expression of the extracellular matrix synthesis enzyme genes chondroitin synthase 1 (chsy1), hyaluronan synthase 3 (has3) and UDP-glucose dehydrogenase (ugdh) was also upregulated in lau mutant ears (Fig. 4EE-JJ). In addition, we found substantial changes in the expression of other semicircular canal marker genes, including aldh1a3, bmp7b and sox9b (Fig. 4KK-PP).

We mapped lau to LG20 by bulked segregant analysis (Michelmore et al., 1991; Bahary et al., 2004). Six hundred and ninety-six mutant embryos from tb233c/WIK incrosses were used for fine mapping of the mutation to the region between SSLP markers Z4627 and Z536 (Fig. 5A). We narrowed this further to a region between SNP markers in slc25a27 and galnt14. In this interval, a SNP marker was found in schnurri2 for which there was no recombination event. There were six genes in the vicinity (Fig. 5A); of these, only gpr126, an adhesion class G protein-coupled receptor gene, was expressed in the inner ear (data not shown; see Fig. 6). We detected a number of alternatively spliced wild-type gpr126 transcripts (data not shown).

Sequencing of gpr126 cDNA and genomic DNA from the four lau mutant alleles indicated two missense mutations in the fourth transmembrane domain, and two nonsense mutations predicting truncated proteins (Fig. 5B-D). In tb233c mutant embryos, a T→A transversion at position 2888 predicted the replacement of Ile963 (isoleucine) by asparagine. In tk256a, a C→T transition at position 2906 led to a replacement of the highly conserved proline Pro969 by leucine. In fr24, a T→A transversion at position 1388 resulted in a nonsense mutation at Leu463. In vu39, a G→A transition at position 2411 resulted in a nonsense mutation at Trp804 in the GPCR proteolytic site (GPS) motif, part of the recently described GAIN (GPCR autoproteolysis-inducing) domain (Araç et al., 2012). This tryptophan residue, which is highly conserved in GPS motifs from other adhesion GPCRs, was present in all wild-type cDNA variants that we isolated, but is reported as a cysteine in the reference sequence (Monk et al., 2009). In both nonsense alleles, the highly conserved CUB (Complement C1r/C1s, UEGF, BMP1) and PTX (Pentraxin) protein domains remained intact in the predicted truncated protein, but the GPS motif and 7-transmembrane (7TM) domain were lost (Fig. 5C). The sequence data indicated the loss or gain of restriction sites in three of the alleles, which we confirmed by genotyping (Fig. 5E).

To provide further confirmation of our sequence data, we used a morpholino to knock down gpr126 function in wild-type embryos. This was designed to target the 19th exon, in which the first 7TM domain helix resides. Wild-type embryos injected with 5 ng of a five-mismatch control morpholino demonstrated normal semicircular canal formation (n=73). Embryos injected with 2.5 ng of the gpr126 splice morpholino phenocopied both the morphological and gene expression defects in lau mutants, even though disruption of splicing was incomplete (Fig. 5F,G). As in lau, canal projections in the morphant ear failed to fuse or grew past one another. The knockdown of gpr126 function persisted until 5 dpf and was dose dependent. A morpholino dose of 0.25 ng had no visible effect on the ear; at 1 ng, 40% injected embryos displayed lau-like ear defects (n=127), whereas at 2.5 ng (1 nl of 0.25 mM), 73% of injected embryos displayed lau-like ear defects (n=133). A 5 ng dose was toxic (data not shown). Sequence analysis indicated that the morpholino caused either skipping of exon 19 or the use of a cryptic splice site in exon 19 (Fig. 5G). In both cases, the reading frame of the mis-spliced transcript was shifted.

The degree of ear swelling in the lau alleles correlated well with the predicted disruption of protein function. Both tb233c and tk256a (missense mutations) showed a milder and more variable ear swelling than the fr24 allele, which has a truncating mutation (Figs 1, 5, 8). Two further alleles of zebrafish gpr126, isolated in a screen for myelination defects, have been described (Pogoda et al., 2006; Monk et al., 2009). We found that our alleles also displayed a reduction or loss of myelin basic protein (mbp) expression in the posterior lateral line nerve and ganglion at 5 dpf, the degree of which also correlated well with the predicted strength of the alleles (supplementary material Fig. S3). We conclude that tb233c and tk256a are hypomorphic alleles, with tb233c the weaker of the two, whereas fr24 and vu39 are stronger, possibly null, alleles.

