The transcription factor RBP-J is essential for retinal cell differentiation and lamination

Background The highly ordered vertebrate retina is composed of seven cell types derived from a common pool of retinal progenitor cells (RPCs), and is a good model for the studies of cell differentiation and interaction during neural development. Notch signaling plays a pivotal role in retinogenesis in mammals, but the full scope of the functions of Notch pathway, and the underlying molecular mechanisms, remain unclear. Results In this study, we conditionally knocked out RBP-J, the critical transcription factor downstream to all four Notch receptors, in RPCs of mouse retina at different developmental stages. Disruption of RBP-J at early retinogenesis resulted in accelerated RPCs differentiation, but only photoreceptors and ganglion cells were overrepresented, with other neuronal populations diminished. Similarly, deletion of RBP-J at early postnatal days also led to overproduction of photoreceptors, suggesting that RBP-J governed RPCs specification and differentiation through retinogenesis. In all the RBP-J deletion models, the retinal laminar structures were distorted by the formation of numerous rosette-like structures, reminiscent of β-catenin deficient retina. Indeed, we found that these rosettes aligned with gaps in β-catenin expression at the apical surface of the retina. By in vivo electroporation-mediated transfection, we demonstrated that lamination defects in RBP-J deficient retinae were rescued by overexpressing β-catenin. Conclusions Our data indicate that RBP-J-mediated canonical Notch signaling governs retinal cell specification and differentiation, and maintains retinal lamination through the expression of β-catenin.


Background
The vertebrate retina consists of seven cell types organized into distinct laminar structures. In mice, retinal neurogenesis begins at embryonic (E) day 11.0 and terminates around postnatal (P) day 11. The six neuronal cell types and one glial cell type that make up the retina all derive from common retinal progenitor cells (RPCs) [1]. RPCs give rise to retinal cells in a conserved chronological sequence: ganglion cells and horizontal cells are born first, followed by cone photoreceptors and amacrine cells during the middle stage of retinogenesis. Rod photoreceptors, bipolar cells and Müller glial cells are the last cell types to be generated, mainly during postnatal stages [2].
The retina has been serving as an excellent model for the studies of cell differentiation and interaction during neural development, attributing to its limited cell types generated in a temporal-spatially defined process. Among molecules and pathways involved in retinal development, the Notch signaling has been demonstrated as an essential regulator of retinogenesis [3,4]. In mice, the canonical Notch pathway includes five ligands (Delta-like [Dll] 1, 3, 4, Jagged1, 2), four receptors (Notch1-4), and the transcription factor recombination signal-binding protein Jκ (RBP-J) (also termed CBF1 in mammals), which regulates the expression of downstream genes such as the Hairy and enhancer of split (Hes) family members. Both Notch ligands and receptors are type I transmembrane proteins mediating direct cell-cell interactions. Upon ligand binding, the Notch intracellular domain (NICD) is released by proteolytic cleavages within the transmembrane domain, and translocates into the nucleus, where it interacts with RBP-J and transactivates the transcription of downstream genes. Because RBP-J binds to NICD of all four mammalian Notch receptors, it serves as the key integrator of canonical Notch signaling [5,6].
Many members of this cascade have been shown to regulate retinogenesis. Transfection of an activated form of Xotch (the homologue of Notch in Xenopus) into cultured Xenopus retinal cells retains their neuroepithelial morphology [7]. Injection of antisense oligoneucleotides of CNotch1 (the chicken Notch1 homologue) into undifferentiated chicken retina increases the recruitment of RPCs differentiating into ganglion cells [8]. Notch1 deficient retinae have more cone photoreceptors and less ganglion cells [9,10], whereas in Hes1 mutant mice, ganglion cells are overproduced [11]. Deleting Hes5 in the mouse retina significantly decreases Müller glial cell population [12]. These findings while revealing the importance of Notch signaling in retinal cell specification and differentiation, demonstrate that the manipulations of individual genes in the Notch pathway can yield a variety of phenotypes, and therefore highlight a need of further studies to fully understand the complexity and molecular mechanisms of Notch signaling during retinal development.
While adopting specific cell fates, differentiating retinal cells migrate to appropriate laminae during retinogenesis. RPCs orient radially along the apical/basal axis and extend their endfeet anchored on either side [13,14]. Retinal cell migration requires the detachment of their endfeet from the apical adherens junctions [15,16]. Retinaspecific inactivation of β-catenin, a component of adherens junctions, has been shown to result in severe retinal lamination defects without affecting cell specification or differentiation [17], suggesting the importance of cell adhesion in retinal lamination. Interestingly, eliminating Notch1 expression from the developing retina also leads to severe lamination defects [9,10], although the underlying molecular mechanisms remain to be elusive.
To gain more insight into the various functions of the Notch signaling pathway in retinal development, we conditionally deleted RBP-J in the mouse retina at both embryonic and postnatal stages. Our results show that RBP-J-mediated canonical Notch signaling not only governs retinal cell specification and differentiation, but also maintains retinal lamination as well, which is achieved probably through the expression of β-catenin.

