Mash1 activates a cascade of bHLH regulators in olfactory neuron progenitors

The process of neurogenesis in the nervous system of vertebrates produces a large variety of neuronal cell types. Neurons are generated from populations of neural progenitor cells, which differ in their proliferative characteristics and differentiation potentials (McConnell, 1995). The genes that control the commitment of ectodermal cells to a neural fate, the proliferation of neural progenitor cells and their differentiation into post-mitotic neurons, are still poorly characterized in vertebrates. In contrast, studies in invertebrates, in particular in Drosophila, have led to the identification of a number of genes, functioning at successive stages in the developmental pathways, leading to differentiated neurons. Proneural genes control the generation of neural precursors from uncommitted ectodermal cells. The main Drosophila proneural genes are the genes of the achaete-scute complex (asc) and atonal (ato), which encode transcription factors of the basic helix-loop-helix (bHLH) class (Campuzano and Modolell, 1992; Jarman et al., 1993). Loss-of-function mutations in proneural genes prevent the generation of neural precursors (Ghysen and Dambly-Chaudière, 1988; Jarman et al., 1993) whereas gain-of-function mutations result in the transformation of ectodermal cells into neural precursors at ectopic positions (Rodriguez et al., 1990; Ghysen et al., 1993).

Genes homologous to Drosophila asc have been identified in vertebrate species. ash1 is a vertebrate asc homolog expressed in subsets of neural progenitors in the central nervous system (CNS) and the peripheral nervous system (PNS), and it has been identified in all species examined (Johnson et al., 1990; Guillemot and Joyner, 1993; reviewed in Guillemot, 1995). A targeted mutation of Mash1 in the mouse results in the elimination of most olfactory and autonomic neurons, demonstrating a role for Mash1 in the development of particular neural lineages (Guillemot et al., 1993; Sommer et al., 1995; Blaugrund et al., 1996).

A number of bHLH genes with similarities to Drosophila ato have also been isolated in vertebrates. Math1, the closest murine relative of ato, is expressed in the dorsal neural tube and the developing cerebellum (Akazawa et al., 1995). NeuroD/Beta2 (Lee et al., 1995; Naya et al., 1995), neuroD2/NDRF/kw8 (Kawakami et al., 1996; Kume et al., 1996; McCormick et al., 1996; Yasunami et al., 1996), Nex1/Math2 (Bartholomä and Nave, 1994; Shimizu et al., 1995) and Math3 (Takebayashi et al., 1997) constitute a subfamily of genes distantly related to ato, and with very similar bHLH domains. These genes are expressed either transiently in differentiating neurons or permanently in terminally differentiated cells. In contrast, neurogenin1/neuroD3 and Math4A/neurogenin2, which define a third subfamily of ato-related genes, are expressed exclusively in progenitors of the CNS and PNS (Gradwohl et al., 1996; Ma et al., 1996; McCormick et al., 1996; Sommer et al., 1996). The functions of NeuroD and a neurogenin gene have been studied by RNA injection in Xenopus embryos. Overexpression of NeuroD accelerates neuronal differentiation in the neural plate and induces neuronal differentiation in ventral ectoderm, demonstrating that NeuroD has neuronal differentiation activity (Lee et al., 1995). Overexpression of the neurogenin homolog X- ngnr-1 has similar effects, and in addition it induces NeuroD and the neurogenic gene delta, which are characteristics of a neural determination gene (Ma et al., 1996).

Most vertebrate neural lineages are still poorly characterized. Development of the lineage of olfactory receptor neurons (ORNs) has been relatively well described, however, owing to the simple structure and accessibility of the olfactory epithelium (OE) in the mouse. This tissue is made up of four cell types that can be distinguished in vivo and in vitro by their morphology, position and antigenic characteristics. They include (1) the ORN themselves, (2) their immediate neuronal precursors (INPs) or globose basal cells, which go through a small number of symmetric divisions before generating post-mitotic neurons, (3) the horizontal basal cells of unknown function and (4) the sustentacular cells, which have supporting and secretory functions (Graziadei and Monti-Graziadei, 1979; Calof and Chikaraishi, 1989; Schwartz Levey et al., 1991; Calof et al., 1996). The OE has the unique ability to maintain neurogenesis into adult stages and to regenerate ORNs when they are destroyed, implying that stem cells of the ORN lineage persist in the OE throughout life (Graziadei and Monti Graziadei, 1978).

