Pushing the envelope of retinal ganglion cell genesis: context dependent function of Math5 (Atoh7)

Abstract

The basic-helix-loop helix factor Math5 (Atoh7) is required for retinal ganglion cell (RGC) development. However, only 10% of Math5-expressing cells adopt the RGC fate, and most become photoreceptors. In principle, Math5 may actively bias progenitors towards RGC fate or passively confer competence to respond to instructive factors. To distinguish these mechanisms, we misexpressed Math5 in a wide population of precursors using a Crx BAC or 2.4 kb promoter, and followed cell fates with Cre recombinase. In mice, the Crx cone-rod homeobox gene and Math5 are expressed shortly after cell cycle exit, in temporally distinct, but overlapping populations of neurogenic cells that give rise to 85% and 3% of the adult retina, respectively. The Crx > Math5 transgenes did not stimulate RGC fate or alter the timing of RGC births. Likewise, retroviral Math5 overexpression in retinal explants did not bias progenitors towards the RGC fate or induce cell cycle exit. The Crx>Math5 transgene did reduce the abundance of early-born (E15.5) photoreceptors two-fold, suggesting a limited cell fate shift. Nonetheless, retinal histology was grossly normal, despite widespread persistent Math5 expression. In an RGC-deficient (Math5 knockout) environment, Crx>Math5 partially rescued RGC and optic nerve development, but the temporal envelope of RGC births was not extended. The number of early-born RGCs (before E13) remained very low, and this was correlated with axon pathfinding defects and cell death. Together, these results suggest that Math5 is not sufficient to stimulate RGC fate. Our findings highlight the robust homeostatic mechanisms, and role of pioneering neurons in RGC development.

INTRODUCTION

The vertebrate retina is a highly ordered structure composed of six major types of neurons and one type of glia. These originate from a common progenitor pool (Holt et al., 1988; Turner and Cepko, 1987; Turner et al., 1990) and include retinal ganglion cells (RGCs), rod and cone photoreceptors, amacrine, horizontal and bipolar interneurons, and Müller glia. At the onset of retinal neurogenesis, embryonic day 11 (E11) in mice, multipotent retinal progenitor cells (RPCs) begin to exit the cell cycle and differentiate.

Birthdating studies, in which nucleoside analogs are used to identify progenitors exiting the cell cycle, have defined a fixed, but overlapping, order for the generation of these major cell classes in vertebrates (Rapaport et al., 2004; Sidman, 1961; Young, 1985a). RGCs are the first to exit the cell cycle, at E11 in mice, with peak birthdates at E14 and termination by P0 (Drager, 1985). This early temporal profile overlaps significantly with those of cone, horizontal, and amacrine neurons. Rods, Müller glia and bipolar cells have characteristically later birthdates. In mice, the distribution of rod births peaks in the neonatal period, but the tails of the distribution extend across most of the histogenic period, from E12.5 to P10 (Carter-Dawson and LaVail, 1979; Swaroop et al., 2010), because rods compose ~80% of the mature retina (Jeon et al., 1998).

Heterochronic co-culture and transplantation experiments, in which early embryonic and late RPCs were cultured in unequal ratios, have suggested that fate determination is largely a cell intrinsic process (Belliveau and Cepko, 1999; Rapaport et al., 2001; Reh, 1992; Watanabe and Raff, 1990). Indeed, progenitors cultured at low density can develop into each of the major cell classes, with similar diversity and proportions as the intact retina, in the absence of environmental feedback signals (Adler and Hatlee, 1989; Cayouette et al., 2003; Reh and Kljavin, 1989). However, extrinsic signals can influence progenitor cell cycle dynamics and override fate decisions in vivo (Cepko, 1999; Ezzeddine et al., 1997; Kim et al., 2005; Yang, 2004). Collectively, these observations are consistent with a temporal, or serial, competence model for retinal development (Cepko et al., 1996; Livesey and Cepko, 2001; Reh and Cagan, 1994; Wong and Rapaport, 2009). According to this model, RPCs pass through discrete competence states over time, in which they can adopt a limited number of cell fates. Within each state, the decision to exit the cell cycle and the final histotypic choice are influenced by extrinsic signals.