Animals
Mice were maintained on the C57BL/6 genetic background. Mice carrying the Chx10Cre BAC [18] or Pet1Cre BAC [19] transgene were crossed with mice carrying a floxed RBP-J allele [5] to obtain Chx10Cre-RBP f/f mice or Pet1Cre-RBP f/f mice. Mice bearing either of these two genotypes survived into adulthood. The expression of Cre in the neural retina of Pet1Cre mice was determined by X-gal staining after crossing to Rosa26 reporter mice [20]. Animal experiments were reviewed and approved by the Animal Experiment Administration Committee of the Fourth Military Medical University.

BrdU labeling and terminal deoxynucleotidyl transferasemediatd biotinylated UTP nick end labeling (TUNEL)
Pregnant mice were given a single intraperitoneal injection of BrdU (60 μg/g body weight) at 12.5 and 13.5 days postcoitum, and were sacrificed 2 hours later. Sections were processed for immunostaining with anti-BrdU, as described above. TUNEL assay was performed using the In Situ Cell Death Detection Kit (Roche).

In vivo electroporation
In vivo electroporation of P0 and P5 mouse pups was performed as described [30]. Five square pulses of 50 milliseconds duration with 1 second intervals were applied using a BTX ECM830 pulse generator (Boston, MA, USA). Pulses with 80 V and 85 V were applied to P0 and P5 pups, respectively. pCAG-EGFP was coelectroporated to visualize transfected cells, and immunostaining was performed to confirm expression of electroporated genes.

Statistics
Cells were counted with Stereo Investigator software (MBF Bioscience), and only cell bodies that were labeled with immunoreactivity were included. Proportions of immunoreactive cells in the total population of retinal cells revealed by Hoechst staining or in total electroporated cells labeled by EGFP or Cre were calculated, and comparisons were made using unpaired Student's t-test. Three sections through the optic disc were selected from each eye, and at least four eyes from each genotype were analyzed. Data were expressed as mean ± s.e.m.

Reduced eye size and disrupted laminar structure of RBP-J deficient retinae
We first used in situ hybridization to determine the expression of RBP-J in the developing retina. RBP-J transcripts were first detected in the neural retina at E10.5, at higher levels at E11.5 and E13.5. During this period, the transcripts of Notch receptors (Notch1, 3) and ligands (Dll1, Jagged1) were also detected (Additional file 1: Figure S1 and data not shown). In addition, RBP-J transcripts were still detectable in the retina at E16.5 and early postnatal stages (P0 and P5; data not shown). To study the putative roles of RBP-J in retinal development, we generated Chx10Cre-RBP f/f mice ( Figure 1A) using RBP-J-floxed mice [5] and Chx10Cre mice [18], in which Cre expression was started in RPCs at approximately E10.5, resulting in almost complete loss of RBP-J in the retinae by E13.5 (Additional file 1: Figure S2). Compared to wild-type controls, the eyes of Chx10Cre-RBP f/f mice were greatly reduced in size at postnatal stages ( Figure 1B). HE staining at P21 showed that in Chx10Cre-RBP f/f mice, the cellular organization of the laminar retinal structure was totally distorted by the appearance of rosette-like structures (Figure 1C and 1D). This is similar to that previously reported in Notch1 deficient retinae [9].
We next examined Chx10Cre-RBP f/f eyes at various embryonic stages to determine the time of onset of these morphological abnormalities. The results showed that rosettelike structures were detectable as early as E13.5 ( Figure 1E-J), while eye size was notably reduced by E14.5 (data not shown).
Pet1Cre mice [19] express Cre in the retina as of E11.0. In contrast to the nearly ubiquitous Cre expression in Chx10Cre mice, Cre activity in Pet1Cre mice was patchy and restricted to a small portion of retinal cells (Additional file 1: Figure S3 A-D). Pet1Cre-RBP f/f mice did not show lamination defects at embryonic stages, although a few rosettes were found at P30 ( Figure 1K and 1L). To clarify the relationship between rosettes and RBP-J mutant cells, we crossed Pet1Cre-RBP f/f mice to Rosa26 reporter mice and observed the distribution of X-gal + cells at P21. Interestingly, X-gal + (mutant) cells were clustered in these rosettes ( Figure 1M-P), suggesting that the lamination defects observed in RBP-J deficient retina is likely a cellautonomous effect of the loss of RBP-J. Taken together, our data indicate that inactivation of RBP-J expression in the retina results in a severe reduction of eye size and abnormal lamination during retinal development.

