The orientation of cell division influences cell-fate choice in the developing mammalian retina

A major challenge in developmental neurobiology is to understand how the enormous diversity of cells in the mammalian CNS is produced. In invertebrates such as Drosophila and C. elegans, asymmetric segregation of cell-fate-determining proteins and mRNAs to daughter cells makes an important contribution to cell diversification(Knoblich, 2001; Lu et al., 2000; Rose and Kemphues, 1998). In the Drosophila PNS, for example, asymmetric segregation of the cell-fate determinant Numb to only one daughter of the sensory organ precursor(SOP) cell is essential to confer distinct fates to the daughter cells(Rhyu et al., 1994). Numb is a cell-cortex-associated protein that can inhibit Notch signaling, which is one way it is thought to influence cell fate(Frise et al., 1996; Guo et al., 1996; Spana and Doe, 1996).

It remains uncertain to what extent asymmetric segregation of cell-fate determinants contributes to cell diversification in vertebrates. Evidence that this mechanism may operate in mammalian CNS development came from a pioneering study by Chenn and McConnell (Chenn and McConnell, 1995) that used videomicroscopy to follow cell divisions in the ventricular zone of explants of developing ferret cortex. They showed that cells dividing with their mitotic spindle aligned horizontally to the plane of the neuroepithelium (we call these `horizontal'divisions) tend to generate two daughters that seem to remain neuroepithelial cells (NECs). By contrast, cells that divide with their mitotic spindle aligned vertically to the plane of the neuroepithelium (`vertical' divisions)tend to generate a basal daughter cell that behaves like a postmitotic neuroblast, whereas the apical daughter seems to remain an NEC. The vertical divisions are asymmetric in that an antigen recognized by anti-Notch-1 antibodies segregates exclusively to the basal daughter cell, which raises the possibility that asymmetric inheritance of cell-fate determinants might help to diversify the daughter cells. However, a limitation of this study is that the daughter cells were not followed for long enough to determine their final fate.

Several vertebrate homologues of Numb have been identified(Verdi et al., 1996; Wakamatsu et al., 1999; Zhong et al., 1996). One of them, called mammalian Numb, is asymmetrically located at the apical pole of dividing NECs in the developing mouse cortex(Zhong et al., 1996) and rat retina (Cayouette et al., 2001; Zhong et al., 1996), and has been shown to rescue the Drosophila numb-mutant phenotype(Zhong et al., 1996). Because some NECs in the developing mammalian cortex(Chenn and McConnell, 1995; Zhong et al., 1996; Zhong et al., 1997) and retina(Cayouette et al., 2001) divide vertically, Numb would be expected to segregate asymmetrically to the apical daughter cell in such divisions, whereas it would be expected to be distributed symmetrically to both daughters in horizontal divisions. Numb can decrease Notch signaling in mouse cells(French et al., 2002) and Notch signaling is known to influence cell fate in both invertebrates and vertebrates (Artavanis-Tsakonas et al.,1999; Justice and Jan,2002). Therefore, it is proposed that the two daughter cells of such asymmetric mammalian divisions have different fates(Zhong et al., 1996; Zhong et al., 1997). In the developing rodent retina, for example, expression of constitutively active Notch promotes the development of Müller glial cells(Furukawa et al., 2000). This raises the possibility that the basal daughter cell of a vertical division(which does not inherit Numb and would, therefore, have unopposed Notch signaling) might be predisposed to become a Müller cell(Cayouette et al., 2001). Recent studies on mouse cortical progenitor cells in culture show that asymmetric segregation of Numb influences the developmental fate of daughter cells (Shen et al., 2002),which supports a role for Numb in binary cell-fate decisions. The importance of Numb and its close relative Numblike in mammalian neurodevelopment is demonstrated by the severe CNS phenotype observed when both genes are inactivated in mice (Petersen et al.,2002).

In the present study, we have used long-term videomicroscopy to follow single NECs expressing GFP in newborn rat retinal explants. In this way, we were able to investigate directly whether the orientation of cell division correlates with the fate of the two daughter cells. We show unambiguously that some NECs in the living newborn retina divide vertically, confirming our previous observations on sections of fixed retina. Most importantly, we show that, at this age, horizontal divisions produce two daughter cells that usually differentiate into the same cell type, whereas vertical divisions produce two daughter cells that usually differentiate into different cell types. We also show that Numb overexpression in newborn rat retinal NECs(RNECs) decreases the number of interneurons and glia and increases the number of photoreceptor cells that develop, indicating that the concentration of Numb in RNECs can influence cell fate in the newborn retina. Last, and surprisingly, we show that at least some RNECs do not retract their basal process when they undergo mitosis. Taken together, these results indicate that the orientation of cell division in the neonatal rat retina affects the fate adopted by the daughter cells and that Numb can influence this fate.

