Abstract
Vertebrate retinal progenitor cells (RPCs) are pluripotent, but pass through competence states that progressively restrict their developmental potential (Cepko et al., 1996; Livesey and Cepko, 2001; Cayouette et al., 2006). In the rodent eye, seven retinal cell classes differentiate in overlapping waves, with RGCs, cone photoreceptors, horizontals, and amacrines forming predominantly before birth, and rod photoreceptors, bipolars, and Müller glia differentiating postnatally. Both intrinsic and extrinsic factors regulate each retinal cell type (for review, see Livesey and Cepko, 2001). Here, we conditionally deleted the transcription factor Rbpj, a critical integrator of multiple Notch signals (Jarriault et al., 1995; Honjo, 1996; Kato et al., 1997; Han et al., 2002), during prenatal mouse retinal neurogenesis. Removal of Rbpj caused reduced proliferation, premature neuronal differentiation, apoptosis, and profound mispatterning. To determine the cell autonomous requirements for Rbpj during RGC and cone formation, we marked Cre-generated retinal lineages with GFP expression, which showed that Rbpj autonomously promotes RPC mitotic activity, and suppresses RGC and cone fates. In addition, the progressive loss of Rbpj−/− RPCs resulted in a diminished progenitor pool available for rod photoreceptor formation. This circumstance, along with the overproduction of Rbpj−/− cones, revealed that photoreceptor development is under homeostatic regulation. Finally, to understand how the Notch pathway regulates the simultaneous formation of multiple cell types, we compared the RGC and cone phenotypes of Rbpj to Notch1 (Jadhav et al., 2006b; Yaron et al., 2006), Notch3, and Hes1 mutants. We found particular combinations of Notch pathway genes regulate the development of each retinal cell type.
Introduction
Notch is a major metazoan signaling pathway and key regulator of retinal neurogenesis (for review, see Perron and Harris, 2000; Baker, 2001; Livesey and Cepko, 2001; Lai, 2002). A Notch signal is transmitted between two cells through ligand-receptor binding, which triggers release of the NOTCH intracellular protein domain (ICD) within the receiving cell (for review, see Fortini, 2009; Kopan and Ilagan, 2009). Subsequently, the NOTCH ICD is transported into the nucleus, where it complexes with RBPJ and MAML and activates target genes such as Hes1 or Hes5. The Notch pathway transmits multiple types of signals. The most well studied of these, lateral inhibition, occurs when an equivalent group of cells initially express ligand and receptor uniformly, until one cell stochastically begins to express more ligand, making it the signaling cell. The classic, canonical pathway: Delta ⇒ Notch ⇒ Rbpj(CSL) ⇒ Hes/E(spl), is widely used during neuronal and glial development (for review, see Baker, 2000; Fortini, 2009; Kopan and Ilagan, 2009).
In the rodent retina, multiple Notch pathway genes are expressed (Weinmaster et al., 1991, 1992; Austin et al., 1995; Ahmad et al., 1997; Bao and Cepko, 1997; Rowan et al., 2004; Nelson et al., 2006). In the frog and chick eye, Delta-Notch signaling controls the temporal development of multiple cell classes. However, in the mouse retina, the early lethality of Notch pathway mutations has hampered a deep examination of these genes (Austin et al., 1995; Dorsky et al., 1995; Ahmad et al., 1997; Dorsky et al., 1997; Henrique et al., 1997; Furukawa et al., 2000; Schneider et al., 2001; Silva et al., 2003). Recently, Cre-lox deletion of Notch1 demonstrated a critical role for this receptor in repressing ectopic cone photoreceptor development (Jadhav et al., 2006b; Yaron et al., 2006). In Notch1−/− eyes, RPCs prematurely exited the cell cycle and differentiated as cone photoreceptors as early as E13.5, upregulating three factors: Otx2, Crx and Thrβ2/Thrb (Jadhav et al., 2006b; Yaron et al., 2006). The extra cones arose at the expense of rod fates, although a loss of retinal ganglion cells (RGCs) was also described (Yaron et al., 2006).
While these Notch1 studies advanced our knowledge of retina neurogenesis, additional questions remain. First, are mammalian cones regulated by a canonical Notch signal (DELTALIKE1 ⇒ NOTCH1 ⇒ RBPJ ⇒ HES1) and if not, which ligand, receptor and downstream effector are involved? Why were cone fates selectively derepressed in Notch1 mutant embryos? Finally, why did RGCs decrease in Notch1 conditional mutants, which differed from other retinal studies demonstrating that Delta-Notch signaling blocks RGC genesis (Austin et al., 1995; Dorsky et al., 1995, 1997; Ahmad et al., 1997; Henrique et al., 1997; Schneider et al., 2001; Silva et al., 2003)? Here, we assessed several roles for the pathway integrator Rbpj during mouse RGC and photoreceptor development, as well as the prenatal RGC and photoreceptor phenotypes of Notch3 and Hes1 mutants.
Materials and Methods
Mice
The Rbpjtm1Hon conditional allele (termed RbpjCKO) was generated by Han et al., maintained on a 129/SvJ background, and genotyped as described (Han et al., 2002). Notch3Gt(PST033)Byg gene trap mutant mice (termed Notch3LacZ) were generated by Mitchell and colleagues, and maintained as a homozygous viable stock in a 129/BL6 mixed background (Leighton et al., 2001; Mitchell et al., 2001; Pan et al., 2004; Demehri et al., 2008). Hes1tm1Fqu mutant mice (termed Hes1−/−) were generated by Ishibashi and colleagues, maintained on an ICR background and genotyped as previously described (Ishibashi et al., 1995; Cau et al., 2000). α-Cre transgenic mice were generated by Marquardt et al., maintained on a CD-1 background and PCR genotyped as described (Marquardt et al., 2001). Chx10-Cre BAC transgenic mice in a CD-1 background were obtained from Jackson Laboratories and genotyped as in (Rowan and Cepko, 2004). Z/EG lineage tracing mice also in a CD-1 background were acquired from Jackson Laboratories and genotyped for GFP per (Novak et al., 2000). Images of adult heads were captured on a Leica dissecting microscope with an Optronics digital camera and software.
