Cross-repressive interactions between Lrig3 and netrin 1 shape the architecture of the inner ear

Precise spatiotemporal regulation of intercellular signaling is crucial for molding tissues into three-dimensional structures during organogenesis. One of the most striking examples of tissue morphogenesis is the development of the three-dimensional architecture of the inner ear, which houses the sensory organs for hearing and balance. Angular acceleration is detected by three fluid-filled semicircular canals that are oriented with respect to the three dimensions of space. Changes in head position result in fluid movement within the canals, thereby activating specialized hair cells in the sensory epithelia located at the base of each canal. Subsequently, vestibular ganglion neurons convey signals from the sensory epithelium to the central nervous system. The inner ear is a small, intricately shaped organ that does not tolerate even subtle changes to its structure; indeed, even the slightest perturbations in the structure of the vestibular canals can result in debilitating dizziness,vertigo and abnormal posture in humans(Sando et al., 2001; Sando et al., 1984).

Morphogenesis of the inner ear is set in motion by early patterning events,which result in the expression of key cell fate determinants within discrete domains of a primordial structure. The inner ear is sculpted from a simple ball of epithelium called the otic vesicle(Fig. 1)(Fekete, 1999). The semicircular canals are derived from two outpocketings, the canal pouches,which are specified by transcription factors such as Otx1 and Dlx5(Merlo et al., 2002; Morsli et al., 1999). A crucial event in canal development is the formation of the fusion plate in the center of the pouch. During this process, the basement membrane breaks down,allowing signaling molecules to induce proliferation in the surrounding mesenchyme and the two epithelial walls to come together(Martin and Swanson, 1993; Pirvola et al., 2004; Salminen et al., 2000; Streeter, 1907). At the same time, fusion plate cells lose their columnar morphology and intercalate to form a single layer of cells (Martin and Swanson, 1993). Importantly, these changes in cell morphology and basal lamina integrity occur only in the vicinity of the fusion plate. By contrast, the epithelium in the perimeter of the pouch remains intact and will eventually form the walls of the mature canal. Hence, the final shape of each canal is determined by when and where fusion occurs.

A key regulator of fusion is the laminin-related molecule netrin 1 (Ntn1). In Ntn1 mutant mice, the inner ear fails to acquire its mature shape due to an arrest in canal morphogenesis(Salminen et al., 2000). Although best characterized for its function as a secreted chemoattractant for axons during neuronal development (Moore et al., 2007), Ntn1 is also crucial for many aspects of tissue morphogenesis, including cell migration and local cell adhesion(Srinivasan et al., 2003; Yebra et al., 2003). In the inner ear, Ntn1 is required for breakdown of the basement membrane surrounding the fusion plate (Salminen et al., 2000). Ntn proteins are localized to the basal lamina(Salminen et al., 2000; Schneiders et al., 2007) and can bind to collagen and laminin in vitro(Schneiders et al., 2007; Yebra et al., 2003), but how Ntn1 signaling promotes basal lamina breakdown remains unclear(Matilainen et al., 2007).

Despite abundant evidence that Ntn1 is a powerful morphogen, little is known about the pathways that restrict Ntn1 expression to highly limited spatiotemporal domains in any developing system(Kennedy, 2000). In addition to its functions during development, Ntn1 is overexpressed in several human cancers (Fitamant et al., 2008; Link et al., 2007),underscoring the importance of understanding how expression of Ntn1is regulated. Here, we demonstrate that Ntn1 expression is controlled by cross-repressive interactions with the Ig superfamily protein Lrig3 that define the boundary between the fusing and non-fusing regions of the lateral canal pouch. This novel feedback loop dictates when and where basement membrane breakdown occurs, thereby ensuring that the inner ear acquires its precise three-dimensional shape.