RBP-J deficient retinae
To determine whether these defects were due to changes in cellular proliferation and/or apoptosis, we performed BrdU and TUNEL labeling between E12.5 and E14.5. BrdU labeling showed significantly reduced cell proliferation at E13.5, while TUNEL assay revealed an increase in the percentage of apoptotic cells at E14.5 in Chx10Cre-RBP f/f retinae compared to controls ( The reduced proliferation and enhanced apoptosis in E13.5-E14.5 Chx10Cre-RBP f/f eyes raised the possibility that RBP-J mutant cells were diminished at later stages of development. To clarify this, we crossed Chx10Cre-RBP f/f mice to Rosa26 reporter mice and observed the distribution of X-gal + cells at P21. In Chx10Cre retinae, X-gal + cells constituted the vast majority of retinal tissue and were homogeneously present throughout the retina ( Figure  2G). Similarly, in Chx10Cre-RBP f/f mice, X-gal + cells predominated and were distributed throughout the retina, albeit less uniformly ( Figure 2H), showing that although increased cell death occured in RBP-J deficient retinae, a substantial number of RBP-J mutant retinal cells survived.

Alteration of cell types in the mature RBP-J deficient retina
The mature neural retina is composed of seven distinct types of cells partitioned into stereotypic layers. To determine whether retinal cell types were altered in the absence of RBP-J, we performed immunolabeling and in situ hybridization of various retinal cell markers in the mature RBP-J deficient retinae. We found that rhodopsin + and On the other hand, the number of ganglion cells was greatly reduced, as shown by Brn3 immunostaining [31] and by Nefl in situ hybridization ( Figure 3I-L and 3A'). Similarly, immunostaining of NF165 and syntaxin [2] indicated that the horizontal and amacrine interneuron populations were also dramatically reduced ( Figure 3M-P and 3A'). In addition, PKCα + rod bipolar cells were decreased in number, while NK3 + cone bipolar cells [32] were nearly completely lost in RBP-J deficient retinae (

X-gal + mutant cells in
Chx10Cre-RBP f/f retinae predominated and were distributed throughout the retina, albeit less uniformly (H), compared to that in Chx10Cre retinae (G). (I, J) Significant difference of percentages of BrdU + cells in the total retinal cells is observed between the two genotypes at E13.5 (I, 25.3 ± 6.3% of the total retinal cells in wild-type, 13.6 ± 4.9% of the total retinal cells in Chx10Cre-RBP f/f ; P < 0.05), and significant difference of TUNEL + cells is found at E14.5 (J, 0.59 ± 0.13% of the total retinal cells in wild-type, 1.15 ± 0.14% of the total retinal cells in Chx10Cre-RBP f/f ; P < 0.05). Scale bars, 100 μm. and 3A'), although GFAP immunostaining showing the endfeet of Müller glial cells [33] appeared to be increased relative to controls ( Figure 3Y and 3Z). It should be noted that Chx10Cre-RBP f/+ mice showed neither reduction of eye size nor changes of retinal cell types compared to wildtype controls, suggesting there is no haplo-insufficiency effect of RBP-J gene in mammalian retinal development. Taken together, these results show that the relative population size of photoreceptors is increased, while that of other neuronal cell types is decrease, in the case of Müller glial cells, unchanged in the mature RBP-J deficient retina.