Very little is known about the genes controlling the development of the ORN lineage. So far, Mash1 is the only transcription factor shown to be required for this process. Mash1 is transiently expressed by ORN progenitors before they reach the INP stage (Guillemot and Joyner, 1993; Gordon et al., 1996), and in mouse embryos mutant for Mash1, most ORNs are missing (Guillemot et al., 1993). In this article, we show that Math4C/neurogenin1 is also expressed in ORN progenitor cells, but after Mash1, and that NeuroD expression in this lineage overlaps with overt neuronal differentiation. In the absence of Mash1, most progenitor cells fail to express Math4C/neurogenin1 and NeuroD and remain undifferentiated in the OE until they die. A subset of progenitor cells present in a ventrocaudal domain of the OE, however, express Math4C/neurogenin1 and NeuroD and differentiate in the absence of Mash1 function. In contrast to the rest of the OE, cells in this domain express Math4A/neurogenin2, another bHLH gene closely related to Math4C/neurogenin1, and do not express Mash1. These results show that Mash1 is required at an early stage in a subset of ORN progenitors to activate expression of Math4C/neurogenin1 and of the neuronal differentiation gene NeuroD. Another gene, possibly Math4A/neurogenin2, activates a similar differentiation program in a region of the OE unaffected in Mash1 mutants.

Genomic DNA and random primed cDNA prepared from E10.5 mouse embryos were subjected to PCR (1 minute, 94°C; 1 minute, 45°C; 30 seconds, 72°C for 5 cycles and 1 minute, 94°C; 1 minute, 60°C; 30 seconds, 72°C for 35 cycles) with the following oligonucleotide primers, corresponding to the amino acid sequences ANNRER and FAHNYI from the basic region and second helix of MATH4A, respectively (Gradwohl et al., 1996): 5′ primer, 5′GAG- GATCCGCIAA(T/C)(G/A)(C/A)I(C/A)GNGA(G/A)(C/A)G3′ and 3′ primer, 5′GGGAATTCAT(G/A)TA(G/A)TT(G/A)TGNGC(G/A)AA3′. In addition to sequences corresponding to the bHLH domains of Math4A and NeuroD, a novel sequence closely related to that of Math4A was amplified, that we named Math4B. The Math4B PCR fragment was used to screen a lambda GEM12 129SV mouse ES cell (D3) genomic library. Four clones were isolated; three of these contained the Math4B gene, and one contained a new gene, with a closely related bHLH sequence, which we named Math4C (Fig. 2B). Math4B and Math4C sequences had uninterrupted open reading frames encoding putative proteins of 194 and 244 amino acids, respectively (not shown). A Math4C cDNA containing the same open reading frame was subsequently isolated by screening a mouse embryonic cDNA phage library. These results suggest that the protein sequence of MATH4C is encoded in a single exon, as is the case for MATH4A (not shown) and other bHLH proteins like MASH1.

Wild-type and Mash1 mutant embryos were obtained from intercrosses of Mash1+/− mice that have been backcrossed on a CD1 genetic background for more than five generations. For the staging of embryos, midday of the day of the vaginal plug was considered as E0.5. The precise stage of E9.5-10.0 embryos was defined by counting somites.

