XOtx5b and XOtx2 regulate photoreceptor and bipolar fates in theXenopus retina

Rod photoreceptors and bipolars are among the last cells to be born and differentiate from the late-progenitor pool(Carter-Dawson and LaVail,1979; Chang and Harris,1998; Holt et al.,1988; Stiemke and Hollyfield,1995; Turner and Cepko,1987; Turner et al.,1990; Wetts and Fraser,1988; Young,1985). As retinal development proceeds fate choices become restricted, such that late progenitors usually differentiate into rods,bipolar cells, Müller glia and certain types of amacrine cells(Belliveau and Cepko, 1999;Belliveau et al., 2000;Fields-Berry et al., 1992;Turner and Cepko, 1987;Turner et al., 1990). Retinal progenitors express many genes that affect cell fate including Notch,Hes1, Pax6, Rax, Optx2, Chx10, p27Xic1 and NeuroD (reviewed byLivesey and Cepko, 2001)Livesey and Cepko, 2001;Perron et al., 1998;Zuber et al., 1999). After differentiating, photoreceptors and bipolars both stop expressing many of these genes yet, in Xenopus laevis, both cell types continue to express XRx1 and NeuroD(Perron et al., 1998). In mice, blue cones and bipolars both express blue opsin transgenes, indicating that these cell types share common gene regulatory mechanisms(Chen et al., 1994;Chiu and Nathans, 1994).

The link between photoreceptors and bipolars is further demonstrated by fate switching experiments. A late precursor may differentiate either as a photoreceptor or a bipolar depending on environmental factors. For example,exposing rat retinal cell cultures to the cytokine, ciliary neurotrophic factor (CNTF), causes a dramatic decrease in the number of opsin-positive cells and a compensatory increase in the number of cells expressing bipolar cell-specific markers, suggesting a change in cell fate choice from rods to bipolars (Ezzeddine et al.,1997). Similarly, when rat retinal cells are cultured at low density there is a decrease in the number of opsin-positive cells and an increase in the number of cells expressing the bipolar cell marker, 115A10(Altshuler and Cepko, 1992). Co-culturing late with early progenitors leads to a similar shift in cell fate, which can be blocked by the CNTF antagonist, hLIF05(Belliveau et al., 2000). The sensitivity of late retinal precursors to extrinsic factors such as CNTF demonstrates that external cues influence the internal switches that differentially specify the fate of these two cell types.

We have found that Xenopus photoreceptors and bipolars also express, XOtx5b, which is a member of the otd/Otx family of paired-like homeodomain transcription factors(Vignali et al., 2000).otd (also known as oc; ocelliless) is the onlyotd/Otx family member identified in Drosophila and it contributes to head formation (Finkelstein et al., 1990). The vertebrate homologues of otd includeOtx1, Otx2, Otx3, Otx4, Otx5, Otx5b (96% similar to Otx5) andCrx, which are all involved in anterior embryo and sensory organ formation (Acampora et al.,1998a; Acampora et al.,1998b; Acampora et al.,1996; Acampora and Simeone,1999; Andreazzoli et al.,1997; Bovolenta et al.,1997; Furukawa et al.,1997; Gammill and Sive,1997; Kablar et al.,1996; Kuroda et al.,2000; Martinez-Morales et al.,2001; Pannese et al.,1995; Sauka-Spengler et al.,2001; Suda et al.,1999; Vignali et al.,2000). The first vertebrate homologues of otd, Otx1 andOtx2, were characterised in mouse(Simeone et al., 1992).Otx2^-/- mice lack fore- and midbrain structures, while a null mutation in Otx1 is less severe, causing, among other defects, a reduction in brain size and lack of ciliary process next to an otherwise normal retina (Acampora et al.,1998a; Acampora et al.,1996; Acampora and Simeone,1999; Martinez-Morales et al.,2001). Overexpression of Xenopus Otx2 (XOtx2)suggests that it plays a similar role in frog(Andreazzoli et al., 1997;Blitz and Cho, 1995;Kablar et al., 1996;Pannese et al., 1995).XOtx5b is also expressed during early embryogenesis. Overexpression of XOtx5b, like XOtx2, produces ectopic cement gland and neural tissue as well as gross embryonic abnormalities in which posterior structures are absent or reduced and the neural tube fails to close(Kablar et al., 1996;Pannese et al., 1995;Vignali et al., 2000). Because of their roles in early embryonic development, the functions of Xenopus XOtx5b and XOtx2 in retinal differentiation have not been previously determined.

Crx was isolated in a yeast one-hybrid screen using the rhodopsin promoter as bait (Chen et al.,1997; Furukawa et al.,1997). Crx can bind the rhodopsin promoter and transactivate its expression, along with a number of other photoreceptor-specific genes(Chen et al., 1997;Livesey et al., 2000). It is expressed in pinealocytes as well as rod and cone photoreceptor cells. The orthological relationships of the mammalian Crx genes with other gnathostome otd-related genes remain controversial. Phylogenetic analyses of the otd/Otx family members, however, support a relationship between the Otx5/5b genes characterised in amphibians or chondrichthyans, and the Crx genes isolated in mammals and in zebrafish (Germot et al.,2001; Sauka-Spengler et al.,2001). These results have led to the hypothesis thatOtx5/5b and Crx genes might belong to a single orthology class. The identification in the zebrafish of a second, unambiguouslyOtx5-related, gene has challenged this view, suggesting that two distinct orthology classes may be present in gnathostomes(Gamse et al., 2001). However,novel phylogenetic analyses including a much wider range of amnioteOtx5 or Crx-related sequences as well as pufferfish sequences, indicate that the zebrafish Otx5 and Crx genes have arisen through a gene duplication, which occurred in the actinopterygian lineage, and support the hypothesis of a single gnathostome Otx5/Crxclass, with an acceleration of evolutionary rate of what became Crxearly in mammalian evolution (S. Mazan, personal communication). This raises the question of whether XOtx5b in lower vertebrates performs the same function as Crx in mammals.