Two prototypical intrinsic factors important for development of mouse RPCs into specific types of neurons are the cone-rod homeodomain (HD) factor Crx and the basic helix-loop-helix (bHLH) factor Math5 (atonal homolog Atoh7). Crx, and closely related factor Otx2, are expressed in during or shortly after the terminal cell cycle in tripotential precursors that give rise to photoreceptors and bipolar cells (Furukawa et al., 1997; Muranishi et al., 2011). Because of high degree of spatiotemporal overlap with Otx2, the precise role of Crx remains unclear. In mice, Crx expression initiates at E12.5 and is necessary for proper development of photoreceptors, and may be partially redundant with Otx2 in conferring competence for photoreceptor specification (Chen et al., 1997; Furukawa et al., 1999; Nishida et al., 2003; Sato et al., 2007). Crx works in concert with other transcription factor to regulate photoreceptor gene expression (reviewed in Hennig et al., 2008; Swaroop et al., 2010), and Crx is abundant in adult rods, cones and bipolar cells. The 5’ regulatory DNA for Crx has been extensively characterized, and a critical 2.4 kb promoter region is thought to faithfully recapitulate the endogenous Crx pattern. This segment has been used to drive Cre, lacZ, and regulators of rod photoreceptor specification in transgenic mice (Cheng et al., 2006; Furukawa et al., 2002; Koike et al., 2005; Nishida et al., 2003; Oh et al., 2007).

Like Crx, the role of Math5 in fate specification has not been fully elucidated. Math5 (Atoh7) is a single-exon gene that is transcribed by retinal progenitors in a spatiotemporal pattern that mirrors RGC births (Brown et al., 1998; Brzezinski et al., 2012; Prasov et al., 2010). This factor is transiently expressed by RPCs during or after their terminal cell cycle (Brzezinski et al., 2012; Feng et al., 2010; Kiyama et al., 2011; Skowronska-Krawczyk et al., 2009) and is required for RGC development. Math5 mutant mice have very few RGCs (<5%) and lack optic nerves (Brown et al., 2001; Wang et al., 2001). Apart from the deficiency of RGCs, all other retinal cell classes are preserved (Brown et al., 2001; Brzezinski et al., 2005). The mRNA profiles of Math5 mutant retinas are altered, with downregulation of genes associated with RGC differentiation (Mu et al., 2005). Lineage tracing experiments have established that Math5-expressing cells contribute to 3% of the adult retina, and every major cell class (Brzezinski et al., 2012; Feng et al., 2010; Yang et al., 2003). Together, these data suggest Math5 acts as an essential competence factor for RGC development. Orthologous genes in zebrafish, chicken, and frog have similar functions, lineage properties, and expression patterns (Kanekar et al., 1997; Kay et al., 2001; Liu et al., 2001; Matter-Sadzinski et al., 2001; Poggi et al., 2005), and mutations in human ATOH7 have been linked to optic nerve aplasia and retinal vascular disease (Ghiasvand et al., 2011; Khan et al., 2011; Prasov et al., 2012).

Despite these expression and phenotypic analyses, the precise role of Math5 in RGC development remains unclear. Gain-of-function studies in frog and chick, and mouse embryonic stem cells, have yielded mixed results. In these systems, Math5 biases proliferating progenitors towards RGC fates when over-expressed during early developmental stages (Brown et al., 1998; Kanekar et al., 1997; Liu et al., 2001; Moore et al., 2002; Yao et al., 2007), but promotes other cell fates when expressed during late development, or in a cross-species context (Brown et al., 1998; Moore et al., 2002). In general, the interpretation of these experiments is confounded by the tendency of proneural bHLH factors to drive cell cycle exit when overexpressed (Farah et al., 2000).