Whole-mount RNA in situ hybridization and immunocytochemistry on E9.5 to E10.5 embryos were performed as described in Wilkinson and Nieto (1993) and Davis et al. (1991), respectively. The same protocols were used for RNA/antibody double-labeling experiments, except that the proteinase K treatment of the RNA hybridization protocol was omitted and replaced by washes in RIPA buffer (Rosen and Beddington, 1993). Following whole-mount antibody or RNA staining, olfactory placodes were dissected with fine needles and flatmounted for microscopic observation. RNA hybridization and immunostaining were performed on cryosections for embryos older than E10.5. The head of E11.5 and the nasal region of E12.5 and older embryos freshly harvested were fixed in 4% paraformaldehyde in PBS at 4°C for 2 hours, rinsed and equilibrated in 20% sucrose in PBS at 4°C for 4-6 hours, mounted in 7.5% gelatin, 15% sucrose in PBS at 37°C, frozen on dry ice, cut in the cryostat, and processed for ³⁵S- or digoxygenin-labeled RNA in situ hybridization as described (Myat et al., 1996; Gradwohl et al., 1996) or for immunocytochemistry as described (Guillemot et al., 1993; Gradwohl et al., 1996). For BrdU incorporation experiments, 2 mg of BrdU were injected intraperitoneally into pregnant mice 2 hours before they were killed. RNA probes were synthesized from the following cDNA clones: a 1674 bp full-length Mash1 cDNA clone (Guillemot and Joyner, 1993); a 695 bp NeuroD cDNA containing residues V127 to D358 (Lee et al., 1995); a 1412 bp full-length Math4A cDNA clone (Gradwohl et al., 1996); a 404 bp Math4C clone containing residues M1 to M106 (Fig. 2B) and a 748 bp Sox2 cDNA (Collignon et al., 1996; kindly provided by Larysa Pevny and Robin Lovell-Badge). Anti-class III β-tubulin and anti-BrdU monoclonal antibodies were from Sigma. Antibody binding was visualized using the Vectastain ABC kit (Vector).

Apoptotic cell death was detected on sections of OE by the TUNEL procedure (Gavrieli, 1992), modified as follows. Frozen sections of fixed tissue, prepared as described above, were rinsed in PBS, overlayed with 200 μl of TUNEL buffer (140 mM sodium cacody-late, 30 mM Tris, pH7.2, 1 mM CoCl₂, 0.25 mg/ml BSA) containing 40 μm digoxygenin-dUTP and 25 units of terminal transferase (Boehringer Mannheim) and incubated in a humid chamber at 37°C for 1 hour. Slides were then washed and processed for immunostaining with an anti-digoxygenin antibody coupled to peroxidase (1:2500, Boehringer Mannheim).

The first aim of this study was to characterize the initial stages of neurogenesis in the olfactory placodes of mouse embryos, using neuronal markers and Mash1 as a marker of ORN progenitors. The early stages of development of the OE in the mouse have been previously described at the morphological level (Smart, 1971; Cuschieri and Bannister, 1975a,b). Olfactory placodes first appear as a thickening of the ectoderm on the ventrolateral sides of the head at about E9.5. At E10.5, the placodes invaginate, forming the olfactory pits, and ORNs start to grow axons. At E11.5, the olfactory pits deepen and form secondary recesses, and the dendrites of olfactory neurons become apparent. The OE becomes organized in layers around E12.5-13.5, with ORN progenitors progressively localized to the basal side of the epithelium and post-mitotic neurons to an intermediate zone. Sustentacular cells, the supporting cells of the OE, start to differentiate at about E15.5 and have their nuclei localized apically. The thickness of the OE increases at later embryonic stages and the tissue reaches its maximum size after birth.

To define the temporal and spatial aspects of neuronal differentiation during the early, less well-characterized stages of olfactory neurogenesis, we examined the expression of class III β-tubulin by immunocytochemistry and SCG10 by RNA in situ hybridization. These two genes are among the earliest characterized markers of neuronal differentiation (Stein et al., 1988; Easter et al., 1993). β-tubulin and SCG10 became first detectable in cells localized in the center of the olfactory placode at around the 27-somite stage, just prior to the invagination of the placode (not shown). At E10.5, β- tubulin+/SCG10+ ORNs are also found in the center of the olfactory pit (Fig. 1D-F and not shown).