Newborn Sprague Dawley rats were killed and the eyes removed and kept in cold HBSS until dissection. Retinal explants containing both the neural retina and retinal pigment epithelium were prepared as previously described(Cayouette et al., 2001). Each explant was placed on a Millipore-CM organotypic filter in a T25 Falcon tissue culture flask in Neurobasal medium containing 20% FCS and penicillin/streptomycin. We made certain that the edge of the explant was folded over so that we could visualize RNECs aligned along the apico-basal axis of the folded retinal neuroepithelium (see Fig. 1A). The explants were allowed to settle for a few hours in a CO₂ incubator at 37°C before they were infected with a retrovirus (see below).

All retroviruses were prepared in Phoenix packaging cells and were concentrated by ultracentrifugation, as previously described(Kondo and Raff, 2000).

For time-lapse videomicroscopy, retinal explants were infected with the LEGFP-N1 retrovirus (Clontech), which encodes GFP. About 25 μl of viral suspension was added directly on top of the explants. After an overnight incubation at 37°C, the medium was changed, and the culture dish sealed with Parafilm and transferred to the stage of an inverted fluorescence microscope enclosed in a custom-made 37°C incubator.

We used one of two retroviruses to label RNECs clones with alkaline phosphatase. The CLE control retrovirus, which encodes alkaline phosphatase alone, has been described previously(Gaiano et al., 1999). The CLE-Numb retrovirus was constructed by cloning the full length mouse Numb cDNA into the CLE retroviral vector, so that an mRNA encoding both Numb and alkaline phosphatase is expressed under the control of the Xenopus EF1α promoter, and alkaline phosphatase is translated using an internal ribosome entry site (IRES; see Fig. 7C). Newborn retinal explants were infected with either CLE or CLE-Numb retrovirus as described above. The next day, the culture medium was changed, and the explants were cultured for 10 days, changing the medium every 2-3 days. The explants were then fixed for 2 hours in 4% paraformaldehyde, rinsed in PBS and the activity of the alkaline phosphatase determined as previously described(Gaiano et al., 1999). After rinsing in PBS, the explants were cryoprotected in 20% sucrose, embedded in OCT, frozen and sectioned at 30 μm in a cryostat, as described previously(Cayouette et al., 2001). Clones of retinal cells were analyzed by counting the number of each cell type present in radial clusters, using morphology and position in the cell layers to identify the different cell types.

An inverted Zeiss Axioscope fluorescence microscope was enclosed in a custom-made 37°C incubator. GFP⁺ cells were viewed with a 10X objective. A field of interest was selected based on the presence of several GFP⁺ cells with neuroepithelial morphology. The field was imaged using a Hamamatsu CCD video camera connected to a Macintosh computer equipped with OpenLab software (Improvision) programmed to capture a frame every 7 minutes for the first 36-48 hours. To minimize light damage, we then reprogrammed the computer to capture a frame every 6-12 hours for the remainder of the recording period. In some cases, after the first 36-48 hours of recording, a mark was made on the filter and the culture flask returned to a conventional CO₂ incubator. Using the mark on the filter, we found the recording area and took a picture of it every day to follow the fate of the cells. By focusing through the explant after the recording period, we excluded the possibility that processes of out-of-focus cells were in the field of view. In some experiments, at the end of the recording period, we fixed and stained the explant with propidium iodide (5μgml^–1 in PBS + 25 U RNase A). In a few cases, we were able to use the mark on the filter to find the original recording area in a MRC 1024 confocal microscope. This allowed us to identify the two daughter cells we had followed to help determine which retinal layer the differentiated cells ended up in.

The surface area of the body of GFP⁺ cells was measured using NIH-Image Software. The software was calibrated for the camera, microscope and objectives used to acquire the images so that measurements were consistent for all cells analyzed.

We used immunofluorescence to visualize Numb in retinal explants infected with the retrovirus encoding either alkaline phosphatase alone or alkaline phosphatase and Numb. Frozen sections of the explants were incubated for 45 minutes in blocking buffer (PBS + 0.1% Triton X-100 + 3% BSA) at room temperature and then overnight at 4°C in rabbit polyclonal anti-Numb antibodies (Upstate Biotech, J. McGlade, or Q. Zhong, diluted 1:200 in blocking buffer) and rabbit anti-alkaline phosphatase antibodies (Accurate,diluted 1:100 in diluting buffer). Bound antibodies were detected using biotinylated goat anti-mouse Ig antibodies (Amersham, 1:100 in PBS), followed by Streptavidin-FITC (Amersham, 1:100 in PBS) and chicken anti-rabbit Ig antibodies conjugated to Alexa Fluor 594 (Molecular Probes, 1:100 in PBS).

