Functional conservation of atonal and Math1 in the CNS and PNS

During evolution primitive cellular machinery is constantly being co-opted for new purposes. An example of this strategy is the development of balance and audition organs. In Drosophila, atonal (ato) controls the development and function of chordotonal organs (CHOs), the sensory elements which provide proprioception and vibration sense (Eberl, 1999; McIver, 1985; van Staaden and Römer, 1998). CHOs populate the peripheral nervous system (PNS) in the body wall and joints (thorax, abdomen, sternum, wings, legs) and antennae (Moulins, 1976), providing the fly with sensory information much as touch and mechanoreceptors do in vertebrates (McIver, 1985; Moulins, 1976). Boyan (1993) proposed that, in the course of evolution, different CHOs became specialized for hearing in different insects. This hypothesis was recently confirmed by van Staaden and Römer (1998). In Drosophila, CHOs in the Johnston organ, located in the second antennal segment, function in near field hearing (Dreller and Kirschner, 1993; Eberl, 1999) and negative geotaxis (B. H. and H. J. B., unpublished data).

During development ato is expressed in a cluster of progenitor cells from which the CHO founder cells are selected (Jarman et al., 1993). It likely functions by regulating the expression of genes necessary for the specification and development of the CHO lineage, as it encodes a basic helix-loop-helix protein (bHLH) that dimerizes with the Daughterless protein and binds to E-box sequences (Jarman et al., 1993). CHO specificity is encoded by the ato basic domain, which is required for DNA binding in bHLH proteins (Chien et al., 1996; Davis et al., 1990; Jarman and Ahmed, 1998; Vaessin et al., 1990). ato is both necessary and sufficient for the generation of CHOs in the fly: loss of ato function leads to the loss of CHOs, while ectopic ato expression causes ectopic CHO formation (Jarman et al., 1993).

Because of ato’s importance in Drosophila peripheral nervous system (PNS) development, a number of laboratories have searched for ato homologs in various species. Interestingly, the mouse homologs (Math4A/neurogenin2, Math4C/neurogenin1, Math3 and Math5) are expressed not just in the PNS, but in the central nervous system (CNS) at various developmental stages and in different anatomical structures (Fode et al., 1998; Ma et al., 1998; Takebayashi et al., 1997).

Mouse atonal homolog 1 (Math1) is one of ato’s closest known homologs, with 82% amino acid similarity in the bHLH domain and 100% conservation of the basic domain that determines target recognition specificity (Ben-Arie et al., 1996; Chien et al., 1996). Math1 is transiently expressed in the CNS starting at embryonic day 9 (E9) in the dorsal portion of the neural tube. Math1 is also expressed in the rhombic lip of the fourth ventricle of the brain, where cerebellar granule cell precursors are born at E13-15 (Alder et al., 1996). Upon proliferation and differentiation, these progenitor cells migrate to form the external granule layer (EGL) of the cerebellar primordia (Hatten and Heintz, 1995). Proliferating EGL cells continue to express Math1 during the first three postnatal weeks, until shortly before they migrate to their final adult destination to generate the internal granule layer (IGL) of the cerebellum (Akazawa et al., 1995; Ben-Arie et al., 1996). Another group of cells, a small population of neuronal precursors in the dorsal spinal cord, expresses Math1 during E10-E14 (Akazawa et al., 1995; Ben-Arie et al., 1996). These precursor cells also express the LIM homeodomain proteins (LH2A and LH2B), markers of the D1 class of commissural interneurons (Lee et al., 1998). Helms and Johnson (1998) reported that lacZ expression under the control of Math1 regulatory elements reproduced Math1 expression patterns in the developing cerebellum and spinal cord, and demonstrated that Math1 is expressed in precursors that give rise to a subpopulation of dorsal commissural interneurons.

To determine the in vivo function of Math1, we previously generated mice (Math1^m1/m1) lacking the MATH1 protein. This null mutation causes major cerebellar abnormalities: lack of granule cell proliferation and migration from the rhombic lip at E14.5, and absence of the entire EGL at birth (Ben-Arie et al., 1997). It is not clear whether the agenesis of cerebellar granule neurons is due to failure of progenitor specification or the cells’ inability to proliferate and/or differentiate. Neonates cannot breathe and die shortly after birth, but there are no gross defects in any cranial nerves or brain stem nuclei that could explain respiratory failure.