Mash1 marks a subset of ORN progenitors (Guillemot and Joyner, 1993; Gordon et al., 1996) and its activity is required for the generation of most ORNs (Guillemot et al., 1993). We studied the expression of Mash1 at early stages of OE development, when mutation of this gene probably has its first phenotypic effects. Mash1 transcripts were first detected at the 24- to 25-somite stage, in clusters of cells scattered in the placode. By the 27- to 28-somite stage, Mash1+ cells were grouped in cell clusters arranged in a ring surrounding a central area of the placode devoid of labeled cells (not shown) and at E10.5, Mash1 labeling was restricted to solid cell clusters at the periphery of the olfactory pit (Fig. 1A).

The distribution of Mash1+ and β-tubulin+/SCG10+ cells in olfactory placodes suggests that olfactory progenitors and differentiated neurons have already segregated at early developmental stages. In double-labeled placodes, Mash1+ and β- tubulin+ cells were indeed spatially segregated in a peripheral and a central domain respectively, and very little intermingling was observed (Fig. 1D,G). These results indicate that the differentiation stage of placodal cells is roughly correlated with their position on a periphery-to-center axis, with the less mature cells located at the periphery and the more differentiated cells in the center of the placode. This organization could be due to the addition of immature progenitors at the periphery of the placode by cell division, while cells in the center of the placode would progressively differentiate.

To futher characterize neurogenesis in the olfactory placode and better define the role of Mash1 in the ORN lineage, we searched for other potential regulators of neural development that would be expressed in ORN progenitors.

We have recently isolated three novel atonal-related bHLH genes named Math4A (Gradwohl et al., 1996), Math4B and Math4C (see Materials and Methods and Fig. 2) encoding proteins with high sequence similarity in their bHLH domains and lower but significant similarity with the bHLH domains of other mouse atonal-related proteins such as NeuroD, MATH1, MATH2 and MATH3 (Fig. 2; Gradwohl et al., 1996). A preliminary analysis by RNA in situ hybridization revealed that these three genes are expressed in embryonic CNS and PNS progenitors (Gradwohl et al., 1996; data not shown). In the OE, Math4C is expressed in a large cell population, whereas Math4A expression is restricted to a limited portion of this tissue (see below) and Math4B is not detectably expressed (not shown). These three genes have been independently isolated in other laboratories, and given the names neurogenin1/NeuroD3 (Math4C), neurogenin2 (Math4A) and neurogenin3 (Math4B) (Ma et al., 1996; McCormick et al., 1996; Sommer et al, 1996). The bHLH gene NeuroD, which has been shown to promote neuronal differentiation in ectodermal cells of Xenopus embryos, is also expressed in olfactory placodes (Lee et al., 1995). We first examined the expression of Math4C and NeuroD during OE development and compared it with the expression of Mash1.

Math4C and NeuroD were first expressed in the olfactory placode at the 24- to 25-somite stage and their transcripts were found in many cells throughout the placode at E9.5 (not shown). The expression of the two genes differed markedly from each other and from that of Mash1 in the olfactory pit at E10.5 (Fig. 1A-C). Because the lateral side of the pit is folded and difficult to observe and Mash1 expression is absent from its ventrocaudal and dorsorostral sides, we compared the expression patterns of the three genes in the medial side of the pit (left in Fig. 1A-C). Math4C+ cells were present both at the periphery and in the center of the pit, while Mash1+ cells were found only at the periphery and NeuroD+ cells only in the center. Interestingly, Math4C was mostly expressed in single cells, whereas Mash1 was expressed in large cell clusters in the medial side of the pit. Based on the differential distribution of neurons and progenitor cells, which suggest a differentiation gradient in the placode, as described above, Math4C appears to mark cells that are at a more advanced stage than those marked by Mash1, and NeuroD to mark cells more advanced than those marked by Math4C. The differentiation state of Math4C+ and NeuroD+ cells was studied by double-labeling with β-tubulin. In contrast to Mash1+ cells, Math4C+ and NeuroD+ cells were present in the center of the placode where they intermingled with β-tubulin+ cells (Fig. 1E,F). There was no apparent overlap, however, between the populations of Math4C+ and β-tubulin+ cells (Fig. 1H), whereas some placodal cells were double-labeled by NeuroD and β-tubulin (arrows in Fig. 1I). These results suggest that Math4C and NeuroD are both expressed in undifferentiated ORN progenitors, with NeuroD expression maintained transiently in differentiating neurons.

