Notch 1 inhibits photoreceptor production in the developing mammalian retina

Studies conducted in vertebrates and invertebrates over the last decade have led to the realization that a few families of signaling molecules are used repeatedly to regulate multiple aspects of development. The Notch transmembrane receptor is one such molecule. Notch was first discovered in Drosophila, and was subsequently shown to be crucial for proper development in vertebrates as well(Artavanis-Tsakonas et al.,1999). However, the precise roles played by Notch in vertebrates have not yet been elucidated. One of the outstanding issues concerns its role in regulating cell fate decisions. Introduction of an activated allele of Notch into the vertebrate nervous system has led to the suggestion that it controls the decision of a cell to become a neuron or a glial cell(Furukawa et al., 2000; Gaiano et al., 2000; Morrison et al., 2000). However, loss-of-function studies have not clearly demonstrated this role. In part, this is due to the fact that mice with a null allele of Notch1die at E10, prior to gliogenesis (de la Pompa et al., 1997). Clarification of the role of Notch in the neuron versus glial cell fate decision, or of the potential role of Notch in regulating which type of neuron might be made, will require the use of a system where one can readily quantify the various neuronal cell fates, as well as exert temporal and spatial control over the loss of Notch.

The retina offers many advantages for the dissection of the roles of a molecule such as Notch. Retinal development is relatively well characterized. The vertebrate retina is an exquisitely light-sensitive tissue capable of transducing the signal from a single photon into a neural stimulus. Moreover,it performs sophisticated information transformations that begin the process of vision. To perform these functions, it requires the correct stochiometric production of seven major cell types during development. Retinal ganglion cells, horizontal cells, rod and cone photoreceptor cells, amacrine cells,bipolar cells, and Muller glial cells arise from multipotent RPCs in a conserved order during development (reviewed by Altshuler et al., 1991). Birthdating studies in the mouse have shown that retinal ganglion cells are the first-born cell type, whereas rods, bipolar interneurons and Muller glial cells are the late-born cell types(Sidman, 1961; Young, 1985).

Although, there is considerable overlap in the generation of the various cell types, there are a limited number of cell fates available to the daughter of an RPC at any given time. For example, while extrinsic factors can alter the ratio of the various cell types produced at one time point (reviewed by Levine et al., 2000),heterochronic mixing experiments have established that progenitor cells from a particular time in development cannot be induced to generate temporally inappropriate cell types (Belliveau and Cepko, 1999; Belliveau et al.,2000; Rapaport et al.,2001). The logic of these observations can be accommodated by the competence model of retinal development(Cepko et al., 1996; Livesey and Cepko, 2001). This model proposes that progenitor cells progress through temporal states; in any particular state, the progenitor is competent to produce only a subset of retinal cell types. A predicted consequence of the competence model is that the failure to maintain RPCs would result in the generation of early born cell types, but not the later born cell types, as progenitor cells would be prematurely depleted and unavailable for the production of later born cell types.

Several gain- and loss-of-function studies of the Notch pathway in the developing retina have suggested a crucial role for this pathway in controlling progenitor multipotency as well as proliferation and apoptosis(Ahmad et al., 1997; Ahmad et al., 1995; Austin et al., 1995; Bao and Cepko, 1997; Dorsky et al., 1997; Dorsky et al., 1995; Furukawa et al., 2000; Henrique et al., 1997; Lindsell et al., 1996; Scheer et al., 2001; Silva et al., 2003; Waid and McLoon, 1998). Expression analysis have further established that Notch1 and Notch3 are both expressed in the developing central retina, whereas Notch2 is expressed in the peripheral retina and the retinal pigmented epithelium (Bao and Cepko,1997; Lindsell et al.,1996). In chick embryos, reduction of Notch1 levels in the early retina enhanced production of the first born retinal cell type – the retinal ganglion cell. In that study, it was also determined that a high percentage of retinal progenitors were competent to adopt the retinal ganglion cell fate, and that Notch signaling was limiting in permitting ganglion cell production (Austin et al.,1995). In complementary studies, the Notch pathway was constitutively activated, which led cells to adopt an undifferentiated progenitor-like state (Dorsky et al.,1995). In fish and rodent, these progenitor-like cells have glial properties, suggesting that Notch1 also may be regulating the glial versus neuronal cell fate decision (Furukawa et al., 2000; Scheer et al.,2001).

Some of the downstream effectors of Notch signaling have been described;the most notable of which is the Hes gene family(Iso et al., 2003). This class of basic helix-loop-helix (bHLH) genes are homologous to Drosophila hairy and Enhancer of split E(spl). In both vertebrates and invertebrates, these transcription factors are known to mediate at least a subset of Notch signaling activities. Evidence for the relationship between Notch1, Hes1 and Hes5 in the retina is supported by misexpression experiments in which these negative bHLH proteins are sufficient to promote the Muller glial/late progenitor-like fate at the expense of neurons(Furukawa et al., 2000; Hojo et al., 2000; Takatsuka et al., 2004). Interestingly, these negative bHLHs do not re-capitulate all of the effects observed with constitutive Notch activation. It was found in the rat retina that activated Notch led to glial formation, concomitant with hyperproliferation, while Hes1 misexpression directed glial formation without promoting proliferation (Furukawa et al.,2000). These data suggest that Notch1 activation engages as yet unidentified downstream effectors, in addition to Hes gene family members.

