The zinc finger transcription factor Gfi1, implicated in lymphomagenesis, is required for inner ear hair cell differentiation and survival

The Gfi1 gene was first identified as causing interleukin-2 (IL2)independent growth in T cells (Gilks et al., 1993) by allowing them to escape G₁ arrest normally induced by IL2 withdrawal (Grimes et al., 1996a) — hence the name growth factor independent(Gilks et al., 1993). Proviral integration in Gfi1 results in upregulated transcriptional activity of Gfi1 and is associated with lymphomagenesis in mice(Gilks et al., 1993;Liao et al., 1997;Scheijen et al., 1997;Schmidt et al., 1996). Gfi1 and its homolog Gfi1b (Tong et al.,1998) are nuclear zinc-finger proteins. Gfi1 functions as a position- and orientation-independent transcriptional repressor through its 20 amino acid N-terminal repressor, or SNAG, domain(Grimes et al., 1996a). Gfi1b appears to function biochemically in a similar manner to Gfi1, as it binds the same DNA recognition site and represses transcription through its SNAG domain.

Recent publications suggest a variety of in vivo functions forGfi1 and Gfi1b. High levels of Gfi1 transgene result in a block of T-cell lymphopoiesis (Schmidt et al., 1998a; Schmidt et al.,1998b). Constitutive Gfi1 expression accelerates entry of resting T cells into S phase of the cell cycle(Karsunky et al., 2002a); and causes decreased levels of apoptosis, increased levels of cell proliferation and a decrease in the levels of negative cell cycle regulators p27^KIP1 and pRb. Gfi1 may also regulate apoptosis through repression of multiple pro-apoptotic regulators(Grimes et al., 1996b). Outside the lymphoid system, Gfi1 is expressed in granulocytes and activated macrophages (H. Hock, M. J. Hamblen, H. M. Rooke, D. Traver, R. T. Bronson, S. Cameron and S. H. Orkin, unpublished)(Karsunky et al., 2002b). Loss-of-function studies in mice mutant for Gfi1 indicate that it is necessary during hematopoesis as it is required for the proper specification and differentiation of neutrophils and macrophages (H. Hock, M. J. Hamblen, H. M. Rooke, D. Traver, R. T. Bronson, S. Cameron and S. H. Orkin, unpublished;Karsunky et al., 2002b). Overexpression of Gfi1b results in inhibition of G1 arrest and differentiation by directly binding the p21^WAF1 promoter and repressing its activity (Tong et al.,1998). Gfi1b can also directly repress the activity of tumor suppressor genes Socs1 and Socs3 by binding their promoters (Jegalian and Wu,2002). The Gfi1b zinc-finger domain may also act as a transcriptional activation domain (Osawa et al., 2002). Thus, Gfi1b may modulate transcription as a repressor or activator depending on promoter and cell type context. Loss of function studies in the mouse indicate that Gfi1b function is required for hematopoiesis as it is required for erythroid and megakaryocytic lineages. Mice deficient for Gfi1b are embryonic lethal and die by E15.5 because of a complete lack of erythrocytes(Saleque et al., 2002).

The Gfi proteins have invertebrate homologues, including senselessin Drosophila (Frankfort et al.,2001; Nolo et al.,2000; Nolo et al.,2001) and pag-3 in C. elegans(Cameron et al., 2002;Jia et al., 1996;Jia et al., 1997). InDrosophila, senseless is required during the development of the embryonic and adult peripheral nervous system (PNS). Embryos that lacksenseless differentiate the majority of PNS cells, but most cells die through apoptosis (Nolo et al.,2000; Salzberg et al.,1997). However, in adult sensory organs, senseless is both necessary and sufficient for their development. Mosaic analysis in imaginal discs shows that senseless mutant clones lack sensory organs. Expression of senseless is dependent on the proper expression of proneural genes, such as atonal, scute, achaete anddaughterless. senseless in turn is required for the upregulation and maintenance of expression of the proneural genes in the sensory organ precursors (SOP), as loss-of-function mutations in senseless abolish the further upregulation and maintenance of proneural gene expression in the SOPs. Ectopic expression of senseless induces ectopic proneural gene expression and ectopic PNS organs. In addition, senseless has been shown to synergize with the proneural genes(Nolo et al., 2000).

PAG-3 is a C. elegans homolog of Senseless and Gfi1.pag-3 is involved in touch neuron gene expression and coordinated movement (Jia et al., 1996;Jia et al., 1997). In addition, null mutations of pag-3 can result in abnormal patterns of apoptosis in the ventral nerve cord as well as abnormal differentiation of certain interneurons and motoneurons. Hence, pag-3 functions in diverse contexts within the developing nervous system. The finding thatpag-3 is expressed in many neuronal subtypes at different points in development suggests that it cooperates with different factors to regulate the expression of cell type- and developmental stage-specific sets of genes to generate the complex pattern of neuronal subtypes seen in C. elegans(Cameron et al., 2002).