The simplest interpretation of the above data is that Mash1, Math4C and NeuroD are sequentially expressed in cells of the same lineage, with the following temporal sequence: Mash1>Math4C>NeuroD>β-tubulin. In particular, Math4C appears to be turned on and off at earlier stages than NeuroD, and continues to be expressed later than Mash1. Although the stages of Mash1 and Math4C activation relative to one another could not be deduced from these experiments, a later onset of Math4C expression with respect to Mash1 was supported by the analysis of Mash1 mutant OE (see below). Whether these genes are transiently co-expressed in ORN progenitors remains to be determined.

We also studied the expression of Mash1, Math4C and NeuroD in the more mature OE, where the distribution of the different cell types has been described in detail. After E11.5, both Math4C and NeuroD were highly expressed in a 1- to 2- cell-thick layer on the basal side of the OE (Fig. 1K,L at E17.5), where ORN progenitors are located (Graziadei and Monti Graziadei, 1979). To confirm that Mash1, Math4C and NeuroD are expressed in dividing progenitors, double-labeling experiments were performed in E17.5 OE with RNA probes and anti-BrdU antibody after a 2-hour incorporation period (Fig. 1M,O). Many of the basal Math4C+ and NeuroD+ cells had incorporated BrdU, demonstrating that they were mitotically active. The results of the double-labeling experiments with BrdU and β-tubulin therefore indicate that Math4C is transiently expressed in dividing progenitors, while NeuroD expression spans the last division of ORN progenitors and the beginning of differentiation of post-mitotic neurons (Fig. 3).

Like Math4C+ and NeuroD+ cells, the majority of Mash1+ cells were localised to the basal side of the OE. Some Mash1+ cells, however, were found in intermediate or apical positions at all embryonic stages examined (Fig. 1J). As shown in Fig. 1M, many of the Mash1+ cells, in both basal and apical positions of a E17.5 OE, had incorporated BrdU after a 2-hour exposure period, indicating that they are dividing. Both basal and apical Mash1+ cells did not express β-tubulin either (not shown), confirming that they are undifferentiated. These results thus suggest that Mash1 marks two distinct stages of ORN progenitor cells, with different locations in the OE. Based on our analysis of the Mash1 mutant phenotype and on expression studies (Gordon et al., 1996), Mash1+ progenitors precede, and are likely precursors of, the basally located INPs/globose cells. Thus, the apically located Mash1+ cells may be early ORN progenitors, possibly stem cells, which give rise to more mature Mash1+ progenitors migrating to the basal side of the OE (Fig. 3). ORN progenitors are indeed located apically in the first few days of OE development (Smart, 1971). Alternatively, these apically located Mash1+ cells could be progenitors for a separate branch of the ORN lineage.

To further characterize the role of Mash1 in the ORN lineage, we studied the effect of the loss of Mash1 function on neuronal differentiation and gene expression in ORN progenitors. We have previously generated a mouse strain carrying a null allele of the Mash1 gene. Animals homozygous for this mutation show a dramatic decrease in number of ORNs at birth (Guillemot et al., 1993). We extended our study of the Mash1 null mutation by analysing in detail the onset of the mutant phenotype in the OE.

The first β-tubulin+ and SCG10+ ORNs appeared as early in mutant as in wild-type placodes, with no clear difference in the number of labeled neurons between mutant and wild-type OE at E9.5 and E10.5 (Fig. 4A,B and not shown). At E11.5, neurons were present in large numbers in a ventrocaudal domain of the mutant OE, but were missing in more rostral regions, which in wild-type OE contained neurons (Fig. 4C,D). At later stages, most regions of the mutant OE were devoid of neurons (Fig. 4E,F), with the exception of the ventrocaudal domain, which retained normal levels of β-tubulin and SCG10 expression (not shown). Thus ORN progenitors fail to differentiate in the mutant OE, except in a ventrocaudal region, which is not affected by loss of Mash1 function.