Although ato and Math1 share a high degree of sequence conservation, there is an apparent discrepancy between their expression patterns and the consequences of their loss of function. Whereas ato is expressed primarily in the PNS of the fly and its absence causes loss of almost all CHOs (Jarman et al., 1993), Math1 is expressed in the CNS and its loss leads to absence of cerebellar granule neurons, the largest neuronal population in the CNS (Ben-Arie et al., 1997). To better understand the functional relations between ato and Math1, we generated a second Math1 null allele in mice (Math^{β-gal/β-gal}) by replacing the Math1 coding region with a β-galactosidase gene (lacZ) and searched for CNS expression of ato in the fruit fly. Our results confirm the functional link between ato and Math1: ato is expressed in the fly brain, and lacZ expression under the control of Math1 regulatory elements (Math1/lacZ) not only replicated the known expression pattern in the CNS (i.e., the neural tube, spinal cord and cerebellum), but appeared in many other cells of the murine PNS. Overexpression of Math1 in Drosophila caused ectopic CHO formation, providing further insight into the relationship between the two genes.

To delete the entire coding region of Math1, we generated a targeting construct that contained the 5′and 3′ genomic flanking fragments as described previously (Ben-Arie et al., 1997) flanking a pSAβgal/PGK-neo cassette (Friedrich and Soriano, 1991). The construct is designed so that lacZ expression is driven by endogenous Math1 control elements, while an independent PGK promoter drives the expression of the selectable marker neo.

The construct was electroporated into ES cells and selection for neo was achieved with G418. Fourteen out of 76 (18%) clones underwent homologous recombination. Genotyping of ES cells, yolk sac and tail DNA was performed using Southern analysis of EcoRI digested DNA and probes previously described (Ben-Arie et al., 1997).

Embryos were staged by designating the morning of the vaginal plug as E0.5. Embryos were dissected out of the uterus, separated from extraembryonic membranes, and placed in cold phosphate-buffered saline (PBS). The embryos were then fixed in 4% paraformaldehyde (PFA) in PBS for 30 minutes, and washed in cold PBS. Yolk sacs or tails were collected before fixation for DNA extraction and genotyping. Equilibration to improve the penetrability of the staining reagents was performed in 0.02% NP40, 0.01% sodium deoxycholate in PBS for 10 minutes at room temperature. Whole-mount staining with X-gal (Bonnerot and Nicolas, 1993) was performed for 16-24 hours at 30°C while shaking in the same equilibration buffer, which also contained 5 mM potassium ferricyanide, 5 mM potassium ferrocyanide, and 40 mg/ml X-gal (dissolved in DMSO). When the desired intensity of staining was achieved, usually within 18 hours, embryos were washed in PBS, postfixed for 30 minutes in buffered formalin, serially dehydrated in 25, 50 and 70% ethanol, and stored at 4°C.

For histological analysis embryos were further dehydrated in 80, 90 and 100% ethanol, treated in Histoclear (National Diagnostics), and embedded in Paraplast (Oxford Labware). 7-20 μm sections were cut using in a microtome (Microme). Counterstaining was performed using nuclear fast red (Vector Laboratories). Immunohistochemistry was performed as detailed previously (Ben-Arie et al., 1997). Antibodies: anti-cytokeratin 18 (DAKO) 1:20; anti-human chromogranin A (DAKO) 1:100; anti-MATH1 (see below) 1:200.

An EcoRI-HindIII fragment encoding the N-terminal 156 amino acids of the MATH1 open reading frame (Math1Δ) was cloned into the pET 28a+ expression vector (Novagen). Math1Δ fragment was expressed as a His tag fusion protein. Soluble MATH1Δ protein was purified according to His-tag kit specifications (Novagen) and 2 mg of protein were used to immunize Chickens (Cocalico Biologicals Inc.).