The above data suggest that this small but specific gene family of Zn-finger transcription factors plays similar as well as different roles in different species and different tissues. These include suppression of apoptosis, suppression of cell cycle checkpoints, as well as promoting cell fate determination and cell differentiation. However, in some lineages it is not obvious what the precise role of these proteins is.

Mammalian inner ear hair cells function as mechanoreceptors to transduce sound and proprioreception. The morphology and development of the mammalian inner ear are very complex in nature. The major structures of the internal ear consist of the utricle and saccule, three semicircular canals, the cochlea,and the endolymphatic duct and sac. The sensory neuroepithelia are innervated by the eighth cranial nerve which consists of two parts, the vestibular and cochlear nerves. Vestibular hair cells are located in the macula of the saccule and the utricle, as well as in the cristae located in the semicircular canals. These hair cells fall into two types, Type I and Type II, and are innervated by the vestibular nerve. The hair cells of the vestibule detect linear acceleration and head position with respect to gravity. They are responsible for the sense of balance and proprioreception. The auditory hair cells in the organ of Corti located in the cochlea also fall into two categories, inner hair cells and outer hair cells. They are innervated by the cochlear nerve. These hair cells are responsible for auditory sensation. The membranous labyrinth of the inner ear first begins to form from the otic cyst and is visible in mice at E10.75. By E17.5 the gross anatomy of the inner ear is mature (Cantos et al.,2000). Many hearing impairments are caused by loss of sensory neurons and inner ear hair cells (for a review, seePetit et al., 2001). Hence, a better understanding of the genetic mechanisms responsible for the development of these structures may help us dissect the mechanisms implicated in hearing impairment or deafness.

A homology between hair cells in vertebrates and chordotonal organs in flies has been recently revealed (Hassan and Bellen, 2000). The bHLH proneural gene atonal was shown to be required for the specification of chordotonal SOPs(Jarman et al., 1993). These chordonal organs function as proprioreceptive organs and hearing devices(Eberl, 1999;McIver, 1985;van Staaden and Römer,1998), much like the hair cells of the balance organs and the auditory system. As mentioned previously, Atonal is required forsenseless expression in the SOP, and ectopic senselessexpression induces atonal expression (H. J. B., unpublished). One of the mouse homologues of atonal, Math1 (Atoh1 — Mouse Genome Informatics) is expressed in the inner ear hair cells during development. Math1-null mice die shortly after birth and lack hair cells in balance organs and cochleae(Bermingham et al., 1999). Interestingly, all the defects associated with loss of Math1 can be rescued by the fly ato gene, suggesting that they are orthologs(Wang et al., 2002). In addition, Math1 overexpression has been shown to induce hair cell growth in inner ear epithelia (Zheng and Gao, 2000). Hence, Math1, like ato, is necessary and sufficient for hair cell development in vertebrates.

Given the similarities between Math1 and atonal and the role of senseless in PNS development, we investigated the expression pattern and role of the senseless homolog, Gfi1, in hair cell development. We find that Gfi1 is expressed in many neuronal precursors as well as differentiating neurons during embryonic development. Consistent with these expression patterns, analysis of Gfi1 function in inner ear development in a previously generated Gfi1-mutant line(H. Hock, M. J. Hamblen, H. M. Rooke, D. Traver, R. T. Bronson, S. Cameron and S. H. Orkin, unpublished) revealed that mutant hair cells are initially specified and express many hair cell markers, including Math1. However, Gfi1 is required for proper differentiation and maintenance of inner ear hair cells. In Gfi1 mutant mice, the vestibular and cochlear hair cells are morphologically abnormal, hair cell organization within the sensory epithelia is aberrant, the outer hair cells in the organ of Corti express a neuronal marker, and cochlear hair cells degenerate after separation. Thus, Gfi1 is expressed in the developing nervous system and is required for the differentiation and survival of inner ear hair cells.

Gfi1 mutant and Math1 mutant mice were generated as previously described in 129/Sv × c57BL/6J backgrounds(Ben-Arie et al., 2000; H. Hock, M. J. Hamblen, H. M. Rooke, D. Traver, R. T. Bronson, S. Cameron and S. H. Orkin, unpublished). For some experiments, lines were crossed to yield mice heterozygous for Math1^β-gal and Gfi1 wild type or null.

Embryos were considered to be E0.5 days on the morning the vaginal plug was observed. To harvest the embryos, pregnant females were sacrificed by cervical dislocation and the embryos dissected out of the uterus. Regions of the yolk sac or tail were saved for genotyping. Embryos were fixed overnight in 4%paraformaldehyde, dehydrated in an ethanol series and embedded in paraffin wax for sectioning according to standard histological protocols. Sections (10μm) were collected and analyzed by in situ hybridization or immunohistochemistry. Ear tissue for postnatal stages was collected by harvesting the temporal bones of the appropriately aged pup, fixing overnight in 4% paraformaldehyde, decalcifying in 1.35 N hydrochloric acid for at least an hour, dehydrating in an ethanol series and embedding in paraffin for sectioning. Sections (10 μm) were collected and analyzed by immunohistochemistry.