Null alleles in Hes1 and Hes5 have been generated in mice and analyzed for retinal defects. Hes5-null retinae have normal retinal size and morphology, but display a 30-40% reduction in Muller glial cell production relative to wild-type littermates(Hojo et al., 2000). Hes1-null mice are more severely affected(Tomita et al., 1996). These retinae are reduced in size and full of abnormal rosettes. Cell fates in these retinae were dramatically altered as there was an increase in early born cell types, such as retinal ganglion cells, horizontal cells and amacrine cells,and a decrease in the later born Muller glial cells(Takatsuka et al., 2004; Tomita et al., 1996). Furthermore, horizontal and photoreceptor cell markers were prematurely expressed. Together, the loss-of-function experiments suggest that Notch-Hes signaling is important for the maintenance of a pool of progenitor cells and the acquisition of the glial fate.

The above data support a model in which Notch1 is expressed by RPCs of each competence state. Notch signaling allows cells to remain undifferentiated and continue proliferating, and progenitor cells downregulate Notch activity in order to produce a postmitotic cell that becomes a neuron. We directly tested this model by conditionally inactivating Notch1 at different times in murine retinal development. Notch1 was found to be crucial for multiple aspects of development, including regulating organ size,morphogenesis and cell fate decisions. Consistent with the competence model,ablating Notch1 early led to enhanced production of an early-born cell type, whereas eliminating Notch1 later led to production of a later born cell type. Surprisingly, progenitors in which Notch1 is deleted do not simply produce all neuronal fates available in their competence state; they predominantly produce the rod or cone photoreceptor fates. These unexpected results demonstrate a crucial role for Notch signaling in inhibiting the photoreceptor fate, and suggest a broader role for Notch signaling in not only inhibiting the neuronal fate, but also in selectively biasing production of one neuronal cell type over another.

Mice carrying the Chx10-CRE BAC transgene(Rowan and Cepko, 2004), Foxg1-CRE knock-in allele (Hebert and McConnell, 2000), R26R allele(Soriano, 1999) or Notch1 floxed allele (Radtke et al., 1999) were crossed to generate mice hemizygous for R26R allele, hemizygous for Chx10-CRE or hemizygous for Foxg1-CRE and homozygous or hemizygous for the Notch1 floxed allele. Notch1 flox/+; CHX10-CRE mice were phenotypically indistinguishable from wild-type mice and used as controls. Mice were genotyped for the Notch1 floxed allele, Chx10-CRE and R26R using DNA extracted from tail preps and PCR amplification as described (Radtke et al.,1999; Rowan and Cepko,2004; Soriano,1999). Longwood Medical Area's Institutional Animal Care and Use Committee (IACUC) approved the animal experiments.

In vitro infection was performed on dissected retinae grown as explants on top of polycarbonate filters in 12-well culture plates with media. The media consisted of 45% DMEM, 45% F12 medium and 10% fetal calf serum (FCS). The retina was immediately infected with virus diluted in a 5 μl droplet of media. In vivo infection was performed by injecting 0.5 μl of highly concentrated virus into the subretinal space of P0 or P3 mice. Mice were allowed to develop for 2-4 weeks before processing. Freshly dissected retinae were cultured with media containing BrdU for 1 hour at 37°C. Labeled cells were detected according to the manufacturer's instructions (Roche).

Myc-tagged cre recombinase (T. Matsuda and C.L.C., unpublished) was cloned directly into the retroviral expression construct pNIN(Dyer and Cepko, 2001a). Virus was produced and concentrated as described(Cepko et al., 1998).

RNA was purified from mouse retinal tissue by Trizol extraction (GIBCO). cDNA was prepared from RNA essentially as described(Tietjen et al., 2003). In brief, DNA was reverse transcribed from RNA with reverse transcriptase (GIBCO)and a polyT primer (TATAGAATTCGCGGCCGCTCGCGAT(24)) (Oligos etc.). PolyA tailing was achieved with terminal transferase (TdT, Roche). DNA was subsequently amplified by PCR using the polyT primer and LA-Taq DNA polymerase(Takara). Between 16 and 22 cycles of PCR amplification was used to obtain the necessary amount of DNA still within the linear range of amplification. For microarray hybridization, cDNAs were labeled by incorporation of Cy3- or Cy5-dCTP (Amersham-Pharmacia). Pairs of labeled probes were hybridized at 42°C overnight microarrays consisting of clones from the Brain Molecular Anatomy Project clone set (kind gift of Dr Bento Soares) and additional clones of interest from the laboratory (Livesey et al., 2004). Slides were washed extensively(Young and Cepko, 2004) before scanning in an Axon GenePix 4000B Scanner (Axon Instruments). Data images were retrieved and analyzed using the GenePix software package (Axon Instruments).