To circumvent these limitations and critically assess the role of Math5 in biasing RGC development, we generated transgenic mice that ectopically express Math5 in a large number of retinal progenitors and newly post-mitotic neurons, under control of a mouse Crx promoter fragment (Crx>Math5 Tg) or bacterial artificial chromosome (Crx>Math5 BAC), with a bicistronic Cre lineage tracer. Although endogenous Crx and Math5 genes mark overlapping populations, and appear to be co-expressed in some cells, profound overexpression of transgenic Math5 did not stimulate RGC production or alter the profile of RGC births. Instead, the number of early-born photoreceptors was reduced. Despite sustained high-level expression of Math5 in photoreceptors and bipolar cells, retinal histology and cell type distribution were grossly normal. Likewise, no substantial RGC bias was observed in retinal explants infected with a Math5-expressing retrovirus. In mutant mice, the Crx>Math5 transgenes rescued RGC development. However, because endogenous Crx expression initiates somewhat later that Math5, early-born RGCs were scarce, and some rescued ganglion cells exhibited pathfinding defects or apoptosis during development. These results suggest that Math5 action is context dependent. Our findings also warrant a re-examination of previous results obtained using conventional Crx transgenes.

MATERIALS AND METHODS

Conventional and Bacterial Artifical Chromosome (BAC) transgenes

To ectopically express Math5 in a wide population of retinal cells, we generated a conventional Crx>Math5-IRES-Cre bicistronic transgene. We assembled mouse Math5 cDNA and Cre recombinase coding sequences, separated by an internal ribosomal entry site (IRES2) and followed by a SV40 polyA signal. The mouse Crx promoter and proximal regulatory region were amplified by PCR and inserted upstream as a 2.4 kb XhoI-SalI fragment (Furukawa et al., 2002; Oh et al., 2007). A matched Crx>Cre transgene was then generated from the Crx>Math5-IRES-Cre plasmid by precise deletion of Math5 and IRES sequences using the single-strand oligonucleotide (ss oligo) recombineering method (Thomason et al., 2007), with a 70 nt antisense oligo (Suppl. Table 1) and AscI selection.

To faithfully express Math5 in the endogenous Crx pattern, we generated BAC transgenes by λRED recombineering (Lee et al., 2001). The targeting construct was assembled, with short (400 bp) 5’ and 3’ homology arms (H) flanking a Math5-IRES-Cre-FRT-amp-FRT cassette. This was equivalent to the conventional transgene, but included an FRT-amp-FRT selection cassette (Gene Bridges, Heidelberg) downstream of the SV40 polyA signal. The 5’ homology arm extends from Crx intron 1 to the exon 2 initiation (ATG) codon, while the 3’ homology arm contains sequence from Crx intron 2. A matched control (Cre-FRT-amp-FRT) was then generated by ss oligo recombineering, with the 70 nt antisense oligo (Suppl Table 1) and AscI selection.

Linearized targeting plasmids were used in parallel to target mouse BAC clone RP23-81H17 by λRED-mediated homologous recombination in strain SW105 (Warming et al., 2005) after heat induction. This 219 kb BAC contains 134 kb 5’ and 69 kb 3’ DNA flanking the Crx gene. Targeted BAC clones were selected on ampicillin and chloramphenicol plates at 30°C, and verified by junctional PCR and DNA sequencing. The amp seletion cassette was then deleted by arabinose induction of Flpe recombinase, leaving a solitary FRT site (Andrews et al., 1985). Homogeneity and integrity of the resulting clones was verified by ampicillin sensitivity, junctional PCRs, restriction mapping, and pulsed-field gel electrophoresis.

Purified circular DNA from BAC transgene constructs or linearized plasmid DNA from conventional constructs was injected into fertilized (C57BL/6J × SJL/2) F2 or R26floxGFP (JAX stock 004077 reporter strain, (Mao et al., 2001)) × B6SJLF1/J oocytes by the UM Transgenic Animal Core. Founders were identified by transgene-specific PCR genotyping (Suppl. Table 1), and lines were maintained by crossing to C57BL/6J or R26floxGFP reporter strains. We analyzed 2 founders and 2 lines for each conventional transgene (Crx>Cre Tg and Crx>Math5 Tg), and ≥3 lines for each BAC transgene. The most extensively characterized transgenes in this report were Crx>Math5 Tg 251, Crx>Cre Tg 352 control, Crx>Math5 BAC 60, and Crx>Cre BAC 764 control.