The LST016 gene trap line has been previously reported(Leighton et al., 2001; Mitchell et al., 2001). The mice have been maintained for over five generations on the C57Bl6/J background. Genotyping was performed using the following primers: LST016F(GAGGTGCCTGATGCTTAAGTTTCG); LST016R (TTCAACCTTGGCTTCCAATGTCCA); and GTR7(CAAGTCTATCCTAGGGAAAGGGTC), which is specific to the Lrig3^LST016 gene trap vector pGT2TMPFS. The Lrig3^flox conditional allele was produced via homologous recombination using a targeted construct with LoxP sites surrounding the ATG-containing exon 1. Heterozygous mice were produced by germline Cre-excision using a global deleter line(Schwenk et al., 1995). Genotyping was performed using the following primers: VEA303(CGGAATTTCCTACAATCTCAGC); VEA304 (GTGCTCCTGGTGGCTCAGT); VEA305(CCCCCTCCAATTTTAACAAA). The Ntn1 gene trap line has been previously reported (Serafini et al.,1996). The mice have been maintained for over five generations on the C57Bl6/J background. Genotyping was performed using the following primers:pGT1_8TM3021R (GTTGCACCACAGATGAAACG); pGTEn2_1723 (TCCCGAAAACCAAAGAAGAAG); and pGT1_8TM1743 (GAACCCTAACAAAGAGGACAAG). The described genotyping screen allows for the detection of the Ntn1-specific gene trap vector pGT1.8TM. Heterozygous and mutant genotypes were confirmed by evaluating the thickness of the ventral commissure in the neural tube following βIII-Tubulin and Neurofilament immunostaining on cryosectioned embryos. These two gene trap lines are annotated as Lrig3^+/- or Ntn1^+/- for the heterozygous condition and Lrig3^-/- or Ntn1^-/- for the mutant condition. Unc5hb mutant mice were maintained on a CD1 background and genotyped as described(Lu et al., 2004). The integrin α6 gene trap mice (Mitchell et al., 2001) were maintained on a C57Bl6/J background and were genotyped using the following primers: LST045F5 (GCCCAAATCCCTTGTGTATG);LST045R1 (ACCCACAGCAACCTTTGTTC); and GTR3 (TCTA GGACAAGAGGGCGAGA). The mice were maintained in accordance with institutional and NIH guidelines approved by the IACUC at Harvard Medical School and the University of Virginia.

cDNA was made from total mouse E11.5 embryonic RNA using random primers. Lrig3 was amplified as described(Abraira et al., 2007) with SYBR Green Supermix (BioRad) and primers flanking the Lrig3 exon3/4 boundary: VEA48 (GAACAACAATGAATTAGAGACCATTC) and VEA49 (AGGGTGGAAAGGCAGTTCTC). Levels were normalized based on amplification of GAPDH from the same samples:GAPDHe-F (CTCATGACCACAGTCCATGC) and GAPDHe-R (GCACGTCAGATCCACGAC). Loss of Lrig3 message from Lrig3 null mice was confirmed by RT-PCR amplification of E12.5 embryonic cDNA using primers VEA48 and VEA49.

E12-E14.5 embryos were fixed overnight at 4°C with Bodian's Fix,dehydrated overnight at room temperature with 100% ethanol, then cleared overnight at room temperature with methyl salicylate. Heads were hemisected,and white latex paint (Benjamin Moore) diluted to 0.025% in methyl salicylate was injected into the cochlea with a pulled glass pipette and a Hamilton syringe filled with glycerol.

Non-radioactive in situ hybridization was performed on 10-12 μm cryosections using the following probes: Ntn1 (NM_008744), Otx1 (NM_011023), Otx2 (NM_144841), Dlx5(NM_010056) and Hmx3 (NM_008257). A detailed protocol is available at http://neuro.med.harvard.edu/site/goodrichweb/.

Staining for β-galactosidase and alkaline phosphatase activity was performed as described (Leighton et al.,2001) except that 10-20 μm frozen sections were used, and the tissue was fixed for 1 hour at 4°C.