Ta/Ta females were kindly provided by Dr P. Overbeek (Baylor College of Medicine, Houston, TX). These were time-mated with Math1+/β-gal males, and embryos were harvested at E16.5. Each pup’s gender was determined by PCR on tail DNA, using primers (forward TGAAGCTTTTGGCTTTGAG, and reverse CCGCTGCCAAA-TTCTTTGG) that yielded a 320 bp product from chromosome X, and a 300 bp product from chromosome Y (Liu et al., 1999). Amplification conditions were: 92°C/1 minute, 55°C/1 minute, 72°C/1 minute for 32 cycles, with an initial denaturation step of 94°C/7 minutes and last extension step of 72°C/7 minutes. Amplification products were separated on 2% agarose gels. X-gal-stained embryos were scored independently by 2 individuals, and only then were results matched with the determined gender.

yw flies were transformed with a UAS-Math1 construct as described by Brand and Perrimon (1993). To overexpress Math1 in wild-type flies, yw; UAS-Math1 flies were mated to HS-Gal4 flies. The progeny were heat shocked as described previously (Jarman et al., 1993). To rescue the loss of chordotonal organs in ato mutant flies, w; UAS-Math1/UAS-Math1; ato1/TM6 flies were crossed to w; HS-Gal4/CyO; ato1/TM6 flies. Embryos were collected for 3 hours, aged for 3 hours, heat shocked for 30 minutes at 37°C and allowed to develop for the next 12-15 hours. Embryos were fixed in 4% formaldehyde in PBS with 50% heptane. Embryos were washed with 100% ethanol, transferred to PBT and stained with mAb 22C10 as described previously (Kania et al., 1995) to detect PNS neurons. Chordotonal neurons were identified by their distinct morphology and position.

Third instar larval brains were dissected and fixed in 10% formaldehyde in PBT. Brains were washed in PBT. In situ hybridization was performed as described by Tautz and Pfeifle (1989).

The targeting construct, containing a lacZ cassette and a PGK-neo cassette (Fig. 1A), was used to replace the Math1 coding region (see Materials and Methods). The targeting construct was electroporated into embryonic stem (ES) cells; 14/76 (18%) clones exhibited correct homologous recombination at the Math1 locus (Fig. 1B).

Three ES cell lines carrying the Math1^+/β-gal allele were injected into host blastocysts to generate chimeric mice. Math1^+/β-gal mice were identified and intercrossed to generate homozygotes (Fig. 1C). The Math1 deletion was confirmed by Southern analysis using both flanking and internal probes (Fig. 1A).

Analysis of the developing cerebellum at E14.5 and postnatal day 0 (P0) in Math1^+/β-gal and Math1Δβ-gal/β-gal Δmice showed that the expression pattern of the lacZ gene faithfully reproduced the Math1 expression pattern observed by RNA in situ hybridization analysis (Akazawa et al., 1995; Ben-Arie et al., 1996) (Fig. 2A,B,E,G). Moreover, the cerebellar phenotype in Math1Δβ-gal/β-gal Δmice (Fig. 2F,H) was identical to that observed in Math1 null mice (Ben-Arie et al., 1997). At E14.5, the precursors of the EGL are present in the rhombic lip from which they migrate over the cerebellar anlage to populate the EGL (Fig. 2E). Mutant mice displayed far fewer of these cells than heterozygous mice (Fig. 2F). At P0, the neurons of the external granule layer (EGL) were completely lacking (Fig. 2H).

Math1/lacZ expression in the developing hind brain and spinal cord similarly reproduced the expression pattern of Math1 (Fig. 2C,D). The only notable difference between the expression patterns established by in situ hybridization and X-gal staining is that β-galactosidase expression persists in differentiating or migrating cells of the spinal cord because of the stability of the β-GAL protein (Fig. 2D). In summary, the neural tissue expression pattern and cerebellar phenotype associated with the replacement of the Math1 coding region by lacZ is consistent with previously published data on Math1 expression (Akazawa et al., 1995; Ben-Arie et al., 1997, 1996; Helms and Johnson, 1998), demonstrating that the endogenous control elements were not disrupted by insertion of the lacZ gene. Moreover, many previously undetected clusters of lacZ-expressing cells became apparent upon X-gal staining of whole embryos and sections in Math1^+/β-gal mice (see below). It is likely that limitations in the spatial resolution of RNA in situ hybridization techniques used to detect the transcript in earlier studies prevented these sites of expression from being discerned (Akazawa et al., 1995; Ben-Arie et al., 1996). Alternatively, the stability of the lacZ gene product and the increased sensitivity due to signal amplification allowed us to identify sites of relatively low expression levels.