Retinal phenotype analyses
Embryonic and postnatal tissues were fixed in 4% paraformaldehyde/PBS for 40–60 min at 4°C, processed through a sucrose/PBS series, cryoembedded and sectioned. Primary antibodies used were anti-βgal (Tom Glaser, University of Michigan, Ann Arbor, MI; 1:1000); anti-BrdU (Abd Serotec; 1:500); anti-cleaved PARP (Cell Signaling Technology; 1:500); anti-CRX (Cheryl Craft, University of California, Los Angeles, Los Angeles, CA; 1:1000), anti-S OPSIN (Cheryl Craft; 1:1000), anti-M/L OPSIN (Cheryl Craft; 1:1000) (Zhu and Craft, 2000; Zhu et al., 2003); anti-RHODOPSIN (Millipore Bioscience Research Reagents; 1:1000); anti-POU4F2 (Santa Cruz Biotechnology; 1:50); anti-SOX9 (Millipore Bioscience Research Reagents; 1:200); anti-RXRγ (Santa Cruz Biotechnology; 1:200); anti-THRB/TRβ2 (Douglas Forrest, National Institutes of Health, Bethesda, MD; 1:2500) (Ng et al., 2009); anti-NR2E3 (Anand Swaroop, National Eye Institute, Bethesda, MD; 1:500) (Cheng et al., 2004); anti-TUBB3 (Covance; 1:1000); anti-CCND1 (Santa Cruz Biotechnology; 1:500), anti-GFP (Invitrogen; 1:1000 or Abcam; 1:1000); anti-HES1 (1:1000) (Lee et al., 2005); anti-ISL1 (Developmental Studies Hybridoma Bank, University of Iowa, Iowa City, IA; 1:20); anti-PROX1 (Covance; 1:1000); anti-CALRETININ (Millipore Bioscience Research Reagents; 1:200); anti-GLUTAMINE SYNTHETASE (Millipore Bioscience Research Reagents; 1:1000); anti-CALBINDIN (Millipore Bioscience Research Reagents; 1:1000); anti-CHX10 (Exalpha Biologicals; 1:1000), DAPI stain (Sigma Chemical; 1:1000). Secondary antibodies used were directly conjugated to Alexa Fluor 488 or Alexa Fluor 594 (Invitrogen/Invitrogen) or biotinylated (Jackson Immunologicals) and sequentially labeled with streptavidin Alexa 488 or 594 (Invitrogen).
In situ hybridization on cryosections was performed as described (Brown et al., 1998) using Math5/Atoh7 (Brown et al., 1998), Hes5 (a gift from Kenny Campbell, Cincinnati Children's Research Foundation, Cincinnati, OH), Nrl and Nr2e3 (gifts from Alan Mears, Ottawa Hospital Research Institute, Ottawa, ON, Canada) cDNA plasmids as templates for digoxygenin-labeled antisense riboprobes. For S-phase analyses, BrdU (Sigma Chemical) was injected intraperitoneally as described in the study by Mastick and Andrews (2001) and animals were killed 1.5 h later for tissue processing that included 2N hydrochloric acid treatment of sections before antibody staining. To visualize LacZ activity in Notch3LacZ embryos, X-gal staining of cryosections followed (Brown et al., 2001). Standard histology on paraffin embedded adult eyes was also performed. All microscopic imaging was performed on a Zeiss fluorescent microscope with Zeiss camera and Apotome deconvolution device. Images were processed using Axiovision (v5.0) and Adobe Photoshop software (v7.0) and electronically adjusted for brightness, contrast and pseudocoloring.
Cell counting
Labeled tissue sections were imaged and cells quantified using Axiovision (v5.0) software. Three or more animals were analyzed per genotype and age, with ≥2 sections from each control or mutant littermate animal. Retinal sections were judged to be of equivalent depth in the eye by anatomical landmarks in the head and other eye tissues, with only the nasal side of the retina imaged for consistency of mutant phenotypes. The cell autonomy of HES1+, BrdU-labeled, cPARP+, POU4F2+, RXRγ+, CRX+, THRB2+ or NR2E3+ cells at E16.5 or P3 was determined within a 200× field of each section containing the distal retina. The percentage of marker+/GFP+; marker+/DAPI or GFP+/DAPI cells ± SEM was determined, with GFP reporting either IRES-GFP or Z/EG expression. The percentages of HES1+/DAPI; POU4F2+/DAPI and CRX+/DAPI nuclei were determined in 200× fields within in the central retinas of E16.5 wild-type and Notch3LacZ/LacZ embryos. The percentage of CRX+/DAPI nuclei was also determined in wild-type and Hes1−/− retinal sections. A two-tailed Student's t test and Welch post hoc test were used to determine p values (Instat Software, v3.0).
Results
Conditional deletion of Rbpj causes microphthalmia and retinal mispatterning
To define the roles of Notch signaling in the mammalian retina more fully, we conditionally deleted the common pathway component, Rbpj, using a conditionally mutant allele (RbpjCKO) (Han et al., 2002) and the α-Cre retinal driver, which initiates Cre and IRES-GFP expression in distal optic cup at E10.5 (Fig. 1A,B) (Marquardt et al., 2001). In our experiments, control animals were littermates of the genotype RbpjCKO/CKO (lacking the Cre transgene), or α-Cre;RbpjCKO/+;Z/EG, which had no histologic or molecular marker abnormalities (E13.5 to P21; n ≥ 6 animals per age). Adult α-Cre;RbpjCKO/CKO eyes were microphthalmic (Fig. 1C–F; supplemental Fig. 1, available at www.jneurosci.org as supplemental material), with severely mispatterned distal retinas that contained large rosettes (Fig. 1F, boxed area; H,J, arrows). These areas of mispatterning were also apparent in Rbpj retinal mutants at E16.5 (Fig. 2B,J), and were more severe than those of Notch1 retinal mutants (Jadhav et al., 2006b; Yaron et al., 2006).