E12 embryos were collected and fixed for 1-2 hours at 4°C in 4% PFA/PBS and then dehydrated in 30% sucrose/PBS overnight at 4°C. Embryos were then embedded in Neg50 (Richard-Allan Scientific). Cryosections (5-10 μm) were blocked and permeabilized in 5% normal donkey serum + 2% BSA + 0.1% Triton X-100 for 1 hour at room temperature. Primary antibodies were added into the above block, without Triton X-100, overnight at 4°C at the following concentrations: collagen IV (1:200, Abcam, ab6586); pan-mouse laminin (1:250,Chemicon, AB2034); βIII-tubulin (1:1000, Covance, PRB-435P); and neurofilament (1:500, DSHB, SH3). The following day, the sections were incubated in secondary antibody (1:2000, Alexa Fluor488 or 568, Jackson Immunoresearch) in block, without Triton X-100. All sections were counterstained with DAPI (1:10,000).

Plastic semi-thin (1 μm) and ultra-thin (90 nm) sections for light and electron microscopy studies were obtained with help from the Harvard Medical School Electron Microscopy Facility(cellbio.med.harvard.edu/research_facilities).

To gain insight into the molecules that drive inner ear morphogenesis, we took advantage of the fact that even subtle changes in the structure of the inner ear cause dramatic behavioral defects. Candidate genes were identified by screening a large set of mice generated by gene trap technology(Leighton et al., 2001; Mitchell et al., 2001). Upon insertion into the intron of a gene, the gene trap vector simultaneously blocks transcription and reports the normal expression pattern of that gene through a β-galactosidase reporter(Fig. 2A). X-gal-stained heterozygous embryos were examined to find genes expressed in restricted domains of the otic vesicle prior to morphogenesis, and then homozygous mutants for the best candidate genes were assessed for defects in hearing and balance.

In this screen, we identified a strain of mice, LST016, withβ-galactosidase activity in the lateral wall of the otic vesicle(Fig. 2B,C), a region fated to give rise to the lateral semicircular canal(Fekete and Wu, 2002). Homozygous mutants exhibit circling and head tossing behaviors, consistent with the presence of an inner ear defect(Fig. 2D). Visualization of the three-dimensional structure of the vestibular apparatus revealed a fully penetrant truncation of the lateral semicircular canal in homozygotes(Fig. 2G-J). Histological studies confirmed that the canal epithelium is missing as early as E13 (data not shown). The lateral ampulla is unaffected, and the sensory epithelium is properly innervated, as determined by myosin VIIa and neurofilament immunostaining (Fig. 2I,J; data not shown). Although the reporter is also active in the developing cochlea(see Fig. S1H in the supplementary material), the mice develop normal hearing as assessed by auditory brainstem response (ABR) assays (see Fig. S2 in the supplementary material). All mutants also display craniofacial deformities and a dramatically shortened snout (Fig. 2E,F).

LST016 mice carry an insertion of the pGT2TMPFS gene trap vector in the third intron of the Lrig3 gene(Fig. 2A), causing a truncation of the wild-type transcript and fusion of the β-galactosidase reporter protein to Lrig3 at amino acid 126(Mitchell et al., 2001). Lrig3 is one of three members in a family of single-pass transmembrane proteins containing 15 leucine-rich repeats (LRR) and three immunoglobulin (Ig) domains in the extracellular domain and intracellular tails that vary in length and composition (Guo et al., 2004; Hedman and Henriksson, 2007). Quantitative RT-PCR of homozygous Lrig3 tissue (n=4 embryos)detected only 3.05±1.59% of the wild-type transcript, indicating that the phenotype is severely hypomorphic. To confirm that the phenotype is due to a loss of Lrig3, a null allele was made by flanking the ATG-bearing exon of Lrig3 with LoxP sites(Fig. 2A). After germline Cre-mediated excision, no Lrig3 messenger RNA remained (data not shown). Null mutant mice exhibit the same lateral canal and craniofacial defects evident in the gene trap allele (data not shown), demonstrating that the phenotypes reflect a complete loss of Lrig3 function.