RNA analysis

Duplex and competitive triplex RT-PCRs were performed as described (Prasov et al., 2010) to compare the levels of transgene-derived and endogenous mRNAs. Total RNA was extracted from embryonic eyes (E14.5) or adult (P21) tissues of transgenic or wild-type animals using Trizol reagent (Invitrogen, Carlsbad, CA). First-strand cDNA was generated by high-fidelity reverse transcription (RT, Transcriptor™, Roche) at 50°C and used as template for PCR, with primers and conditions in Suppl. Table 1. Triplex competitive RT-PCRs used a common 6-carboxyfluorescein (FAM)-labeled forward primer in Crx exon 1, and two reverse primers (Suppl. Table 1). Dual products were closely matched for size and G+C content, and analyzed using a 3730XL capillary electrophoresis unit (Applied Biosystems, Carlsbad, CA) and Gene Marker software (SoftGenetics, State College, PA). Expression copy-number levels for Crx>Math5 BAC transgenes were determined relative to endogenous Crx values by direct analysis of peak areas (k).

Quantitative RT-PCRs were performed using custom Taqman probes and Universal Taqman Mastermix (Applied Biosystems), and were analyzed on the ABI 7600 Real Time PCR System. Critical cycle threshold levels were normalized to Gapdh internal controls, using two-fluorophore (VIC and FAM) detection. Fold activity was calculated using the ddCt method (Livak and Schmittgen, 2001) and reported relative to the Crx>Math5 BAC expression level. Expression copy-number levels for conventional transgenes were calculated by normalizing to endogenous Crx values (k × a/b), using the mean ratio determined in triplex competitive RT-PCRs (k) and the relative ratios of Math5 (a) and Crx (b) transcripts determined by qPCR. Measurements were obtained using independent RNA pools from 2–5 mice of each genotype.

Histology

For section immunostaining, eyes or embryonic heads were fixed in 2–4% paraformaldehyde (PFA) 0.1 M NaPO₄ pH 7.3 for 30–60 min at 22°C, processed through a 10–30% graded sucrose series, embedded in OCT (Tissue-Tek, Torrence, CA) and cryosectioned at 10 µm. For flatmount preparations, eyes of P1 or adult mice were removed and fixed in 4% PFA for 5 min. The optic nerves were then transected, and the retinas were teased apart from other ocular tissues, fixed in 4% PFA for 25 min. After immunostaining, retinas were incised with 6–8 radial cuts and flattened with the ganglion cell layer (GCL) facing upward.

For immunodetection, slides or whole retinas were blocked in a solution of 10% normal donkey serum (NDS), 1% bovine serum albumin (BSA) in PBTx (0.1 M NaPO₄ pH 7.3 0.5% Triton X-100) for 1–4 hrs. To reduce mouse-on-mouse background associated with mouse monoclonal primary antibodies, donkey anti-mouse IgG Fab fragments were added at 0.8 mg/mL to some blocking reactions. Primary antibodies were applied overnight at 4°C and diluted in 3% NDS 1% BSA in PBTx. Sections or retinas were then washed in PBS, incubated for 2 hrs at 22°C with Dylight-conjugated secondary antibodies and 4',6-diamidino-2-phenylindole (DAPI), and mounted in Prolong Gold Antifade (Invitrogen, Grand Island, NY). Slides were imaged using the Zeiss LSM510 Meta confocal system or an Olympus BX-51 epifluorescence microscope.

