Differential expression of mammalian Numb, Numblike and Notch1 suggests distinct roles during mouse cortical neurogenesis

During mammalian neurogenesis, neural progenitors and postmitotic neurons occupy distinct regions of the neurogenic epithelium which form the walls of the embryonic neural tube. Neural progenitors occupy the inner layer (ventricular zone), whereas newborn neurons migrate outwards, away from the luminal surface, and settle in the outer layer for terminal differentiation. Such cellular events have been well documented in the developing neocortex and appear to be conserved in other regions of the central nervous system (CNS) (reviewed by Rakic, 1988; McConnell, 1995). Although various genes have been shown to be specifically expressed either in ventricular precursors or in differentiating neurons, little is known about the molecular mechanism that regulates whether ventricular daughter cells remain as progenitors or opt for terminal differentiation.

In the developing neocortex, cortical progenitors undergo mitosis with their nuclei at the ventricular surface. Retroviral lineage tracing experiments in rodents and primates have shown that a given cortical progenitor can generate neurons over multiple divisions at different times of development, suggesting that cortical neurogenesis is, at least partly, achieved through asymmetric divisions, in which one daughter cell of a cortical progenitor becomes a postmitotic neuron while the other remains as a precursor cell (Luskin et al., 1988; Price and Thurlow, 1988; Walsh and Cepko, 1988; Kornack and Rakic, 1995; Reid et al., 1995; reviewed by Rakic, 1988; Caviness et al., 1995; McConnell, 1995). Imaging of DiI-labeled ventricular cells in living ferret brain slices using time-lapse confocal microscopy has provided more direct evidence that cortical progenitors may indeed divide asymmetrically and that cell cleavage planes may predict whether a cortical progenitor divides symmetrically or asymmetrically (Chenn and McConnell, 1995). When a cortical progenitor divides with the cell cleavage plane roughly parallel to the ventricular surface, it generates two daughter cells that differ in their overall morphology and in the speed with which their nuclei move away from the luminal surface. The basal daughter cell loses its luminal contact and its nucleus moves rapidly towards the cortical plate, at a rate similar to that of the migration of young postmitotic neurons. The apical daughter cell, however, retains its luminal contact and its nucleus moves more slowly. It has been proposed that the basal daughter cell differentiates into a postmitotic neuron in the cortical plate while its apical counterpart remains as a progenitor cell within the ventricular zone (Chenn and McConnell, 1995). Recent findings indicate that some evolutionarily conserved molecular mechanisms may be responsible for specifying distinct daughter cell fates in such asymmetric divisions (Zhong et al., 1996).

During Drosophila neurogenesis, d-numb and Notch are two of the genes involved in distinct daughter cell fate specification during asymmetric divisions of neural precursors. d-numb encodes a membrane-associated protein that is asymmetrically localized to one pole of dividing neural precursors so as to be predominantly segregated to only one of the daughter cells after division. Differential segregation of d-Numb is required for the two daughter cells to adopt distinct fates (Uemura et al., 1989; Rhyu et al., 1994; Spana et al., 1995). Notch is a transmembrane receptor that mediates cell-cell interactions in various developmental pathways (reviewed by Artavanis-Tsakonas et al., 1995). Antagonistic roles of Notch and d-numb in asymmetric divisions of neural precursors are suggested by the finding that Notch mutant phenotypes are opposite to d-numb mutant phenotypes (Hartenstein and Posakony, 1990; Guo et al., 1996; Spana and Doe, 1996; Frise et al., 1996). Furthermore, d-Numb can interact physically with the Notch intracellular domain (Guo et al., 1996). It has therefore been suggested that d-Numb functions, at least partly, by causing a bias in the Notch mediated cell-cell interaction, possibly via direct physical interaction with Notch (Jan and Jan, 1995; Guo et al., 1996). Thus it appears that asymmetric cell divisions are achieved through interactions between cell-intrinsic and cell-extrinsic mechanisms.