The semicircular canals develop from two epithelial outpocketings called canal pouches, with the anterior and posterior canals arising from the dorsal pouch and the lateral canal forming from the lateral pouch(Fig. 1). Early patterning events that define the axes of the otic vesicle result in restricted expression of transcription factors that specify the two pouches. Because Lrig3 expression is restricted to the lateral pouch during these early patterning stages, it seemed possible that Lrig3 acts in a signaling pathway that ensures restricted expression of transcription factors required for specification of the lateral canal. To test this, we examined the expression patterns of Otx1, Otx2, Hmx3 and Dlx5,transcription factors required for normal development of the vestibular system(Merlo et al., 2002; Morsli et al., 1999; Wang et al., 1998). However, Lrig3 mutant embryos exhibited no obvious changes in early gene expression, indicating that the lateral pouch is specified in the right place and at the right time (see Fig. S3 in supplementary material).

As canal patterning was unaffected, we asked whether subsequent morphogenesis events proceed normally in Lrig3 mutant embryos. We visualized pouch outgrowth and fusion by paintfilling inner ears between E11.5 and E12.5 (Fig. 3). The precise stage of canal development was determined by evaluating the extent of fusion in the anterior and posterior canals, which develop earlier than the lateral canal (Martin and Swanson,1993). In control embryos, the lateral pouch grows out at E12 and fusion begins at E12.5 (Fig. 3A-C). No changes in the size or shape of the early canal pouch occur in E12 Lrig3 mutant embryos, consistent with correct patterning of the otic vesicle (Fig. 3D). However, fusion in the lateral pouch begins several hours earlier than normal and over a larger area, extending to the perimeter of the lateral pouch epithelium (Fig. 3E). The lateral canal is truncated by E12.5, when fusion is just beginning in wild-type littermates (Fig. 3F). Thus, fusion is expanded and accelerated in Lrig3mutants.

Although fusion occurs prematurely in Lrig3 mutants, fusion is arrested in Ntn1 mutants, raising the possibility that these two genes cooperate to determine the timing and extent of fusion. To gain more insight into the relationship between Lrig3 and Ntn1 during canal morphogenesis, we compared their expression patterns on adjacent sections from E12 to E12.5. We found that Lrig3 expression is enriched in the lateral pouch epithelium during the canal pouch outgrowth stage, when Ntn1 is not yet expressed(Fig. 3G,G′). Then, just before fusion begins, Lrig3 expression becomes reduced in the center of the pouch (Fig. 3H),concomitant with the initiation of Ntn1 transcription in the nascent fusion plate (Fig. 3H′). As fusion progresses, the domain of Ntn1 transcription expands(Fig. 3I′), whereas Lrig3 is downregulated in Ntn1-positive cells(Fig. 3I) but is maintained in the surrounding, non-fusing epithelium. After the canal is fully formed, Lrig3 and Ntn1 continue to be expressed in non-overlapping domains of the canal epithelium (data not shown). Hence, consistent with their opposing activities during canal development, Lrig3 and Ntn1are expressed in complementary domains of the lateral pouch.

These results suggest that Lrig3 is required in the non-fusing epithelium to prevent fusion from occurring by balancing the activity of Ntn1 in the fusion plate. Malformation of the semicircular canals in Ntn1 mutant mice is preceded by a failure in basement membrane breakdown (Salminen et al.,2000). Therefore, we asked whether the early and ectopic fusion event in Lrig3 mutant mice is also accompanied by changes in the integrity of the basal lamina that normally separates the epithelium from the surrounding mesenchyme. We found that dramatic changes occur in the basal lamina of Lrig3 mutant embryos. In control embryos, laminin and collagen networks are intact prior to fusion plate formation, as shown by immunostaining and electron microscopy analysis(Fig. 4A-E). By contrast, in Lrig3 mutants, the basal lamina is missing or disrupted throughout the lateral pouch, including regions where fusion normally never occurs(Fig. 4F-J). Moreover, the otic epithelium throughout the lateral pouch is abnormally thin, such that even cells in the perimeter of the pouch resemble fusion plate cells(Fig. 4A′,F′). Neither caspase 3 immunostaining (see Fig. S4 in the supplementary material)nor electron microscopy analysis (data not shown) revealed an increase in apoptosis, suggesting that these morphological changes are not associated with cell death. Thus, Lrig3 inhibits basement membrane breakdown in the non-fusing epithelium.