Crx, Otx1, Otx2 and Otx5b are all expressed in the developing retina prior to retinal differentiation. Some otd/Otxfamily members are expressed in both photoreceptors and bipolar cells while others are expressed in one cell type or the other. Zebrafish Crx(Liu et al., 2001), mouseOtx2 (Baas et al.,2000; Bovolenta et al.,1997) and Xenopus Otx5b (see below), for example, are expressed by both bipolars and photoreceptors, while mouse Crx is expressed only in photoreceptors cells(Chen et al., 1997;Furukawa et al., 1997) andXenopus Otx2 is found in bipolar cells but not photoreceptors(Kablar et al., 1996;Perron et al., 1998). We were therefore interested to know what roles XOtx5b and XOtx2play in retinal development, and specifically whether they contribute to the switch between these two cell fates. In this study, we examined the expression of XOtx5b and XOtx2 during development and determined the effect of each on retinal cell differentiation. We found that bothXOtx5b and XOtx2 are involved in retinal cell fate specification. Next, we identified the domains of these proteins that were important in determining photoreceptor verses bipolar cell fates. Finally, we used the Xenopus opsin promoter to examine the relationship betweenXOtx5b and XOtx2 in regulating photoreceptor cell specificity.

All embryos were the products of in vitro fertilizations of wild-type adultXenopus laevis (from Nasco) maintained in the Department of Anatomy at Cambridge University.

DNA isolated by Qiagen maxi preps was diluted in nuclease-free water to a concentration of 1.5 μg/μl. These stocks were spun down for at least 10 minutes at 4°C prior to use. 1 μl of each construct was mixed with 1μl of pCS2⁺ green fluorescent protein (GFP) DNA, to label transfected cells. GFP with pCS2⁺ vector alone was the control. 9μl of DoTAP (Roche) was added to 3 μg of DNA and injected into stage 17-18 embryos. At stage 41, embryos were fixed for 1 hour at room temperature,sunk in 2% sucrose overnight at 4°C and cryostat sectioned (10 μm). The samples were then rehydrated with two washes of 1× PBS for 5 minutes,mounted in FluorSave (CalBioChem) containing 2% DABCO (Sigma) and dried overnight at room temperature.

Cryostat sections (10 μm) were washed three times for 5 minutes in 1× PBS + 0.1% Triton X-100 (PBST), blocked 30 minutes with PBST + 5%heat-inactivated goat serum then incubated overnight at room temperature with primary antibody: XAP-1 (DSHB) at 1:1, bovine anti-rhodopsin at 1:500 (R2-12N;from Paul Hargrave) and anti-calbindin (Oncogene) 1:500. Sections were washed in PBST three times for 5 minutes then three times for 20 minutes and incubated with a 1:500 dilution of the appropriate Cy3-conjugated secondary antibody for 2 hours. The sections were washed again in PBST and mounted.

The construction of pCS2.XOtx5b has been described previously(Vignali et al., 2000). pCS2.XOtx2 was generated by PCR cloning of the XOtx2 coding region into the EcoRI site of pCS2 with 5′XOtx2 and 3′XOtx2 (Table 1).XOtx5bN+HD-EngR was obtained by cloning theEcoRI/TaqI fragment of XOtx5b (amino acids (aa)1-107) into the EcoRI site of pCS2EngN. XOtx2N+HD-VP16 was obtained by PCR cloning the XOtx2 region spanning aa residues 1-109 into theClaI site of pCS2VP16-N with 5′ClaX2 and 3′ClaX2.XOtx5N+HD-VP16 was obtained by PCR cloning the XOtx5b region spanning aa residues 1-107 into the ClaI site of pCS2VP16-N using 5′Cla and 3′Cla.

Swapped domain constructs (XOtx2/5b andXOtx5b/2+3′UTR) were generated as follows. TheBglII/SpeI fragment of T7TSXotx2(Pannese et al., 1995)spanning the coding region plus 3′ untranslated region (3′UTR) was cloned into the BamHI/XbaI site of pCS2⁺. By site-directed mutagenesis, an Asp718 restriction site was generated into this clone and XOtx5b at the equivalent aa residue 61 codon,causing the CGT sequence to change to CGG. This site allowed precise swapping of the coding region C-terminal to aa 61. Because the homeodomains (HD) are identical in the region between aa 61 and the polyQ stretch at the HD end,swapping is only effective beyond aa 99 of both sequences (aa 99 to end). The 3′UTR of XOtx5b/2+3′UTR was removed by PCR cloning the coding region of XOtx5b/2+3′UTR into the ClaI/EcoRI site of pCS2⁺ using 5′Cla and 3′XOtx2, to generateXOtx5b/2.

Whole-mount in situ hybridisations were performed on bleached embryos(Broadbent and Read, 1999). Dioxigenin-labelled antisense RNA probes were generated from the full-lengthXOtx2 coding sequence and the 3′ untranslated region ofXOtx5b [clone 13F8 (Gawantka et al., 1998)] as described previously(Shimamura et al., 1994). In situ hybridisation on sections was done using the same protocol with the following modifications: rehydrated sections were fixed to slides using 100%methanol for 10 minutes, then rinsed in 1× PBS for 2 minutes and washed 3 times for 5 minutes in PBST. Sections were treated with 20 μg/ml proteinase K for 30 seconds with subsequent wash times reduced by half.