We have recently isolated a highly conserved mammalian homologue of d-numb, m-numb (Zhong et al., 1996). During mouse cortical neurogenesis, m-Numb is asymmetrically localized to the apical membrane of dividing ventricular cells. Therefore, in an asymmetric horizontal division, m-Numb is segregated predominantly to the apical daughter cell that remains as a progenitor. Furthermore, like its Drosophila counterpart, m-Numb can physically interact with the intracellular domain of the mammalian Notch homologue, Notch1, which is also present in cortical progenitors. We have postulated that mechanisms for asymmetric divisions during neurogenesis are evolutionarily conserved and that the presence of m-Numb in apical daughter cells suppresses Notch activity in these cells, thereby allowing them to remain as progenitors. Presumably, the higher Notch1 activity in the basal daughter cell allows it to take on a fate different from that of a neural progenitor. Once it has migrated to the cortical plate, the basal daughter cell may require other factors to regulate Notch signaling before it can terminally differentiate into a neuron, since Notch signaling usually directs cells towards nonneuronal fates (reviewed by Artavanis-Tsakonas et al., 1995).

Here we report the cloning and characterization of mouse numblike (nbl), which has extensive sequence similarity to m-numb and d-numb. We show that, when expressed in dividing neural precursors in Drosophila, Nbl is symmetrically distributed in the cytoplasm. We further show that, while it fails to enable neural precursors to divide asymmetrically, nbl allows both their daughter cells to adopt the fate determined by d-Numb. In mice, nbl and m-numb have distinct but overlapping expression patterns both in the adult and during embryogenesis. We show that in the developing neocortex, Nbl is expressed in terminally differentiating neurons in the cortical plate, but not in progenitor cells within the ventricular zone, where both m-Numb and Notch1 are expressed. We also show that, in dividing cortical progenitors, while m-Numb is asymmetrically localized to the apical membrane, Notch1 is distributed around the entire cell membrane. These results may have important implications for mechanisms of progenitor cell proliferation and neuronal differentiation during mammalian neurogenesis.

A 2.6 kb rat cDNA, which encodes an open reading frame of 607 amino acids, including a stretch of 290 amino acids that shares 63.7% identity with d-Numb (Zhong et al., 1996), was used to screen a mouse midgestation cDNA library (a gift from G. Martin, UCSF) under moderate stringency [hybridization at 65°C overnight in a buffer (Church and Gilbert, 1984) containing approx. 10⁶ cpm/ml of ³²P-labeled probe with the highest stringency wash in 2× SSC/0.1% SDS at 60°C for 1 hour]. A 2.6 kb mouse nbl cDNA was isolated. The deduced amino acid sequence shares 98% identity with that of the rat cDNA, suggesting that they are homologues.

For antiserum production, a peptide corresponding to amino acids 570-603 of Nbl was synthesized. A cysteine residue was added at the N terminus for coupling to carrier keyhole limpet hemocyanin. Antibodies were raised in rabbits using standard procedures by Caltag Corporation, South San Francisco, California. The resultant antiserum (anti-MNBL1) was affinity purified using its immunogen coupled to Sulfolink columns (Pierce). However, while the antiserum strongly recognizes Nbl, it also cross-reacts with m-Numb due to the high level of conservation between these two proteins in this region. On immunoblots using adult mouse tissue extracts and in vitro translated proteins, anti-MNBL1 specifically recognized both Nbl and m-Numb, but no other proteins (W. Zhong, unpublished observation). Therefore, Nbl protein expression patterns are recognized by comparing the staining pattern of anti-MNBL1 to the staining pattern of anti-MNBR1, which specifically recognizes m-Numb (Zhong et al., 1996), and nbl mRNA expression patterns. The anti-Notch1 antibody is a rabbit antiserum against the intracellular domain of Notch1 and affinity purified.

Immunofluorescence staining of mouse sections or Drosophila embryos was as described by Zhong et al. (1996). Affinity-purified anti-MNBL1, anti-Notch1 and anti-Kv1.4 were diluted 1:1000 and immunoreactivities were visualized using biotinylated goat anti-rabbit IgG secondary antibodies and streptavidin-coupled fluorescein. Images were recorded using a Bio-Rad confocal microscope and merged using Adobe Photoshop.

The 2.6 kb rat nbl cDNA, which starts at amino acid 5, was fused inframe with the Xenopus β-globin 5′UTR and MYC epitope (MEEK-LISEED). The resultant DNA was cloned into pUAST and injected into embryos to generate transgenic flies (Brand and Perrimon, 1993). For analysis of nbl expression, UAS-nbl/Cyo flies were crossed to hairy-GAL4/TM3 flies (Brand and Perrimon, 1993) and embryos collected. For rescue experiments, UAS-nbl/TM3, UAS-d-Numb/TM3 (provided by S. Wang) or hairy-GAL4/TM3 flies were crossed to numb¹/Cyo flies (Uemura et al., 1989). The resultant numb¹/+; UAS-nbl/+ or numb¹/+; UAS-d-Numb/+ flies were crossed to numb¹/+; hairy-GAL4/+ flies and embryos collected. Rescued embryos (numb¹/numb¹; hairy-GAL4/UAS-nbl) were recognized by double labeling with anti-MNBL1 and mAb22C10 or anti-Prospero.