Deletion of RBP-J leads to overproduction of ganglion cells during early retinogenesis
Canonical Notch signaling inhibits neurogenesis [3,34], and therefore the blockade of Notch pathway in the retina is expected to result in enhanced production of neuronal cells, especially of the first-born ganglion cells [8,35]. However, the population of ganglion cells was decreased Alteration of cell types in mature RBP-J deficient retinae in the mature Chx10Cre-RBP f/f retinae (Figure 3). To gain more insights into this inconsistency, we focused our examination of RBP-J deficient retinae at stages prior to E14.5, at which abnormal cell death was first detected (Figure 2). The differentiation of ganglion cells was accelerated in Chx10Cre-RBP f/f embryos at E12.5 and E13.5, as shown by the up-regulation of Math5, which promotes the ganglion cell fate [34], and two ganglion cell marker genes, Nefl and Islet1 [9] (Figure 4A-F). Furthermore, we found that the proportion of Brn3 + ganglion cells in the total Hoechst-counterstained retinal cells was increased in Chx10Cre-RBP f/f retinae at E13.5 and E16.5 ( Figure 4H, J and 4M), but was reduced by P0 ( Figure 4L and 4M). Thus, inactivation of RBP-J in retina leads to overproduction of ganglion cells at embryonic stages, but does not persist postnatally. This phenotype is distinct from that of Notch1 deficient retinae in which ganglion cells are reduced at both embryonic and postnatal stages [9,10].
Up-regulation of ganglion cell differentiation-related genes and increase of ganglion cells in early retinogenesis of Chx10Cre-RBP f/ f mice

Precocious differentiation and ultimate reduction of interneurons in
Chx10Cre-RBP f/f retinae Interneurons are underrepresented in mature Chx10Cre-RBP f/f retinae (Figure 3). To determine whether there was a concomitant change in the generation of interneurons, we stained Chx10Cre-RBP f/f retinae for the bHLH proneural gene Math3 that promotes the differentiation of all three interneuron cell types (horizontal, amacrine and bipolar cells) [25,34]. At E13.5, the expression domain of Math3 was restricted to the outer part (close to the pigmented epithelium) in control retinae, whereas it was remarkably expanded into the inner part of Chx10Cre-RBP f/f retinae ( Figure 5A and 5B). Furthermore, the bHLH proneural genes Neurod1, which promotes amacrine cell and photoreceptor fates [34,36], Mash1, which is involved in the differentiation of bipolar cells and photoreceptors [37], and Ngn2 [38] were all up-regulated in RBP-J deficient retinae at E13.5 ( Figure 5C-H). We also examined the expression of Notch downstream Hes genes in RBP-J deficient retinae. We found that Hes1 expression was dramatically decreased in Chx10Cre-RBP f/f retinae at E13.5, when initial morphological change occurred ( Figure 5I and 5J), while Hes5 and Hesr2 showed no obvious change compared to controls ( Figure 5K, L and Additional file 1: Figure S4).
Although these proneural genes were up-regulated at E13.5, the population of retinal interneurons was reduced at later embryonic stages. We found that the number of horizontal cells, the first interneuron type generated, was greatly decreased at E16.5, as determined by NF165 immunostaining (Figure 5M and 5N). Postmitotic precursors of GABAergic interneurons [39][40][41] were decreased as well, as shown by Ptf1a in situ hybridization ( Figure 5O and 5P). In addition, the mRNA levels of the HMG box transcription factor Sox2 and the homeodomain transcription factor Pax6, which are involved in the maintenance of RPCs [2], were also decreased at E13.5 ( Figure  5Q-T). Furthermore, Cyclin D1 and Fgf15 were decreased as well ( Figure 5U-X). Taken together, these results suggest that in the absence of RBP-J, RPCs likely differentiate prematurely into interneuron precursors, and the eventual reduction in the number of interneurons in later retinogenesis might be due to subsequent lowered cell proliferation and/or excessive cell death.