All tissue sections were prepared by equilibrating retinae in 30%sucrose/PBS at 4°C, followed by equilibration in a 1:1 mix of 30%sucrose/PBS and OCT and finally embedded in OCT. All sections were 20 μm. For immunostaining, sections were hydrated in PBS and then blocked in 5% goat serum in 0.1% Triton X-100/PBS (PBST). Primary antibody was applied for 2 hours at room temperature, followed by three PBST washes and secondary antibody incubation at room temperature for 2 hours. Antibodies used were:anti-β-tubulin III (mouse monoclonal, 1:200, Upstate), anti-RMO 270.7 directed against low-molecular weight neurofilament(Carden et al., 1987) and goat anti-mouse or goat anti-rabbit Cy3 or Cy5 (Jackson Immunoresearch Laboratory 1:200). Cells were counterstained with DAPI and washed several times. For alkaline phosphatase staining, retinal tissue was harvested and fixed with 4%PFA/PBS for 15 minutes at room temperature. After extensive washing in PBS,endogenous alkaline phosphatase activity was heat inactivated by placing the tissue at 65°C for 1 hour. Retinae were then stained as whole mounts for 1-4 hours at room temperature (Fekete and Cepko, 1993). For X-gal staining, retinal tissue was harvested and fixed with a 1% PFA/0.5% glutaraldehyde mix as described(Kwan et al., 2001). Retinae were stained as whole mounts overnight at 37°C. Section in situ hybridization was performed on retinal cryosections as described(Chen and Cepko, 2002). The following probes were used: Crx (Furukawa et al., 1997), Otx2 (Ang et al., 1994), NeuroD1 (Morrow et al., 1999), S-opsin and M-opsin(Corbo and Cepko, 2005), and genes listed below in the GenBank Accession Numbers section. All images were taken on a Nikon Eclipse E1000 microscope using a Leica DC200 digital camera. For further descriptions see also http://www.stjude.org/faculty/0,2512,407_2030_10417,00.html.

PNR, BC017521; NRL, BF464350; islet 1, AI845893; NF-L, BE953485; Hes1, BI557608; Hes5, BE952148; Hey1, AI851652; Chx10, BF461223; Delta-like 1, AW047187; Math3, AI846749; clusterin, BE996359; p57, BF464158; Pax6, BE953199; Notch1, BE981557.

The requirement for Notch1 in early retinal development was investigated by breeding floxed Notch1 mice(Radtke et al., 1999) to mice expressing cre recombinase under the Chx10 promoter(Rowan and Cepko, 2004) (see Fig. 2A). This strategy circumvented the embryonic lethality observed in Notch1-null mice(de la Pompa et al., 1997). Chx10 is a relatively retinal specific transcription factor that is expressed in progenitor cells starting at ∼E10. The Chx10-CRE line has been shown to have cre activity beginning around the onset of Chx10expression (Rowan and Cepko,2004). Floxed Notch1; Chx10-CRE mice had wild-type viability and survived to adulthood, consistent with the finding that Chx10 expression is relatively specific to the retina. Most notably, reduced eye size was observed in all Notch1 conditional knockout mice (CKO) relative to their wild-type littermates, as early as E13.5. Hematoxylin and Eosin staining of wild-type and mutant littermates was performed at P21 when retinal development is complete. Relative to retinae from wild-type littermates, retinae from Notch1 mutant mice displayed severe morphological defects and reduced optic nerve thickness(Fig. 1A-B and not shown). Cells in the outer nuclear layer (ONL) of the mutant mice were aberrantly organized into rosette-like structures. The number of cells in the inner nuclear layer (INL) was severely reduced in the mutant retinae compared with the wild-type retinae.

As Notch signaling is known to have cell-autonomous and cell non-autonomous effects on cell proliferation and apoptosis(Artavanis-Tsakonas et al.,1999), it was unclear whether the reduction in retinal size was due to a cell autonomous reduction in Notch1 in proliferating progenitors or to a cell non-autonomous effect. To distinguish between these possibilities, Notch1 flox/flox retinae were explanted at E14.5 and infected with a low titer replication incompetent retrovirus encoding nuclearβ-galactosidase without (NIN) or with cre recombinase (NIN-CRE). Retinae were allowed to grow as explants for 10 days and then processed forβ-galactosidase histochemical detection using X-gal. Single clones were easily identified as patches of X-gal+ cells, which were assayed for clone size (Fig. 1C). NIN-infected control clones varied from 1-52 cells per clone, with an average of 5.9±1.3 cells per clone. By contrast, NIN-CRE infected clones ranged from 1-28 cells per clone, with an average of 3.7±0.6 cells per clone. There was a slight increase in 1-3 and 4-6 cell clones, and a slight decrease in 7-9 and greater than 10 cell clones, in NIN-CRE versus NIN clones(Fig. 1D). The overall reduction in clone size in the Notch1-deleted progenitors suggests that cell-autonomous effects account for at least some aspects of progenitor cell proliferation.