RNA preparation, northern blot analysis and T2 RNase protection assay were as described by Zhong et al. (1994). For the northern blot, a ³²P-labeled rat nbl fragment of 1.2 kb was used to detect nbl mRNA. The same blot was also probed with GAPDH as a control. For the T2 RNase protection assay, a ³²P-labeled antisense RNA probe of 0.5 kb (corresponding to the exon encoding amino acids 37-133 and the flanking intron sequences) was used.

Mouse embryos were fixed in 4% paraformaldehyde by overnight immersion and sectioned (10-15 μm, cryostat sections for digoxigenin [DIG] probes and paraffin sections for ³³P probes). For in situ hybridization, antisense RNA probes corresponding to mouse nbl or m-numb (approx. 1.1 to 1.6 kb for DIG-labeled probes and approx. 0.5 kb for ³³P-labeled probes) were synthesized to detect nbl or m-numb transcripts on mouse tissues, using methods described elsewhere with minor modifications (Schaeren-Wiemers and Gerfin-Moser, 1993; Duncan et al., 1994).

The production of GST fusion protein of the intracellular domain (ICD) of Notch1, in vitro translation of Numblike (full length MYC-rat-Numblike fusion used in transgenic fly production), protein binding assays and the yeast two-hybrid assay were as described (Zhong et al., 1996).

We isolated mouse nbl from a midgestation embryonic cDNA library (see Materials and Methods). This 2.6 kb cDNA encodes a protein of 603 amino acids (Fig. 1). The N-terminal portion of the protein (amino acids 42 to 331) shows strong sequence similarity to both m-Numb and d-Numb, with 76% and 63.7% identity, respectively (Fig. 2A). Like m-Numb and d-Numb, this region of the protein contains all key residues predictive of a phosphotyrosine binding (PTB) domain (Fig. 1 underlined; Zhou et al., 1995; Zhong et al., 1996). The C-terminal portion of Nbl also shows significant sequence similarity to m-Numb (46.7% identity, Fig. 2A). This segment of Nbl contains a stretch of 15 glutamine residues (Fig. 1), which are also present in its rat counterpart (23 residues) but not in m-Numb or d-Numb. Expansion of poly-glutamine repeats in various proteins has been implicated in the pathogenesis of several neurodegenerative diseases (reviewed by Ross, 1995). Overall, Nbl is more similar to m-Numb than d-Numb. The similarity between Nbl and m-Numb is present throughout the protein while the homology between Nbl and d-Numb is primarily confined to the N-terminal portion of the protein (Fig. 2B).

Unlike m-Numb and d-Numb, which are membrane associated proteins and can be asymmetrically localized in dividing neural precursors of Drosophila so as to be segregated primarily to one of the daughter cells (Rhyu et al., 1994; Zhong et al.,1996), Nbl was not localized asymmetrically in neural precursors in Drosophila. As shown in Fig. 3, the expressed Nbl was symmetrically distributed throughout the cytoplasm in non-dividing neuroblasts of the CNS (Fig. 3A) and remained so even when these neural precursors underwent mitosis (Fig. 3B-D).

To test whether Nbl is also a cytoplasmic protein in mice, we performed immunofluorescence assays on cultured murine C2C12 myoblasts that had been infected with retroviral vectors expressing either nbl or m-numb (W. Zhong, unpublished data). Indeed, the expressed Nbl was distributed in the cytoplasm while m-Numb was localized to the plasma membrane (Fig. 3E and F, respectively).