To investigate directly the relationship between the orientation of cell division along the apico-basal axis and cell fate in the newborn rat retina,we needed a system in which we could both visualize the orientation of division of individual RNECs along this axis and follow the fate of the daughter cells over days. First, we tried cultures of slices of newborn retina, but extensive cell death precluded their use. Therefore, we used newborn retinal explants. By folding over the edge of the explant, cell divisions along the apico-basal axis could be observed at the edge of the fold(Fig. 1A). These explants seemed to develop normally, with mitosis occurring at the apical surface(Fig. 1B) and distinct layers of photoreceptors and interneurons developing after 6 days in culture(Fig. 1C). We infected the explants overnight with a retrovirus encoding GFP and then placed them on the stage of an inverted fluorescence microscope that was enclosed within a 37°C incubator. We followed the behavior of GFP⁺ RNECs(typically 1-10 cells per field) by time-lapse videomicroscopy.

GFP⁺ RNECs were readily identified by their typical morphology(Fig. 2A) and their characteristic interkinetic nuclear migration (basal to apical rate =1.2±0.6 μm minute^–1; mean±s.d., n=49). The plane of the interkinetic nuclear migration established the apico-basal axis of the neuroepithelium and allowed us to determine the plane of division with respect to this axis(Fig. 2B). The location of the cell division along this axis established the apical surface of the retina. We only studied cells that were strongly GFP⁺, in focus, and displayed interkinetic nuclear migration.

We recorded >150 cell divisions in 30 explants and found that RNECs divided at various orientations to the apico-basal axis(Fig. 3; Movie 1). Most of the 150 cell divisions occurred partly out of the plane of focus,making it impossible to determine their orientation. We could, however,unambiguously determine the orientation of division in 49 cases. Of these, 58%were horizontal, 34% vertical and 8% intermediate in orientation(Fig. 3B). Thus, a substantial proportion of RNECs divided vertically in the living, newborn retina,confirming our previous findings on sections of fixed retinal preparations(Cayouette et al., 2001).

To determine if the orientation of cell division correlated with the fate of the daughter cells, we followed the two daughter cells of each GFP⁺ cell that divided either horizontally or vertically for up to 5 days after the division. Most daughter cells did not divide again: they did not undergo interkinetic nuclear migration and they adopted a morphology that indicated that they had started to differentiate.

Of the 28 horizontal divisions that we followed, 14 produced daughter cells that we could not follow long enough to assess their fates. In two of the remaining 14 cases, one of the daughter cells died and the other divided again; in neither case could we follow the daughter cells of the second division to determine their fates. The remaining 12 horizontal divisions were terminal in that both daughter cells did not divide again. More than 90% (11 out of 12) of these divisions produced two daughter cells that acquired similar morphologies. Five examples, seen 60-90 hours after division, are shown in Fig. 4A-E. In one case, it was possible to stain the explant with propidium iodide (after the end of video recording) and then find the two GFP⁺ daughter cells in a confocal microscope 70 hours after the cell division that produced them. The nuclei of the two cells were located in the photoreceptor layer and resembled those of their GFP^– neighbors, which had the distinctive heterochromatin pattern characteristic of photoreceptors(Neophytou et al., 1997)(Fig. 4E). In all cases, the two GFP⁺ daughter cells resembled GFP⁺ photoreceptor cells in confocal sections of a newborn retinal explant infected with a GFP-encoding retrovirus and examined after 10 days in culture (compare Fig. 4A-E with Fig. 4F). In the single case where the two daughter cells of a horizontal division acquired different morphologies, one acquired a photoreceptor-like morphology, whereas the other became an interneuron-like cell with a large body and small processes (not shown). Thus, at this developmental age, it seems likely that most horizontal divisions produce daughter cells that become photoreceptors.