The cDNA probe for Brn3c (Pou4f3 — Mouse Genome Informatics) was kindly provided by Bill Klein. Probes for each of the genes were transcribed in the antisense direction and labeled with digoxigenin using the Dig RNA Labeling Kit from Roche. Probes were hybridized to paraffin sections and detected by anti-digoxigenin antibody coupled to alkaline phosphatase. Hybridization and stringent posthybridization wash steps were performed at 65°C.

Anti-Myosin VI/VIIa was kindly provided by Tama Hasson. Anti-TUJ1 was obtained from Babco. Anti-activated Caspase 3 was obtained from R&D Systems. An anti-Gfi1 antibody was generated in guinea pig. This antibody was raised against the domain of Gfi1 between the SNAG domain and the zinc fingers(amino acids 20-256), cloned into pET28a. This domain does not display homology to Gfi1b or other proteins. It is a specific nuclear antigen that is not present in Gfi1 mutant mice and recognizes a specific band of the appropriate molecular weight on western blots of Gfi1-expressing yeast extracts and of mouse thymus protein extracts (data not shown). We used antibodies to Myosin, TUJ1 and Caspase 3 at a 1:1000 dilution, and Gfi1 at a 1:2000 dilution, and followed the ABC Vectastain directions with secondary anti-rabbit antibody (Myosin VI/VIIa and Activated Caspase 3), anti-mouse antibody (TUJ-1) or anti-guinea pig antibody (Gfi1) followed by DAB staining. Briefly, paraffin wax embedded sections were blocked in 1%H₂O₂ in methanol for 20 minutes at room temperature,rehydrated in a series of ethanols, boiled in citrate antigen retrieval solution in a microwave for 5-10 minutes, and blocked with horse serum(Vectastain) in PBS for 30 minutes at room temperature. Primary antibody was diluted in blocking solution and incubated on the section overnight at 4°C. Slides were rinsed in PBS and incubated in secondary for 30 minutes at room temperature. The slides were rinsed in PBS and incubated in Vectastain ABC solution for 30 minutes. The slides were again rinsed in PBS and the signal was detected with 2 mg/ml DAB, 0.02% H₂O₂ in PBS. Some slides were counterstained with Hematoxylin.

Mice heterozygous for the β-galactosidase cassette in the place ofMath1-coding region were bred to Gfi1 heterozygous mice to generate double heterozygotes, which in turn were crossed to obtain genotypes that were Math1 heterozygous and either Gfi1 wild type orGfi1 null. This allowed us to visualize hair cells in theGfi1 null mutants by staining the tissues for β-galactosidase. Appropriately staged mice were harvested, fixed briefly in 4% paraformaldehyde and stained overnight at 37°C for β-galactosidase according to established procedures (Ben-Arie et al.,2000). The tissue was then fixed overnight in 4% paraformaldehyde,and processed for paraffin wax embedding and sectioning or imaged immediately for whole mounts.

Staged cochleae were dissected and fixed in 0.1 M cacodylate buffer, 1%glutaraldehyde and 4% formaldehyde at 4°C for 2 hours. Specimens were then rinsed in 0.1 M cacodylate buffer and post fixed in 1% osmium tetroxide in cacodylate buffer at 4°C overnight. Samples were again washed in cacodylate buffer and rinsed with distilled water. Specimens were stained with 4% uranyl acetate for 3 hours and again washed in distilled water. Specimens were then dehydrated for 15 min each in a series of ethanols: 50%, 70%, 80%,90%, 95% and 100% (twice), and then finally 100% overnight at room temperature. Samples were then rinsed in ethanol followed by rinsing in propylene oxide and embedded in scipoxy 812 resin with dodenyl succinate anhydride and nadric methyl anhydride. Semithin sections (0.5 μm) were obtained and then thin sections (50 nm) were obtained and grid stained with 4%uranyl acetate and 2.66% lead acetate and observed on an electron microscope.

As Gfi1 and Gfi1b have not been reported to be expressed in neurons or neuronal precursors, we first searched for othersenseless homologues that are expressed during embryonic mouse development. Two lines of evidence suggest that there may be no additionalsenseless homologs to Gfi1 and Gfi1b. First, we did not identify another gene with similar sequence homology in BLAST searches of the human and mouse genomes. Second, RT-PCR with degenerate primers corresponding to the highly conserved zinc-finger sequences allowed us to identify transcripts in E10.5, E11.5, E12.5 and E13.5 embryos and adult thymus. Sixty-four clones were sequenced and all corresponded to Gfi1and Gfi1b (data not shown). These data show that Gfi1 andGfi1b are expressed in embryos and that it is unlikely that othersenseless homologues exist in mice.