To better understand the early events that may have contributed to the abnormal retinal morphology and size observed in Notch1-null adult mice, Notch1-deficient mice were examined at embryonic timepoints. In the course of using the Chx10-CRE transgenic mice, variable levels of CRE activity have been observed (Rowan and Cepko, 2004). As a result, variable degrees of the mutant phenotype were observed depending on the percentage of cells with deletion of Notch1. To track the degree of cre recombinase activity more precisely and fate map the individual cells in which Notch1 was probably deleted,the R26R cre reporter allele(Soriano, 1999) was crossed into the Chx10-CRE; Notch1 flox/flox mice(Fig. 2A). Retinae from mutant mice with a very high percentage of X-gal+ cells were consistently very small in size and highly rosetted as early as E13.5. Mutant retinae with a much lower percentage of X-gal+ cells were more wild type in size and displayed normal lamination.

Fate mapping of wild-type littermates at E16.5 and E18.5 by X-gal staining revealed cells with morphology consistent with progenitor cells in that they had radial processes that spanned the retina. In addition, there were X-gal+cells in the outer neuroblastic layer (ONBL) and in the inner neuroblastic layer (INBL) (Fig. 2B,C). These data are consistent with previous experiments that examined the morphology of individual X-gal+ retinal cells in Chx10-CRE/+; R26R/+ mice. That study demonstrated the potential of Chx10-expressing progenitor cells to produce all retinal cell types(Rowan and Cepko, 2004). In mutant littermates with a relatively low percentage of X-gal+ cells, very few of the X-gal+ cells exhibited radial or progenitor-like morphology. There was an increase in more compact X-gal+ cells lining the scleral surface of the ONBL and in the INBL (Fig. 2D,E). The locations and morphologies are suggestive of cells that have exited cell cycle and begun to differentiate prematurely.

Markers of RPCs were examined in Notch1-deficient retinae at E13.5. Fgf15 and cyclin D1 are normally expressed by RPCs at this age(Blackshaw et al., 2004; Sicinski et al., 1995). These genes were markedly decreased in the Notch1 CKO(Fig. 3C,D,G,H). Previous studies in invertebrates and vertebrates have demonstrated that at least a subset of Notch activities are mediated by members of the E(Spl)-related Hes genes (Iso et al., 2003). To address the consequence of deleting Notch1 on these canonical downstream effectors, in situ hybridization for Notch1, Hey1, Hes1 and Hes5 at E13.5 was performed. Both Notch1 and Hey1were noticeably reduced (Fig. 3A,B,E,F), but changes in Hes1 or Hes5expression were not seen (data not shown). Interestingly, Hes5, along with Notch1 and Hey1, was found to be lower by microarray expression profiling (see below).

A decrease in Notch signaling, as well as in expression of progenitor cell markers, might be due to a depletion of RPCs resulting from cell death or precocious neurogenesis. To distinguish between these possibilities,expression of markers of neuronal differentiation, as well as proneuronal genes, were examined. The Notch ligand, Delta-like 1 (Dll1)(Fig. 3I,M), and proneuronal bHLHs Hes6, Neurod1 and Math3 were significantly upregulated in the CKO retinae (Fig. 3J-P). Interestingly, expression profiling and in situ hybridization (data not shown)failed to detect a change in Math5, a proneuronal bHLH that is essential for ganglion cell production. To examine cell death, TUNEL staining was performed on CKO versus wild-type retinae at early stages (E12.5 and E13.5) and later stages (P4 and P10). No differences in TUNEL labeling were observed (data not shown). Together, these data are consistent with an increase in neurogenesis in Notch1-deficient retinae, probably as a consequence of progenitor cells producing a higher ratio of postmitotic to mitotic progeny, relative to control retinae.

Previously, it was observed that reduction of Notch1 levels early during chick retinal development with antisense oligonucleotide-mediated knockdown led to an enhancement of ganglion cell production(Austin et al., 1995). As deletion of Notch1 in the embryonic mouse retina led to the precocious onset of neuronal markers, it was next determined whether completely removing Notch1 results in a similar increase in retinal ganglion cell production. In addition, it was of interest to determine if removal of Notch1 led to a change in proliferation at this age. Retinae were harvested from wild-type and Notch1 CKO mice at E14.5 and pulsed with BrdU for 1 hour. Retinae were dissociated and then immunostained with anti-BrdU and two antibodies that recognize retinal ganglion cells, anti-β-tubulin III and anti-neurofilament (antibody#270.7) (Fig. 4A). The percentage of DAPI-positive cells that incorporated BrdU was 32±8% in the wild-type retinae when compared with 17±2% in the Notch1CKO retinae. This reduction in BrdU-positive cells in the Notch1 CKO is consistent with mitotic cells either prematurely exiting cell cycle and producing neurons or an alteration of cell cycle kinetics. Surprisingly, the percentage of DAPI-positive cells expressing β-tubulin III and NF was also reduced in Notch1 CKO retinae relative to wild-type retinae. The percentage of β-tubulin III-positive cells in the wild-type retinae was 18±1%, whereas in the Notch1 CKO retinae it was 8±2%. Similarly, the percentage of NF-positive cells was reduced in the mutant retinae from 17±1% cells in the wild-type retinae to 9±3% cells in the Notch1 CKO retinae.