To understand the molecular basis of the basement membrane phenotype, we explored the possibility that Lrig3 modulates Ntn1 activity by regulating one of its known receptors. This hypothesis is supported by the facts that Lrig3 is expressed complementary to Ntn1 and that Lrig proteins regulate degradation of many different transmembrane receptors(Hedman and Henriksson, 2007). Ntn1 is best known as an axon guidance molecule, which signals through the Ig superfamily of receptors DCC (deleted in colorectal cancer) and neogenin 1, as well as the Unc5 family of receptors (Unc5Ha-d)(Moore et al., 2007). Outside of the nervous system, Ntn1 signals through neogenin(Srinivasan et al., 2003),Unc5hb (Lu et al., 2004) and integrin α3/6 receptors (Yebra et al., 2003) to regulate cell adhesion, migration and other aspects of tissue morphogenesis.

To identify relevant receptors, we asked whether any of the known Ntn1 receptors is required for inner ear morphogenesis. In situ hybridization screens by our laboratory and others revealed Unc5hb, neogenin and integrin α3/6 as potential candidates(Matilainen et al., 2007). However, although Unc5hb is expressed together with Lrig3 in non-fusing epithelium, the inner ear forms normally in Unc5hb mutant mice (see Fig. S5 in the supplementary material). Moreover, none of the known Ntn1 receptors that are expressed in the developing otic epithelium (neogenin,integrin α3/6) or surrounding mesenchyme (Unc5hc) are required for canal morphogenesis (see Fig. S5 in the supplementary material)(Matilainen et al., 2007).

Interestingly, the pro-angiogenic activities of Ntn1 have been proposed to be independent of known receptors, suggesting an alternative binding partner in this system (Wilson et al.,2006). Although Lrig3 is a cell-surface protein, Lrig3 does not appear to be the missing receptor, as tagged versions of Lrig3 and Ntn1 do not colocalize in cultured cell lines (see Fig. S6 in the supplementary material). Moreover, equal amounts of Ntn1 are secreted from cells in the presence and absence of Lrig3 (see Fig. S6 in the supplementary material). Thus, Ntn1 appears to act through a non-canonical pathway in the inner ear to control basal lamina integrity, either via a novel receptor or a receptor-independent mechanism.

As Lrig3 does not appear to act through the Ntn1 pathway, we considered alternative explanations for why basement membrane breakdown is expanded in Lrig3 mutants. Based on their complementary expression patterns and opposing activities, a simple idea is that the main function of Lrig3 is to restrict expression of Ntn1 to the fusion plate. As hypothesized, we found that Ntn1 expression is expanded to encompass the entire lateral pouch epithelium in Lrig3 mutants(Fig. 5A,E). As Lrig3 appears to regulate Ntn1 expression at the transcriptional level, we asked whether the transcriptional downregulation of Lrig3 in the wild-type fusion plate (Fig. 3) is similarly a result of Ntn1 activity. To do this, we took advantage of the fact that a placental alkaline phosphatase (PLAP) reporter gene is present in the gene trap vector used to generate Lrig3 mutants(Fig. 2A), but not in Ntn1 gene trap mutants (Serafini et al., 1996). As seen by β-galactosidase histochemistry(Fig. 3I), PLAP expression is reduced in the fusion plate in Lrig3 heterozygous embryos(Fig. 5C). By contrast, in Ntn1 mutants, PLAP expression persists throughout the lateral pouch(Fig. 5G). Surprisingly, we also observed sustained expression of Lrig3 itself in Lrig3mutants, indicating that Lrig3 is required for its own downregulation(Fig. 5B,F). Therefore, we asked whether Ntn1 transcription is also autoregulated. Indeed,Ntn1-βgeo expression is lost from the lateral pouch in Ntn1mutants (Fig. 5D,H). Ntn1 autoregulation appears to be limited to the lateral pouch, as no changes in expression occur in the dorsal pouch. Together, these experiments show that mutually antagonistic interactions between Lrig3 and Ntn1restrict Ntn1 expression to fusion plate cells and Lrig3expression to non-fusing epithelium. This feedback loop seems to be uniquely important for lateral canal development, suggesting that new mechanisms operate in this most recently evolved semicircular canal.