A striking aspect of the gpr126 expression pattern was in the developing ear (Fig. 6; supplementary material Fig. S4). A trace of expression was seen in the otic vesicle at 24 hpf (Fig. 6A,I,M), and strong expression appeared in the canal projections as they grew out at 48 hpf (Fig. 6B-D,J,K,N). gpr126 was also expressed in the supporting cells of the anterior macula (Fig. 6D,K). Expression levels were similar in both wild-type and lau mutant embryos (Fig. 6E-H); however, expression in lau persisted at slightly higher levels in the unfused projections at 96 hpf (Fig. 6H). In the canal projections, expression overlapped with that of vcanb at the projection tips, but gpr126 expression was more widespread (Fig. 6O,P). After projection fusion in wild-type embryos, gpr126 expression persisted in the pillars at lower levels, whereas vcanb was no longer detectable in the pillars (Fig. 6O). Otic expression of gpr126 decreased in both wild-type and mutant embryos at later stages (not shown).

In addition to its expression in the ear, gpr126 was expressed dynamically in other tissues during early development (Fig. 7; supplementary material Fig. S4). We detected gpr126 expression in neural crest-derived glia, mesoderm, heart, olfactory epithelium, and head and pectoral fin chondrocytes. This expression pattern was very similar to that of sox10, a SoxE family transcription factor gene expressed in the neural crest and ear (Fig. 7A-D) (Dutton et al., 2001; Dutton et al., 2009). This led us to ask whether Sox10 regulated gpr126 expression. In colourless (cls) (sox10^-/-) mutants at 24 hpf, gpr126 was specifically missing from a group of cells posterior to the otic vesicle (Fig. 7E,F,J,K). Comparison with expression of foxd3, a neural crest marker (Odenthal and Nüsslein-Volhard, 1998; Kelsh et al., 2000), confirmed that these were Schwann cells associated with the posterior lateral line ganglion and nerve (Fig. 7E-I). This fits with the known requirement for Gpr126 in the initiation of myelination by Schwann cells (Monk et al., 2009; Monk et al., 2011; Glenn and Talbot, 2013). As foxd3-expressing glia are reduced but still present at 27 hpf in cls mutants (Kelsh et al., 2000), the complete loss of gpr126 expression in these cells at 24 hpf indicates that gpr126 expression here is sox10 dependent. In the cls mutant ear, expression of gpr126 was also severely downregulated, even in ears containing rudimentary epithelial projections (Fig. 7L-O). Expression of gpr126 in the heart and posterior mesoderm appeared to be independent of Sox10 function (Fig. 7J,K).

Many GPCRs bring about their downstream effects via cAMP signalling (Jalink and Moolenaar, 2010) (and references within). Forskolin, which activates adenylyl cyclase and raises intracellular cAMP levels (Seamon and Daly, 1981), has been shown to rescue the myelination defect in gpr126^st49 mutants (Monk et al., 2009). To show whether increased cAMP could also rescue the ear defects in lau, we applied two agonists of cAMP signalling to mutant embryos. In addition to forskolin, we used IBMX, which acts as a competitive, non-selective phosphodiesterase inhibitor to raise intracellular cAMP and cGMP levels and activate PKA (Schultz et al., 1973; Elks and Manganiello, 1985). Forskolin and IBMX can each act on other pathways, but their only common target is cAMP.

Both forskolin and IBMX were found to ameliorate the ear phenotype in tb233c embryos when applied during a crucial window of ear development spanning 60-90 hpf (Fig. 8; supplementary material Fig. S5). The extent of ear swelling was assayed visually in a dorsal view and categorized as ‘not swollen’, ‘mild’ or ‘severe’. Treatment with DMSO alone had no significant rescuing effect (Fig. 8M-P,CC). However, treatment with either forskolin (Fig. 8Q-T) or IBMX (Fig. 8U-X) resulted in a significant rescue of the tb233c allele, reducing swelling (compare Fig. 8D,X) and vcan expression (Fig. 8Y-BB), and restoring fusion and pillar formation (asterisks, Fig. 8T,X; supplementary material Fig. S5). The rescue was dose dependent (Fig. 8CC; supplementary material Fig. S5). Pillar formation was not perfect, however, with occasional small protrusions of tissue present on the fused pillars (arrowheads, Fig. 8T,X; supplementary material Fig. S5). In the stronger fr24 allele, forskolin was effective, but IBMX treatment resulted in only a partial rescue (data not shown). Interestingly, we also found that incubation at a lower temperature throughout the fusion period had a partial rescuing effect in the tb233c allele (supplementary material Fig. S6).