Microarray analysis was carried out on CKO versus wild-type E13.5 mouse retinae to determine more broadly the molecular changes that resulted from deleting Notch1 (Fig. 4B). Consistent with in situ hybridization analysis, components of the Notch pathway (Notch1, Hey1) as well as progenitor markers (cyclin D1, Fgf15, Sfrp2) were lower in mutant retinae(Fig. 4C). In addition, higher levels of the proneuronal bHLHs Neurod1 and Math3 were seen. Consistent with the dissociated cell immunostaining analysis, there were lower levels of the retinal ganglion cell markers, islet 1 and Gap43. By contrast, higher levels of the photoreceptor transcription factors Crx and Otx2, and of the photoreceptor specific gene cone transducin α were observed (Fig. 4D).

To confirm these results, in situ hybridization was carried out on E13.5 Notch1 mutant retinae with the photoreceptor genes, Otx2 and Crx (Fig. 5A,B,E,F),and retinal ganglion cell markers, neurofilament light (NF-L; Nefl– Mouse Genome Informatics) and islet 1(Fig. 5C,D,G,H). As predicted by the microarray results, a significant increase in photoreceptor markers concomitant with a decrease in retinal ganglion cell markers were observed. Thyroid hormone receptor β 2 and retinoid x receptor gene γ are two additional genes involved in early photoreceptor development(Forrest et al., 2002; Hoover et al., 1998). Both of these genes were also observed to be upregulated in Notch1 CKO retinae at E13.5 by in situ hybridization (data not shown). These data suggest that enhanced neurogenesis in E13.5 Notch1-deficient retinae included an increased production of photoreceptors. Interestingly, not all early-born neurons were increased; in fact, there was a decrease in expression of retinal ganglion cell markers.

As previous data in other species had suggested that reduction of the Notch pathway would result in enhanced production of neuronal cell types, and in particular of the first-born retinal ganglion cell(Austin et al., 1995), the early requirement for Notch1 in the mouse was further determined by breeding floxed Notch1 mice to mice expressing cre recombinase under the Foxg1 promoter (Hebert and McConnell, 2000). Foxg1 is a winged helix transcription factor that proceeds Chx10 expression in the retina. Fate-mapping experiments have determined that Foxg1 descendants are uniformly labeled in the nasal embryonic retina(Hebert and McConnell, 2000; Pratt et al., 2004). Deletion of Notch1 in Foxg1-CRE expressing cells therefore should exclude the timing or variable expression of Chx10-CRE as a potential explanation for why photoreceptor production is enhanced, while ganglion cell production is reduced.

To confirm that the loss of Notch1 reduces progenitor status and enhances neurogenesis, in situ hybridization was carried out on E13.5 Notch1 flox/flox; Foxg1-CRE mutant retinae for cyclin D1 and Dll1. As predicted, cyclin D1 expression(Fig. 5I,M) was reduced,whereas Dll1 expression was enhanced (data not shown). In situ hybridization was next performed for the photoreceptor genes Otx2 and Crx and retinal ganglion cell markers, Nefl and islet 1. Both photoreceptor genes were upregulated(Fig. 5J,N and not shown),whereas both ganglion cell genes were downregulated(Fig. 5K,L,O,P). These data further corroborate a requirement for Notch1 in the production of retinal ganglion cells. The Notch1 flox/flox; Foxg1-CRE mice are perinatal lethal (Mason et al.,2005) and so later developmental aspects of Notch signaling were examined only in the Notch1 flox/flox; Chx10-CRE mice.

Ablation studies of Notch1 in other regions of the nervous system have demonstrated a requirement for Notch1 in the completion of neuronal differentiation. Premature differentiation of Notch1-deleted cells in the cerebellum led to an upregulation of proneuronal bHLH proteins at embryonic stages, but fate mapping of these mutant cells into the adult stages revealed a decrease in cell survival. As a result, many Notch1-deficient neurons were eliminated by apoptosis and very few of them were detected in the mature cerebellum(Lutolf et al., 2002).