When in situ hybridisation was followed by immunohistochemistry, we performed the in situ hybridisation protocol described previously(Myat et al., 1996). After the coloration with NCT and BCIP, we washed with PBST for 5 minutes and then followed the immunohistochemistry protocol as described above.

To determine if XOtx5b or XOtx2 transcripts were expressed in dividing cells, stage 41 embryos were injected with BrdU(5-bromo-2′-deoxyuridine; Labelling and Detection Kit I, Roche) in the gut, fixed 30 minutes later and cryostat sectioned. In situ hybridisation was performed on sections as follows: DIG-labelled probes (2 ng/ml in hybridisation buffer) (Shimamura et al.,1994), were heated to 70°C for 10 minutes then incubated on sections overnight at 60°C. The rest of the protocol was as previously described (Myat et al., 1996). Following NBT/BCIP staining, sections were stained for BrdU using the protocol below.

To determine if the XOtx5b or XOtx2 constructs effect proliferation, lipofected embryos were injected with BrdU in the gut at stage 31. Embryos were fixed and cryostat sectioned (10 μm) at stage 41. The sections were stained with both BrdU and GFP antibodies. To do this, the cryostat sections were washed with 2 N HCl for 45 minutes then neutralised with several PBST washes. The antibodies for anti-mouse BrdU and anti-rabbit GFP (Molecular Probes, Eugene, OR) were added at dilutions of 1:10 and 1:500,respectively, and incubated at 37°C for 30 minutes. After washing with three changes of PBST, secondary antibodies [Alexa 488-goat anti-rabbit(Molecular Probes) and Cy3-goat anti-mouse (Chemicon)] were added together,both at a dilution of 1:500, and incubated for 30 minutes at 37°C. The sample were again washed three times with PBST and then stained with 15μg/μl Hoechst solution for 3 minutes at room temperature to visualise nuclei. After a final three washes in PBST the sections were mounted in FluorSave (Calbiochem) containing 2% DABCO (Sigma Genosys).

Embryos were lipofected with GFP and pCS2, XOtx2 orXOtx5b-VP16, grown to stage 31, 33/34 or 37, fixed and cryostat sectioned. We used Intergen's ApopTag Kit to stain apoptotic cell nuclei and counterstained with Hoechst (15 mg/ml). Retinas with at least one lipofected,apoptotic cell were counted. Five retinas were counted for each construct at each stage. The percentage of apoptotic lipofected cells versus all lipofected cells was determined.

XOtx5b and XOtx2 are expressed during and after retinal differentiation (Pannese et al.,1995; Sauka-Spengler et al.,2001; Vignali et al.,2000). In situ hybridisation on whole-mount embryos is shown inFig. 1. In stage 20 embryos,XOtx2 expression is detected in the two eye primordia and forebrain,while XOtx5b expression is only observed in the epiphysis and cement gland (Fig. 1A,D). By stage 25,XOtx2 expression remains largely unchanged, while XOtx5b is no longer expressed strongly in the cement gland but remains in the presumptive pineal (Fig. 1B,E). By stage 31, when cellular differentiation is underway in the eye, the expression patterns of XOtx2 and XOtx5b overlap in the retina (Fig. 1C,F).

In order to compare their relative expression patterns at a more cellular level in the retina, we performed in situ hybridisation on retinal sections. At stage 25, XOtx2 is found throughout the presumptive retinal pigment epithelium (RPE) and retina, while only a few cells in the central retina express XOtx5b (Fig. 2A,F). A few hours later, at stage 28, XOtx2 expression has narrowed to the central retina and RPE, while XOtx5b expression has expanded (Fig. 2B,G). By stage 33, XOtx2 and XOtx5b expression patterns are indistinguishable; transcripts are found throughout the developing retina except in the most peripheral regions, corresponding to the ciliary marginal zone (CMZ; Fig. 2C,H). At stage 37, XOtx2 expression is detected in all layers of the peripheral retina, but is becoming restricted in the central retina to the outer edge of the inner nuclear layer or INL (Fig. 2D). Similarly, XOtx5b expression, in the central retina is narrowed to just the outer (ONL) and inner nuclear layers(Fig. 2I). In the mature, stage 41 retina, XOtx2 is only found in the outer edge of the INL and inner edge of the CMZ (Fig. 2E).XOtx5b is expressed in the inner edge of the CMZ, the outer edge of the INL and the ONL (Fig. 2J).

It is difficult to identify by in situ hybridisation alone which INL retinal cell types express XOtx2 and/or XOtx5b. In mouse,bipolar-specific markers helped show that Otx2 is expressed in bipolar cells (Baas et al.,2000). Unfortunately, no bipolar cell-specific antibody is available for Xenopus and our attempts at staining Xenopusretinas with mouse protein kinase C(Greferath et al., 1990) and Chx10 (Burmeister et al.,1996) antibodies were unsuccessful. The most salient distinguishing feature of a bipolar cell, however, is its characteristic bipolar morphology with terminal arbours in the two plexiform layers. Therefore, we lipofected cDNA for GFP into Xenopus retinas, followed by immunohistochemistry to GFP, to reveal the morphology of a random selection of cells. We then performed in situ hybridisations on these lipofected retinas using XOtx2 or XOtx5b probes to identify the cell types expressing these otd/Otx family members(Fig. 3). We found that messenger RNA from both of these genes is expressed in bipolar cells(Fig. 3A-F). Of the GFP-labelled bipolar cells 55%±8% were co-labelled withXOtx5b, while 82%±5% were XOtx2 positive(Table 2). Double in situ hybridisation with XOtx5b and XOtx2 revealed that all the INL cells that express XOtx5b also express XOtx2 (3G-I),indicating that 55%±8% of bipolar cells co-express XOtx5b andXOtx2 (Table 2).