Histological analysis of Mash1−/− embryos had previously suggested that cell death was occurring in the mutant OE (Guillemot et al., 1993). To determine if apoptotic cell death was responsible for the lack of differentiated neurons in mutant OE, wild-type and mutant OE were subjected to the TUNEL procedure (Gavrieli, 1992). At E11.5 and earlier stages, very few cells were labeled in both mutant and wild-type OE, indicating that little cell death occurred at these stages even in the absence of Mash1 function (Fig. 5A,B and not shown). At E12.5, TUNEL-labeling above wild-type levels was observed in the dorsorostral extremity of the mutant OE (data not shown). At E13.5, dying cells were present in large numbers in most regions of the mutant OE, whereas TUNEL-labeling remained very low in the wild-type OE (Fig. 5C,D). Thus, cell death is abundant only after E12.5, i.e. after the initial differentiation arrest of Mash1 mutant cells, indicating that death of progenitors is not a direct cause of the lack of ORNs observed in mutant OE at early stages.

To confirm that mutant progenitor cells were present in the OE before the onset of cell death, we studied the expression of molecular markers in the mutant OE. Using a Mash1 RNA probe, we detected in Mash1−/− embryos a mutant Mash1 transcript with a distribution similar to the wild-type transcript. In particular, Mash1+ cells were present in the same positions in Mash1 mutant and in wild-type OE at E11.5 and E12.5 (Fig. 7A,E). We also examined expression of Sox2, a marker of neural progenitors in the CNS and in olfactory placodes (Uwanogho et al., 1995; Collignon et al., 1996). We detected similar expression of Sox2 in wild-type and Mash1 mutant embryos between E9.5 and E12.5, including in the areas of mutant OE devoid of differentiated neurons (not shown). Thus, until at least E12.5, Mash1-expressing cells are present in the mutant OE but fail to differentiate.

Expression studies suggest that Math4C and NeuroD are expressed after Mash1 in ORN progenitors (see above). As ORN progenitors are present in Mash1 mutant OE until E12.5, it was possible to study transcription of Math4C and NeuroD in these cells before their elimination by apoptosis.

Both Math4C and NeuroD expression appeared at similar stages in Mash1 mutant and in wild-type placodes. Expression of NeuroD and Math4C was reduced in mutant placodes, however, and confined to the center of the placode, suggesting that some cells of the placode failed to transcribe Math4C and NeuroD (not shown). At E10.5 and later stages, expression of both Math4C and NeuroD was absent in most cells of the mutant OE, except in a ventrocaudal domain (Fig. 6). This loss of Math4C and NeuroD expression around E10.0 is the first defect observed in mutant OE cells.

Normal levels of Math4C and NeuroD expression were maintained in a ventrocaudal domain of mutant olfactory pits (Fig. 6A-H). A comparison of Mash1 and Math4C expression patterns in wild-type OE (Fig. 1A,B at E10.5; Fig. 7A,C at E11.5; Fig. 7I,K at E12.5) showed that, in contrast to other regions of the OE where Mash1 and Math4C are co-expressed, only Math4C is expressed in this ventrocaudal domain. The lack of expression of Mash1 is a clear indication that Math4C expression must be regulated differently in this domain from in the rest of the OE. Interestingly, Math4A, a gene closely related to Math4C (Gradwohl et al., 1996; Fig. 2), is expressed in a region of the OE that overlaps with the Mash1-independent domain of Math4C expression and neurogenesis (Fig. 7B). This Math4A expression domain persists in Mash1 mutant OE (Fig. 7F). Math4A may be activated in parallel with Math4C in this ventrocaudal domain of the OE, or it may be involved itself in the activation of Math4C expression in Mash1-independent ORN progenitor cells.

In this article, we describe the expression of four bHLH transcriptional regulators, Mash1, NeuroD, Math4A/neurogenin2 and the novel gene Math4C/neurogenin1, at early stages of olfactory neurogenesis, and we characterize the consequences of the loss of Mash1 activity on gene expression and differentiation in the ORN lineage.