To test for a similar requirement for Notch1 in the retina, the fates of X-gal+, presumed Notch1^-/-, cells in the fully formed P15 retina were determined. The wild-type littermate retinae revealed X-gal+ cells contributing to all of the cellular layers(Fig. 6A,D). By contrast,X-gal+ cells contributed only to a subset of the mature retinal cell layers in the Notch1 mutants (Fig. 6B,E). In CKO retinae with a very high level of X-gal+ cells,abnormal morphology precluded proper identification of the cell types that were fate mapped (Fig. 6C,F). By contrast, a lower expression of cre recombinase in some Notch1mutants led to a more normal retinal morphology and size(Fig. 6B,E). In these mice,X-gal staining was most prominent in the scleral region of the outer nuclear layer. Staining was observed also in the scleral portion of the inner nuclear layer, and/or possibly in the outer plexiform layer, and in a subset of cells in the ganglion cell layer. The staining in the scleral region of the outer nuclear layer is most consistent with cone photoreceptor cells. These data suggest that Notch1 deficient progenitors predominantly give rise to cone photoreceptor cells at the expense of all other cell types.

To further clarify the adult cell composition of Notch1-deficient retinae, in situ hybridization was performed with markers on P15 retinae. Markers of cone photoreceptor cells (M-opsin, S-opsin, Pde6h) were significantly higher in Notch1-deficient retinae(Fig. 7A,B,F,G and data not shown). Markers of other early born retinal neurons, including Gap43and Nefl (retinal ganglion cells), Pax6 (amacrines, retinal ganglion cells, horizontals) and p57 (amacrines)(Dyer and Cepko, 2001b) were significantly lower (Fig. 7C-E,H-K,P). Markers of late-born cell types also were lower. The rod photoreceptor specific transcription factors, Nrl and Pnr, were downregulated (Fig. 7L,M,Q,R) as were Chx10 (mature bipolar marker) and clusterin (Muller glial cells) (Fig. 7N,O,S,T).

To probe whether late RPCs require Notch1 for proper cell fate determination, Notch1 was clonally inactivated in postnatal day 0(P0) and postnatal day 3 (P3) Notch1 flox/flox mice. Newborn or 3 day old mice were injected with a retrovirus encoding alkaline phosphatase without(LIA) or with cre recombinase (LIA-CRE). Retinae were harvested and processed for alkaline phosphatase staining after two or more weeks of development. At these stages, retinal cell fate decisions are complete and the fates of infected cells can be readily identified by alkaline phosphatase staining(Fig. 8A)(Fields-Berry et al., 1992; Turner and Cepko, 1987).

When P0 retinae were infected with LIA virus, the cells labeled were 80%rod photoreceptor cells, 2% amacrine cells, 13% bipolar interneurons and 5%Muller glial cells. By contrast, infection with LIA-CRE virus resulted in 93%rod photoreceptor cells, 2% amacrine cells, 5% bipolar interneurons and 0%Muller glial cells (Fig. 8B). Similarly, infection of the P3 retina with LIA-CRE led to enhanced rod photoreceptor production from 78% to 93%. Again, this occurred at the expense of other cell types, as bipolar interneurons decreased from 11% to 3% and Muller glial cell production was reduced from 10% to 1%(Fig. 8C). Furthermore, the percentage of clones with more than one cell consisting of only rods increased from 37% (LIA) to 76% (LIA-CRE) in clones initiated at P0 and from 32% (LIA)to 69% (LIA-CRE) in clones initiated at P3. Clones of more than one cell are those in which viral integration occurred in a cell that went on to divide at least once more; these clones tend to be more diverse in terms of cell type,and thus are a more sensitive indicator of an effect of loss of Notch1 on cell fate decisions.

In this study, the cellular and molecular consequences of inactivation of Notch1 in the retina at early and late developmental time points were analyzed. Notch1 was found to be essential for proper proliferation,morphology and cellular diversification through inhibition of the photoreceptor fate.

Chx10 driven expression of cre recombinase most probably eliminates Notch1 between E11 and E12(Rowan and Cepko, 2004),coincident with the first Notch1 expression observed by in situ hybridization in the mouse retina (data not shown). By E13.5, in situ hybridization and microarray analysis showed a reduction in early progenitor markers, such as Fgf15 and cyclin D1, and BrdU incorporation at E14.5 revealed a decrease in proliferating cells. Understanding which genes are directly sensitive to Notch signaling will be crucial to furthering our understanding of how Notch influences progenitor cells. Clonal inactivation of Notch1 revealed that cell-autonomous effects account for at least some of the retinal size defects. The known Notch targets, Hes1, Hes5 or Hey1, may account partly for the reduction in size, as well as the morphological defects. Hes1 null mice have severely reduced retinal size and form rosettes (Tomita et al.,1996). However, we were unable to detect a reduction in Hes1 mRNA in the Notch1-deficient retinae at E13.5 by in situ hybridization and microarray analysis. By contrast, both Hey1and Hes5 were reduced in the microarray analysis, and these genes also may be partially responsible for the reduction in size. Further clarification may require compound inactivation of both Notch1 and Hes family members. Interestingly, previous experiments examining the status of Hes genes in Notch1-deficient mice also have shown that some Hes genes are sensitive to Notch signaling whereas others are not(de la Pompa et al., 1997; Lutolf et al., 2002). For example, En1-CRE mediated excision of Notch1 led to a reduction of Hes5, but not Hes1, in the ventral cerebellum(Lutolf et al., 2002).