We also wanted to know which photoreceptors express XOtx5b. Using cone- or rod-specific antibodies on XOtx5b-stained in situ hybridisation sections, we found XOtx5b was expressed in all cone and rod photoreceptors (Fig. 3J-O). Thus, XOtx5b is preferentially expressed in photoreceptors and a subset of bipolar cells, while XOtx2 is preferentially expressed in bipolar cells.

The expression of XOtx5b in all photoreceptors and the combined expression of XOtx5b and XOtx2 in a subset of bipolar cells led us to hypothesise that these genes may play a role in retinal cell fate. To test this, we lipofected stage 18 retinoblasts in vivo with eitherXOtx5b or XOtx2 DNA expression constructs(Fig. 4). GFP DNA was co-injected with the experimental DNA in order to identify the lipofected cells. GFP DNA co-injected with vector only DNA was used as a control.XOtx5b significantly increased (P<0.0001) the proportion of photoreceptors in lipofected cells (Fig. 4A compared with Fig. 4B; quantified in Fig. 4D) when compared to the control. A slight decrease in the proportion of ganglion (P=0.0002), amacrine (P=0.047),Müller (P=0.0016) and horizontal (P=0.014) cells was also observed in the XOtx5b lipofected cell population, while bipolar cell numbers were unaffected. In contrast, XOtx2 overexpression increased the proportion of bipolar cells (P<0.0001) at the expense of photoreceptor (P=0.003) and Müller cells(P=0.0004; Fig. 4A compared with 4C; quantified in Fig. 4E). Staining of XOtx5b lipofected retinal sections with the XAP-1 (anti-photoreceptor) antibody confirmed that the cells counted were indeed photoreceptors (data not shown).

Since XOtx5b is normally expressed in both rod and cone photoreceptors (Fig. 3J-O), we wondered if overexpression of XOtx5b increased the proportion of rods, cones, or both. To address this question, we used the cone photoreceptor-specific anti-calbindin antibody to identify cones inXOtx5b lipofected retinas. To increase the number of photoreceptors in this analysis, we used XOtx5bN+HD-VP16 for our in vivo lipofections (see below for further details of this construct).XOtx5bN+HD-VP16 dramatically increased the proportion of photoreceptor cells from 27% to 71% (Fig. 5C). When we stained these same sections for the cone marker,calbindin, we found approximately equal numbers of calbindin-positive and-negative photoreceptors (Fig. 5D). In control, GFP and vector only lipofected retinas,we also found approximately equal numbers of rods and cones(Fig. 5D), which is consistent with previous reports (Chang and Harris,1998). Together these results demonstrate thatXOtx5bN+HD-VP16 increases the proportion of both rods and cones in equal numbers.

Transcription factors can act as activators, repressors, or both. Mouse Crx has been shown to activate rhodopsin but repress NeuroD,either directly or through a cascade of other factors(Furukawa et al., 1999;Livesey et al., 2000). We were interested to know if XOtx5b and XOtx2 worked primarily as repressors or activators when modulating retinal cell fates. Therefore, we fused the powerful repressor domain of engrailed to the N terminus and homeodomain (HD) of XOtx5b (XOtx5bN+HD-EngR) and XOtx2(XOtx2N+HD-EngR) and determined the effect of these fusions on cell fate(Fig. 5E-G). Lipofection of theXOtx5bN+HD-EngR led to fewer lipofected photoreceptor cells than controls (P=0.001; Fig. 5G). This suggests activation of XOtx5b target genes stimulates transcription necessary for photoreceptor cell specification.XOtx2N+HD-EngR lipofected retinas showed a decrease in the percentage of lipofected bipolar cells compared to controls (P=0.004;Fig. 5G), suggesting that native XOtx2 also acts as an activator of genes necessary for bipolar cell specification. These results also suggest some control of cell fate specification by the N-terminal and HD sequences.

Engrailed fusion constructs of XOtx5b and XOtx2 also gave some unexpected results. XOtx5bN+HD-EngR did not affect ganglion,amacrine or Müller cell populations, while the proportion of bipolar cells was slightly increased (5%, P<0.05). XOtx2N+HD-EngRincreased the proportion of ganglion cells (by 4%, P<0.02) but did not alter the photoreceptor cell population. These results demonstrate that although full-length XOtx2 and XOtx5b may modulate photoreceptor levels, they both may act in more complex ways than as dedicated repressors or activators.