The sensory organs of the inner ear were among the newly identified sites of Math1/lacZ expression. Expression in the otic vesicle was first detected at E12.5 and continued until E18.5 throughout much of the sensory epithelia (Bermingham et al., 1999) (Fig. 3A,B). Null mutants displayed Math1/lacZ expression in the inner ear throughout embryogenesis (Fig. 3C). Math1 null mutants lack hair cells in all of the sensory organs (Bermingham et al., 1999), but maintain supporting cells, the other sensory epithelia-derived cells (Fig. 3C). These supporting cells seem to be functional, based on their morphology and the presence of overlying membranes secreted in part by these cells. Although the expression of Math1 in inner ear sensory epithelia was not demonstrated by RNA in situ hybridization analysis, the complete lack of inner ear hair cells in the null mutants leaves little doubt about the authenticity of the Math1/lacZ expression pattern.

In the brainstem Math1/lacZ expression appeared from E18.5 to P7 in the pons in the regions corresponding to the pontine nuclei (Fig. 3D and inset). This finding is consistent with the hypothesis of Akazawa and colleagues that Math1-positive cells in the developing hind brain are precursors to the bulbopontine neurons (Akazawa et al., 1995). No staining in these regions appeared in the null mutants (Fig. 3E and inset). These data raise the possibility that the absence of X-gal staining may be due to failure of precursor neurons to migrate, proliferate, and/or differentiate. Haematoxylin and eosin staining of brain stem sections from wild-type and null animals showed that null mice lack the pontine nuclei (Fig. 3F,G).

Expression of Math1/lacZ was detected in the developing proximal joints, such as those of the hip and shoulder, as early as E12.5 (Fig. 4A). X-gal-positive staining was detected at subsequent developmental stages in a progressive proximal-distal pattern that paralleled the normal development of joints (Fig. 4B). In the joints, Math1/lacZ expression immediately follows mesenchymal condensation, which begins at E11.5. Condensed mesenchyme cells differentiate into chondrocytes (Bi et al., 1999; Horton et al., 1993; Karsenty, 1998).

Chondrocytes differentiate into three subtypes during bone formation: resting, proliferating and hypertrophic. The resting chondrocytes that populate the articular cartilage are referred to as articular chondrocytes (Buckwalter and Mankin, 1998; Poole, 1997). Prior to birth, resting chondrocytes constitute the entire chondrocyte population in joints. To establish which cells expressed Math1/lacZ, sections from E18.5 and P7 Math1^+/β-gal mice were stained with X-gal. Math1/lacZ is expressed in the resting chondrocytes of all joints analyzed at E18.5; resting chondrocytes in the elbow joint are shown in Fig. 4C, and Fig. 4D shows the resting, proliferating and articular chondrocytes of a P7 mouse. We also examined joints of E18.5 embryos with anti-MATH1 antibody and found expression in resting chondrocytes, whereas no expression was observed in null embryos (data not shown). It should be noted that not all articular cartilage cells express Math1/lacZ (Fig. 4E). Math1/lacZ expression in Math1 null mutants is similar to that in heterozygous mice at E18.5, suggesting that Math1 is not required for resting chondrocyte development. Since Math1 null mice die at birth, however, we cannot asses the role of Math1 in the development of proliferating and articular chondrocytes or ossification.

By E14.5 Math1/lacZ-positive cells were apparent around the vibrissae and in the skin of much of the body (Fig. 4B). In the trunk, the stained cells were arranged in a striped pattern defined by the epidermal ridges. This staining was apparent only in the hairy, not the glabrous, skin. All the primary (mystical) vibrissae, including the lateral nasal, maxillary and four large hairs, were positive for Math1/lacZ. Staining was also detected in the secondary vibrissae, including the labial, submental, rhinal and isolated orbital vibrissae (supra-, infra-and post-orbital) (Yamakado and Yohro, 1979). By E15.5 staining appeared in clusters of cells in the foot pads (Fig. 4B).