We tried repeatedly to stain the explants as wholemounts with antibodies after time-lapse recording to confirm the cell-types of the daughters, but it proved to be impossible to find the daughter cells and assess their staining. In an alternative way to help identify the differentiated cell types, we measured the size of the cell body. First, we measured the size of the cell body of GFP-infected photoreceptors and cells of the interneuron layer in cryosections of newborn retinal explants. As shown in Fig. 4G, photoreceptor cells had a cell-body size of 58±8 arbitrary units (mean±s.d., n=35), whereas cells located in the interneuron layer had a cell-body size of 113±25 arbitrary units (n=28). This difference in mean cell-body size between interneuron layer cells and photoreceptors was highly significant when analyzed by Student's t-test (P<0.0001). In the 11 out of 12 horizontal divisions that produced daughter cells with similar morphology, the cell-body size of the two daughters was similar(Fig. 4H). Daughter cells of horizontal divisions had a cell-body size of 54±16 arbitrary units(n=24), which was similar to that of photoreceptors (58±8 arbitrary units, Fig. 4G,H). In the single case where the two daughter cells of a horizontal division acquired different morphologies, the daughters had different cell-body sizes, one similar to that of photoreceptors and one similar to that of cells in the interneuron layer (Fig. 4G,H). These results strongly suggest that the majority of daughter cells of horizontal divisions in the newborn rat retina become photoreceptor cells.

Of the 17 vertical divisions that we followed, six produced daughter cells that we could not follow long enough to assess their fates. Of the remaining 11 vertical divisions, >80% (nine out of 11) generated two daughter cells that acquired different morphologies. Five examples, seen 70-96 hours after division, are shown in Fig. 5A-E. In one case, it was possible to stain the explant with propidium iodide (after the end of video recording) and identify the two GFP⁺ daughter cells in a confocal microscope 84 hours after the division that produced them. One of the daughters was large and found among cells that had a large nucleus, whereas the other was smaller and among cells that had a small nucleus. This suggests that they were in the interneuron and photoreceptor layers, respectively (Fig. 5E). In four cases, we were able to follow the fate of the apical and basal daughters separately and, in each case, the basal daughter developed as a Müller-like cell, similar to those seen in confocal sections of a newborn retinal explant infected with a GFP-encoding retrovirus and examined after 10 days in culture (compare Fig. 5D with 5F). The apical daughter cell in these four divisions became either neuron-like (Fig. 5A) or photoreceptor-like (Fig. 5D). Thus, it seems that at this developmental age most vertical divisions produce two daughter cells that become different cell types.

This conclusion was supported by measurements of cell-body size. In the 9 out of 11 vertical divisions that produced daughter cells with different morphologies, the cell-body size was different(Fig. 5H). The smaller daughters had a cell-body size of 52±13 arbitrary units(n=11), which was similar to that of photoreceptors (58±8 arbitrary units), whereas the bigger daughters had a cell-body size of 119±66 arbitrary units (n=11), which was similar to that of interneuron layer cells (113±25 arbitrary units, Fig. 5G,H). In the two cases in which the two daughter cells of a vertical division acquired similar morphologies, the body size of the daughters was similar, and also similar to that of photoreceptors (Fig. 5G,H). These results indicate that the majority of vertical divisions in the newborn rat retina produce one daughter that becomes either an interneuron or a Müller cell and one daughter that becomes a photoreceptor.

Based on electron microscopic studies, it was thought that mitotic RNECs round up and retract their basal process. However, recent studies of fluorescently labeled radial glial cells in the developing cortex suggest that these cells do not loose their basal process during division(Miyata et al., 2001; Noctor et al., 2001). Moreover, the daughter cell that inherits the basal process tends to become a neuroblast (Miyata et al.,2001). Do RNECs also maintain their basal process during division?

In most cases, we could not detect a basal process during mitosis. However,in nine out of 48 cells analyzed (19%) we detected a thin basal process that persisted throughout mitosis. One such cell is shown in Fig. 6A-D. Some of these nine cells divided horizontally, whereas others divided vertically. In one favorable case, we were able to confirm the presence of a basal process by confocal microscopy (Fig. 6E-G).

In each case in which a basal process was detected in a mitotic cell, it seemed to be inherited by only one of the two daughter cells(Fig. 6D). The daughter cell that inherited the process in either a horizontal or a vertical division always moved to take up a position basal to the other daughter cell (not shown).

One reason why vertical divisions tend to produce two daughter cells that become different cell types may be because Numb is asymmetrically segregated to the two daughters in such divisions. We(Cayouette et al., 2001) and others (Dooley et al., 2003)previously provided evidence that Numb is asymmetrically distributed on the apical side of the rat retinal neuroepithelium. However, a recent paper raised the possibility that the anti-Numb monoclonal antibody (Transduction Lab) used in these two studies recognizes another protein, called Mortalin, which has the same molecular weight as Numb (Rivolta and Holey, 2002). We therefore repeated the staining for Numb in the newborn rat retina using rabbit anti-Numb antibodies that were either obtained from J. McGlade or bought from Upstate Biotech. In both cases, we found that Numb localized asymmetrically on the apical side of the retinal neuroepithelium. The staining with the antibodies from Upstate Biotech is shown in Fig. 7A.