However, if cAMP agonists were applied to wild-type embryos over the same 60-90 hpf time window, high doses also induced an ear swelling when compared with DMSO controls, which could reflect a direct cAMP-mediated stimulation of endolymph production (supplementary material Fig. S7; see Discussion). Fusion was also disrupted, and expression of vcana remained high in any unfused projections, suggesting that there may be an additional contact-mediated requirement for its transcriptional downregulation.

Formation of semicircular canals in the zebrafish ear requires outgrowth of epithelial projections, contact recognition, adhesion and formation of a fusion plate, resulting in pillars of tissue spanning the otic vesicle. We have shown here that an orphan adhesion class G protein-coupled receptor gene, gpr126, is expressed in the outgrowing projections of the zebrafish ear, and that its function is required for correct fusion plate and pillar formation. This fits well with the proposed roles of other adhesion class GPCRs in adhesion, signalling and cell-cell or cell-matrix interactions (Bjarnadóttir et al., 2004; Bjarnadóttir et al., 2007; Yona et al., 2008).

The 7-transmembrane (7TM) region in adhesion class GPCRs tends to be divergent from that of other GPCRs, and has been functionally implicated in cellular migration and dimerization of the protein [(Yona et al., 2008) and references within]. The two missense mutations in the tk256a and tb233c alleles suggest that this region, and especially TMIV, is important for Gpr126 function. In particular, the substitution in tk256a of Pro969, which is highly conserved throughout all adhesion GPCRs (Bjarnadóttir et al., 2004; Bjarnadóttir et al., 2007), may affect insertion of the protein into the membrane or disrupt function.

The partial rescue of pillar formation in hypomorphic lau mutants by artificial elevation of cAMP levels indicates that Gpr126 signalling in the ear is mediated through a cAMP/Protein Kinase A pathway, as previously shown for Schwann cell myelination (Monk et al., 2009). As rescue of the ear phenotype was not efficient in the stronger fr24 allele, this is still consistent with a putative role for the extracellular domain of the protein in mediating contact or adhesion at the fusion plate. The rescue data also suggest that Gpr126 functions through association with a stimulatory G protein alpha subunit (Gα_s), although this has not yet been identified. Three Gα_s genes have been identified in the zebrafish genome, and a fourth gene belonging to a novel class of G alpha subunit, gnav1, is expressed in the inner ear (Oka et al., 2009). These four genes are potential candidates for the Gα subunit that interacts with Gpr126.

In the lau mutant ear, we found dramatic alteration of expression of genes coding for ECM core proteins or modifying enzymes, suggesting that Gpr126 signalling is required directly or indirectly for their regulation. Outgrowth of the canal projections in Xenopus and zebrafish is driven by ECM production, in particular of the glycosaminoglycan hyaluronan (HA) (Haddon and Lewis, 1991; Busch-Nentwich et al., 2004) (and this work). We have shown here that other ECM components are expressed dynamically in zebrafish canal projections, including chondroitin sulphate proteoglycans (CSPGs) and proteoglycan link proteins, where they might influence both the mechanical and signalling properties of the projections. This correlates with the observation that the epithelial cells of the growing projections are rich in rough endoplasmic reticulum, indicating active protein synthesis (Waterman and Bell, 1984). In the dysmorphic, unfused canal projections of the lau mutant ear, expression of several ECM genes remains at high levels, suggesting that the projections remain in an ‘outgrowth-like’ state, and fail to undergo the normal changes associated with fusion and pillar formation. We infer that Gpr126 normally provides the signal that triggers this change in cell behaviour.

The persistence of vcan (a CSPG core protein gene) and chsy1 (chondroitin synthase) expression in the unfused projections of the lau mutant ear is of considerable interest. Versicans interact with a range of ECM and cell-surface components, including both HA (via link proteins) and EGFR. They are generally known to be anti-adhesive molecules, and are upregulated in a number of malignant tumours (reviewed by Ricciardelli et al., 2009). Regulation of versican expression and/or processing is known to be essential for mammalian palate fusion (Enomoto et al., 2010) and cardiac ventricular septal formation (Hatano et al., 2012), both of which require fusion and remodelling events similar to those found in semicircular canal pillar formation. In Xenopus, immunoreactivity to chondroitin sulphate (CS) is high on the epithelium of the semicircular canal projections in the ear, but is absent from the core of the projections (Haddon and Lewis, 1991). Interestingly, CS is a ligand for two human adhesion class GPCRs, EMR2 and CD97 (Stacey et al., 2003; Kwakkenbos et al., 2005). Morpholino-mediated knockdown of chsy1, or overexpression of human CHSY1 mRNA in the zebrafish, disrupts canal pillar formation, but leaves development of sensory cristae intact (Li et al., 2010).