In contrast to m-Numb and d-Numb, the symmetric cytoplasmic localization of Nbl in dividing neural precursors in Drosophila results in the partition of Nbl to both daughter cells of a neural precursor. We then asked whether in each daughter cell Nbl could substitute for d-Numb in cell fate specification. Loss of numb function in Drosophila embryos results in a severe loss of neurons in the peripheral nervous system (PNS), which can be visualized by staining with the marker mAb22C10 (Fig. 4B; Uemura et al., 1989; Rhyu et al., 1994). This severe neuronal loss is caused by a change in the fate adopted by one of the sensory organ precursor (SOP) daughter cells, the IIb cell, which normally inherits d-Numb and produces a neuron and a sheath. In the absence of d-Numb, the IIb cell adopts the fate of its sister cell (IIa, see below). We expressed Nbl in these mutant embryos using a method which strongly promoted its expression in alternate parasegments (Brand and Perrimon, 1993; see Materials and Methods). As a result, the corresponding segments (t2, a1, a3, a5 and a7) no longer exhibited the neuronal loss characteristic of the Drosophila numb mutant (Fig. 4C).

Whereas the effects of Nbl expression superficially resembled those due to expression of d-Numb in the same segmental pattern (Fig. 4D), differences between Nbl and d-Numb actions became apparent upon further analysis of the formation of the two dorsal-most external sensory (es) organs in abdominal hemisegments, where the effect of d-Numb had been well characterized (Uemura et al., 1989; Rhyu et al., 1994). As schematized in Fig. 4E, each wild-type es organ consists of four cells, a neuron, a sheath cell and two outer support cells. During PNS neurogenesis, these four cells are the descendants of a single SOP cell through two rounds of asymmetric divisions. The SOP first divides into two cells, named ‘IIa’ and ‘IIb’. The IIa cell then divides to form the two outer support cells while the IIb cell gives rise to the neuron and the sheath cell. d-Numb is asymmetrically localized to only one half of the SOP cell membrane before division and segregated primarily to the IIb cell. In the absence of d-Numb, the IIb cell is transformed into a IIa cell, generating a mutant es organ with four outer support cells (Uemura et al., 1989; Rhyu et al., 1994). We used the anti-Cut antibody as a marker to recognize all eight cells of the two es organs (Fig. 4G, bracketed) and the anti-Prospero antibody to reveal the two sheath cells within this eight-cell cluster (Fig. 4F, arrows). In numb mutant embryos, the sheath cells (as well as the neurons, not shown) were transformed into outer support cells, resulting in the loss of Prospero staining (Fig. 4I) without altering the number of Cut-positive cells (Fig. 4J, bracketed). In segments of mutant embryos where Nbl was expressed, however, there were four Prospero-positive cells (Fig. 4L, arrows), in contrast to the presence of two such cells in wild-type embryos or in segments of embryos that express m-Numb or d-Numb in the same segmental pattern (Zhong et al., 1996; data not shown). The number of Cut-positive cells remained at eight (Fig. 4M). This phenotype of Nbl expression is similar to that of d-Numb misexpression, driven by a heatshock promoter, in both daughter cells (Rhyu et al., 1994). To summarize, whereas asymmetric localization of d-Numb and m-Numb results in their segregation into one of the two daughter cells of the SOP so that it adopts the fate of the IIb cell, symmetric localization of Nbl causes both daughters of the SOP to inherit Nbl and become IIb cells (Rhyu et al., 1994; Zhong et al., 1996). This indicates that the IIb cell fate can be specified by introducing d-Numb, m-Numb or Nbl into that cell.

nbl is expressed both during mouse embryogenesis and in the adult (Fig. 5A and B, respectively). The functional differences between nbl and m-numb when assayed in Drosophila neural precursor cells (Fig. 4) suggest that nbl may play roles somewhat different from m-numb during mammalian neurogenesis. To gain insight into such nbl functions, we further characterized nbl expression during mouse embryogenesis by performing in situ hybridization and immunohistochemistry on embryo sections (Fig. 6). While little nbl mRNA could be detected at E8.5 (E0.5 is noon of the day when the vaginal plugs are observed), it was clearly present in trigeminal ganglia and dorsal root ganglia by E9.5 (data not shown). At E12.5, nbl was highly expressed in the developing nervous system, with little or no expression elsewhere in the embryo (Fig. 6A-E). This is very different from m-numb, which is expressed at low levels in most embryonic structures, including the developing nervous system (Fig. 6F-H; Zhong et al., 1996). nbl mRNA was detected throughout the neural tube, from rostral telencephalon to caudal spinal cord. It was also expressed in all of the cranial sensory ganglia (V to XI), the dorsal root ganglia (DRG), the sympathetic ganglia, and the developing sensory organs such as nasal epithelia and retina (Fig. 6A-C, E and data not shown).