These results also suggest that the N-terminal and HD regions of these proteins are not sufficient to direct cell fate specification in the same way as the full-length proteins. This fits with a recent study that compares otd/Otx family members (Sauka-Spengler et al., 2001) and the fact that Xenopus XOtx5b and XOtx2 have very similar homeodomains and N-terminal regions. Indeed,Xenopus XOtx5b and XOtx2 homeodomains (HD) have only one residue difference (XOtx2=A; XOtx5b=S) while the region N-terminal to the HD, which we called the N terminus, is 73% identical in these two proteins. If the N terminus and homeodomain were mostly responsible for the differences we observed in the lipofection results between these two full-length constructs, and if, as the above results suggest, that some aspects of the phenotype were due to their activity as activators rather than repressors, then a construct which fuses the N terminus and HD to the strong activator VP16 should mimic a full-length construct. To test this, we fused the N terminus and HD to the strong activator domain of VP16 for each of these genes (XOtx2N+HD-VP16 and XOtx5bN+HD-VP16) and lipofected each with the tracer GFP into stage 18 embryos(Fig. 5H,I). In each case, the number of GFP-positive photoreceptor cells increased by about three times that of the control lipofected retinas (both P<0.0001) at the expense of all the other retinal cell types(Fig. 5J). These results suggest that the N terminus and homeodomain may be involved in regulating photoreceptor cell specification but show a striking lack of specificity between photoreceptors and bipolars. The increased percentage of photoreceptor cells in retinas lipofected with the VP16 activator fusion constructs(XOtx5bN+HD-VP16=80% and GFP=27%) was even more than retinas lipofected with full-length XOtx5b (XOtx5b=37% and GFP=28%;Fig. 4D compared withFig. 5J). We therefore had to consider the possibility that the VP16 activator alone may be responsible for photoreceptor cell fate. To test this, we injected a VP16 activator construct fused to a nuclear localization sequence and GFP into developing embryos and found no effect on cell fate (data not shown). It could also be that the N terminus and homeodomain can activate photoreceptor-specific genes on their own and adding the VP16activator domain made the constructs even stronger transactivators. Hence, we lipofected retinas with just the N terminus and homeodomain of either XOtx5b or XOtx2, but in neither case was there an effect on cell fate (data not shown). These results suggest that the N-terminal and HD of both XOtx5b and XOtx2 may have a common function in retinal cell determination, and be part of the shared regulatory mechanisms of photoreceptors and bipolars(Chen et al., 1994;Chiu and Nathans, 1994). The lack cell type specificity of these domains when fused to an activator suggested that the region C-terminal to the HD may be more important in modulating cell fate.

C-terminal to the homeodomain, otd/Otx family member genes have several unique domains: a glutamine (Gln) rich region, a WSP domain and an OTX tail sequence (review by Morrow et al.,1998). Comparison of the entire XOtx2 and XOtx5b C termini sequence shows that they are 60.1% identical (data not shown). To test if the difference we observed in the way XOtx2 and XOtx5boverexpression affects retinal cell fate maps to the C terminus, we did domain swap experiments. The N terminus and HD of XOtx2 was fused to the C terminus of XOtx5b (XOtx2/5b) and the N-terminal and HD sequence of XOtx5b was fused to the C terminus of XOtx2(XOtx5b/2; Fig. 6A). Each domain swapped construct was then co-lipofected with GFP into developing embryos as above (Fig. 6C,D) and the ratios of each cell type calculated(Fig. 6E,F). XOtx5b/2,like XOtx2 itself, increased bipolar cells (P=0.0001) at the expense of photoreceptor (P=0.031) and Müller(P=0.0001) cells while XOtx2/5b, like XOtx5bitself, increased photoreceptor cells (P=0.0004) at the expense of ganglion (P=0.004), amacrine (P=0.018), Müller(P=0.005) and horizontal (P=0.021) cells. These results clearly suggest that the C-terminal region of XOtx2 andXOtx5b is much more critical for cell fate specification affects than the N terminus and HD.

Increases in photoreceptor or bipolar cell percentages in the population ofXOtx5b or XOtx2 lipofected cells, respectively, could be due cell type-specific changes in proliferation or death. Previous studies have shown that XOtx2 is expressed in proliferating retinoblasts(Perron et al., 1998) butXOtx5b expression in dividing cells has not been described. To test if XOtx5b is normally expressed in mitotic cells, we injected embryos with BrdU and fixed them 30 minutes later. We sectioned their retinas and stained for XOtx5b or XOtx2 expression by in situ hybridisation, and afterwards, for BrdU by immunostaining to look forXOtx5b- or XOtx2-positive cells that were also BrdU positive(Fig. 7A-F). We found that most of the cells that expressed XOtx5b or XOtx2 were not BrdU positive. However, a few cells in the CMZ that were XOtx5b orXOtx2 positive, were also labelled with BrdU. This confirmed thatXOtx2 is in some proliferating cells and indicated thatXOtx5b can also be found in proliferating cells of the retina, as suggested by their expression during early stages of retinal development.

As these transcripts are present in BrdU-positive cells, it is possible that they cause an increase in proliferation of the respective cell types they influence. To test this, we lipofected developing retinas with GFP and each one of the following constructs in separate experiments: XOtx5b,XOtx5bN+HD-VP16, XOtx5bN+HD-EngR, XOtx2, XOtx2N+HD-VP16 and the swapped domain constructs (XOtx2/5b and XOtx5b/2) at stage 18. At stage 31 (the beginning of photoreceptor cell differentiation), we injected the embryos with BrdU. At stage 41 (retinal maturation), the embryos were fixed and the retinas sectioned and stained for GFP and BrdU. Cells expressing just GFP or GFP and BrdU were counted. We found that none of the constructs we tested showed an increase in proliferation in any of the cell types at this stage, suggesting that these constructs do not influence proliferation(Fig. 7G,H).