The presence of cross-repressive interactions between Lrig3 and Ntn1 suggests that a slight reduction in Ntn1 levels might be sufficient to compensate for the loss of Lrig3. To test this idea,we reduced the dose of Ntn1 in Lrig3 mutant mice by intercrossing Ntn1^+/-; Lrig3^+/-transheterozygotes and collecting E14.5 embryos (n=81) for paintfilling (Fig. 6). Wild-type (n=4/4) and transheterozygous (n=23/23) embryos developed normal canals (Fig. 6A,B,G), whereas littermates homozygous for Lrig3 but wild type for Ntn1 exhibited lateral canal truncations, as expected(n=11/12 ears) (Fig. 6C,G). By contrast, no truncations occurred in Lrig3homozygotes that were also heterozygous for Ntn1 (n=12/12 ears) (Fig. 6D,G), indicating that the Lrig3 phenotype is caused by a failure to properly regulate Ntn1 levels in fusion plate cells. Consistent with the complete rescue, adult Lrig3^-/-;Ntn1^+/- mice did not display circling behavior (n=16). As the size and shape of the canal pouch is similar in Ntn1 (Fig. 6E,G) (n=7) and Ntn1;Lrig3 double mutant embryos(Fig. 6F,G) (n=4), Lrig3 probably acts upstream of Ntn1.

We conclude that Lrig3 plays a pivotal role in lateral canal morphogenesis by restricting Ntn1 expression to the fusion plate. Cross-repressive interactions between Lrig3 and Ntn1coordinate the timing and location of fusion, thereby determining the shape of the lateral canal.

A central challenge in developmental biology is understanding how signaling pathways cooperate to sculpt tissues with complex three-dimensional shapes. Here, we describe the presence of a novel feedback loop that restricts the expression of two genes with opposing functions to discrete domains of the otic vesicle during inner ear morphogenesis. Because the pattern of Ntn1 expression determines when and where fusion occurs, the spatiotemporal regulation of Ntn1 through cross-repressive interactions with Lrig3 ensures perfect morphogenesis of the elaborate structure of the inner ear.

Historically, much emphasis has been placed on the identification of determinants that define domains within the anlage of a developing structure. More recently, however, it has become clear that an additional level of control is needed to restrict the spatiotemporal activities of each factor. A common solution to this problem is the production of a feedback-induced antagonist that dampens the activity of a signaling pathway after activation,as is the case for Sprouty and Sef proteins in the FGF pathway(Shim et al., 2005; Tsang and Dawid, 2004). Signaling activity is also modulated by the basal lamina, which can limit dispersal of the protein and its ability to bind to its receptor(Relan and Schuger, 1999). The Lrig3/Ntn1 feedback loop incorporates both of these features: the induction of an antagonist, Lrig3, which controls the timing and extent of Ntn1-dependent basal lamina breakdown.

The Ntn1/Lrig3 feedback loop uncovered by this analysis is a novel mechanism for modulating expression of Ntn1 during development. In addition to the developing canals, Lrig3 and Ntn1 are also expressed in many other regions of the embryo, including the cochlea, neural tube and somites (see Fig. S1 in the supplementary material). Although no alterations of these tissues are obvious in Lrig3 mutant mice (data not shown), this may be due to compensation by the close family member Lrig1. Indeed, our expression studies indicate that the only two places where Lrig1 and Lrig3 do not overlap are the lateral canal pouch and the branchial arches, consistent with the lateral canal truncation and craniofacial abnormalities evident in Lrig3 mutant mice. Conversely, psoriasis is the only salient defect reported in Lrig1 mutant mice (Suzuki et al.,2002). Hence, Lrig1; Lrig3 double mutants may exhibit additional phenotypes related to misregulation of Ntn1 and may reveal a general mechanism for Ntn1 regulation.