If asymmetric segregation of Numb influences cell-fate in the retina, one would expect to see a change in cell fate if Numb were overexpressed so that it was inherited by both daughter cells of vertical divisions. To test this expectation, we used retroviral vectors that encoded either alkaline phosphatase alone or alkaline phosphatase and Numb(Fig. 7C). We infected newborn retinal explants overnight and cultured them for 10 days. We then fixed and stained the explants to reveal clones of cells that had alkaline phosphatase activity. We analyzed the composition of 1704 Numb-infected clones and 1891 control clones using morphology and retinal layer position to identify the cell types (Fig. 7D-G). In explants infected with the retrovirus encoding Numb, Numb-immunostaining showed that many cells expressed Numb basally as well as apically(Fig. 7H); even in vertical divisions of these cells, both daughters would be expected to inherit Numb. Using the same antibody (from W. Zhong, Yale University), cells in explants infected with the control vector that encodes alkaline phosphatase alone did not express Numb basally (data not shown).

In accordance with previous studies by others, we detected clones of various sizes and composition that contained photoreceptors, amacrine cells,bipolar cells and Müller cells in various combinations (see Fig. 7D-G). However,Numb-infected explants contained fewer cells with a complex morphology than did control explants expressing alkaline phosphatase alone(Fig. 7I,J). As shown in Fig. 8A, overexpression of Numb had little effect on clone size, indicating that it did not significantly influence either RNEC survival or proliferation at this stage of development. Quantitative analysis of frozen sections indicated that clones in the control explants were similar in size and composition to those reported previously using similar retroviruses to infect newborn RNECs in vivo(Turner and Cepko, 1987; Turner et al., 1990). This indicates that our explants developed normally. By contrast, in frozen sections of Numb-infected explants, there was an increase in both the proportion of clones that contained photoreceptor cells only and the proportion of photoreceptor cells among all the infected cells(Fig. 8B,C). In these explants,there was a corresponding decrease in the proportion of clones that contained at least one nonphotoreceptor cell (amacrine, bipolar and Müller cell),as well as in the proportion of nonphotoreceptor cell types among all the infected cells (Fig. 8D,E). To compare the Numb-overexpression findings with our time-lapse imaging results,we analyzed separately the clones that contained only two cells. As shown in Fig. 8F, there were significantly more two-cell clones containing two photoreceptors and significantly fewer two-cell clones containing a mixture of cell types in Numb-infected explants than in control explants(Fig. 8F). Thus, Numb overexpression in newborn retinal explants apparently increased the development of photoreceptors at the expense of the development of interneurons and Müller cells.

We have used long-term videomicroscopy to visualize the division of GFP-labeled RNECs in explants of newborn rat retina and to follow the subsequent differentiation of the daughter cells that result. We found that RNECs that divide horizontally tend to give rise to two daughter cells that both become photoreceptor-like cells, whereas RNECs that divide vertically tend to give rise to two daughter cells that become different from each other. We also found that overexpression of Numb in RNECs in newborn retina increases the proportion of photoreceptors and decreases the proportion of nonphotoreceptor cells that develop from the RNECs. Together with our previous findings that Numb is normally segregated to the apical daughter cell in vertical RNEC divisions (Cayouette et al.,2001), these results are consistent with the hypothesis that asymmetric segregation of Numb during vertical divisions influences cell-fate choice in the newborn rat retina.

There is increasing evidence that some NECs in vertebrate neuroepithelia rotate their mitotic spindle and divide vertically during CNS development. This evidence comes from studies of the developing cortex in mouse(Estivill-Torrus et al., 2002; Zhong et al., 1996), ferret(Chenn and McConnell, 1995) and chick (Wakamatsu et al.,1999), and the developing retina in rat(Cayouette et al., 2001) and chick (Silva et al., 2002). Molecules that are asymmetrically distributed along the apico-basal axis during NEC division would segregate asymmetrically to the two daughter cells in vertical divisions.