Although Notch1 loss-of-function studies in other parts of the central nervous system have also demonstrated an enhanced initiation of the neurogenesis program, these studies did not specifically address a potential role for Notch activity in determining particular neuronal cell fates(de la Pompa et al., 1997; Lutolf et al., 2002). Previously, our laboratory showed that knockdown of Notch1 in the developing chicken retina led to increased production of retinal ganglion cells (Austin et al., 1995). Although other cell types were not assayed, these data were consistent with the hypothesis that Notch1 blocks cell differentiation and release from this signal allows enhanced production of the early-born cell types. Consistent with our previous results, we found that ablation of Notch1 in the early mouse retina also leads to enhanced production of early-born neurons.

Surprisingly, we found that not all early-born cell fates were enhanced in the mature retina at P15. In fact, markers for all retinal cell types were dramatically reduced, except for the markers of cone photoreceptor cells. As precociously produced neurons can be eliminated by apoptosis, we checked for early neuronal markers at E13.5 by gene profiling and in situ hybridization. These approaches showed that both the proneuronal bHLH genes, Math3and Neurod1 as well as early photoreceptor genes, Otx2 and Crx, were upregulated. By contrast, the early ganglion cell markers, Gap43, islet 1 and Nefl, were downregulated. Similar in situ hybridization results were observed when Notch1 was deleted by cre driven by the earlier expressed and uniformly acting Foxg1 promoter. Furthermore, TUNEL staining revealed no differences in cell death in the wild-type and CKO retinae at E12.5 and E13.5. A reduction of β-tubulin III and Nfel-positive cells on dissociated E14.5 CKO retinae further confirmed a loss of retinal ganglion cells and other non-photoreceptor neuronal cells. Together, these data suggest that ganglion cell production is significantly reduced in the absence of Notch1.

The reasons for overproduction of ganglion cells in the chick following Notch1 knockdown, versus cone photoreceptor cell overproduction in the mouse following removal of Notch1, are not clear. However,overproduction of photoreceptor cells was also observed in Xenopuswhen Delta1 was misexpressed at early and late timepoints(Dorsky et al., 1997). Presumably, cells overexpressing Delta1, which were surrounded by wild-type cells, escaped Notch inhibition through negative feedback. Interestingly, the Delta1-expressing cells predominantly became or produced photoreceptor cells, even though other cell fates were available at both timepoints tested. When Delta1 was misexpressed early (16-cell stage),retinal ganglion cell production increased slightly, whereas cone photoreceptor production increased dramatically from ∼7% to ∼61% at the expense of other cell types, including the early-born amacrine and horizontal cells. When Delta1 was introduced at stage 18, rod and cone photoreceptor production increased dramatically from ∼26% to greater than 50% at the expense of both bipolar and Muller glial cells(Dorsky et al., 1997). Similarly, addition of the γ secretase inhibitor DAPT in the early developing chick retina led to a significant increase in visinin positive photoreceptor cells (Kubo et al.,2005). By contrast, introduction of a dominant negative form of Delta1 led to enhanced production of ganglion and amacrine cells at the expense of photoreceptor cells in both Xenopus and chick(Dorsky et al., 1997; Henrique et al., 1997; Kubo et al., 2005). The relative levels of Notch activity in DAPT treated or Delta1- and dnDelta1-misexpressing cells is unclear. However, one interpretation consistent with all of the data from chick, mouse and Xenopus is that the level of Notch signal is crucial for cell fate determination(Fig. 9). Very low levels of Notch would favor photoreceptor production, probably through induction of Otx2(Nishida et al., 2003), while slightly higher levels of Notch would favor production of other neuronal cell types (Hatakeyama and Kageyama,2004). Very high levels of Notch would prohibit exit from cell cycle and inhibit production of any postmitotic cell type(Dorsky et al., 1995) (A.P.J. and C.L.C., unpublished).

By viral transduction of cre recombinase, Notch1 was inactivated in clones initiated at P0 and P3. If removing Notch1 simply forces cells to exit and differentiate, it might be expected that ablation of Notch1 would promote generation of the types of neurons peaking in production at the relevant time point. Birthdating studies in mouse have established that at P0, rod photoreceptor production peaks, whereas at P3,bipolar interneuron production peaks(Young, 1985). Surprisingly, Notch1 deleted progenitor cells almost exclusively produced rod photoreceptors at both P0 and P3. We conclude from these studies that Notch1 is required for progenitor cells and/or newly postmitotic cells to sort out the photoreceptor versus non-photoreceptor cell fate. As predicted by the birthdating curve, ablation of Notch1 early in development promotes the first born photoreceptor cell type, the cone photoreceptor, while ablation of Notch1 later in development promotes the later born photoreceptor cell type, the rod photoreceptor.