To determine when and where Gfi1 is expressed, we carried out in situ hybridization and immunohistochemistry in embryos. As shown inFig. 1A, Gfi1 mRNA is expressed in E12.5 embryos in the PNS, CNS and many sensory tissues. More specifically, Gfi1 mRNA is expressed in the developing brain, optic epithelia, dorsal root ganglia, otic vesicle, trigeminal and vestibulo-cochlear ganglia, and gut epithelia(Fig. 1A). At later stages expression is also seen in many sensory organs such as the developing eye(Fig. 1B), presumptive Merkel cells (Fig. 1D,E), cells of the nasal epithelia (Fig. 1G),epithelia of the tongue (Fig. 1H), as well as in small clusters of neuroepithelial precursor cells in developing lung (Fig. 1I; D. W., unpublished), and many cells of the developing thymus(Fig. 1K). Hence, Gfi1mRNA is widely expressed in epithelia in which sensory cells are specified(tongue, nasal epithelia, gut, lung and eye), as well as in the developing brain and PNS ganglia. However, the Gfi1 protein has a more restricted expression pattern and localizes to several specialized sensory cells of the PNS. Gfi1 protein is present in the eye(Fig. 1C), the presumptive Merkel cells (Fig. 1F) and the lung (Fig. 1J). We did not detect Gfi1 protein expression in the brain or any of the ganglia, which is where we see Gfi1 mRNA expression (see Discussion).

Whereas Gfi1 is expressed in a variety of tissues, we chose to focus our analysis of the role of Gfi1 in the developing ear. As shown in Fig. 2, Gfi1mRNA and protein are expressed in the ear throughout development.Gfi1 mRNA is expressed in the developing otic vesicle and several ganglia, including the vestibulo-cochlear ganglia as early as E12.5(Fig. 2A). Gfi1 protein can be detected at E12.5 in a more restricted pattern in the hair cell precursors in the otic vesicle, where Math1 is expressed(Bermingham et al., 1999). This timing is concomitant with the initial generation of hair cells during inner ear development as sensory hair cell precursors undergo their terminal mitosis between E11 and P2 (Ruben,1967). Specification and differentiation of hair cells is initiated just after terminal mitosis between E12-P2(Zheng and Gao, 1997), and hair cells then mature and grow sterociliary bundles during differentiation. By E14.5, we find Gfi1 mRNA and protein in the developing vestibular organs in positions analogous to the newly developed hair cells(Fig. 2B,E). By E16.5 to E18.5,when the sensory structures are well defined, Gfi1 mRNA and protein are clearly detected in the vestibular organs(Fig. 2C) as well as in the organ of Corti in the cochlea (Fig. 2D,F arrows), where it localizes to hair cells. Gfi1 mRNA expression is also detected in the cochlear ganglia(Fig. 2D, asterisk); however,Gfi1 protein is not expressed in the cochlear ganglia at any time (data not shown). Hence, the temporal and spatial expression pattern of Gfi1 protein in the ear correlates with the specification of hair cells and is very similar if not identical to the expression pattern of Math1(Bermingham et al., 1999;Chen et al., 2002).

We previously established a mouse line deficient for Gfi1 (H. Hock, M. J. Hamblen, H. M. Rooke, D. Traver, R. T. Bronson, S. Cameron and S. H. Orkin, unpublished). This Gfi1 allele has part of the 5′ UTR along with the entire first and second exon, as well as part of the third exon deleted. This deleted region contains the entire SNAG transcriptional repression domain but not the zinc fingers and creates a severe loss-of-function or null allele. We have found heterozygous mice to be phenotypically indistinguishable from wild-type littermates at all stages in our assays. The mutant mice are viable for 3-6 months. The mutants look similar to wild type and heterozygous littermates until about postnatal day 10(P10). By P10, the mutants no longer continue to grow at the rate of their littermates (H. Hock, M. J. Hamblen, H. M. Rooke, D. Traver, R. T. Bronson, S. Cameron and S. H. Orkin, unpublished;Karsunky et al., 2002b) and ataxic behavior becomes apparent. This ataxia and the differences in size between mutant and heterozygous littermates increase in severity with age (see video athttp://flypush.imgen.bcm.tmc.edu/lab/deeann/mouse-video1.avi). The mutant animals display several behavioral abnormalities suggestive of inner ear defects including hyperactive circling, head tilting, ataxia and lack of a proper startle response to loud noises. This phenotype coupled withGfi1 expression in the inner ear sensory epithelia suggests inner ear defects.