Our present results confirm our previous findings that a proportion of RNECs divide vertically in sections of fixed retina(Cayouette et al., 2001). We showed previously that the proportion of vertical divisions is <5% in the embryonic rat retina, ∼20% in the newborn retina and ∼10% in the postnatal day 4 (P4) retina. In the present study, the proportion of vertical divisions we observed by time-lapse recording in newborn retina was 34%. There are several possible explanations why we found a higher proportion of vertical divisions in the present study. First, in our previous study, we included both metaphase and anaphase cells in the analysis. The proportion of vertical divisions was greater when we analyzed anaphase cells only, indicating that some `horizontal' metaphase cells reorient their spindle and divide vertically, as has been shown to occur in Drosophila embryos(Kaltschmidt et al., 2000; Roegiers et al., 2001). The proportion of vertical divisions in the present study is based on analyses of actual cell division, rather than spindle orientation, and is, therefore, a more accurate measure. Second, although here we studied explants of newborn retina, most of the cell divisions analyzed occurred at an age equivalent to P2. It is possible that the proportion of vertical divisions peaks at P2 and decreases by P4. Third, and perhaps most important, our time-lapse analysis is almost certainly biased toward vertical divisions, because many more horizontal than vertical divisions would be expected to occur out of the plane of focus. This bias would result in an apparent increase in the proportion of vertical divisions. Thus, we suspect that the true proportion of vertical divisions in the first two postnatal days of rat retinal development is 20-30%.

Although we mainly followed cells that divided either vertically or horizontally, four out of the 49 divisions in which we could follow the fate of the two daughters occurred with an intermediate apical-basal orientation. Such intermediate divisions could segregate cell components asymmetrically to the two daughter cells. For example, because the apical domain of NECs constitutes such a small portion of the total plasma membrane, even a slight rotation of the mitotic spindle away from the horizontal could lead to unequal distribution of apical components to the daughter cells during division(Huttner and Brand, 1997).

Chenn and McConnell (1995)were the first to use videomicroscopy to provide evidence that the plane of NEC division can influence cell-fate choice. Studying the ventricular zone of developing ferret cortex, they found that horizontal divisions generated two daughter cells that seemed to remain NECs, whereas vertical divisions generated a basal daughter cell that seemed to become a postmitotic neuroblast and an apical daughter that seemed to remain an NEC. Because the proportion of vertical divisions increased with developmental age, they proposed that early in development horizontal divisions expand the NEC pool, whereas later in development, vertical divisions produce one cell that differentiates and another that remains an NEC and continues to divide. It was not clear whether late cell divisions that produce two neuroblasts and deplete the NEC pool occurred by horizontal, vertical or both types of cell division.

In our study of the newborn rat retina, we find that the great majority of both horizontal and vertical divisions are terminal divisions, with both daughter cells differentiating. In only two out of 49 divisions (both horizontal), one of the two daughter cells divided again. In both cases, the other daughter cell died and we were unable to follow the fate of the daughter cells produced by the second division.

The most important finding of the present study is that horizontal divisions of NECs in the newborn rat retina usually produce daughter cells that acquire the same photoreceptor-like morphology and size, whereas vertical divisions usually produce two daughter cells that acquire different morphologies and sizes. The probability that chance alone accounts for 11 out of 12 horizontal divisions producing daughter cells that acquired the same morphology and size, and nine out of 11 vertical divisions producing daughters that acquired distinct morphologies and sizes is less than 0.005, assessed by chi-square analysis. In several cases, we were able to show that the two differentiated daughters of horizontal divisions have a photoreceptor-like morphology and size, and that they end up in the photoreceptor layer. However,the two differentiated daughters of vertical divisions not only usually look different and have different sizes, but they also end up in different retinal layers.

For technical reasons, it was not possible to confirm our cell-type assignments by immunostaining. However, daughter cells of horizontal divisions tend to have similar cell-body size, which is also similar to the size of GFP-infected photoreceptor cells in cryosections of retinal explants, which strongly supports our photoreceptor assignments. Thus, it seems likely that,at this stage of retinal development, horizontal divisions almost always generate two photoreceptors. These findings are consistent with previous results. Rod photoreceptors are the main cell type produced in the neonatal rat retina and make up >70% of the cells in the adult rat retina. Two-rod clones account for 79% of all the two-cell clones observed following retroviral infection of newborn rat retina in vivo(Turner and Cepko, 1987), and,as discussed above, we estimate that about 70-80% of cell divisions in the newborn rat retina are horizontal.

Analysis of cell-body size also supports our conclusion that vertical divisions in the newborn rat retina usually generate two daughters that become different. One daughter usually becomes a cell with the typical morphology and body size of a photoreceptor cell, whereas the other daughter usually becomes a cell with a morphology and body size similar to that of cells in the interneuron layer – either interneurons or Müller cells. At this stage of development, it is likely that the interneurons produced are either amacrine cells or bipolar cells. It is important to devise ways to combine time-lapse analysis with molecular identification of the differentiated cells.