One important feature of nbl expression within the developing neural tube is the presence of nbl mRNA only in regions outside, but not within, the mitotically active ventricular zone. In the developing midbrain (Fig. 6D), for example, nbl mRNA was only present in the mantle zone, which contains postmitotic neurons that are undergoing active differentiation. In developing spinal cord, neurogenesis does not proceed uniformly, with a rostral (older) to caudal (younger) neurogenetic gradient. As a result, the rostral E12.5 spinal cord, which was near the end of neurogenesis, had a very thin ventricular zone and appeared to have a more diffuse pattern of nbl expression (Fig. 6E, top). In contrast, in the caudal spinal cord, where active neurogenesis was still underway, nbl mRNA (Figure 6C and E, bottom) and protein (Fig. 6K) were limited to the mantle zone, including the differentiating motor neurons of the ventral horn. m-Numb, however, was also very abundant in the germinal ventricular zone (Fig. 6L). Such differential expression of nbl and m-numb could also be observed in other regions of the developing neural tube and sensory organs (nasal epithelia and retina; Figs 6, 7 and data not shown).

The potential interaction between numb and Notch functions during neurogenesis in Drosophila and mice prompted us to further investigate the distribution of Nbl, m-Numb and Notch1 during cortical neurogenesis. In the developing mouse neocortex, Nbl immunoreactivity was mainly present in the cortical plate (Fig. 7A), which contains only postmitotic neurons that are undergoing active differentiation. The low level of staining in the ventricular zone is likely due to cross-reaction to m-Numb, since the part of the Nbl protein used for immunogen is well conserved between m-Numb and Nbl and the antiserum recognizes both proteins on immunoblots (see Materials and Methods and Fig. 2). This is consistent with the observation that nbl mRNA is only detected in cells at the cortical plate (Fig. 7A, insert). In contrast, Notch1 immunoreactivity was higher in cells of the ventricular zone, with less protein present at the cortical plate (Fig. 7G), consistent with the distribution of Notch1 mRNA observed in both mice and rats (Weinmaster et al., 1991, 1992; Lardelli et al., 1994). However, m-Numb was expressed in all layers of the cortical neuroepithelium, from the ventricular zone to the cortical plate (Fig. 7B; Zhong et al., 1996).

In interphase cells within the ventricular zone, both Notch1 and m-Numb staining appeared to be symmetrically distributed. As expected, the Notch1 immunoreactivity appeared to be membrane-bound. Since the extra-nuclear space is very thin in non-dividing cortical progenitors and the anti-Notch1 antiserum we used is against the intracellular domain of the protein, it is difficult to completely rule out the possibility that the antiserum is also recognizing Notch1 proteins that are being synthesized and transported in the cytoplasm. However, in all the late metaphase to anaphase cortical progenitors we have examined (n=75), strong Notch1 immunoreactivity was always associated with the cell membrane and never filled the cytoplasmic space created by the breakdown of nuclear membranes during cell divisions.

In dividing cells at the ventricular surface, however, while m-Numb became asymmetrically localized to the apical or apical-lateral membrane (Fig. 7B-E; Zhong et al., 1996), the Notch1 immnuoreactivity remained distributed around the entire circumference of the cell membrane (Fig. 7G-M). Since apparently asymmetric immunostaining could arise from potential section artifacts, we randomly scored m-Numb or Notch1 staining in late prophase to anaphase ventricular cells (Table 1). Only staining covering less than 75% of the cell circumference was considered as being asymmetric.

We first used an antiserum against a transmembrane potassium channel, Kv1.4, as a control to test whether we could detect consistent asymmetric localization of its immunoreactivity at either apical or basal membrane of dividing cells. In E12.5 mouse neocortex sections, strong Kv1.4 immunoreactivity was detected both in the cortical plate (data not shown) and in a significant number of cells in the ventricular zone (Fig. 7F). In most of the dividing cells at the ventricular surface, Kv1.4 staining appeared to be symmetric (88.3%, n=86, late prophase to anaphase cells). A small percentage of the cells (11.7%) did show asymmetric Kv1.4 staining. However, the location of concentrated immunoreactivity appeared to be random (3.5% basal, 3.5% lateral and 4.7% apical). We then scored similar dividing cells for m-Numb staining with an anti-m-Numb antiserum. 41.7% of such dividing cells (n=94) exhibited an apical (or apical-lateral) crescent of m-Numb immunoreactivity. In these cells, m-Numb staining only covered less than half of the cell membrane with the basal or basal-lateral cell membrane exhibited no detectable m-Numb immunoreactivity. Only 3.2% of the cells showed basal (or basal-lateral) m-Numb localization and the rest showed little m-Numb staining throughout the entire circumference of the cell membrane, in contrast to the consistently detectable m-Numb around interphase cells. When similar E12.5 neocrotex sections were stained for Notch1, 93.4% of the dividing cells (n=106) exhibited symmetric Notch1 immunoreactivity around the cell membrane. The Notch1 staining in these cells was as intense as that in interphase cells of the ventricular zone. A small number of cells also showed localized Notch1 immunoreactivity (4.7% basal or basal-lateral and 1.9% lateral).