To identify the Math1/lacZ-positive cells in the vibrissae, footpad and hairy skin, we examined histological sections from Math1^+/β-gal mice (Fig. 5A-D). Sections through the vibrissae showed that the stained cells are localized to the more apical half of the hair shaft, but are not in the hair itself. Cross sections through the foot pad illustrated staining of clusters of cells in the epidermal layer (Fig. 5B,C). As shown in Fig. 5D, sections through the truncal skin identified clusters of Math1/lacZ-stained cells. The stained cells were arranged in a horseshoe-shaped pattern centered within an elevated button-like structure in the hairy skin. These button-like structures were identified as touch domes or Haarscheiben (Pinkus, 1905), which are characterized by a thickened epidermis and an elevated dermal papilla with a capillary network. Touch domes are associated with large guard hairs dispersed between other hair types in the coat. The spatial distribution of Math1/lacZ-stained cells, the timing of their appearance at E14.5, and their localization within the mystical pads of the vibrissae and the touch domes in the hairy skin suggest that these cells correspond to Merkel cells, specialized cells in the epidermis that form slow-adapting type I mechanoreceptor complexes with neurites (see below and Munger, 1991).

The results of comparative analysis of the Math1/lacZ expression pattern in heterozygous and homozygous E16.5 animals are shown in Fig. 5E-L. Math1Δβ-gal/β-gal Δembryos displayed a staining pattern similar to that of Math1^+/β-gal littermates in the vibrissae and footpads (Fig. 5E and I, G and K), although the staining was more intense because of the lacZ dosage effect. In contrast, staining in the touch domes of the hairy skin was barely detectable in Math1Δβ-gal/β-gal Δembryos (Fig. 5F and J, H and L). The reduction of staining in null animals was also obvious at E18.5 (not shown).

To further define Math1/lacZ-positive cells in the skin, Math1^+/β-gal mice were mated to Tabby mice. Tabby (Ta) is a spontaneous X-linked mutation displaying a similar phenotype in hemizygous males and homozygous females (Ferguson et al., 1997). Tabby mutants lack guard hair follicles (tylotrich), a subset of follicles that are associated with touch domes in the hairy skin of the trunk (Vielkind et al., 1995), and some of the five secondary vibrissae on the head (Gruneberg, 1971). Hence, in a cross of Ta/Ta females with a heterozygous Math1^+/β-gal male, 50% of the male progeny are Ta/Y: Math1^+/β-gal, allowing us to assess whether the Math1/lacZ-positive cells correspond to Merkel cells.

Both Tabby females and males carrying the Math1^+/β-galallele displayed X-gal staining in the vibrissae and foot pads (Fig. 6A,B and data not shown). The effect of the Tabby mutation on the number of secondary vibrissae was quite clear: hemizygous males completely lacked Math1/lacZ-positive cells around the missing secondary vibrissae (typically lacking in Ta mutants) and on the trunk (Fig. 6E). Females that are heterozygous for Tabby showed patchy staining in the touch domes (although less than wt), as should be anticipated in female carriers of a mutation in a gene that undergoes random X chromosome inactivation (Fig. 6C,D). The localization and distribution of the positive cells, as well as their absence in selected vibrissae and the trunk of Tabby males, strongly indicate that Math1 is expressed in the Merkel cells associated with guard follicles in the touch domes of the hairy skin.

To ascertain whether the Math1/lacZ expression pattern reflects the normal Math1 expression pattern, immunohistochemical analysis of MATH1 was performed on sections from abdominal skin. As seen in Fig. 7A and B, MATH1-positive cells were detected around the hair follicles of Math1^+/+ but not Math1Δβ-gal/β-gal Δmice. Antibodies against two Merkel cells markers were chosen for further analysis: anti-cytokeratin18, expressed in simple epithelia, and chromogranin, localized to secretory granules of neuroendocrine, endocrine, and neuronal tissues. Both cytokeratin 18 (Fig. 7C,D) and chromogranin A (Fig. 7E,F) confirmed the identity of the Math1/lacZ-positive cells as Merkel cells, but did not reveal staining abnormalities in Math1Δβ-gal/β-gal Δmice. Thus, Math1 does not seem to be essential for the genesis of the neuroendocrine Merkel cells. Because Math1 null mutants die at birth, we cannot assess the functional integrity of Merkel cells in these mutants.

Given the remarkable similarity in expression patterns of ato and Math1, and their identical basic domains, we wondered if Math1 would mimic the effects of ato overexpression by producing ectopic chordotonal organs. Expressing Math1 during pupal development by heat shock using the UAS-Gal4 system (Brand and Perrimon, 1993) resulted in supernumerary external sense organs on the notum (Fig. 8A,B) and the wing blade (data not shown), as reported for ato (Jarman et al., 1993) and the Achaete-Scute complex (AS-C) genes (Rodriguez et al., 1990). Math1 expression, like ato, produced ectopic chordotonal organs (Fig. 8G), although with less efficiency than ato (A. Jarman, personal communication). Overexpression of the AS-C genes does not, however, result in ectopic chordotonal organs (Jarman et al., 1993). Math1 thus has a similar functional specificity to ato.