As the overall gross morphology of the inner ear appeared normal at P0 inGfi1 mutant mice (data not shown), we immunocytochemically stained ear epithelia with several hair cell markers to identify differentiation defects. Myosin VI/VIIa is an early marker for hair cell differentiation and initiation of expression of this marker occurs properly at E13.5 in the mutant balance organs and cochlea (Fig. 3A,B) (Hasson et al.,1997). However, as shown inFig. 3A,B, the mutant hair cells in the utricle are thinner and more elongated than the wild-type cells,and there are two to three layers of myosin VI/VIIa-positive cells in the mutant instead of the single layer observed in wild-type epithelia. Anti-myosin VI/VIIa staining of the utricle at E14.5(Fig. 3C,D) also shows abnormal hair cell morphology and layering in the mutant. The vestibular organs normally show a straight line of hair cells at the edge of the lumen, but the mutant hair cells do not form this straight line as the hair cells are present in the support cell layer. Similarly, the organization of the auditory hair cells in the organ of Corti is also aberrant. The characteristic three rows of outer hair cells and single row of inner hair cells are not present in the mutant. Serial sections of the organ of Corti often show that one of the outer rows of cells is lacking as shown with anti-myosin VI/VIIa staining at E16.5(Fig. 3E,F). Similarly,Math1 in situ hybridization at E17.5(Fig. 3G,H), and Brn3cin situ hybridization at E18.5 (Fig. 3I,J) each show one inner hair cell and two outer hair cells. Thus, though some hair cells appear to be missing, hair cell-specific markers are expressed and maintained throughout inner ear hair cell development inGfi1 mutant mice. In addition, we find that the remaining mutant hair cells in the organ of Corti are not properly innervated. As shown inFig. 3K,L, staining with anti-TUJ1, a marker for β-tubulin in neurons, reveals a very different staining pattern in mutant animals. In wild-type hair cells(Fig. 3K), the cochlear neurons synapse with the base of the three outer hair cells forming a cup-like staining pattern at the base of each outer hair cell (indicated by green arrows). In the mutant outer hair cells, the anti-TUJ1 labels the entire cell body including the cytoplasm. However, the base of the cells where synaptic sites are normally seen as cup-like structures in wild-type embryos are not,or are barely visible (indicated by green arrows inFig. 3L). This aberrant pattern is not observed for the inner hair cells. Hence, as the cytoplasm of the outer hair cells in Gfi1 mutants stain with TUJ1 antibody, the outer hair cells express a neuronal marker that is not normally expressed in these cells. We conclude that based on aberrant morphology of vestibular hair cells and ectopic expression of the neuronal marker TUJ1 in outer hair cells, hair cell differentiation is affected in Gfi1 mutants.

In flies, proneural gene expression is required for initiation ofsenseless expression, and senseless expression is required for maintenance of proneural gene expression(Nolo et al., 2000;Frankfort et al., 2001). To determine if a similar relationship exists between Math1 andGfi1, we investigated the expression of Gfi1 inMath1 mutants and Math1 in Gfi1 mutants. AsMath1 and Gfi1 expression overlap temporally and spatially in the inner ear epithelia, Math1 may be required for Gfi1expression. We therefore tried to assess Gfi1 expression at the earliest stage of development, just prior to hair cell formation when hair cell precursors are specified (E12.5). At this stage, Gfi1 mRNA expression in the Math1 mutant is present in the entire otic epithelia similar to wild-type controls(Fig. 4A,B). However, Gfi1 protein expression is drastically reduced or absent(Fig. 4C,D). These observations indicate that initiation of Gfi1 mRNA expression is not dependent onMath1, but that Gfi1 protein expression is Math1 dependant. At later stages in development (E14.5, E16.5 and E18.5), Gfi1 mRNA and protein are both drastically reduced or absent in Math1 mutants(Fig. 4E,F and data not shown). This may imply that Math1 is required for Gfi1 expression,or, alternatively, that the cells in which Gfi1 is expressed are not specified (Chen et al., 2002). It also indicates that Gfi1 mRNA and protein expression is confined to hair cells at later developmental stages.

Is Gfi1 required to maintain Math1 expression? To monitorMath1 expression in Gfi1 mice, we used a mouse containing the β-galactosidase-coding region that replaced the entireMath1-coding region. β-galactosidase staining of heterozygousMath1 animals faithfully mimics Math1 expression, whereas hair cell specification appears normal(Ben-Arie et al., 2000). Mice that were Math1 heterozygous(Math1^+/βGal) and Gfi1 wild-type or null mutant were derived. In the Gfi1 mutants,Math1^βGal expression is present in all inner ear sensory epithelia (Fig. 4G-J, Fig. 5A-L). However, sections of P0 saccules stained with β-galactosidase show some disorganization and some cells appear to be present in the supporting cell layer (Fig. 4G,H). The mutant hair cells also seem to have less well organized stereocilli than do wild-type cells. Note, however, that the cristae appear to be less affected or unaffected as they have a very similar morphological appearance to wild-type cristae (Fig. 4I,J). Similar data were also observed with Math1 in situ hybridization inGfi1 mutant organ of Corti (Fig. 3G,H). In summary, we found no obvious changes in Math1expression pattern in Gfi1 mutants. These data suggest that there is no dependence of Math1 on Gfi1 in mouse. This is in contrast to what we observed in fruit fly between atonal andsenseless (Frankfort et al.,2001; Nolo et al.,2000).