Inactivation of RBP-J promotes photoreceptor specification
The transcription factors Otx2 and Crx are essential for photoreceptor specification and terminal differentiation, respectively [42]. We examined their expression in RBP-J deficient retina. In normal retinae, Otx2 and Crx were expressed only in the outer part at E13.5, but their expression domains were expanded into the inner part of Chx10Cre-RBP f/f retinae ( Figure 6A-D). Furthermore, while Crx, Otx2 and the cone precursor-specific gene Thrβ2 were restricted to the outermost layer in control retinae at E16.5, their expression was found throughout Chx10Cre-RBP f/f retinae, and specifically enriched at the center of rosettes ( Figure 6E-H, data not shown).
Unlike cone photoreceptors, no obvious increase of rod precursors were detected as measured by the expression of NR2e3 in Chx10Cre-RBP f/f retinae at embryonic and early postnatal stages (Additional file 1: Figure S5 and data not shown). Because rod precursors differentiate relatively late in retinogenesis, and rod photoreceptors are increased in the mature Chx10Cre-RBP f/f retinae (Figure 3), our failure to see an increase of the number of Nr2e3 + cells might be due to possible secondary defects during retinal development. To further determine the role of RBP-J in rod photoreceptor specification, we performed in vivo electroporation of Cre in RBP f/f retinae at P0 and P3 to inactivate RBP-J within the peak stage of active generation of rod photoreceptors as well as bipolar cells [34]. pCAG-Cre and pCAG-EGFP plasmids were co-electroporated into the retinae of RBP f/+ , RBP f/f and wild-type mice. EGFP + cells showed Cre immunoreactivity (see below). By P14, most electroporated RPCs in RBP f/+ and wild-type retinae had differentiated into rod photoreceptors in the outer nuclear layer and, to a lesser degree, bipolar and Müller glial cells in the inner nuclear layer ( Figure 6I, J and data not shown), consistent with previous data [30]. However, in RBP f/f retinae, the number of EGFP/rhodopsin-colabeled cells was obviously increased, whereas EGFP/PKC-colabeled bipolar cells and EGFP/glutamine synthetase-colabeled Müller glial cells were greatly reduced ( Figure 6L, M, and data not shown). Quantitative analysis of P0-and P3electroporation data revealed that the percentage of rhodopsin + rod photoreceptors in the total number of EGFP + cells was increased in RBP f/f retinae compared with controls ( Figure 6O). Correspondingly, the percentages of PKCα + bipolar cells and glutamine synthetase + Müller glial cells in the total number of EGFP + cells were decreased ( Figure 6O). These results indicate that RPCs are more biased to take on the rod photoreceptor fate in the absence of RBP-J in the retina. We also observed that the inactivation of RBP-J at P0 and P3 resulted in the appearance of rosettes ( Figure 6L, data not shown). Thus, our data show that the deletion of RBP-J in RPCs during early retinogenesis results in an increase of the two types of photoreceptors, and postnatal inactivation leads to an overproduction of photoreceptors at the expense of bipolar and Müller glial cells.

Postnatal inactivation of RBP-J at P5 also decreases Müllerglial cells
Notch signaling promotes gliogenesis [4,12]. We found that inactivating RBP-J at P0 and P3 impeded the generation of Müller glial cells (Figure 6), but their number was Up-regulation of interneurons differentiation-related genes but decrease of interneurons in Chx10Cre-RBP f/f retinae at embry-onic stages   Figure  3). To further dissect the role of RBP-J in retinal gliogenesis, we inactivated RBP-J by in vivo electroporationinduced Cre expression at P5, the stage at which Müller glial cell differentiation is mostly active [34]. In P14 RBP f/ + and wild-type mice, EGFP-labeled cells were distributed in the inner nuclear layer and the photoreceptor-enriched outer nuclear layer. Furthermore, most of EGFP-labeled cells in the inner nuclear layer extended two processes to the basal and apical surfaces, the typical morphology of mature Müller glial cells ( Figure 6K). In contrast, none of EGFP-labeled mutant cells in the inner nuclear layer possessed this morphological characteristic of Müller glial cells in RBP f/f retinae ( Figure 6N). Glutamine synthetase immunostaining confirmed these morphological observations. Approximately 10.51 ± 1.20% of EGFP-labeled cells were glutamine synthetase + in RBP f/+ retinae, compared to only 0.40 ± 0.04% in RBP f/f retinae ( Figure 6K, N  and 6P). The percentage of PKCα + bipolar cells in all EGFP + cells increased from 30.52 ± 1.01% in RBP f/+ retina to 37.40 ± 1.74% in RBP f/f retina. In addition, the percentage of rhodopsin + rod photoreceptors changed from 58.24 ± 0.72% in RBP f/+ retina to 61.74 ± 1.81% in RBP f/f retina, albeit not with statistical significance ( Figure 6P). These results indicate that deleting RBP-J at P5 impairs the generation of Müller glial cells.