The assessment of cell autonomous gene function has been an instrumental tool in the fruit fly eye for deciphering multiple roles for Notch signaling (for review, see Baker, 2000, 2001; Lai, 2002). Because this pathway is inherently more complex in the mouse retina, we wished to approach the genetic rigor of the Drosophila experiments. Moreover, many Cre transgenes (including α-Cre) are mosaically expressed, thereby complicating phenotypic analyses. To address these issues, we integrated the Z/EG transgene in our mouse breeding scheme, to mark and follow Rbpj−/− retinal cells (Novak et al., 2000). In retinal progenitor cells (RPCs) with α-Cre activity, GFP expression was permanently activated by removal of a flox-stop cassette within the Z/EG transgene, simultaneous with Cre-mediated deletion of Rbpj (Fig. 1A) (Novak et al., 2000; Han et al., 2002). Figure 1, I and J, show the distribution of α-Cre lineage (GFP+ cells), within the distal retina of adult control and Rbpj mutant eyes. Normally, α-Cre+ RPCs produce all seven cell classes that span across the laminated retina (Figs. 3⇓–5; supplemental Fig. 2, available at www.jneurosci.org as supplemental material). However, all distal optic cup cells do not express GFP, due to mosaic transgene expression (Marquardt et al., 2001; Yaron et al., 2006; Riesenberg et al., 2009). Interestingly, when α-Cre cells (GFP+) were mutant for Rbpj, they were inappropriately separated from Rbpj+/+GFP-negative (neg; wild-type) cells, such that wild-type cells largely resided in the retinal rosettes (Fig. 1J, arrows), and Rbpj−/−GFP+ cells had an abnormal, rounded morphology.
Rbpj promotes RPC proliferation and suppresses differentiation
To understand when and how the severely mispatterned retinas of Rbpj mutants arose, we surveyed embryonic eyes from E13.5–E16.5. By E16.5, removal of Rbpj had already disrupted retinal lamination and caused excess differentiated TUBB3+ (βIII Tubulin+) neurons, which surrounded the rosetted areas (Fig. 2A,B) (data not shown). In younger, E13.5 α-Cre;RbpjCKO/CKO eyes, TUBB3+ neurons were already disorganized and mispositioned at the outer retina (Fig. 2, compare C, D). Mitotically active RPCs were also examined by BrdU pulse labeling and CCND1/CyclinD1 expression. In E16.5 control eyes, RPCs expressed these markers broadly (Fig. 2I), but in Rbpj mutants there was an eightfold loss of S-phase Rbpj−/−GFP+ cells, with the remaining proliferative RPCs mislocalized to the periphery of forming rosettes (Fig. 2J,T) (control 35 ± 1.5%, n = 6758 GFP+ cells from 3 eyes; Rbpj mutants, 4.2 ± 0.2%, n = 4345 GFP+ cells from 3 eyes; p < 0.0001). Earlier at E13.5, we also observed fewer Rbpj−/−GFP+BrdU+ (data not shown) or Rbpj−/−GFP+CCND1+ RPCs, compared to littermate control eyes (Fig. 2K,L). This simultaneous loss of proliferating RPCs and increased neuronal differentiation in the absence of Rbpj were consistent with the phenotypes of Notch1 conditional mutants (Jadhav et al., 2006b; Yaron et al., 2006).
Next we asked how the removal of Rbpj affected the expression of Hes1 and Hes5, two known transcriptional targets of the NOTCH-ICD/MAML/RBPJ complex (Kageyama and Ohtsuka, 1999; Kopan, 2002). Normally, many E16.5 GFP+ RPCs coexpress HES1 (Fig. 2E,S), but in Rbpj mutants there was an 11-fold loss of GFP+HES1+ cells (Fig. 2F,S) (control, 59 ± 0.3%, n = 4586 GFP+ cells from 3 animals; Rbpj mutants, 5 ± 0.2%, n = 3401 GFP+ cells from 3 mutants; p < 0.0001). Moreover, an Rbpj-dependent reduction in HES1+ cells was already evident by E13.5 (Fig. 2G,H). We also examined Hes5 mRNA expression, and found it downregulated in E12.5-E16.5 α-Cre;RbpjCKO/CKO peripheral retinas (Fig. 2O,P) (data not shown).
We then tested whether Rbpj mutant cells undergo apoptosis, by double labeling E13.5-P3 retinal sections with anti-GFP and anti-cPARP. Apoptotic cells are normally very rare in E13.5-E16.5 control retinas (Fig. 2M, arrows). Although there was no significant difference in cPARP+ cells between E13.5 control and Rbpj conditional mutant eyes (n = 3 animals/per age and genotype, data not shown), by E16.5 apoptosis was cell autonomously increased in the Rbpj−/−GFP+ population (Fig. 2N, arrows) (controls, 0.2 ± 0.04%, n = 4970 GFP+ cells from 3 animals; Rbpj−/−, 0.9 ± 0.1%, n = 3311 GFP+ cells from 3 animals; p = 0.006). At P3, excess cPARP+GFP+ cells were still obvious in α-Cre;RbpjCKO/CKO eyes (data not shown). We conclude that without Rbpj function, embryonic RPCs autonomously downregulate Hes1 and Hes5, prematurely exit the cell cycle and differentiate. These defects are accompanied by disorganization of the developing retinal architecture and cell autonomous death of a portion of the α-Cre;Rbpj−/− retinal lineage. Reduced RPC proliferation, increased differentiation, and to a lesser extent increased apoptosis, all contribute to a significant reduction in the GFP+Rbpj mutant lineage by E16.5 (Fig. 6A).
Rbpj cell autonomously blocks RGC formation
Because embryonic retinal neurogenesis is derepressed in the absence of Rbpj, we were interested to learn how RGC would be affected, since this cell class differentiates first. Therefore, we tested the expression of two genes critical for RGCs formation, Atoh7/Math5 and Pou4f2/Brn3b (Gan et al., 1999; Brown et al., 2001; Wang et al., 2001). The bHLH gene Atoh7 is required for RGC genesis, presumably because it activates Pou4f2 and additional RGC factors, although paradoxically terminally mitotic Atoh7+ RPCs give rise to all seven retinal cell fates (Brown et al., 2001; Hutcheson and Vetter, 2001; Liu et al., 2001; Wang et al., 2001; Yang et al., 2003; Brzezinski, 2005). In E13.5 Rbpj mutants, Atoh7 expression was upregulated (Fig. 3, compare A, B), along with an analogous increase in Pou4f2+ RGCs (Fig. 3E,F) (data not shown). At E16.5, the Atoh7 expression pattern was extremely disrupted in Rbpj conditional mutants, with distal-most RPCs displaying an intense area of Atoh7 expression (Fig. 3C,D). Derepression of Atoh7 in Rbpj mutants correlated with the loss of Hes1 (Fig. 2F,H), which normally represses Atoh7 activation (Brown et al., 1998; Takatsuka et al., 2004; Lee et al., 2005).