The Math1^βGal/+; Gfi1 mice provided us with a convenient tool to follow hair cell development in wholemounts of organ of Corti and asses differences in apical and basal areas of the cochlea. Normally, by E15.5 Math1/β-galactosidase expression is visible in the developing hair cells of the organ of Corti and rows of hair cells are beginning to differentiate in a basal-to-apical gradient. As shown inFig. 5A,B, by E15.5,Math1/β-galactosidase positive cells are present in wild-type and mutant embryos. However, the rows are not as clearly defined in the mutant as in the wild type. By E17.5, the wild-type hair cells have formed the characteristic one row of inner hair cells and three parallel rows of outer hair cells (Fig. 5C). At E17.5,the mutant hair cells are disorganized and less numerous in the basal cochlea(Fig. 5D). As shown inFig. 5E-H, by P0 the loss of hair cells has progressed in a basal to apical gradient.Fig. 5E shows the orderly arrangement of wild-type hair cells at P0, whereas the basal cochlea of theGfi1-null mice has lost the majority of its hair cells as gauged byMath1 expression (Fig. 5F). The medial cochlea is also severely affected, but the inner hair cells are still present (Fig. 5G). At P0 the apical cochlea shows little to no loss of hair cells, but does exhibit a disorganization similar to that seen in basal cochlea as early as E15.5 (Fig. 5H, compare with Fig. 5B). By P3 the loss of hair cells in the basal to apical gradient is more severe in the mutants (Fig. 5I-L). The basal cochlea has few hair cells by P3(Fig. 5J). The medial cochlea still retains the majority of inner hair cells but has lost almost all outer hair cells (Fig. 5K). Even the apical cochlea is beginning to show drastic reduction in the number of hair cells by P3, though again it appears as though the outer hair cells degenerate first. Hence, it appears that the inner and outer hair cells are initially specified by E15.5 but are subsequently lost in a basal-to-apical gradient. In all cases, the outer hair cells in a given region degenerate prior to the inner hair cells.

As the mutants age, the organ of Corti becomes unrecognizable. Analysis of serial sections stained with Hematoxylin and Eosin indicates that by P14 all cochlear hair cells and most support cells have disappeared in mutant animals(Fig. 6A,B). However, despite the rapid degeneration of cochlear hair cells, the hair cells in the vestibular organs do not degenerate, but remain unorganized(Fig. 6C,D). Note the separation of hair cells and support cells in the saccule of the wild-type mouse (Fig. 6C). This layering is again not as clearly defined in the mutant when compared with wild type(Fig. 6D).

As shown in Fig. 2D(asterisk), Gfi1 mRNA is also expressed in the cochlear ganglion neurons, although we did not observe Gfi1 protein expression. We therefore examined number and morphology of the cochlear ganglion neurons. As shown inFig. 6E,F, at P7, both wild-type (Fig. 6E) and mutant(Fig. 6F) cochlear ganglion neurons show similar cell densities and a low level of apoptosis as indicated by anti-activated-caspase 3 (C3) staining. Low levels of apoptosis at P7 in wild-type cochlear ganglion neurons has been previously observed(Kamiya et al., 2001). At P21,the cochlear ganglion neurons in the wild-type(Fig. 6G) and mutant mice(Fig. 6H) show slightly different cell densities and quite a few mutant cells express activated-caspase 3, suggesting that these cells undergo cell death by apoptosis. We did not observe apoptosis in the wild-type mouse at P21. By 5 months of age there is a dramatic reduction of neurons in the cochlear ganglion, as seen by Hematoxylin and Eosin staining(Fig. 6I,J). Hence, cochlear ganglion neuron degeneration occurs after hair cell loss and is progressive.

To determine the ultrastructural defects in the cells of the organ of Corti, we carried out transmission electron microscopy (TEM). TEM shows abnormal hair cell morphology at E18.5 in the mutant mice and confirms the disorganization of hair cells in the organ of Corti. The stereocilli are nicely preserved and easily identifiable in the wild type outer hair cells(Fig. 7A,C), but the stereocilli of the mutant hair cells are not well preserved and are barely visible in some cells. Some of the mutant outer hair cells also display typical morphological signs of apoptosis, such as shrinkage of the cell body,extensive blebbing and vacuolization (Fig. 7B,D). Also, consistent with an apoptotic mechanism of cell death,the mutant mitochondria appear indistinguishable from the wild-type mitochondria, in contrast to what happens when cells die by necrosis. These data suggest that the outer and inner hair cells die by apoptosis.

As both senseless and pag-3 play a role in the development of the nervous system, we wished to determine if Gfi1 andGfi1b play a role in nervous system development in vertebrates. However, expression of Gfi genes in nervous system development had not been documented. We therefore determined the developmental expression patterns for both Gfi1 and Gfi1b. We found Gfi1 mRNA to be expressed in a variety of CNS, PNS and sensory epithelia, whereasGfi1b seems to be predominately expressed in the fetal liver with minimal or no expression in the nervous system (data not shown). Hence, we chose to focus our analysis on Gfi1.

Gfi1 mRNA is present in a variety of tissues during development. It is expressed in the CNS, a variety of ganglia and many specialized sensory cells of the PNS. However, the Gfi1 protein expression pattern is more restricted. We detect Gfi1 protein primarily in specialized sensory cells of the PNS. However, we did not detect Gfi1 protein in the CNS or any ganglia. The difference in Gfi1 mRNA and protein expression may have several possible explanations. First, senseless mRNA is also more widespread than its protein expression pattern in the fly(Nolo et al., 2000). Thus, it is likely that this is a real phenomenon and not just an artifact of in situ or immunohistochemical analysis. Alternatively, the RNA and/or the protein stability may vary from cell to cell type. Third, it is possible that Gfi1 is only translated under specific conditions, i.e. in the presence of Math1.