The OE of Mash1−/− embryos presents three defects which appear sequentially: (1) loss of Math4C and NeuroD expression, (2) block of neuronal differentiation, and (3) cell death. What are the causal links between loss of Mash1 and these different phenotypes?

Lack of Math4C and NeuroD expression are the earliest defects observed in Mash1 mutant ORN progenitors. This, added to the temporal and spatial distribution of gene expression in the placode, points to the following developmental pathway in the OE: Mash1>Math4C>NeuroD>β- tubulin. The precise timing and the degree of overlap of expression of these genes at the single cell level remain to be defined, but the current data raise the possibility that in the ORN lineage, Mash1 directly activates the transcription of Math4C, and that Math4C activates the transcription of NeuroD. An indirect activation or even a non-cell autonomous effect of Mash1 on Math4C and NeuroD expression cannot, however, be excluded. X-ngnr-1, a Xenopus gene related to mouse Math4C/neurogenin1, has been demonstrated to have a role in inducing NeuroD expression and neurogenesis by RNA injection in Xenopus embryos (Ma et al., 1996). Thus there is evidence for an interaction between members of these two gene families in other tissues.

Expression of Mash1, Math4C and NeuroD precedes neuronal differentiation in the ORN lineage. In the most parsimonious hypothesis, the main function of Mash1 and/or Math4C could be to activate NeuroD, which would alone trigger neuronal differentiation. Experiments in Xenopus embryos have demonstrated that ectopic expression of NeuroD is sufficient to promote neuronal differentiation in ectodermal cells (Lee et al., 1995). Alternatively, each of these genes could control the expression of a distinct subset of genes contributing to the neuronal phenotype. The generation of loss-of- function and gain-of-function mutations in these genes should help to distinguish between these possibilities.

Cell death in Mash1 mutant OE is first detected at E12.5, more than 2 days after the initial block in Math4C and NeuroD expression. This delay suggests that loss of Mash1 only indirectly leads to cell death in the OE. Thus, Mash1 could regulate genes that mediate the survival of ORN progenitors, for example by controlling their response to trophic factors. Experiments of bulbectomy in adult mice have provided evidence that ORN progenitors are dependent on extrinsic signals for their survival (Holcomb et al., 1995). Therefore, Mash1 could both promote neuronal differentiation and regulate certain characteristics of progenitor cells, such as their trophic dependence or proliferative properties.

Analysis of the ORN lineage in mice and in cell cultures has provided evidence that ORN progenitors pass through a succession of developmental stages, which can be distinguished in particular by their proliferative properties (Graziadei and Monti-Graziadei, 1979; Calof and Chikaraishi, 1989). A population of rapidly dividing cells, called globose cells, has been described at the basal side of the OE (Graziadei and Monti Graziadei, 1979; Mackay-Sim and Kittel, 1991; Schwartz Levey et al., 1991). Explant cultures (Calof et al., 1996), lineage tracing (Caggiano et al., 1994) and ³H-thymidine labeling experiments (Graziadei and Monti Graziadei, 1979) have demonstrated that globose cells constitute a population of ORN progenitors with characteristics of immediate neuronal precursors.

Expression studies in OE explants and in regenerating OE have shown that Mash1 is expressed at an early stage of the ORN lineage by progenitors that are distinct from, and are likely to give rise to, INPs/globose cells (Gordon et al., 1996). Since the expression of Math4C follows that of Mash1 and precedes neuronal differentiation, it could mark the main population of INPs/globose cells. It has been proposed that each stage of ORN progenitor development responds to a distinct set of mitogens, survival factors and differentiation factors, and thus represents a checkpoint at which neuron production can be controlled (Gordon et al, 1996; Calof et al., 1996). Thus Mash1 and Math4C could regulate distinct responses to extrinsic factors in the successive stages of the ORN progenitor cells in which they are expressed.