Hes1-null mutants show precocious and enhanced development of many early born cell types, including retinal ganglion cells, horizontal cells and amacrine cells, and a decrease in the later born Muller glial cells. Hes5-null mice are also deficient in Muller glial cells, but the production of neuronal cell types is unaffected. As neither Hes1- nor Hes5-null mice exhibit a bias in producing the photoreceptor versus non-photoreceptor cell fate (Hojo et al.,2000; Takatsuka et al.,2004; Tomita et al.,1996), this aspect of Notch activity is most likely Hes independent. Single or compound mutations in various positive bHLH transcription factors also lead to a relatively specific loss of particular non-photoreceptor cell fates. Null mutations in Math5 lead to loss of ganglion cells (Brown et al.,2001; Wang et al.,2001), mice lacking both Mash1 and Math3 fail to produce bipolar interneurons (Hatakeyama et al., 2001), while mice lacking both Neurod1 and Math3 are missing amacrine cells(Inoue et al., 2002). The loss of non-photoreceptor cell fates in the Notch1 CKO mice is most probably not due to the loss of specific bHLH transcription factors, as a decrease in Math5 expression could not be appreciated by gene profiling or in situ hybridization (data not shown). Furthermore, Math3 and Neurod1 are both required for the amacrine and bipolar fates, and expression of both these genes was significantly upregulated in the Notch1 CKO retinae. Another gene significantly increased in E13.5 Notch1 ablated retinae was Otx2, a transcription factor recently shown to be expressed in newly postmitotic neurons and essential for photoreceptor differentiation(Nishida et al., 2003). In that study, misexpression of Otx2 in postnatal day 0 rat increased the production of rod photoreceptor-only clones from 80.7% to 95%(Nishida et al., 2003). Our results suggest that Notch signaling directly or indirectly regulates the level of Otx2, which is crucial for ensuring the proper ratio of photoreceptor versus non-photoreceptor fates produced. High levels of Otx2 might be dominant to the activities of bHLH and other genes that can induce non-photoreceptor neuronal fates.

In the developing Drosophila retina, Notch signaling is recruited for multiple activities including: (1) setting up a dorsal/ventral boundary,(2) establishing planar polarity, (3) upregulating early atonal levels to promote neurogenesis and (4) antagonizing late atonal activity to inhibit neurogenesis (Frankfort and Mardon,2002). This study and previous reports suggest similarly diverse roles for Notch activity in the vertebrate retina. First the early developing vertebrate retina expresses Notch1 initially only in the central region of the retina (Lindsell et al.,1996) (data not shown), suggesting that Notch-Delta signaling is necessary only when neurogenesis begins. As ablation of Notch1dramatically influences organ size and introducing activated Notch can lead to hyperproliferation (Bao and Cepko,1997), Notch1 most probably engages the cell cycle machinery. Interestingly, when Notch1 was deleted in E14.5 retinal progenitor cells, smaller clones were observed, but a total loss of more than three cell clones was not observed. The modest effect of loss of Notch on clone size is in keeping with data from Drosophila. Loss of Notch from the entire eye disc leads to a dramatic reduction in size, whereas loss of Notch in clones does not dramatically reduce clone size. A non-autonomous ligand,unpaired, that is activated by Notch is thought to mediate these effects(Chao et al., 2004). In vertebrates, it is not known if there is such a non-autonomous ligand. Nonetheless, in mice, Notch appears to be one regulator of cell proliferation,but it is not essential for cell cycle progression. It is, however, sufficient for at least some period of time, as sustaining Notch activity in late retinal progenitor cells leads to formation of very large clones(Furukawa et al., 2000). Notch activation also leads to a block in neuronal fates and an increase in cells with some glial characteristics. The data presented here also clearly indicate a cell fate role for Notch1 in the vertebrate retina. The specific promotion of the photoreceptor fate at the expense of both glial and other neuronal cell fates following deletion of Notch1 reveals that it normally represses the photoreceptor fate. If photoreceptors were the initial cell type when the retina first evolved, one speculative interpretation of these data is that Notch was recruited to allow preservation of a pool of RPCs for later cellular diversification

The authors thank the Aguet (ISREC) laboratory for Notch1 conditional mice;S. Rowan for the gift of Chx10-CRE mice; S. Rowan, M. Samson, T. Young, A. Shaw, A. Singla, J. Trimarchi and D. Kim for support and technical assistance;S. White (Beth Israel Medical Center Histology Core Facility) for Hematoxylin and Eosin staining; J. Trimarchi and G. Fishell for critical reading of the manuscript and members of the Cepko, Tabin and Dymecki laboratory for helpful discussions and advice. This work was supported by funds from HHMI and NIH EYO 86767. A.P.J. was supported by T32EY007110 and C.L.C. is an Investigator of the HHMI.