POU4F2 encodes a POU-domain transcription factor that is expressed by the majority of differentiating RGCs (Erkman et al., 1996; Gan et al., 1996). Without Rbpj function, POU4F2+ cells were moderately expanded at E13.5 and increased by 1.5-fold at E16.5 (Fig. 3E–K, arrows in J) (control, 30.3 ± 0.9%, n = 1060 GFP+ cells from 3 animals; Rbpj−/−, 46.4 ± 2.7%, n = 828 GFP+ cells from 3 animals; p < 0.001). Like TUBB3+ neurons, ectopic POU4F2+ RGCs were excluded from forming rosettes (Fig. 3H, arrow). While some Rbpj mutant RGCs were mispositioned at the outer retina (Fig. 3F, arrows), others correctly migrated to the ganglion cell layer (gcl) (Fig. 3G–J), implying that the initiation of lamination does not require Rbpj activity. However, the subsequent mispatterning of α-Cre;RbpjCKO/CKO retinas ultimately affected the gcl, since it was bent around the forming rosettes in mutant eyes (Fig. 3H,J). Despite the formation of ectopic embryonic RGCs from E13.5-E16.5, very few POU4F2+ RGCs were found in P21 α-Cre;RbpjCKO/CKO eyes, which had thinner optic nerves (supplemental Fig. 1, available at www.jneurosci.org as supplemental material). The loss of RGCs was correlative with increased apoptosis of Rbpj−/− GFP+ cells (Fig. 2M,N).
Rbpj suppresses cone photoreceptor fates autonomously
Next, we wished to understand to what extent Rbpj regulates cone photoreceptor fates. The orphan retinoic acid receptor RXRγ and its heterodimeric partner thyroid hormone receptor, THRB/TRβ2, are two of the earliest markers of cone photoreceptors (Hoover et al., 1998; Mori et al., 2001; Roberts et al., 2005; Ng et al., 2009). RXRγ is also expressed by prenatal RGCs, but RGCs and cones are easily distinguishable by their locations on opposite sides of the optic cup and distinct morphologies (Fig. 4A,I). In E13.5 Rbpj mutants, the RGC and cone populations were properly located at the inner and outer retina respectively, with some disorganization distally (data not shown). At E16.5, we quantified the percentage of GFP+RXRγ+ cells in control and Rbpj conditional mutants, and found a threefold, autonomous increase in RXRγ+ Rbpj−/− cells (Figs. 3L, 4C,D) (controls, 25 ± 1.1%, n = 6381 GFP+ cells from 3 animals; Rbpj−/−, 77 ± 3.8%, n = 4261 GFP+ cells from 3 animals; p < 0.0001). This represented a 3.2-fold increase in embryonic RGCs (controls 13.5% versus Rbpj−/− 43.5%) and 2.9-fold increase of embryonic cones in the outer retina (controls 11.5% versus Rbpj−/− 33.5%). Importantly, the RXRγ+ cones were always abnormally clustered in the center of forming rosettes within α-Cre;RbpjCKO/CKO;Z/EG eyes (Fig. 4C,D). To verify that mispositioned RGCs had not skewed our RXRγ+ cone quantification, we also assayed E16.5 THRB+ cones, which displayed a fourfold increase in the absence of Rbpj (Figs. 4E,F, 6B) (controls 13 ± 0.7%, n = 2234 GFP+ cells from 3 animals; Rbpj−/− 52 ± 3%, n = 1320 GFP+ cells from 3 animals).
Cones and rods are thought to originate from CRX+ bipotential precursor cells (Furukawa et al., 1997; Chen et al., 2002). The decision of a CRX+ precursor to adopt either a cone or rod fate is partly regulated by the rod genes Nrl and Nr2e3 (Swaroop et al., 1992; Liu et al., 1996; Chen et al., 1999; Kobayashi et al., 1999; Mears et al., 2001). In the distal optic cup of E13.5 Rbpj conditional mutants, CRX+GFP+ cells (Fig. 4A,B) were disorganized and expanded, just like RXRγ+ cones. The percentage of CRX+GFP+ in E16.5 control and Rbpj mutant retinas was also determined (Figs. 5A–D, 6B) (data not shown). Here, CRX+ postmitotic photoreceptors were increased by 2.1-fold, autonomously within the Rbpj−/− population (Figs. 5A–D, 6B) (controls, 22.3 ± 1.1%, n = 5616 GFP+ cells from 3 eyes; Rbpj−/−, 47.3 ± 1.1%, n = 5278 GFP+ cells from 3 eyes). We also observed this same outcome for the marker OTX2 (data not shown), which acts upstream of CRX during retinal development (Nishida et al., 2003). Therefore, we assume that Rbpj is genetically required for cone genesis at the level of Otx2/Crx gene activation. To understand how the loss of Rbpj affected mature cone photoreceptors, S OPSIN (blue) and M/L OPSIN (red/green) expression were examined in adult eyes (Fig. 4G–J) (Zhu and Craft, 2000; Zhu et al., 2006). Both types of cones were present in the rosettes of Rbpj mutants but, we were unable to score their cell autonomy, due to severe mispatterning and aberrant outer segment morphologies (Fig. 4H,J). However, in Rbpj retinal mutants M/L OPSIN+ cones were more prevalent than S OPSIN+ cones (Fig. 4I,J, arrows), which differed from Notch1 conditional mutants, where the S cones outnumbered M/L cones (Jadhav et al., 2006b). Overall, we conclude that Rbpj normally acts cell autonomously within embryonic RPCs to block cone photoreceptor formation.