RBP-J participates in retinal lamination through apical βcatenin expression
In addition to abnormal development of the different retinal cell types, Chx10Cre-RBP f/f mice also exhibited retinal lamination defects. It has been shown that conditional deletion of β-catenin disrupts lamination and causes many rosette-like structures to form [17], and that Notch activity regulates the cytoplasmic level of β-catenin in vertebrates [43] and invertebrates [44,45]. We thus examined the expression of β-catenin at E13.0 and E13.5 by immunohistochemistry. In normal retinae, β-catenin was enriched at both apical and basal surfaces ( Figure 7A and 7C). However, in RBP-J deficient retinae, β-catenin immunoreactivity was absent from the apical surface in a few regions at E13.0, and these gaps were expanded and increased in number at E13.5 ( Figure 7B and 7D). Counterstaining with Hoechst at E13.5 revealed that apical extents lacking β-catenin overlayed the rosettes ( Figure 7B and 7D).
The precocious differentiation of RPCs already appeared at E13.5 (Figure 4 and 6), coexisted with the disturbance of apical β-catenin expression. To gain more insights into their relationship, we examined the expression of cell markers in RBP-J deficient retinae at E13.0, the earliest stage when β-catenin gap was observed. Double-immunostaining showed that Brn3 + ganglion cells were increased locally in the retinal regions lacking β-catenin. Accordingly, Sox2 + RPCs were initially decreased within these gaps, as compared with those in normal β-catenin expressing regions ( Figure 7E-H). At E12.5, prior to the βcatenin expression defects, Math5 and Neurod1 were upregulated ( Figure 4A and 4B; Figure 7I and 7J), while the level of Math3, Mash1, Pax6, Hes1 and Hes5 showed no differences compared with controls ( Figure 8K-P, data not shown).
We further inactivated RBP-J at P0 by electroporating Cre into RBP f/f retinae. We observed a similar phenotype: βcatenin expression at the apical surface was lost by P3 in areas proximal to clusters of EGFP + cells ( Figure 8A-D). Expression of Cre in RBP f/+ or wild-type retinae, in contrast, did not affect β-catenin expression ( Figure 8A and 8C, data not shown). Although the proportion of rhodopsin + cells in the total population of EGFP + cells was not altered (Additional file 1: Figure S6) at P3, Sox2 + RPCs in the total population of Cre + cells was decreased in the electroporated regions of RBP f/f retinae, as compared with Cre-electroporated RBP f/+ or wild-type retinae ( Figure 8E, G and 8O, data not shown).
To further clarify that the loss of apical β-catenin expression contributes to defective lamination in RBP-J deficient retina, we co-transfected P0 RBP f/+ and RBP f/f retinae with pCAG-Cre, pCAG-β-catenin and pCAG-EGFP. EGFP- expressing cells showed enhanced accumulation of cytoplasmic β-catenin at P14 (Figure 8M and 8N). Interestingly, the expression of β-catenin in RBP-J deficient retinae prevented the formation of rosettes, restoring lamination to normality in transfected areas of RBP f/f retinae ( Figure 8J and 8M). However, expression of β-catenin did not rescue the retinal cell specification defects, as Sox2 + RPCs were decreased at P3, rod photoreceptors were present in excess at P14, and the number of bipolar cells was still reduced at the same stage ( Figure 8H, O and 8P).