Abnormal regulation of Ntn1 by Lrig may have broad implications in disease as disruption of the laminin network is a major factor in tumor invasion (Gupta and Massague,2006). Consistent with this idea, all three Lrig genes are downregulated in specific human cancers, with Lrig1 studied most extensively (Guo et al., 2006; Hedman and Henriksson, 2007; Hedman et al., 2002; Lindstrom et al., 2008; Ljuslinder et al., 2007). Strikingly, the loss of Lrig1 correlates with worse outcomes in advanced stage cervical cancer (Lindstrom et al., 2008) and squamous cell carcinomas(Tanemura et al., 2005). In addition, the presence of Lrig1 is sufficient to inhibit the invasive behavior of bladder cancer cell lines and to prevent breakdown of fibronectin in a cell-matrix adhesion assay (Yang et al., 2006). Conversely, increased Ntn-1 induces tumor growth and invasion in human colorectal cell lines(Rodrigues et al., 2007) and is associated with poor prognoses for adenocarcinoma(Link et al., 2007). Although the effect of Ntn-1 on tumor behavior is mediated in part by inhibition of apoptosis through the DCC and Unc5h dependence receptors(Fitamant et al., 2008; Mazelin et al., 2004; Rodrigues et al., 2007), our data raise the possibility that in some tumors, Ntn1 may act downstream of Lrig3 to facilitate basement membrane breakdown and tumor metastasis.

Based on the current understanding of this poorly characterized family of proteins, it is likely that Lrig3 represses Ntn1 transcription by regulating activity of a receptor tyrosine kinase signaling pathway. The best-studied family member, Lrig1, is induced by EGF signaling and antagonizes downstream signaling events by causing degradation of all four ErbB receptors(Gur et al., 2004; Laederich et al., 2004). Lrig1 has also been implicated as a negative regulator of Met and Ret receptor tyrosine kinases (RTKs) (Ledda et al.,2008; Shattuck et al.,2007), raising the possibility that Lrig proteins serve as general antagonists of RTKs. Consistent with this idea, Lrig3 can bind to the FGF receptor tyrosine kinase receptor in vitro and inhibits FGF signaling in the developing neural crest in Xenopus(Zhao et al., 2008). This function may be conserved in mice, as Lrig3 morphant tadpoles exhibit craniofacial defects similar to what is observed in Lrig3 mutant mice. As FGF signaling plays a prominent role in canal morphogenesis(Chang et al., 2004; Pauley et al., 2003; Pirvola et al., 2004), the Lrig3 inner ear phenotype is also likely to be caused by aberrant FGF activity. However, although FGF ligands and receptors have been implicated,the downstream signaling events in the fusion plate remain elusive, with no known target genes and multiple feedback-induced antagonists that are not required for canal development (Abraira et al., 2007; Shim et al.,2005). Hence, until we have a better understanding of how FGF signaling acts specifically in the fusion plate, it will be difficult to determine whether and how Lrig3 influences FGF activity.

Together with what is known about Lrig function, the simplest interpretation of our results is that Lrig3 titrates activity of a receptor tyrosine kinase signaling pathway that normally induces expression of Ntn1, as well as Lrig3 itself(Fig. 7). Prior to fusion, Lrig3 is expressed throughout the lateral pouch, where it inhibits RTK activity and prevents fusion from beginning(Fig. 7, step 1). Subsequently,we hypothesize the presence of a fusion plate inducing signal, which overcomes Lrig3-mediated inhibition (Fig. 7, step 2) to activate the RTK pathway and allow expression of Ntn1 (Fig. 7, step 3). Ntn1, in turn, enhances activity of this same pathway, most probably by promoting breakdown of the basal lamina(Fig. 7, step 4). Hence, in Lrig3 mutants, increased RTK signaling results in early and expanded Ntn1 expression, as well as persistent Lrig3 expression(Fig. 7A). Conversely, in Ntn1 mutants, RTK signaling occurs at low levels, both because of the presence of Lrig3 and the failure to potentiate the pathway through Ntn1-mediated breakdown of the basal lamina. Owing to the low level of RTK signaling, canal development arrests at the canal pouch stage, such that expression of Ntn1 is lost and Lrig3 is never reduced(Fig. 7A).