Rod photoreceptor development in Rbpj conditional mutants
At E16.5, cone differentiation has peaked within the CRX+ retinal cell population, while the apex of rod genesis is still days away. Thus, at this age the CRX population might be expected to contain more cones than rods, but because rods are the largest retinal cell class in rodents, they already outnumber cones (Cepko et al., 1996; Rapaport et al., 2004). The expanded CRX-expressing population of Rbpj conditional mutants prompted us to ask whether rod development was also affected, for which we foresaw several possible mechanisms. First, the removal of Rbpj autonomously depleted the RPC pool (Figs. 2S,T, 6A), resulting in fewer RPCs to adopt a rod fate. Complicating this idea is the potential for the remaining RPC population to autonomously require Rbpj to block rod development, resulting in most (or all) of the Rbpj−/− RPCs becoming rods, at the expense of bipolar and Müller glial fates. To explore this idea further, the mRNA expression patterns of Nrl and Nr2e3, two transcription factors critical for rod differentiation, were examined (Swaroop et al., 1992; Liu et al., 1996; Chen et al., 1999; Kobayashi et al., 1999; Mears et al., 2001). At P0, Nrl and Nr2e3 are normally present in postmitotic rods within the outer nuclear layer (Fig. 5E,G). In Rbpj conditional mutants both genes were expressed by a large proportion of the rosetted cells (Fig. 5F,H). At E13.5 and E16.5 neither Nrl nor Nr2e3 mRNAs were precociously expressed in the absence of Rbpj (data not shown). Thus, although Nrl+ and Nr2e3+ rods were mispositioned within the rosettes of Rbpj mutants, it was not obvious whether the number of rods produced was abnormal.
The other retinal cell classes were also surveyed in the α-Cre lineage of adult Rbpj conditional mutants (supplemental Fig. 2, available at www.jneurosci.org as supplemental material). There were no noticeable alterations in amacrines, using the markers CALBINDIN (marks horizontals and a subset of amacrines; supplemental Fig. 2A–C, available at www.jneurosci.org as supplemental material) and CALRETININ (marks A2 amacrines; supplemental Fig. 2J–L, available at www.jneurosci.org as supplemental material). Calbindin+ horizontal neurons were also unaffected in Rbpj conditionally mutant eyes (right arrows in supplemental Fig. 2C,available at www.jneurosci.org as supplemental material). In addition, CHX10+ bipolars (supplemental Fig. 2D–F, available at www.jneurosci.org as supplemental material) and SOX 9+ or CRALBP+ Müller glia (supplemental Fig. 2G–I, available at www.jneurosci.org as supplemental material) (data not shown) were compared in control versus Rbpj conditionally mutant eyes. Here, we observed fewer CHX10+GFP+ or SOX9+GFP+ cells within the Rbpj mutant α-Cre lineage (supplemental Fig. 2D–I, available at www.jneurosci.org as supplemental material). While this supported the idea that postnatal RPCs erroneously adopt an rod fate, it was also consistent with a role for Rbpj in autonomously promoting Müller glial fates, like Notch1 regulation of this cell class (Furukawa et al., 2000; Bernardos et al., 2005; Roesch et al., 2008). Therefore, we concluded it was not possible to discriminate among these possibilities further here, since the autonomy of potential Rbpj mutant bipolar and Müller glia phenotypes must be examined by conditional deletion of Rbpj specifically after RGC and cone genesis is completed.
Wild-type cells compensate for shifts in Rbpj mutant photoreceptor cell populations
Yet another potential mechanism by which rod fates could be affected in Rbpj−/− retinas might occur if postmitotic cones non-autonomously signal RPCs to adopt a rod fate. Intriguingly, we noticed that in E16.5 Rbpj conditional mutants, CRX+GFP-neg cells usually resided next to CRX+GFP+ cells, particularly around the periphery of forming rosettes (Fig. 5D inset, arrowheads). However, the E16.5 CRX+GFP-neg population was not significantly increased (Fig. 6C), and expansion of CRX+ retinal cells was solely attributable to the cell autonomous increase in THRB+ cones (Fig. 6B–D). Nevertheless, Rbpj−/− cones may non-autonomously influence the fate that later RPCs adopt, for which the outcome is not immediately evident. To test this possibility, we quantified nascent THRB+ cones and NR2E3+ rods in early postnatal control and Rbpj conditionally mutant eyes (Figs. 5I–L, 6E–H) (data not shown). At P3, the α-Cre lineage normally encompasses 80% of the distal retina (Fig. 6E), of which 30% are NR2E3+ rods and 4% are THRB+ cones (Fig. 6F). In contrast, the Rbpj mutant lineage was reduced eightfold (Fig. 6E), but with a sixfold increase in THRB+ cones (control 4 ± 0.4%, n = 15,441 GFP+ cells from 3 animals; Rbpj−/− 25 ± 4%, n = 818 GFP+ cells from 3 animals; p = 0.01). Remarkably, although the mutant lineage was much smaller (Fig. 6E), it contained the correct proportion of rods (Fig. 6F) (control, 29.3 ± 0.2%, n = 18,459 GFP+ cells from 3 animals; Rbpj−/−, 32 ± 0.3%, n = 15,560 GFP+ cells from 3 animals; p = 0.43). Furthermore, the total population of rods and cones was the same between controls and Rbpj conditional mutant eyes (Fig. 6H), suggesting that the percentages of cones and rods were adjusted within the wild-type population, to correct for abnormal numbers of each cell type within the α Cre;Rbpj−/− lineage. Consistent with this hypothesis, we observed that NR2E3+ rods were significantly increased, and THRB+ cones significantly reduced, only in the GFP-neg population (Fig. 6G). We conclude that wild-type RPCs compensated for two simultaneous abnormalities in the α-Cre;Rbpj−/− lineage: (1) cell autonomous overproduction of cones that began around E13.5, and (2) dramatic loss of the postnatal α-Cre lineage, which contained the correct ratio, but not the proper number of GFP+ rods (Fig. 6E–H).