Since several ato enhancers are ato-dependent (Sun et al., 1998), they may be activated by Math1, which would then lead to ectopic CHO specification. To determine whether Math1 can substitute for ato function in the fly, and to rule out the possibility that production of CHOs by Math1 is due to ato activation, we expressed Math1 in ato mutant embryos. The mutants lack all chordotonal neurons (Fig. 8D), but overexpressing Math1 partially rescues the loss of these neurons (Fig. 8E) in a manner similar to ato (Chien et al., 1996).

Over the past few years significant progress has been made towards unraveling the roles of bHLH proteins in vertebrate neurogenesis. Neural vertebrate bHLH-encoding genes were isolated and characterized because Drosophila homologues such as ato or the AS-C genes had been previously shown to be required for neurogenesis (Anderson, 1995; Guillemot, 1995; Lee, 1997; Takebayashi et al., 1997). Indeed, several genes were shown to be proneural because their absence caused a failure of neuroblast or sensory organ precursor (SOP) specification, whereas their overexpression lead to the recruitment of supernumerary neuronal precursors (Ghysen and Dambly-Chaudiere, 1989). With the exception of neurogenin (Ngn) 1 and 2 (Fode et al., 1998; Ma et al., 1998), it remains uncertain which of the vertebrate homologues play roles similar to their Drosophila counterparts, and what precise role different bHLH proteins play in neural development. In Drosophila, ato is required for the development of a specific subset of sense organs, the chordotonal organs (Jarman et al., 1993). CHOs are internal mechanosensors of the PNS (McIver, 1985). Thus, ato and the CHOs provide an excellent system in which to ascertain not only the molecular and developmental relationship between invertebrate and vertebrate neurogenesis vis-à-vis the function of the proneural genes, but also the evolutionary conservation of sensory organ function and specification.

Seven ato homologues have been cloned and analyzed in the mouse: Mouse Atonal Homologues (MATH) 1, 2, 3, 4A (also known as Ngn2), 4B (Ngn3), 4C (Ngn1), and 5 (Akazawa et al., 1995; Bartholomä and Nave, 1994; Ben-Arie et al., 1997, 1996; Fode et al., 1998; Ma et al., 1998; McCormick et al., 1996; Shimizu et al., 1995; Takebayashi et al., 1997). Most are expressed during neurogenesis in both the CNS and PNS. These homologues vary in the degree of their sequence conservation, and may be divided into three groups. The most distantly related group, the neurogenins, includes Ngn 1, 2 and 3. These gene products share, on average, 53% identity in the bHLH domain with ATO. They are expressed largely in mitotic CNS and sensory ganglia progenitor cells. Recent work suggests that these genes may play a role in neuroblast determination, and may therefore be true proneural genes (Fode et al., 1998; Ma et al., 1998). The second group includes MATH2 and MATH3, which share 57% identity in the bHLH domain with ATO. These proteins have been postulated to function in postmitotic neural cells (Bartholomä and Nave, 1994; Shimizu et al., 1995). Math2 expression is confined to the CNS, while Math3 is expressed in both the CNS and the trigeminal and dorsal root ganglia. The third group includes MATH1 and MATH5, arguably the only true ato homologues by amino acid sequence criteria, sharing 67% and 71% identity with the bHLH domain of ATO, respectively. It is noteworthy that both genes encode a basic domain identical to that of ATO. Interestingly, the basic domain of ATO was shown to be sufficient, in the context of another proneural protein (SCUTE), to substitute for the loss of ato function (Chien et al., 1996). Math1 was initially shown to be expressed in the precursors of the cerebellar EGL and in the dorsal spinal cord (Ben-Arie et al., 1997, 1996). Math5 is expressed in the dividing progenitors in the developing retina and in the vagal ganglion (Brown et al., 1998).