Although we have focused on the orientation of cell division along the apico-basal axis of the retina, it is also possible that the orientation of division in the plane of the retina could influence cell fate. In flies and worms, for example, asymmetric divisions along various axes can influence cell-fate choices (Hyman and White,1987; Roegiers et al.,2001; Rose and Kemphues,1998). A recent study showed that the orientation of cell division within the plane of the neuroepithelium changes along the radial axes of the developing retina in both zebrafish and rat; in zebrafish, this switch correlates with neurogenesis, which is consistent with the possibility that cell-fate determinants may segregate asymmetrically along axes oriented in the plane of the vertebrate retina (Das et al.,2003).

How does the plane of cell division along the apico-basal axis influence cell fate? One mechanism may involve the asymmetric segregation of Numb during vertical division. Because Numb is localized apically in the rat RNECs(Fig. 7A), only the apical daughter of vertical divisions should inherit Numb(Cayouette et al., 2001). Because Numb can inhibit Notch signaling(French et al., 2002; Frise et al., 1996; Guo et al., 1996; Spana and Doe, 1996), and Notch signaling can influence cell fate in the developing rat retina(Bao and Cepko, 1997; Furukawa et al., 2000), it is possible that the two daughters of vertical RNEC divisions develop differently because of differences in Notch signaling associated with differences in the levels of Numb. Our finding that the overexpression of Numb in RNECs in newborn retinal explants promotes photoreceptor development at the expense of interneuron and Müller cell development is consistent with this. Although we cannot exclude the possibility that overexpression of Numb decreases the probability that a RNEC will divide vertically, experiments in flies makes this unlikely. Our Numb overexpression results are consistent with our observation that horizontal divisions, in which Numb is expected to be distributed symmetrically to daughter cells, produce two photoreceptor-like cells, whereas vertical divisions, in which Numb is expected to be asymmetrically distributed produce two daughters that become different from each other. These findings are also similar to the results seen following overexpression of Numb in SOP cells in Drosophila(Rhyu et al., 1994). In dividing SOP cells, Numb is normally asymmetrically segregated to the daughter cell that becomes the IIb cell. The cell that does not inherit Numb becomes the IIa cell, apparently because of unopposed Notch signaling(Frise et al., 1996; Guo et al., 1996). When Numb is overexpressed in the SOP cell, both daughters become IIb cells(Rhyu et al., 1994). Thus, in both SOP cells and RNECs, Numb overexpression reduces cell diversification. Similarly, mouse cortical progenitor cells in culture can asymmetrically distribute Numb when they divide, thus influencing the fate of the daughter cells (Shen et al., 2002). Moreover, cortical progenitors isolated from Numb-knockout mice show a 50% reduction in their ability to produce daughter cell pairs that adopt different fate (Shen et al.,2002). In the same study, it was found that in terminal divisions that produced two neurons in which Numb was asymmetrically distributed, the neurons acquired different morphologies. This observation is similar to our finding that vertical divisions, in which Numb would be asymmetrically inherited, produce two daughters that adopt different morphologies.

Whereas Numb overexpression in newborn rat RNECs inhibits Müller-cell development, it has been shown previously that excessive Notch signaling in RNECs increases Müller-cell development(Furukawa et al., 2000). On the basis of the Müller-cell-promoting effect of excessive Notch signaling and apical localization of Numb in RNECs, we hypothesized previously that the basal daughter of a vertical RNEC division would tend to differentiate into a Müller cell(Cayouette et al., 2001). Our present results are consistent with this hypothesis but indicate that it is an oversimplification. In the four vertical divisions in which we were able to follow the fate of the apical and basal daughter cell separately, the basal cell acquired a Müller-cell-like morphology, whereas the apical cell acquired either a photoreceptor-like or a neuron-like morphology. However, not all vertical divisions produce Müller-like cells (see, for example, Fig. 5B). Thus, it seems likely that the apical daughter cell of a vertical division of a newborn RNEC that inherits Numb has an increased probability of becoming a photoreceptor cell,whereas the basal daughter that does not inherit Numb has an increased probability of becoming either an interneuron or a Müller cell. Numb might act positively, to promote photoreceptor development, negatively, to inhibit interneuron and Müller cell development, or in both ways.