Finally, differentiated rods were examined by comparing the expression of the terminal differentiation marker RHODOPSIN at P10 and P21, in both α-Cre;RbpjCKO/+;Z/EG and α-Cre;RbpjCKO/CKO;Z/EG eyes (Fig. 5M–P) (data not shown). Although control retinas had an abundance of RHODOPSIN+GFP+ rods (Fig. 5O), only rare RHODOPSIN+GFP+ rods were observable in Rbpj conditional mutants (Fig. 5P, arrow), indicating that most of the rosettes were almost entirely comprised of wild-type rods. Importantly, the predominantly rod-filled rosettes of Rbpj mutants (Fig. 5P) drastically differed from Notch1 retinal mutants, whose rosettes contained mostly cones (Yaron et al., 2006), implying that Rbpj might act either outside of the Notch pathway, or complexes with a different activated receptor, to control photoreceptor cell population dynamics.
Broad removal of Rbpj with Chx10-Cre causes analogous embryonic retinal phenotypes
To independently verify each Rbpj retinal phenotype, we also deleted Rbpj with a different retinal Cre driver, Chx10-Cre, which is expressed by the vast majority of embryonic RPCs, albeit with some mosaicism (Rowan and Cepko, 2004; Jadhav et al., 2006a). The Z/EG transgene was also included in these experiments, to demonstrate the cell autonomy of each retinal cell class. At E13.5 and E16.5, we observed identical shifts in the expression of HES1 (supplemental Fig. 3A,B), POU4F2 (supplemental Fig. 3C,D), CRX (supplemental Fig. 3E,F), and RXRγ (supplemental Fig. 3G,H, all available at www.jneurosci.org as supplemental material) in Rbpj−/−GFP+ cells. Thus, HES1+ RPCs decreased cell autonomously while RGC and cone differentiation also increased cell autonomously. These shifts were found as early as E13.5 (data not shown). We also tested NR2E3 expression at P3, and found that the rod precursor cells were overwhelmingly GFP-neg, in CHX10-Cre; RbpjCKO/CKO;Z/EG eyes (supplemental Fig. 3I,J, available at www.jneurosci.org as supplemental material).
Notch3 and Hes1 suppress RGC, but not cone development
Notch1 and Rbpj each suppress cone formation, yet their mutant phenotypes suggest they perform distinct functions during RGC genesis (Jadhav et al., 2006b; Yaron et al., 2006). Therefore, we hypothesized that distinct combinations of Notch pathway genes regulate each retinal cell type. To address this idea, we compared the functions of Rbpj to those of the Notch3 receptor and downstream effector, Hes1. First, we better defined the expression pattern of Notch3, which was already known to be present in the prenatal rodent retina (Lindsell et al., 1996), and then searched for RGC or photoreceptor phenotypes in Notch3 mutants. Here, we took advantage of a Notch3 gene trap mutant allele, in which an in-frame βgal-neo insertion into the Notch3 coding region causes a 99% loss of mRNA in homozygotes (Leighton et al., 2001; Mitchell et al., 2001). Moreover, the resulting Notch3-βgal fusion protein is localized to the endoplasmic reticulum and secretory vesicles of Notch3-expressing cells (Fig. 7A–I) (Mitchell et al., 2001). We observed Notch3LacZ in E11.5, E13.5 and E16.5 retinal progenitor cells and nascent RGCs, although expression in the latter cell class was downregulated at E16.5 (Fig. 7A–C). The Notch3LacZ expression pattern in E11.5–E13.5 RPCs is consistent with the published mRNA expression pattern (Lindsell et al., 1996), although additional expression in nascent RGCs presumably reflects βgal perdurance. Indeed, there was extensive Notch3LacZ expression within POU4F2+ RGCs at E13.5 (Fig. 7E), although βgal+HES1+ RPCs (Fig. 7D) and a few βgal+CRX+ photoreceptor precursors (Fig. 7F) were also evident. Interestingly, by E16.5 only the youngest RGC cell bodies at the GCL-neuroblastic boundary expressed Notch3LacZ (Fig. 7H, arrows), along with some residual βgal expression in RGC axons (Fig. 7C,H). At this older age, there were numerous βgal+HES1+ progenitor cells (Fig. 7G), but only a few, scattered βgal+CRX+ photoreceptors (Fig. 7I). By comparing E16.5 wild-type and Notch3 mutant retinas, we found a 1.3-fold increase in POU4F2+ RGCs (Fig. 7K) (control 16.8 ± 0.9%, n = 1930 cells from 3 animals; Notch3−/− 21.2 ± 0.6%, n = 1405 cells from 3 animals; p = 0.01), which correlated with the 1.1-fold decrease in HES1+ progenitor cells (Fig. 7J) (control 48.4 ± 1.1%, n = 1785 cells from 3 animals; Notch3−/− 43.9 ± 1%, n = 1152 cells from 3 animals; p < 0.01). Interestingly, the loss of Notch3 did not affect CRX+ photoreceptor precursors (Fig. 7L) (control 25.2 ± 0.7%, n = 1756 cells from 3 animals; Notch3−/− 23.3 ± 1.1%, n = 1371 cells from 3 animals; p = 0.11). We conclude that Notch3 normally suppresses RGC, but not cone formation.