The ability of Nbl to mimic d-Numb misexpression phenotypes in Drosophila embyros suggests that Numb can interact with downstream effectors of d-Numb. Since d-Numb may function, at least partly, by causing a bias in the Notch mediated cell-cell interaction via direct physical interaction with Notch (Jan and Jan, 1995; Guo et al., 1996), we tested whether Numblike, like m-Numb, could also directly interact with the intracellular domain (ICD) of Notch1, a mammalian homologue of Notch (Fig. 8).

We first used the yeast two-hybrid assay (Bartel et al., 1993). We constructed two plasmids, one carrying a gene encoding a fusion protein of the LexA DNA-binding domain and Nbl and the other carrying a gene encoding a fusion protein of Notch1 ICD and the GAL4 transcription activation domain. As shown in Fig. 8A, cotransformation of both plasmids into a yeast reporter strain resulted in the detection of significant β-galactosidase reporter activity, whereas control experiments using one vector and one plasmid carrying a fusion gene had little effects (Fig. 8A).

We then performed a biochemical analysis as an independent means of verifying the Nbl-Notch1 interaction observed in the two-hybrid assay. We used a fusion protein of glutathione S-transferase (GST) and Notch1 ICD that was present in protein extracts containing an excessive amount of crude bacterial proteins. Mixing such extracts with an in vitro translated Nbl protein resulted in the specific co-purification of Nbl by glutathione agarose beads (Fig. 8B), but not the vast majority of bacterial proteins, further confirming that Nbl can bind to Notch1 directly.

We have reported here the cloning and characterization of numblike, a mammalian gene with sequence similarity to d-numb and m-numb. nbl encodes a cytoplasmic protein that, when expressed in Drosophila, is symmetrically distributed in dividing neural precursors and presumably partitioned into both daughter cells to cause the daughter cells to both adopt the fate normally determined by d-Numb. During mouse cortical neurogenesis, nbl is expressed in terminally differentiating neurons in the cortical plate, but not in progenitor cells within the ventricular zone where m-numb and Notch1 are expressed. We have also found that, in dividing cortical progenitors, Notch1 appears to be symmetrically distributed around the membrane. We propose that an interplay between cell-intrinsic mechanisms (executed by m-numb and numblike) and cell-extrinsic mechanisms (mediated by Notch) may be involved in both progenitor cell proliferation and neuronal differentiation during mammalian cortical neurogenesis.

The high degree of sequence similarity between Nbl and d-Numb (Fig. 2) suggests that, like m-Numb (Zhong et al., 1996), Nbl may mimic the function of d-Numb. Indeed, when expressed in Drosophila numb mutant embryos, Nbl allows SOP daughter cells to both adopt the fate normally determined by the presence of d-Numb (Fig. 4) (Rhyu et al., 1994). This is likely the result of Nbl being unable to localize asymmetrically in dividing neural precursors (Fig. 3) while still being able to interact with down-stream effectors of d-Numb function.

When expressed in dividing neural precursors in Drosophila, m-Numb, the previously identified mammalian homologue of d-Numb, can be asymmetrically localized in a way similar to d-Numb (Zhong et al., 1996). Since Nbl and m-Numb share extensive sequence similarity throughout the protein, sequence comparison between these three Numb proteins may provide some clues to the require-ment for asymmetric Numb localization.

The ability of both Nbl and m-Numb to allow daughter cells of Drosophila neural precursors to adopt the fate normally determined by the presence of d-Numb suggests that they have overlapping functions in mice. nbl and m-numb are both expressed in differentiating neurons outside the ventricular zone during development and in adult neurons throughout the nervous system (Fig. 6 and data not shown). Therefore, it is likely that nbl and m-numb have redundant functions in specifying and maintaining neuronal differentiation.