The temporal and spatial distribution of Gfi1 transcripts in many cells and epithelia overlaps with that of many bHLH gene expression patterns. In most tissues in which Gfi1 is expressed, there is a corresponding bHLH gene that may regulate/control Gfi1 expression. For example,Mash1 (Ascl1 — Mouse Genome Informatics) is expressed in the developing olfactory epithelia (Cau et al., 1997), the neuroendocrine cells of the lung(Borges et al., 1997;Ito et al., 2000) and the tongue (Seta et al., 1999),where we observe Gfi1 mRNA expression. Math1 is expressed in the developing ear epithelia (Bermingham et al., 1999), gut epithelia(Yang et al., 2001) and Merkel cells (Ben-Arie et al., 2000)where we observe Gfi1 mRNA and protein expression. Math5 is expressed in the developing eye and retinal ganglion cells(Wang et al., 2001).Neurod1 is expressed in the ear epithelia, as well as the ganglia that innervate the ear (Liu et al.,2000). These bHLH genes are homologous to the DrosophilabHLH proteins Achaete, Scute, Atonal or Amos and are required for the specification of subtypes of cells. Similarly, other bHLH proteins such as the neurogenins have been shown to be expressed and required in some of the PNS ganglia where Gfi1 mRNA is expressed(Ma et al., 1997;Ma et al., 2000b). These observations suggest that, similar to fruit flies, Gfi1 expression may be regulated by bHLH genes.

Because we found Gfi1 to be expressed in the developing ear, we chose to focus our analysis on the developing ear to test potential interactions of Gfi1 with the bHLH gene Math1(Bermingham et al., 1999). We assessed the expression pattern of Gfi1 in Math1 mutants andMath1 expression in Gfi1 mutants. Interestingly,Math1 expression is unaffected in Gfi1 mutants. As atonal positively regulates its own expression in the fly(Sun et al., 1998), it is possible that Math1 may also regulate its own expression in the mouse. This provides a potential explanation as to how Math1expression may be maintained in the Gfi1-deficient mouse. This data may also suggest that Gfi1 is not required for maintenance ofMath1 expression. However, Math1 is required for Gfi1 protein expression, but not required for initial Gfi1 mRNA expression. Hence, it remains to be established how Gfi1 mRNA and protein expression is precisely controlled.

Our data support a model where Gfi1 is downstream of Math1, but may not support a model in which Gfi1 functions in a positive feedback loop with Math1. However, it is possible that another bHLH gene expressed in the vertebrate ear epithelium prior to Math1expression is required for Gfi1 mRNA expression. The identity of this putative bHLH protein is unknown. However, the existence of such factor is suggested because in Math1-null mutants, hair cell precursors form a zone of non-proliferating cells that delineate the sensory primordium within the cochlear anlage, and a significant subpopulation of these precursors die because of apoptosis in a basal-to-apical gradient(Chen et al., 2002). The fact that these cells die instead of becoming support cells indicates that these cells have a different fate than their surrounding cells in the absence of or prior to Math1 expression. We surmise that this difference is induced by the presumptive factor. Our data are consistent with the idea that this factor is upstream of both Math1 and Gfi1. Such a factor could function similar to a proneural gene as it might be initially expressed in a cluster of cells rendering them competent to become neural cells and then refine to a specific cell that also expresses Math1 and Gfi1to become a hair cell (Chen et al.,2002; Hassan and Bellen,2000). This factor could be responsible for the initial expression of Gfi1 and explain why in a Math1 mutant we observeGfi1 mRNA expression early on. Candidate bHLH transcription factors expressed prior to Math1 in ear development include Neurod1(Liu et al., 2000) and neurogenin 1 (Ma et al.,2000a). Both are required for proper development of the inner ear,but Neurod1 and neurogenin 1 mutant mice display very different phenotypes from the ones we observe in Gfi1 mutants(Liu et al., 2000;Ma et al., 1998), suggesting that neither Neurod1 nor neurogenin 1 corresponds to the proposed factor.