Next, we surveyed RGC and early photoreceptor development in Hes1 germline mutants (Tomita et al., 1996). Hes1 was previously shown to promote RPC proliferation, repress RGC, horizontal and rod neurogenesis and promote Müller glia differentiation (Tomita et al., 1996; Furukawa et al., 2000). In addition, Hes1−/− retinas are mispatterned by E15.5, with retinal rosettes that contain too many rod photoreceptors (Tomita et al., 1996; Takatsuka et al., 2004). Somewhat surprisingly, cone development has not been examined in Hes1 mutants. Therefore, we assayed the prenatal RGC and photoreceptor phenotypes of Hes1 mutants. Previous reports showed both precocious and an expanded domain of TUBB3+ neurons in Hes1−/− embryonic retinas, during RGC genesis (Takatsuka et al., 2004; Lee et al., 2005). We revisited this by comparing the RGC markers POU4F2, ISL1 and RXRγ in E13.5 and E15.5 Hes1 control and mutant eyes (Fig. 8A–D) (data not shown). Indeed, all three RGC markers were greatly expanded in the absence of Hes1 (Fig. 8B,D,F,J). In some embryos, we also found RGCs inappropriately located at the outer optic cup (Fig. 8B, arrows), like Rbpj conditional mutants (Figs. 2D, 3F). Unexpectedly, the increase in RGCs present in Hes1 mutants was accompanied by a loss of outer RXRγ+ cones (Fig. 8F, arrows). We also found a 1.5-fold reduction in E13.5 CRX+ cells in Hes1 mutants, further confirming the loss of cones (control, 11.9 ± 0.007%, n = 5300 DAPI+ cells from 3 embryo eyes; Hes1−/−, 8.2 ± 0.01%, n = 5712 DAPI+ cells from 3 embryo eyes; p = 0.02). Interestingly, by E15.5, the proportion of CRX+ cells in Hes1−/− eyes appeared to rebound to that of controls, although this is likely due to the precocious onset of rod development as previously reported (Tomita et al., 1996). Unfortunately Hes1 germline mutant embryos could not be recovered beyond E15. We conclude that RGC neurogenesis is normally suppressed by Notch3, Rbpj and Hes1, while cone formation is regulated by Notch1-Rbpj signaling through a different downstream effector, instead of Hes1.
Discussion
Although the Notch pathway clearly has multiple roles in the vertebrate retina, the requirements for each gene remain largely unknown, and the cell autonomy of those gene functions already investigated has not been very well determined. Nevertheless, loss of Delta-Notch signaling results in excess embryonic RGCs and cone photoreceptors; while overexpression of Delta1, activated Notch or Hes1 prolongs the mitotic activity of RPCs (Austin et al., 1995; Dorsky et al., 1995; Henrique et al., 1995; Tomita et al., 1996; Ahmad et al., 1997; Dorsky et al., 1997; Schneider et al., 2001; Silva et al., 2003; Takatsuka et al., 2004; Jadhav et al., 2006b; Yaron et al., 2006). In this paper, we demonstrate that the Notch signal integrator Rbpj autonomously promotes progenitor cell growth, and suppresses RGC and cone photoreceptor development. In the embryonic retina, both Rbpj−/− RGCs and cones were overproduced and mispatterned, including mislocalization of RGCs at the outer retina, and cones in the center of forming rosettes. But, only the ectopic cones fully persisted into adulthood in Rbpj conditionally mutant eyes. An increase in apoptotic Rbpj−/− cells occurred where mutant RGC cell bodies reside, during the period of normal RGC axon outgrowth and connectivity, which correlated with the loss of RGCs and thinner optic nerves in Rbpj−/− adult eyes. Although Notch signaling can regulate cell survival (Oishi et al., 2004), we suggest the additional possibility that Rbpj−/− RGCs die potentially during the normal corrective process of RGC overproduction. Alternatively, the ectopic RGCs may produce misrouted axons that fail to reach the optic nerve.
Our interest in RGC cell fate specification drew us to the mechanism of how Notch signaling controls the timing of RGC differentiation. Notch regulation of RGC neurogenesis is less complex, since these cells initiate differentiation in the absence of extrinsic signals from other neurons (Waid and McLoon, 1998; Silva et al., 2003; Dakubo and Wallace, 2004; Liu et al., 2006). Based on previous work in the vertebrate and Drosophila eye, the prevailing model holds that an ‘equivalence group’ of mitotic RPCs coexpressing DELTA and NOTCH, subsequently undergo lateral inhibition to produce one or more postmitotic cells, which downregulate Notch/Rbpj/Hes activity and upregulate bHLH proneural expression thereby controlling the sequential onset of each retinal neuron class (Cepko, 1999; Kageyama and Ohtsuka, 1999; Kageyama et al., 2008). However, Notch1, Notch3, Rbpj, and Hes1 mutant mice exhibited separate and overlapping retinal phenotypes, provoking the question of which combinations of ligands, receptors and downstream effectors regulate RGC versus cone formation. Here, we tested the embryonic roles of Rbpj, which integrates input from all combinations of Notch ligand and receptors. We found that Rbpj represses RGC fates, consistent with the function of Hes1, as well as a Notch-mediated blockade of RGC formation in other vertebrate eyes. But, Notch1 conditional mutants had reduced numbers of RGC marker+ cells (Jadhav et al., 2006b; Yaron et al., 2006), and although Notch1 and Rbpj each block cone fates, neither Notch3 nor Hes1 participate in this process. Therefore, we delineated two branch points in the Notch pathway through which Rbpj regulates RGC and cone formation simultaneously, namely variable receptor input, and/or the activation of different downstream effectors (Fig. 9). The 1.3-fold increase in POU4F2+ RGCs in Notch3 mutants is similar to the 1.5-fold RGC increase that occurred without Rbpj. However, because conditional deletion of Rbpj underestimates its full requirements during retinal cell type specification, it remains plausible that Notch1 and Notch3 act synergistically, or cross-regulate one another, during RGC formation. Thus, the total requirement for Notch receptors during RGC neurogenesis should become evident through simultaneous removal of both receptors. We are optimistic that future Notch1;Notch3 mutant analyses will finally unify the Notch RGC phenotypes among different vertebrate model organisms.
Because Rbpj can activate either Hes1 or Hes5, Hes5 is the obvious candidate to respond to Notch1-Rbpj during cone photoreceptor genesis. This raises interesting questions about the spatial and temporal overlap of HES1 and HES5 expression and resulting retinal lineages, about the influences of other signaling pathways, such as shh, in modulating Hes1 or Hes5 gene activity (Wall et al., 2009), and whether these transcriptional repressors act separately, or with partially overlapping cell autonomous functions. Because cone differentiation was reduced in Hes1 mutants, Hes1 may normally repress a negative regulator of cone fates. Interestingly, Hes1 and Hes5 genetically repress one another in particular contexts (Hatakeyama et al., 2004). Alternatively, Rbpj may regulate the timing of cone precursor cell formation directly, or act cell autonomously through another transcriptional target (Iso et al., 2003). Future experiments that establish the cell autonomy of Deltalike1, Notch3, Hes1 and Hes5 gene functions will distinguish among these possibilities. Interestingly, yet another level of Notch signal complexity is likely to exist, since the requirement for Deltalike1 and Deltalike4 ligands was recently suggested for RGC development (Rocha et al., 2009) (Fig. 9).