However, several lines of evidence suggest that nbl and m-numb also have distinct functions in mice. First, nbl and m-numb have different spatial and temporal expression patterns. m-numb is expressed in most tissues whereas nbl is highly enriched in the nervous system (Figs 5 and 6). Furthermore, in the developing nervous system where both genes are expressed, nbl mRNA is only expressed in differentiating neurons outside the ventricular zone, whereas m-numb mRNA is also expressed in dividing neural progenitors (Figs 6 and 7). Second, Nbl and m-Numb exhibit different subcellular localization. While m-Numb is a membrane associated protein like d-Numb (Fig. 3; Zhong et al., 1996), Nbl appears to be a cytoplasmic protein (Fig. 3). Third, when expressed in Drosophila neural precursors, m-Numb can localize asymmetrically (Zhong et al., 1996), whereas Nbl is unable to do so (Fig. 3). This suggests that, even if Nbl were expressed in mammalian neural progenitors, it would be unable to fully substitute for m-Numb (Zhong et al., 1996).

It has been recently reported that, in the ferret, the Notch1 immunoreactivity is localized to the basal membrane of dividing cortical ventricular cells and can be differentially segregated to only the basal daughter cells (Chenn and McConnell, 1995). However, we have found that, in the great majority of neural progenitor cells in mice, Notch1 is distributed around the entire circumference of the cell membrane and segregated to both daughter cells. The anti-Notch1 antibody we used is a polyclonal antiserum raised against the intracellular domain of Notch1, which is highly conserved in other mammalian Notch proteins (Weinmaster et al., 1991, 1992; del Amo et al., 1993; Lardelli et al., 1994). In principle, it is possible that the Notch1 immunoreactivity we have observed may also be contributed to by Notch2 and Notch3, both of which are also expressed in the embryonic neocortex. On immunoblots, however, the antiserum fails to recognize in vitro synthesized Notch2 protein (G. Weinsmaster, unpublished observation). Furthermore, the staining pattern in the E12.5 embryonic brain appears to reflect the distribution of Notch1 mRNA, but not surface, whereas we have found that Notch1 remains symmetrically distributed around the membrane (Fig. 7).

It has been suggested that when a cortical progenitor divides horizontally with the cell cleavage plane parallel to the ventricular surface, it generates an apical daughter cell that remains as a progenitor and a basal daughter cell that leaves the ventricular zone for terminal differentiation (Chenn and McConnell, 1995). Such asymmetric divisions are, at least partly, responsible for cortical neuron production. It has also been suggested that, in Drosophila, d-Numb functions by physically interacting with the intracellular domain of Notch to cause a bias in Notch mediated cell-cell interactions (Jan and Jan, 1995; Guo et al., 1996). Furthermore, both m-Numb and Nbl can physically interact with the intracelluar domain of Notch1 (Zhong et al., 1996 and Fig. 8).

Notch2 or Notch3 mRNA, as previously reported using in situ hybridization (Lardelli et al., 1994). For example, Notch1 immunoreactivity could not be detected in the choroid plexus, where Notch2 mRNA is specifically expressed, nor in the diencephalon, where Notch3 expression is significant (Fig. 7N).

Chenn and McConnell (1995) observed little Notch1 staining in most interphase cells of the ventricular zone or at the apical surface of dividing cells in the ferret. It is possible that the presence of a significant amount of Notch1 immunoreactivity throughout the entire circumference of dividing ventricular cells in mice prevented us from observing quantitative differences in the amount of Notch1 protein present in apical and basal cell surface. It is also possible that only specifically modified Notch1 protein that is undetectable by our antibody is localized in mice as in ferrets. A third possibility is that there are species differences in Notch1/Numb localization; ferrets may localize both Notch1 and Numb (or only Notch1) for its cortical neurogenesis.