The hair cells of the inner ear seem to be specified properly as they express many of the typical hair cell markers such as myosin VI/VIIa,Math1 and Brn3c. Thus, Gfi1 is not required for the specification of hair cells as they are formed in both the vestibule and the cochlea. However, the loss of Gfi1 seems to affect the vestibular and cochlear hair cells differently. In the vestibule, the hair cells are morphologically abnormal at the earliest stages of hair cell differentiation and at all subsequent stages. In addition, hair cells are not specifically localized to a lumenal monolayer, and are more variable in size and shape. This disorganization of hair cells in the vestibule may account for the ataxic behavior of the mice. In the cochlea, Gfi1 is required for the organization and maintenance of both inner and outer hair cells. Although the mutant hair cells seem to be specified in the developing organ of Corti as early as E15.5 and express typical hair cell markers, they are disorganized. In addition, the outer hair cells express the neuronal marker TUJ1 at E17.5. This abnormal/ectopic TUJ1 expression may indicate a partial transformation of outer hair cells into neurons, or the de-repression of a single neuronal marker. It is thus possible that these cells are part hair cell and part neuron, and this ambiguity could trigger apoptosis. In fact, the outer cochlear hair cells are the first to disappear starting at E17.5. Based on TEM analysis, we see some of the classical morphological signs of apoptosis in the mutant hair cells at E18.5, including shrinkage of the cell body, blebbing and vacuolization. Whole-mount analysis of the cochlea indicates that this loss of hair cells occurs in a basal to apical gradient and affects outer hair cells prior to inner hair cells in any given region of the cochlea. The hair cells and support cells of the organ of Corti continue to disappear until the entire organ of Corti has been destroyed by P14. Because wild type mice do not perceive sound until after P12 (Kamiya et al., 2001), and because Gfi1 null mice have no hair cells by P14, we assume that these mice are deaf, which is in agreement with the lack of a startle response to loud noises.

Upon degeneration of the organ of Corti, the cochlear ganglion neurons also degenerate. This degeneration is progressive, beginning after P7 but fairly extensive by five months of age. As Gfi1 mRNA is expressed at low levels in the neurons, it may be directly required for neuronal survival. However, Gfi1 may not be directly responsible for neuronal loss because Gfi1 protein is not expressed in the cochlear ganglion, and cochlear neurons normally die after degeneration of hair cells, presumably because of the withdrawal of trophic support (Dodson,1997; Lefebvre et al.,1992). Thus, it seems likely that the loss of cochlear ganglion neurons is secondary to the loss of hair cells in the organ of Corti.

There are other mutant mice with similar, yet distinct phenotypes.Brn3c-deficient mice have a similar vestibular phenotype with a small number of hair cells retained in the support cell layer in the vestibular sensory epithelia (Xiang et al.,1998). The hair cells in Brn3c-null mice are also initially specified, but fail to mature and form stereocilli. TheBrn3c-deficient hair cells then rapidly degenerate by apoptosis(Xiang et al., 1997). The loss of hair cells occurs in the organ of Corti as early as E17.5 with nearly complete loss by P5. This is similar to the Gfi1 mutant, but unlike in the Gfi1 mutants, this degeneration is also detected in the vestibule as early as E18.5. In Brn3c mutants, the loss of hair cells is then followed by a loss of the cochleo-vestibular neurons with a substantial loss as early as P4, earlier than in the Gfi1 mutants. Note that Brn3c expression is maintained in all inner ear sensory epithelia of Gfi1 mutant mice(Fig. 3I,J; data not shown).Barhl1-deficient mice also show a progressive degeneration of cochlear hair cells (Li et al.,2002). This degeneration is much slower than in Gfi1mutants, occurring roughly from P6 to 2 months of age for outer hair cells and between six months and ten months for inner hair cells. Interestingly, the outer hair cells degenerate first in an apical-to-basal gradient, and the inner hair cells degenerate second in a reverse basal-to-apical gradient.

Gfi1 has a novel phenotype with respect to its effect on the inner ear. The fact that different types of hair cells expressing Gfi1 in the different sensory organs are affected differently is mirrored inDrosophila senseless mutants and C. elegans pag-2 mutants. In the Drosophila embryonic PNS, some mutant neuronal subtypes undergo apoptosis like the auditory hair cells in Gfi1 mutants(Nolo et al., 2000). In C. elegans, other neuronal subtypes are improperly differentiated or abnormal such as the BDU interneurons, the Pn.aa neuroblasts, and the VA and VB motoneurons similar to the vestibular hair cells in the Gfi1mutants (Cameron et al.,2002).

Hence, it is possible that the function of Gfi1 and its homologs is dependent on the tissue in which it is expressed as it may have a variety of functions depending on its environment. It is most likely thatGfi1 plays a variety of different roles as a transcriptional repressor or activator. Thus, different genes are repressed or activated in different tissues resulting in a variety of functions. A more precise explanation as to the function of Gfi1 will have to await the further analysis of Gfi1 function in other tissues and the identification of direct Gfi1 target genes and interaction partners.

We thank Fred Pereira, Vincent Wang, Qi (Melissa) Yang, Bill Klein, Tama Hasson, Ruth Anne Eatock, Ming-Jer Tsai, Loretta Doan, Hamed Jafar-Nejad,Mahira Zaheer and Suzanne Paylor for discussions, reagents and technical assistance. Funding was provided by NASA grants NAG2-1496 and NAG2-1355 to H. J. B., and NIH NRSA grant F32 DC 5109 to D. W. Support was provided by the cores of the Baylor Mental Retardation Research Centers HD 24064 to H. Y. Z.,a Centers of Excellence in Hematology Award to S. H. O., and by a Cancer Research Foundation of America Hope Street Kids grant to H. L. G. H. Y. Z., H. J. B. and S. H. O. are investigators of the HHMI.