With the exception of Math5 expression in the neural retina, these observations pose a paradox: none of the vertebrate homologs appeared to be expressed in peripheral organs or tissues similar to those where ato is expressed. Jarman et al. (1993) reported that ato is expressed in the CNS. In this study we show that, in addition to the inner proliferation center of the optic lobe, ato is expressed in a small anteriomedial patch of cells in each brain lobe (Fig. 8F). Because it remains unclear, however, precisely what role ato plays in Drosophila CNS development, it has been difficult to argue that ato and its vertebrate homologues display functional conservation.

Our experiments reveal sites of previously uncharacterized Math1 expression. As expected, we found that Math1/lacZ expression in the CNS corresponds to that of Math1, but we also found that Math1 is expressed in the skin, the joints, and the inner ear, in striking parallel to ato expression in the fly. Moreover, the expression in the ear (sensory epithelium) and the skin (Merkel cells) is restricted to sensory structures whose function is to convert mechanical stimuli into neuronal electrochemical signals. It is important to point out that in Drosophila, ato appears to play two roles simultaneously. It is required not only to select the precursors of the CHOs (proneural role), but also to specify these precursors as CHO precursors (lineage identity role) (Jarman and Ahmed, 1998; Jarman et al., 1993). The specificity of Math1 expression in the periphery makes it tempting to speculate that it, too, may endow specific cells with very specific lineage identities to distinguish them functionally from other sensory structures. The ability of Math1 to induce ectopic CHO formation and to restore CHOs to ato mutant embryos supports the notion that Math1, and particularly its basic domain, encodes lineage identity information not unlike that encoded by ato. This suggests that the mammalian cells expressing Math1, at least in the ear and the skin, are functionally similar and perhaps evolutionarily related to Drosophila cells that require ato. Furthermore, Math5 expression in the neural retina suggests that the functions of atonal in the fly are carried out by two genes in the mouse: the development of some mechanoreceptors is under the control of Math1 and retinal development is possibly under the control of Math5. It is interesting to note that in the fully sequenced nematode C. elegans, only one homolog of atonal, lin-32, was identified (Zhao and Emmons, 1995). Mutants with the u282 allele are touch-insensitive, which strengthens the argument for evolutionary conservation of atonal function in mechanoreception.

The pattern of Math1/lacZ expression in the pontine nuclei led us to carefully evaluate this region in null mutants. Although we did not previously detect defects in the pons of Math1 null mice (Ben-Arie et al., 1997), closer analysis revealed the lack of pontine nuclei at this site. These neurons derive from the rhombic lip (Altman and Bayer, 1996) as do the EGL neurons, which are also lacking in Math1 null mice.

While it is possible to draw parallels between Math1 and ato expression in the skin and ear, it is not clear that such is the case for the joints. ato expression in the fly joints is required for the formation of leg CHOs. In contrast, Math1 is expressed in resting and articular chondrocytes that do not have any described neural function, and for which no parallels exist in the fly. It may be that Math1 expression in cartilage indicates a novel role for a mechanosensory gene, or it may simply reflect similarities in the molecular events underlying the development of the various Math1-expressing cell types. Alternatively, CHOs may also function as joint structural elements in the fly, or articular cartilage may have a mechanoreceptive or transducive capacity yet to be described. There is no evidence at this point to support one or another of these possibilities. We must await the generation of mice with a joint-specific Math1 mutation.

It will be interesting to determine whether Math1 and ato have similar roles and regulate similar molecular pathways. For example, is Math1 a proneural gene in at least some of the tissues in which it is expressed? Does it play a determining role in some tissues, but a differentiating role in others? Does Math1, like ato, function in a Notch-Delta-dependent pathway? Does it activate the epidermal growth factor receptor pathway in neighboring cells? Analyzing the functions of ato and Math1 will enhance our understanding of neural development and the evolutionary conservation of sensory function. The sites and specificity of Math1 expression may make it suitable as a tool of gene therapy or gene activation approaches to illnesses such as hearing loss and osteoarthritis that are due to age-related or environmental damage.

The authors gratefully acknowledge the expert assistance of Barbara Antalffy in the Neuropathololgy Core of the Mental Retardation Research Center at Baylor College of Medicine. We thank P. Overbeek for Tabby mice, and A. Grossler for the selection cassette. B. A. H. is funded by an NIH postdoctoral fellowship. Portions of this work were supported by a grant from NASA. H. J. B. and H. Y. Z. are investigators of the Howard Hughes Medical Institute.