During mammalian cortical neurogenesis, cortical progenitors occupy the inner, ventricular zone, whereas newborn neurons migrate to the outer, cortical plate where they differentiate into various neurons (reviewed by Rakic, 1988; Caviness et al., 1995; McConnell, 1995). In the developing mouse neocortex, the nbl mRNA is only detected in cells at the cortical plate while Nbl immunoreactivity is also mainly observed in such cells. Therefore it is likely that nbl is only expressed by differentiating neurons at the cortical plate. Notch1, in contrast, is primarily expressed in proliferating cells of the ventricular zone, while m-Numb is expressed in all layers of the cortical neuroepithelium. Furthermore, in dividing cells at the ventricular surface, m-Numb is asymmetrically localized to the apical

Based on these observations, we propose that the dynamic distribution of m-Numb, Notch1 and Nbl proteins provides a molecular basis for decisions concerning neural progenitor proliferation and neuronal differentiation during cortical neurogenesis (Fig. 9). During a horizontal cell division, Notch1 is partitioned into both daughter cells while m-Numb is segregated preferentially to the apical daughter cell. m-Numb suppresses the Notch1 activity in the apical daughter cell, thereby allowing it to remain as a progenitor (Zhong et al., 1996). The basal daughter cell, in the absence of inherited m-Numb, exhibits a relatively higher Notch1 activity and, as a result, is not a progenitor, nor does it become a fully differentiated neuron until it reaches the cortical plate (Chenn and McConnell, 1995). Once there, newly accumulated Nbl and m-Numb suppress the residual Notch1 activity in the cell and allow it to fully differentiate into a neuron.

Such a scenario for cortical neurogenesis is consistent with functions proposed for Numb and Notch in cell fate determination during Drosophila development. Numb functions not to specify one particular fate but rather to ensure that daughter cells of asymmetric divisions acquire distinct fates (Jan and Jan, 1995; Zhong et al., 1996). In SOP cell divisions during adult es organ formation, for example, d-Numb is involved in specifying the IIb daughter cell of SOP, as well as the socket daughter cell of the IIa cell and the neuron daughter cell of the IIb cell (Rhyu et al., 1994). Notch, however, may play a restrictive rather than instructive role in cell fate decisions (reviewed by Artavanis-Tsakonas et al., 1995). For instance, transient expression of an activated form of Notch does not permanently endow cells with particular fates. Once Notch signaling subsides, those cells are able to recover and develop properly or respond to later developmental cues (Fortini et al., 1993; Struhl et al., 1993). It is worth noting that, in our model, only one of the proteins, either m-Numb or Notch1, is required to be localized asymmetrically to provide a molecular basis for differential daughter cell fate specification, and that localization of both proteins to opposite poles of the cell would further ensure an asymmetric division.

Recently, Verdi et al. (1996) reported the immunostaining of m-Numb in mice. The m-Numb peptide used as an immunogen by Verdi et al. (1996) is derived from a conserved region between m-Numb and Nbl (14 out of 16 amino acid residues are identical to the corresponding portion of Nbl). A cross-reactivity of this antiserum to Nbl and m-Numb may account for their immunostaining of the developing mouse neural tube, showing stronger immunoreactivity in the mantal zone than in the ventricular zone, as well as the symmetric cytoplasmic staining they oberved in developing DRG neuroblasts (Verdi et al., 1996).

It is likely that asymmetric cell divisions are not the only means for cortical neuron production, especially during late cortical neurogenesis when other mechanisms may also be involved (Kornack and Rakic, 1995; Reid et al., 1995; and reviewed by McConnell, 1995; Rakic, 1995). It is also likely that other factors are also required for various developmental decisions during cortical neurogenesis. Indeed, diffusible factors such as bFGF and NT3 have been shown to affect proliferation of neural progenitors and neuronal differentiation, respectively (Ghosh and Greenberg, 1995; Vicario-Abejon et al., 1995). Future experiments are required to test differential effects of m-numb/nbl and Notch functions, as well as how cell-intrinsic (mediated by m-numb and nbl) and cell-extrinsic mechanisms (mediated by Notch) are integrated during cortical neurogenesis.

We thank S. Wang for the UAS-d-Numbflies and G. Martin for the mid-gestation mouse cDNA library. We also thank S. McConnell, T. Doniach, L. Luo and M. Utset for helpful comments, C.-T. Chien, T. Doniach and members of the Jan lab for discussions, C.-T. Chien and S. Younger-Shepherd for assistance with fly embryo injection, P. Tsai for assistance with the yeast two-hybrid assay, and E. Frise for assistance with computer graphic work. This work was supported by a NIMH grant to the Silvio Conte Center for Neuroscience Research at UCSF. W. Z. was supported by a NIH neuroscience postdoctoral training grant and is currently supported by a postdoctoral fellowship from the American Cancer Society. L. Y. J. and Y. N. J. are HHMI investigators.