A key feature of this model is the fact that Lrig3 is present prior to any of these events, but is also crucial for the subsequent emergence of mutually exclusive domains of Lrig3 and Ntn1 expression. As Lrig3 activity is required for the feedback loop to function properly, the initial Lrig3 expression also fails to be downregulated in Lrig3mutants. Hence, the subsequent interactions between Lrig3 and Ntn1 serve to simultaneously reduce Lrig3 transcription and increase Ntn1 only in the regions that receive the inducing signal.

The proposed model fits both with the results reported here as well as the known activities of Lrig3 and Ntn1, but also raises several questions for future consideration. Unfortunately, the ability of Lrig proteins to bind to widely divergent members of the receptor tyrosine kinase family in vitro will make it difficult to pinpoint a single binding partner in vivo, especially in a structure as small as the lateral pouch. The best candidate is the FGF receptor, not only because Lrig3 is known to bind to and inhibit FGF receptor,but also because of the known importance of FGF signaling during inner ear development. In the developing vestibular system, FGF signaling cooperates with the BMP pathway to define sensory and non-sensory domains of the inner ear (Chang et al., 2004; Pauley et al., 2003) and is subsequently required for proliferation of the periotic mesenchyme(Pirvola et al., 2004). Moreover, mesenchymal proliferation is reduced in Ntn1 mutants,consistent with the idea that Ntn1-dependent breakdown of the basal lamina promotes the ability of the FGF ligand to act during canal morphogenesis. Indeed, it is well established that FGF signaling levels are modified by interactions with the basement membrane during other types of tissue morphogenesis (Lonai, 2003; Patel et al., 2007). Although it remains unclear how Ntn1 mediates its effects, the extent of basement membrane breakdown correlates strongly with the amount of Ntn1:breakdown does not occur in Ntn1 mutants, is expanded in the presence of ectopic Ntn1 in Lrig3 mutants, and proceeds normally in rescued embryos that have only one copy of Ntn1. Additional experiments will be needed to understand the specific function of Ntn1 in the basement membrane, as well as how Ntn1-induced changes influence the activity of FGF or other signaling ligands in the extracellular matrix.

Because modest perturbations in the structure of the inner ear cause severe behavioral deficits, our genetic studies were able to reveal the consequences of slight changes in signaling activity that may be undetectable by in vitro methods. Indeed, lateral canal truncations also occur in BMP4heterozygotes, emphasizing the unusual sensitivity of the developing inner ear to modest changes in signaling levels(Chang et al., 2008). Similarly, in humans, the lateral canal is the most common site of inner ear anomalies (Sando et al., 2001; Sando et al., 1984),emphasizing the importance of identifying the molecular players that make this canal unusually susceptible to developmental insults. Hence, the Lrig3/Ntn1 feedback loop may provide an additional safeguard for tissues whose function depends crucially on the perfect morphogenesis of complex structures.

We thank X. Lu for providing Unc5hb mutant tissue, D. Goodenough for insights into the EM results, N. Hyun and N. Pogue for genotyping assistance, and R. Segal, N. Andrews, X. Lu, M. Scott and members of the Goodrich laboratory for advice on the project and comments on the manuscript. We are also grateful to S. Walker and J. Brugge for assistance with the acinar cultures. This work was supported by grants R01 DC7195 (L.V.G.) and F31 DC008450 (V.E.A.) from the N.I.H./N.I.D.C.D., and by funding from the C.I.H.R.(J.S.), the Smith Family New Investigators Program (L.V.G.) and the Mathers Charitable Foundation (L.V.G.).