The otic swelling in lau mutants appears to be, at least in part, a secondary consequence of the failure of projection fusion, rather than its cause. We have shown that the ears swell when canal projection outgrowth is disrupted by genetic, enzymatic or chemical means. A similar ear swelling is seen in dfna5 morphants (Busch-Nentwich et al., 2004), where there is insufficient production of ECM and a lack of projection outgrowth, rather than overgrowth and failure of fusion as in lau mutants. We propose that any process that causes the failure of fusion, whether an overgrowth or an agenesis of the projections, can lead to ear swelling. This might be due to a lack of physical support that would normally counteract turgor pressure from the endolymph. It is also possible that the hygroscopic properties of both Versican and HA (Chen and Abatangelo, 1999; Wight, 2002) act to draw fluid into the otic vesicle, which might contribute to swelling in lau mutant ears.

An alternative explanation is a disruption to endolymph homeostasis. Although we did not find direct evidence to support this conclusion, semicircular canal duct epithelium is known to be an endolymph-generating tissue (Milhaud et al., 2002; Abbas and Whitfield, 2009; Pondugula et al., 2013), and so any defects in canal tissue might have secondary effects on endolymph production. cAMP signalling can stimulate endolymph production directly (e g. Sunose et al., 1997a; Sunose et al., 1997b; Milhaud et al., 2002; Pondugula et al., 2013), which could explain the swelling seen after treatment of wild-type ears with cAMP agonists.

Morphogenesis of the semicircular canal system shows both similarities and differences between amniotes and anamniotes. In the chick and mouse, flattened pouches are formed from the otocyst, rather than finger-like projections: it is the sides of these pouches, rather than the tips of the fingers, that recognize and fuse with each other (Martin and Swanson, 1993; Bissonnette and Fekete, 1996). The resultant fusion plate is thus much larger than that in Xenopus or zebrafish, and cell death is thought to be the primary mechanism for clearance of cells in this region, at least in the chick (Fekete et al., 1997; Lang et al., 2000; Cecconi et al., 2004; Kobayashi et al., 2008). In addition, the dfna5/ugdh/HA pathway, which is implicated in semicircular canal projection outgrowth in the fish, does not appear to be conserved in the mouse (Busch-Nentwich et al., 2004; Van Laer et al., 2005). By contrast, other genes involved in canal formation, such as Otx1, are highly conserved between fish and mammals (Hammond and Whitfield, 2006). Despite the differences, formation of the canal ducts in all groups requires the recognition, adhesion and remodelling of cells at the fusion plate.

In the zebrafish, the otic vesicle is a major site of expression of gpr126, and the defects in the lau mutant ear indicate a crucial function for Gpr126 in semicircular canal morphogenesis. In the mouse, initial studies using a Gpr126^LacZ reporter did not detect expression in the ear (Waller-Evans et al., 2010). Instead, embryonic expression of murine Gpr126 (also known as DREG) has been reported only in somites, heart, branchial arches, frontonasal process, kidney collecting duct and Bergmann glia of the cerebellum (Moriguchi et al., 2004; Koirala and Corfas, 2010; Pradervand et al., 2010). However, recent work by Engel and colleagues describes strong expression of gpr126 in the mouse otic vesicle at E9.5 and E11.5 (Patra et al., 2013), suggesting possible functional conservation of this signalling pathway in otic development in mammals.

Homozygous loss of function of Gpr126 in the mouse results in mid-gestational or perinatal lethality (depending on the allele), with a few pups surviving to 1-2 weeks of age (Waller-Evans et al., 2010; Monk et al., 2011). In surviving mutant animals, a detailed analysis of ear structures has yet to be determined; however, no gross morphological defects in the ear have been reported. Analysis of the auditory nerve revealed a lack of peripheral myelination, with aberrant extension of central myelination by oligodendrocytes into the periphery (Monk et al., 2011), indicating that Gpr126 function in peripheral myelination is conserved between fish and mammals. It will be of interest to determine whether Gpr126 has a similarly conserved function in the development of the semicircular canal system.

We thank M. Cotterill and L. Murphy for technical assistance; G. Cooper, R. Kelsh, H. Roehl, F. van Eeden and K. Whitlock for discussion and advice; H. G. Frohnhöfer for providing fish; the Sheffield aquarium staff for expert animal care; and members of the zebrafish community for providing probes. We thank F. B. Engel for sharing information before publication, G. Corfas for advice on Gpr126 expression in the mouse ear, and J. Jen for sequence analysis of human GPR126.