Inactivation of RBP-J enhances production of photoreceptors
Co-transfection of β-catenin and Cre at P0 did not cause any detectable morphological changes or cell type change in RBP f/+ or wild-type retinae at P3 ( Figure 8E, F and 8O) and P14 ( Figure 8K, N and 8P, data not shown). These results indicate that RBP-J-modulated β-catenin expression in the apical retina is necessary for the formation of retinal laminar structures.
Apical disturbance of β-catenin expression and localization is accompanied by precocious differentiation of RPCs in Chx10Cre-RBP f/f retinae

Discussion
In the present study, the transcription factor RBP-J, which integrates signals from the four mammalian Notch receptors, was specifically inactivated in mouse RPCs, thereby circumventing potential functional redundancy among Notch receptors during retinal development. We found that disruption of RBP-J in early retinogenesis resulted in overproduction of ganglion cells and photoreceptors at the expense of retinal interneurons, while postnatal deletion of RBP-J at P0/P3 promoted the generation of rod photoreceptors at the expense of bipolar and Müller glial cells, and deletion at P5 enhances the production of bipolar cells at the expanse of Müller glial cells. These results suggest that RBP-J-mediated signaling controls cell specification and differentiation in retinogenesis in a stepwise manner. Furthermore, the formation of rosette-like structures in RBP-J deficient retina is likely a consequence of disturbed apical β-catenin expression, because restoring βcatenin expression rescued the lamination defects. In summary, our results show that RBP-J signaling plays multiple roles during retinal cell specification, differentiation and lamination.

Roles of RBP-J in retinal neurogenesis
Among the six neuronal cell types, only the population of photoreceptors was found to be increased in both embryonic and mature Chx10Cre-RBP f/f retinae (Figure 3 and 6). The result also get support from the observation that postnatal inactivation of RBP-J in the retina at P0 and P3 also leads to an overproduction of photoreceptors and this occurs at the expense of bipolar and Müller glial cells (Figure 6). Therefore, it is likely that in the absence of RBP-J, RPCs are biased to take on photoreceptor fate in the retinogenesis. This is consistent with the recent finding that the suppression of Notch signaling promotes ES cell-derived RPCs to differentiate into Crx + photoreceptor precursors [46].
In Notch1 deficient retina, only cone photoreceptors are increased [9,10], whereas both two types of photoreceptors as well as ganglion cells are overproduced in RBP-J deficient retinae (Figure 3 and 4). These differences might underscore compensatory influences of other Notch receptors expressed in Notch1 deficient retinae. For instance, Notch3, which is also expressed in the neural retina (Additional file 1: Figure S1), could inhibit the expression of specification-and/or differentiation-related genes and thereby prevent the superfluous generation of ganglion cells and rod photoreceptors (Brown et al., 2001;Wang et al., 2001). The differences between the Notch1 and RBP-J deficient retinae also suggest that different Notch receptors may regulate the generation of distinct types of retinal cells.
During early stages of retinogenesis, the number of ganglion cells was increased in Chx10Cre-RBP f/f retinae, but ultimately reduced by P0 (Figure 3 and 4). This may be explained by an increase in cell death at later stages of retinogenesis in the absence of RBP-J. On the other hand, the expression of several bHLH proneuronal genes that are known to promote retinal interneuron differentiation is up-regulated at E13.5, but the number of interneurons is decreased at embryonic and postnatal stages (Figure 3 and 5). This suggests that the rate at which RPCs differentiate into retinal interneurons is accelerated in the early retinogenesis, and this precocious differentiation, which in turn leads to the depletion of their precursor pool, together with elevated levels of apoptosis and the preferential generation of photoreceptors, would lead to a net decrease of these neurons in RBP-J deficient retinae. This could happen in the Notch dependent or independent manner. A recent study has shown that RBP-J forms a complex with the transcription factor Ptf1a in spinal cord to promote the differentiation of GABAergic interneurons, independent of the canonical Notch pathway [47]. Interestingly, Ptf1a, which is also expressed in horizontal and amacrine cell precursors and is involved in the differenti- ation of the two types of interneuron [39][40][41], is downregulated in RBP-J deficient retinae. Further experiments are needed to determine if RBP-J and Ptf1a interact in retinal precursors and if the RBP-J-Ptf1a complexes regulate horizontal and amacrine cell differentiation.