To produce an increase in the percentage of lipofected bipolar cells,XOtx2 could cause cell death in other retinal cell types, like photoreceptors or Müller cells. Similarly, XOtx5b lipofected retinal cells that are not photoreceptors might be more likely to undergo programmed cell death. To test if either of these constructs caused apoptosis,we lipofected developing retinas as above, fixed at stages 31, 33/34 and 37 and stained the sections for apoptotic nuclei using a TUNEL assay. The number of GFP-positive cells in each of the retinas was counted as well as those apoptotic cells that were GFP positive. In each case, the percentage of GFP-positive cells that were also apoptotic was around 2-4%. Statistical analysis using a Student's t-test showed no differences (data not shown). This suggests that selective cell death is also not the basis for the overexpression phenotypes. The most likely explanation for the phenotypes observed, therefore, is that these constructs cause a fate switch or conversion.

Crx binds to the OTX-binding consensus sequence found in the rhodopsin promoter and transactivates its expression(Chen et al., 1997;Furukawa et al., 1997). Like Crx, Otx2 also binds this sequence and has been shown to both activate and repress genes by binding to this site(Kelley et al., 2000;Morgan et al., 1999). SinceXOtx5b has been suggested to be part of the same subfamily of otd/Otx transcription factors as Crx (Germot et al., 2001; Sauka-Spengler et al., 2001), we reasoned that XOtx5b might also regulateopsin expression. To test this hypothesis, we injected one dorsal cell of a 4-cell stage blastomere with an expression vector of GFP driven by the Xenopus opsin promoter [Xop-GFP(Mani et al., 2001)], cappedXOtx5b RNA and β-gal RNA. We allowed the embryos to grow to stage 17-18, they were then fixed and the fluorescence quantified. As a negative control, Xop-GFP DNA and β-gal RNA were injected alone(Fig. 7I). All embryos that were injected with Xop-GFP DNA and XOtx5b RNA on the dorsal side(Fig. 7J), expressed GFP at levels higher than the negative control embryos(Fig. 7I,P<0.0001), suggesting that XOtx5b does activate the opsinpromoter.

Since XOtx5b is expressed in a subset of bipolar cells withXOtx2, we were interested to know whether XOtx2 affected the ability of XOtx5b to regulate opsin expression. Co-injection of equal amounts of XOtx2 and XOtx5b RNA plus Xop-GFP DNA andβ-gal RNA (Fig. 7K) showed that XOtx2 buffers the effect of XOtx5b on Xop-GFP; i.e. levels of GFP expression are similar to those of the uninjected side(Fig. 7N). Staining forβ-gal confirmed that these embryos were successfully injected with the RNA/DNA mixture. By adding double the concentration of XOtx2 thanXOtx5b RNA plus Xop-GFP DNA and β-gal RNA(Fig. 7L), we still observed a suppression of the XOtx5b-driven opsin activation. However,when we added the Xop-GFP DNA and β-gal and XOtx2 RNA alone(Fig. 7M), we found thatopsin expression was enhanced by XOtx2. This was not surprising since either Crx and Otx2 alone activates the promoter of another photoreceptor-specific gene, interphotoreceptor retinoid binding protein (IRBP) (Bobola et al.,1999). Our results with the Xenopus opsin promoter suggest that the relative levels of XOtx2 and XOtx5b are important in determining the activation or repression of opsin. It appears that XOtx2 plays a role in suppressing the effect ofXOtx5b on opsin expression when equal or double the amounts of XOtx2 to XOtx5b are present in the cell.

The XOtx2 suppression of the activation of a photoreceptor-specific gene in the above assay may partly explain why bipolar cells do not normally express opsins. A similar question raised by this study is why cells that express both XOtx5b and XOtx2 appear to differentiate into bipolars rather than photoreceptors. Could it be thatXOtx2 suppresses the ability of XOtx5b to induce photoreceptors? To test this hypothesis, we colipofected XOtx2 andXOtx5b with GFP into the eye field of stage 18 embryos as described above. Controls were lipofected with XOtx5b and GFP or GFP and vector alone. XOtx5b-lipofected retinas had the expected increase in photoreceptor cells (P=0.001). However, when XOtx2 andXOtx5b were colipofected, there was no increase in photoreceptors but rather an increase in the population of lipofected bipolar cells(Fig. 7O-R; P=0.007). These results suggest that when XOtx2 and XOtx5b are both present in progenitors, cells are more likely to take on bipolar cell fates.

Crx, is expressed in the pineal, both rods and cones in the retina(Chen et al., 1997;Furukawa et al., 1997;Liu et al., 2001) and has recently been localised to bipolar cells(Chen et al., 2000). It has also been demonstrated to be an activator of many photoreceptor-specific genes(Chen et al., 1997;Furukawa et al., 1997) and involved in photoreceptor cell specification (for a review, seeMorrow et al., 1998). In this paper, we demonstrate in Xenopus that XOtx5b, which is highly related to mammalian Crx(Germot et al., 2001;Sauka-Spengler et al., 2001)has an identical expression pattern to Crx in later stages of embryogenesis. We confirmed its expression in the pineal(Sauka-Spengler et al., 2001;Vignali et al., 2000) and identified its expression in both rod and cone photoreceptor and bipolar cells. Misexpression of XOtx5b in vivo demonstrated that it is also involved in photoreceptor cell specification. Misexpression of XOtx2or XOtx5b fused to the VP16 activator in developing retinoblasts suggests that (1) XOtx2 and XOtx5b target genes are involved in photoreceptor cell determination and (2) these target genes require activation in order to promote photoreceptor cell fate. Overexpression ofCrx by retroviral infection of retinal progenitors is reported to increase the percentage of clones that contain only rods, suggesting thatCrx influences cell fate(Furukawa et al., 1997),however Crx knockouts do make rods and cones(Furukawa et al., 1999)suggesting that Crx by itself is not a key determiner of photoreceptor cell fate. In the Xenopus retina, there are an equal number of rods and cones (Chang and Harris, 1998)unlike rodent retinas, which consist mostly of rods(Young, 1985). We have also found that the population of photoreceptors increased by overexpresssion of aXOtx5b activator construct (XOtx5bN+HD-VP16), is composed of both rods and cones equally. This is not surprising since it is expressed in both types of cells in Xenopus and zebrafish(Liu et al., 2001).