In a recent study, Silva et al. (Silva et al., 2002) found little correlation between cell-fate choice in the chick retina and either the orientation of cell division or the distribution of Numb. Although this seems at odds with our findings, there are a number of reasons, apart from species difference, why the two studies come to such different conclusions. One is that Silva et al.(Silva et al., 2002) studied earlier stages of retinal development. Because there are very few vertical divisions in the rat retina this early age, the mechanism we propose is unlikely to operate. Another is that they use the term asymmetric division to describe a division in which one daughter cell continues to divide and the other differentiates, whereas we use it to describe a division in which one or more molecules are asymmetrically segregated between the two daughter cells. Silva et al. use the term symmetric division to describe a division in which both daughter cells continue to divide but, without following the fate of the two daughters, it is impossible to know whether they are the same or not; they could have inherited different components from the mother cell, for example,and have different fates. Similarly, cell divisions in which both daughter cells differentiate can be asymmetric if the two daughters become different,as shown in this study. Even at the early stages of retinal development studied by Silva et al., the plane of division could, apparently, influence cell-fate choice: in the rare divisions with a vertical spindle, the apical daughter cell never expressed Numb and never differentiated into a retinal ganglion cell.

Because Notch signaling has different functions in different tissues during development, and at different developmental times in the same tissue, it seems likely that Numb will also have different functions in different tissues and at different developmental times (Cayouette and Raff, 2002; Zhong,2003). Indeed, experiments in both flies and vertebrates indicate that Numb proteins help make the two daughter cells of a division develop along different pathways, rather than to specify a particular cell type. For example, in Drosophila Numb functions in binary cell-fate decisions but does not induce one particular fate; it can help a daughter cell to either remain a progenitor cell or to develop into a neuron, glial cell or other cell type (Gho et al., 1999; Rhyu et al., 1994; Roegiers et al., 2001; Spana et al., 1995; Van De Bor et al., 2000). Similarly, in rodent cortical progenitor cells dividing in culture, asymmetric segregation of Numb is observed in divisions that produce two different cell types, but it can be inherited by a cell that remains a progenitor cell or by a cell that becomes a neuron (Shen et al.,2002).

Although it was generally believed that NECs round up and retract their basal process during division, it was shown recently that radial glial cells maintain a basal process during mitosis(Miyata et al., 2001; Noctor et al., 2001). This is inherited asymmetrically by the daughter cell that becomes a postmitotic neuroblast (Miyata et al.,2001). Moreover, in the fish retina, it has been shown recently that mitotic cells always retain a thin basal process that can be inherited asymmetrically (Das et al.,2003). In agreement with these results, we find that at least some RNECs retain their basal process during mitosis. Although it is possible that only a proportion of mitotic RNECs retain their basal process, it seems more likely that they all do, but that some were too thin to be detected.

We saw RNECs that maintain their basal process during both horizontal and vertical divisions. In favorable cases, the basal process was asymmetrically inherited by one of the two daughter cells, in both kinds of divisions. We could not determine whether the inheritance of the basal process influences cell fate because the two daughter cells often stay so close to each other that it is difficult to follow them separately for long periods. When we were able to follow the two daughter cells separately during the first few hours after division, the cell that inherited the process always migrated to lie basal to its sibling. The differentiated cells in retinal clones are aligned radially along the apico-basal axis(Turner and Cepko, 1987; Turner et al., 1990) and it may be that the retained basal process helps cells in a clone achieve this radial organization.

In conclusion, it seems likely that the asymmetric segregation of cell-fate determinants during division plays a much greater part in influencing cell-fate decisions in vertebrate development than is generally appreciated. There are increasing numbers of examples where a proportion of vertebrate epithelial cells reorient their mitotic spindle to divide vertically to the plane of the epithelium. These include ependymal cells(Johansson et al., 1999), stem cells of the oesophageal epithelium (Seery and Watt, 2000) and several NECs(Cayouette et al., 2001; Chenn and McConnell, 1995; Chenn et al., 1998; Silva et al., 2002; Wakamatsu et al., 1999; Wakamatsu et al., 2000; Zhong et al., 1996). The challenge is to identify the putative cell-fate determinants and to determine how they influence cell fate.

We are grateful to Weimin Zhong for the Numb cDNA and anti-Numb antibodies, Gord Fishell for the CLE retroviral vector, Jane McGlade for anti-Numb antibodies, Mark Shipman for help with confocal microscopy and time-lapse videomicroscopy, Bill Harris and Ben Barres for comments on the manuscript, and members of the Raff lab for stimulating discussion and support. M.C. was funded by a Long-Term Fellowship from the Human Frontier Science Program Organization and M.R. was funded by the Medical Research Council.