Gliogenesis in RBP-J deficient retinae
Notch signaling promotes glial cell differentiation. A recent study has shown that knocking out RBP-J in the dorsal root ganglia results in the loss of glia cells [48], but we did not observe obvious change of Müller glial population size in the mature Chx10Cre-RBP f/f retinae ( Figure  3). However, postnatal inactivation of RBP-J at P0, P3 and P5 led to a decrease in the number of glia cells in the retina (Figure 6), suggesting that RBP-J acts as a positive regulator of retinal gliogenesis. No obvious change of Müller glia cells in the mature RBP-J deficient retinae is likely to be a consequence of neuronal loss and/or global distorted lamination, which may lead to reactive gliogenesis. On the other hand, Hes5 and Hesr2, two Notch-RBP-J effectors that promote retinal gliogenesis [12,28], remained unchanged in the RBP-J deficient retina ( Figure 5 and Additional file 1: Figure S4). It is therefore possible that the sustained expression of Hes5 and Hesr2 contributes to normal generation of Müller glial cells in RBP-J deficient retina. This finding also suggests the existence of a RBP-J independent pathway that activates or maintains Hes5 and Hesr2 expression in the RBP-J deficient retina.

β-catenin expression
Lamination defects always appear in mouse retinae when Notch signaling is impaired. For example, Notch1 deficient and Hes1 mutant retinae both display similarly distorted morphologies, including the appearance of numerous rosette-like structures [9,11,49]. Three main factors are thought to contribute to retinal lamination: the retinal pigmented epithelium, Müller glial cells, and cell adhesion among RPCs [50]. Lamination defects in Chx10Cre-RBP f/f retinae are most likely due to defective cell adhesion among RPCs, because Chx10Cre is not expressed in the pigmented epithelium, and furthermore, the defects precede Müller glial cell differentiation. RPC adherens junctions include the homophilic adhesion molecule N-cadherin and the intracellular binding partner β-catenin, which links the cytoplasmic domain of Ncadherin to the actin cytoskeleton. Eliminating N-cadherin expression in Zebrafish disrupts retinal lamination [51], while conditional inactivation of β-catenin in the mouse retina also results in severe retinal lamination defect [17].
Adherens junctions form at both the apical and basal surfaces of the developing retina, where the endfeet of RPCs are anchored, with the apical junctions are especially important for retinal lamination [17,50]. We found that the morphological changes in RBP-J deficient retinae were correlated with the disturbance of apical expression of βcatenin, and that overexpression of β-catenin could rescue the lamination phenotype (Figure 7 and 8). Thus, normal expression of β-catenin in the retina requires RBP-J, while defective β-catenin expression leads to the formation of rosette-like structure in RBP-J deficient retina. Notch activity has been shown to regulate the cytoplasmic level of βcatenin [43][44][45], and RBP-J has been found to bind to βcatenin in mouse neural precursor cells [52]. Thus it could be interesting in future studies to explore if such interactions also exist in RPCs and if loss of apical β-catenin is caused directly by deficiency of RBP-J. On the other hand, since the up-regulation of proneural genes Math5 and Neurod1 precedes the disturbance of apical expression of β-catenin, and an increase of Brn3 + ganglion cells and a decrease of Sox2 + RPCs are present in the retinal region with defective apical expression of β-catenin (Figure 7), it is possible that this disturbance of β-catenin expression is due to the change of bHLH expression profiles in the absence of RBP-J. Further studies are needed to elucidate whether the up-regulation of Math5 and Neurod1 leads to defective expression of β-catenin in RBP-J deficient retina.
Supplying β-catenin by in vivo electroporation could rescue the lamination defects but not precocious differentiation or cell type mispecification in RBP-J deficient retina, while overexpression of β-catenin in wild-type and RBP f/+ retinae did not affect the differentiation of RPCs ( Figure  8). These results suggest that the restoration of lamination defect by β-catenin is not achieved by inhibition of precocious differentiation of RPCs in RBP-J deficient retina. In addition, conditional deletion of β-catenin in the retina leads to abnormal lamination, without affecting either cell specification or differentiation [17]. Therefore, it seems most likely that the role of β-catenin in retinal lamination is independent of retinal cell differentiation.

Conclusions
In our present study, we conditionally inactivated transcription factor RBP-J which integrates all Notch receptor signals in mouse retina, and found that RBP-J mediated Notch signaling inhibits ganglion cell and photoreceptor differentiation, and promotes Müller glial cell differentiation during retinogenesis. In addition, RBP-J regulates retinal lamination via maintaining apical β-catenin expression. Altogether, our results indicate that RBP-J mediated Notch signaling not only governs cell-type differentiation, but also participates in cellular organization during retinal development, which ensure the exquisite process of retinogenesis.