Contrary to mammalian and zebrafish Crx studies(Chen et al., 1997;Furukawa et al., 1997;Liu et al., 2001), Xenopus XOtx5b is expressed at much earlier stages of development and plays a role similar to XOtx2 in early embryogenesis(Vignali et al., 2000).XOtx5b is expressed in dividing retinal cells and in all layers of the undifferentiated retina during development, which is similar to zebrafishCrx, but unlike mammalian Crx, which is expressed in photoreceptor cells just after they are born(Chen et al., 1997;Furukawa et al., 1997;Liu et al., 2001;Morrow et al., 1998). We also found that XOtx5b-engrailed constructs resulted in clones with a decrease in photoreceptor cells. This is also different from theCrx-engrailed repressor fusion construct results, which produced clones with rods lacking outer segments(Furukawa et al., 1997). The ability of XOtx5b to decrease the number of photoreceptors, albeit a small amount in comparison to the activation seen by the VP16 activator fusion constructs, suggests that XOtx5b may be playing an earlier role in photoreceptor specification than Crx. Phylogenetic analysis suggests that XOtx5b and Crx may be orthologous(Germot et al., 2001;Sauka-Spengler et al., 2001)suggesting that one might consider renaming Xenopus Otx5b, XCrx. Our findings that XOtx5b functions similarly to mammalian Crx,supports this idea. However, since there are clear differences between the roles played by mammalian Crx and Xenopus XOtx5b before eye formation, and since it is still possible that a Xenopus Crx gene with more sequence similarity exists as was found in teleosts, we would suggest that Xenopus XOtx5b is not renamed.

To date, zebrafish is the only species reported to have separate genes encoding Crx and Otx5 (Gamse et al., 2002). Xenopus also has two genes in the hypothesised Otx5/Crx orthology class, XOtx5(Kuroda et al., 2000) andXOtx5b (Vignali et al.,2000). Although zebrafish Crx and Otx5 retinal and pineal expression are the same, their protein sequences are only 64%identical (data not shown) and they regulate pineal circadian gene expression differently (Gamse et al., 2002). This suggests that they are different genes. Although XOtx5 retinal expression has not yet been described, the early embryonic XOtx5 and XOtx5b expression patterns are virtually identical (Kuroda et al.,2000; Vignali et al.,2000) and comparison of their protein sequences show they are 96%identical (data not shown). Injections of the blastomere with eitherXOtx5 or XOtx5b results in embryos of reduced size with posterior deficiencies and ectopic cement gland structures(Kuroda et al., 2000;Vignali et al., 2000),suggesting that the two genes function similarly in the embryo. Given the pseudotetrapoid Xenopus laevis genome, these results strongly suggest that XOtx5 and XOtx5b arose from a gene duplication event. This is further supported by phylogenetic analysis unambiguously placingXOtx5 and XOtx5b in the same orthology class(Germot et al., 2001;Sauka-Spengler et al.,2001).

Previously, XOtx2 has been shown to be involved in anteriorizing the embryo and to have a role in cement gland formation(Andreazzoli et al., 1997;Blitz and Cho, 1995;Gammill and Sive, 1997;Gammill and Sive, 2001;Pannese et al., 1995). Here,we show that it is involved in specifying the bipolar cell. Recent studies on bipolar cell specification have examined several transcription factors using transgenic and mutant mice. Mice carrying a null mutation in Chx10produce retinas lacking differentiated bipolar cells(Burmeister et al., 1996). Similarly, Mash1-Math3 double knockout mice retinas are also missing bipolar cells, yet neither mutation alone affects bipolar cell production(Tomita et al., 2000;Tomita et al., 1996). Viral infection experiments show that together, Chx10, Mash1 andMath3, are all involved in bipolar cell genesis(Hatakeyama et al., 2001). However, not all cells that co-expressed Chx10 and eitherMash1 or Math3 become bipolars, suggesting that other factors are also necessary. Like Chx10, XOtx2 is expressed throughout the undifferentiated neural retina and later is restricted to differentiated bipolar cells. Like Chx10 and either Mash1 or Math3overexpression (Hatakeyama et al.,2001), overexpression of XOtx2 in developing retinal cells increases the number of bipolar cells in that population. Consistent with this, we found that lipofection of the N terminus and DNA binding domain of XOtx2 fused to the engrailed transcription repressor decreases the percentage of lipofected bipolar cells, suggesting thatXOtx2 acts as an activator to generate an increase in lipofected bipolar cells. It will be interesting to know if these four genes together,XOtx2, Chx10, Mash1 and Math3, are sufficient for directing all retinoblasts to a bipolar cell fate.