Analysis of the Mash1 mutant phenotype has revealed that different genetic mechanisms are used to activate neurogenesis in two domains of the OE. Mash1 is required to activate Math4C transcription and promote neuronal differentiation in most of the OE. In contrast, another bHLH gene, Math4A, is expressed exclusively in a ventrocaudal domain of the OE where Mash1 is not expressed. Whether Math4A activates Math4C expression in this domain, or whether the two genes are activated in parallel, remains to be determined. It will be important to ascertain whether the genetically distinct populations of Mash1-dependent and -independent ORNs also have distinct functions. Each ORN expresses one or a few odorant receptor genes from a family of about 1000 genes (Ressler et al., 1993; Vassar et al., 1993). In rodents, the OE is divided into spatial zones in which different sets of odorant receptor genes are expressed (Ressler et al., 1993; Vassar et al., 1993). It is an intriguing possibility that the Math4A+ ventrocaudal domain of the OE represents such a zone where Mash1-independent ORNs express a distinct set of odorant receptor genes. Math4A, Math4C and NeuroD have similar expression patterns in most of the CNS and PNS, with NeuroD consistently expressed in more differentiated cells than Math4A and Math4C (Fig. 7J,K,L; Sommer et al., 1996; our unpublished data), suggesting that Math4A and/or Math4C may be activating NeuroD in a number of neuronal lineages. In contrast, the domain of expression of Mash1 and those of Math4A and Math4C are distinct in most regions of the CNS and mutually exclusive in the PNS, where Mash1 is expressed in autonomic precursors and the Math4 genes in sensory precursors (Fig. 7I,K,L and unpublished results). The lack of co-expression of Mash1 and Math4C in most neural tissues indicates that Mash1 must have distinct target genes in the OE and in other neuronal lineages. The comparison of the Mash1 mutant phenotypes in the ORN and sympathetic neuron lineages strongly supports this conclusion. The analysis of primary cultures and immortalized cell lines derived from Mash1 mutant neural crest cells have demonstrated that Mash1 is required at a stage of sympathetic precursor development that already expresses class III β-tubulin and other neuronal markers, for further differentiation into sympathetic neurons (Sommer et al., 1995). In contrast to this late differentiation function, Mash1 is expressed early in the olfactory placode, where it activates the expression of Math4C and of the neuronal differentiation gene NeuroD. Furthermore, Mash1 is expressed in large cell clusters in the olfactory placode, whereas the downstream gene Math4C is mainly expressed in isolated cells, suggesting that a process of lateral inhibition (Artavanis-Tsakonas and Simpson, 1991) may restrict Mash1 activity and neurogenesis to a subset of placodal cells. Altogether, these results point to a function of Mash1 in the initiation of a differentiation program in ORN progenitors, or even in the determination of the ORN lineage. The early expression of Sox2 by placodal cells (Uwanogho et al., 1995) suggests a degree of commitment to a neural fate, which precedes Mash1 activity. Mash1 expression may thus be necessary for placodal cells to make a further decision towards neuronal development, for example in a bipotential progenitor cell common to ORNs and epithelial sustentacular cells.

In summary, Mash1 is required at different stages, and serves distinct functions, in the ORN and sympathetic lineages. As previously suggested by ectopic expression experiments with Drosophila proneural genes and vertebrate myogenic genes (Brand et al., 1993; Wang et al., 1996), the specific function of a bHLH protein in a cell lineage appears to depend on the particular developmental context in which it is expressed.

We are grateful to Larysa Pevny and Robin Lovell-Badge for the generous gift of the Sox2 cDNA clone, Valérie Meyer for histological preparations, Serge Vicaire for DNA sequencing, Adrien Staub for oligonucleotide synthesis, and to Siew Lan Ang, Pat Simpson and Uwe Strahle for comments on the manuscript. This work was supported by grants from the Fondation pour la Recherche Médicale, the Ministère de l’Enseignement et de la Recherche and the Association pour la Recherche sur le Cancer, and funds from the Institut National de la Santé et de la Recherche Médicale, the Centre National de la Recherche Scientifique and the Centre Hospitalier Universitaire Régional. C.F. was supported by a long term fellowship from the Human Frontier Science Program. The Math4B and Math4C DNA sequences have been deposited in the EMBL Nucleotide Sequence Database, with accession numbers Y09166 and Y09167, respectively.