In comparing photoreceptor development between marked Rbpj−/− and control retinal lineages, we obtained clear evidence that retinal cells adjust their production of rods and cones, when confronted with population shifts in a neighboring lineage. Therefore, the developing retina monitors both the overall production of photoreceptors to non-photoreceptors, and the correct proportion of rods to cones. Previous in vitro studies with mixed-age retinal cultures showed that rod precursors can induce nearby embryonic RPCs to differentiate as rods (Wantanabe and Raff, 1990; Reh, 1992). In mixed pellet cultures of embryonic and postnatal retinal cells, the embryonic RPCs had a higher propensity to differentiate as rods (Wantanabe and Raff, 1990). In the second study, embryonic RPCs were introduced to retinal monolayers containing photoreceptor-filled rosettes. Here too, the embryonic RPCs were induced to adopt the rod fate, especially when situated next to rod-containing rosettes (Reh, 1992). Exogenous growth factor addition could influence the rate of rod production, but not the fate chosen by RPCs. In addition, embryonic RPCs could not be induced to become rods when likewise cultured with a monolayer of cortical cells. Together, these studies suggested that a local cue, emanating from closely situated rod precursors, directs prenatal RPCs to adopt the rod fate.
Our discovery of non-autonomous compensation by wild-type retinal cells for the shifts in rod and cone numbers in Rbpj conditionally mutant retinas, raises the obvious question of whether Rbpj regulates some, or all, aspects photoreceptor homeostasis during development. RBPJ, either within or outside of the context of Notch signaling, for example in a complex with PTF1A (Masui et al., 2007; Hori et al., 2008), could non-autonomously influence the choice of bipotential CRX+ cells through a mechanism that maintains the balance of rod to non-rods, cones to non-cones and/or total photoreceptor to non-photoreceptor populations. Theoretically, such a signal might be transduced from cell-to-cell in a subsequent round of Notch signaling, or use other signaling pathways. Importantly, Notch regulates tissue homeostasis in different organs of the body, although it does so by controlling a variety of physiologic processes (Lin and Kopan, 2003; Lewis, 2008; Okuyama et al., 2008; Robinson, 2008; Brabletz et al., 2009). In addition, Notch is a key regulator of normal tissue growth, and Notch activity is inappropriately upregulated during tumor cell overgrowth (for review, see Kopan, 2002; Gridley, 2003; Lasky and Wu, 2005; Sjölund et al., 2005; Louvi and Artavanis-Tsakonas, 2006). If the photoreceptor homeostasis highlighted in our Rbpj conditional mutant analysis is Notch-dependent, it might act through a different receptor, since Notch1−/− cells autonomously overproduced cones, but without an analogous appearance of rod photoreceptors within the forming rosettes (Jadhav et al., 2006a; Yaron et al., 2006). On the other hand, we observed that Rbpj−/−GFP+ cells autonomously maintained the correct ratio of rods (Fig. 6F), despite a profound loss of the α-Cre lineage. Furthermore, we found a loss of cone photoreceptors in Hes1 germline mutants, at the same age that RGC development was both precocious and expanded. This suggests that since all the retinal cells lacked Hes1 activity, at least some RPCs were shunted away from the cone fates to maintain the correct overall number of photoreceptors, perhaps because rod fates are expanded in this mutant background. At present our data implicate but do not clearly demonstrate whether Notch signaling regulates photoreceptor cell population dynamics. Alternatively, the quantification of cell autonomy for each mutant phenotype, coupled with the reduced mutant RPC pool in this mutant, may have identified an Rbpj-independent retinal process for regulating photoreceptor cell numbers. To understand the genetic hierarchy that controls this important process, future experiments will compare both the cell biological characteristics and gene profiles of the wild-type and Rbpj−/− marked cell populations, within the period cone and rod development examined here (E16-P3).
The ability of tissues to sense and regulate their overall size, and the proportion of each cell type, was first hypothesized more than two decades ago (Gurdon, 1988). Both characteristics are critical for normal development, and presumably are affected during tumor formation. These elusive homeostatic mechanisms are still intensely investigated, with multiple signaling pathways implicated as the inducers of this process (Gurdon et al., 1998, 1999; Standley et al., 2001; Piddini and Vincent, 2009). Here, we demonstrate that during a critical developmental period the mammalian retina keeps track of, and can correct, the size of its photoreceptor populations. This finding is directly relevant for embryonic stem cell or retinal progenitor cell therapies, which aim to restore reduced or missing vision (MacLaren et al., 2006; Lamba et al., 2009). Although much progress has been made in this area, several significant hurdles remain, including the ability to produce pure populations of photoreceptor precursors for reintroduction and improving their efficiency of tissue integration. It is exciting to speculate that the future identification of molecular pathways that monitor photoreceptor population dynamics will contribute beneficially toward these unresolved cell therapy issues.
Footnotes
-
This work was supported by National Institutes of Health (NIH) Grants GM55479, DK66408, and HD44056 to R.K. and NIH Grants EY13612 and EY18097 to N.L.B. We thank Tasuku Honjo and Tom Gridley for RbpjCKO mice; Ruth Ashery-Padan and Peter Gruss for alpha-Cre transgenic mice; Anand Swaroop, Tom Glaser, Cheryl Craft, and Doug Forrest for antibody reagents; Kenny Campbell and Alan Mears for cDNA plasmids; Ashley Riesenberg, Tien Le, and Kevin Conley for technical support and Valerie Wallace, Tiffany Cook, Masato Nakafuku, Kenny Campbell, Rob Hufnagel, and Tom Glaser for valuable discussion or critical evaluation of this manuscript.
- Correspondence should be addressed to Nadean L. Brown, Division of Developmental Biology, Cincinnati Children's Research Foundation, ML 7007, 3333 Burnet Avenue, Cincinnati, OH 45229. nadean.brown{at}cchmc.org