Analysis of deletion constructs of Crx and their ability to transactivate the bovine rhodopsin promoter lead to the conclusion that regions C-terminal to the homeodomain are important for gene activation(Chau et al., 2000). Moreover,they showed that the N terminus, homeodomain and basic region (aa 1-107) are necessary for binding the BAT-1 site on the rhodopsin promoter but are poor transactivators of rhodopsin (Chau et al.,2000). Consistent with this, we found that lipofection of this same region of XOtx2 (aa 1-109) or XOtx5b (aa 1-107) does not have any effect on photoreceptor cell fate, but when fused to theengrailed repressor or VP16 activator, lipofection of these fusion constructs influences the cell fate of the lipofected cells. Even though this region (aa 1-107) can successfully target the rhodopsin promoter in vitro(Chau et al., 2000), it may be that removal of such a large area of the protein (aa 110-289 for XOtx2; aa 108-290 for XOtx5b) could alter the ability of the fusion constructs to correctly target DNA cis-acting elements in vivo. However, when we swapped the C-terminal region of these proteins onto the opposite N-terminal and homeodomain sequence, we found that the region C-terminal to the homeodomain of XOtx5b and XOtx2 are capable of producing the same phenotype as XOtx5b and XOtx2, respectively,independently of which N terminus and HD they are fused to.

Clearly, the region C-terminal to the homeodomain of XOtx5b andXOtx2 is not simply required for activation or repression of target genes, as overexpression of the DNA binding domain and the sequence N-terminal to the homeodomain of either XOtx5b (aa 1-107) or XOtx2 (aa 1-109) fused to a VP16 activator or engrailed repressor produced a different labelled retinal cell distribution than overexpression with either the XOtx2 or XOtx5b alone. It has been shown that XOtx2 directly activates some genes like that of the Xenopus transcription factor Clock (Green et al.,2001), and indirectly activates the secreted protein, XAG, while it directly represses others, like the transcription factor Brachyury and the secreted factor, Wnt-5a (Latinkic et al.,1997; Morgan et al.,1999) (reviewed by Boncinelli and Morgan, 2001).

A basic leucine zipper protein, NRL, has been shown to work synergistically with Crx to transactivate the bovine rhodopsin promoter(Chen et al., 1997). Studies done on the binding of Crx with NRL suggest that the basic domain and homeodomain of Crx are necessary for this interaction(Mitton et al., 2000). Deletion analysis on Crx shows that the region C-terminal to the homeodomain is necessary for photoreceptor-specific gene activation with or without NRL. The fact that the VP16 activator fused to either XOtx2 orXOtx5b N-terminal region and homeodomain induces photoreceptor cells,suggest that these regions may be necessary for targeting genes involved in photoreceptor specification even if they are not sufficient to specify cell fate effectively. Deletion analysis also revealed that half of the transactivation function of Crx was lost upon removal of the Otx tail(Chau et al., 2000). Comparison of the Otx tail of all the Crx sequences listed to eight different species of Otx2 homologues showed an extra threonine (T) in the second repeat of the Otx tail of all the Otx2 proteins. Thus, site directed mutation analysis may clarify which residues in the C terminus of XOtx2 and XOtx5b are critical for the unique function of each of these closely related family members.

How are XOtx2 and XOtx5b involved in retinal cell fate specification? Our results confirm other studies done on chick(Bovolenta et al., 1997) that early in retina formation, XOtx2 transcripts are found in all layers of the neural retina. Yet, if XOtx2 is involved in bipolar cell specification in Xenopus, why do not all cells become bipolar cells?Similarly, XOtx5b is expressed in dividing retinoblasts in all layers of the early retina during its formation but is involved in specification of photoreceptor cells. This suggests that different transcription factors may be competing or cooperating in cell fate specification in a combinatorial way. The results with the Xenopus rhodopsin promoter may shed light on this. Both XOtx5b and XOtx2 alone can activate opsin expression but together they block activation of opsin. A similar mechanism has been shown for Barrier-to-autointegration-factor (Baf), which is expressed in bipolars with Crx in the mammalian retina. Baf interacts directly with Crx to repress Crx-mediated transactivation of a rhodopsin reporter(Wang et al., 2002). A possible mechanism for how the Otx transcription factors may interact to produce a similar result comes from recent studies of mouse Otx3, which is expressed in the developing eye, binds to the same consensus binding sequence as Crx and yet appears to act as a repressor rather than an activator(Zhang et al., 2002). This idea of cooperative specification is supported by the results of the co-lipofection experiment showing that misexpressed XOtx5b on its own induces photoreceptors but the combination of the XOtx2 andXOtx5b overrides this activity and induces bipolar cells. We hope that these findings may stimulate insight or speculation into the developmental and evolutionary mechanisms whereby overlapping patterns of homologous genes within a tissue specify the fates of similar and possibly homologous cell types within that tissue, much as Hox genes specify segmetal identity at the tissue level.

We would like to thank Drs Rod McInnes and Lynda Plodder for kindly sending us their mouse Chx10 antibody, Dr Daniel Kessler for kindly sending us hispCS2+Eng and pCS2VP16 DNA fusion constructs and DSHB for providing the XAP-1 antibody. We would also like to acknowledge Louise McKenna, Asha Dwivedy and Jessica Springbett for their excellent technical assistance and to thank Dr Sylvie Mazan for generously sharing her insights and thoughts, and Tilak Das and Jeremy Skepper for assisting in taking the confocal images. This work was supported in part by NEI (NRSA grant EY-07051)to A. S. V., a Hitchings-Elion fellowship from the Burroughs Wellcome Fund to M. E. Z. and grants from the Wellcome Trust and the EC to W. A. H.