The primary antibodies were mouse anti-AP2α (1:1000, DSHB, Iowa City, IA); rabbit anti-βgal (1:5000, ICN Cappel, Aurora, OH); rat anti-β-galactosidase (1:500, (Saul et al., 2008)); rat anti-BrdU (BU1/75, 1:100, Harlan Seralab, Indianapolis, IN); mouse anti-calbindin (CB-955, 1:500, Sigma, St. Louis, MO); rabbit anti-cleaved-caspase 3 (1:100, Cell Signaling, Beverly, MA); sheep anti-Chx10 (1:250, Exalpha, Shirley, MA); mouse anti-Cre (clone 7.23, 1:300, Covance, Princeton, NJ); rabbit anti-Crx (1:1000, (Zhu and Craft, 2000)); chicken anti-GFP (1:2000, Abcam, Cambridge, MA); mouse anti-hPLAP (monoclonal 8B6, 1:250, Sigma); mouse anti-PKC (MC5, 1:100, Sigma); rabbit anti-mCar (1:500, Millipore, Billerica, MA); mouse anti-syntaxin (HPC-1, 1:1000, Sigma); rabbit anti-M-opsin (1:1000, Millipore); rabbit anti-S-opsin (1:5000, (Applebury et al., 2000)); rabbit anti-phosphohistone H3 (1:400, Upstate, Lake Placid, NY); rabbit anti-rhodamine (1:500, Invitrogen); rabbit anti-Sox9 (1:250, Millipore); rabbit anti-TuJ1 (MRB-435P, 1:2000, Covance). The Crx antibody appears to cross-react weakly with Otx2 antigen (Brzezinski et al., 2010), most likely through a shared LDYKDQ sequence in the Crx 14-residue peptide immunogen (Zhu and Craft, 2000).

For detection of BrdU (5-bromo-2-deoxyuridine) and other antigens, cryosections were fully stained with primary and secondary antibodies to the other markers. Sections were then treated with 2.4 N HCl in PBTx for 1hr at 22°C, and immunostained for BrdU. Likewise, EdU (5-ethynyl-2-deoxyuridine) was detected after immunostaining, using an azide-alkyne cycloaddition reaction (Buck et al., 2008) and with Click-iT-647 reagents (Invitrogen).

For fine histology, mice were perfused transcardially with 2% PFA and 1.25% glutaraldehyde. The eyes were removed, post-fixed overnight at 22°C, dehydrated, embedded in glycol methacrylate plastic resin (JB-4, Polysciences, Warrington, PA), sectioned at 4 µm with a Leitz 1512 rotary microtome, and stained with basic fuchsin and methylene blue. Paraffin or cryosections (5–10 µm) of eyes or optic nerves were stained with hematoxylin and eosin as described (Brown et al., 2001).

Retrograde axon labeling of RGCs

RGCs were definitively marked by retrograde axon labeling with rhodamine dextran (Brzezinski et al., 2012; Rachel et al., 2002). Eyes from adult or P1 mice were removed and immersed in Hank’s balanced salt solution containing calcium, magnesium and 1 mM glucose (HBSSG). Optic nerves were transected within 1 mm of the sclera, and lysine-fixable tetramethyl rhodamine dextran 3,000 MW powder (Molecular Probes, Eugene, OR) was applied directly to the cut site. The eyes were positioned with severed optic nerves facing downward against cubes of surgifoam (Ethicon, Somerville, NJ) saturated with 3% L-α-lysophosphatidyl choline (LPC, Sigma) and rhodamine dextran. These were sealed with 1 % agarose, and incubated en bloc in aerated HBSSG for 1 hour at 22°C. The surgifoam was then removed, and the eyes were incubated overnight in HBSSG under the same conditions. Rhodamine-labeled eyes were fixed in 4% PFA for 4 hrs at 22°C and processed for sectioning or stained as whole retina preparations. In some experiments, the signal was enhanced by indirect immunofluorescence staining with anti-rhodamine antibody.

Cre lineage and dual reporter concordance analysis

To trace the descendants of cells expressing Cre recombinase, transgenic mice were crossed to R26floxGFP or Z/AP (JAX stock 003919, (Lobe et al., 1999)) reporter strains, which activate cytoplasmic GFP (green fluorescent protein) and membrane-tethered hPLAP (human placental alkaline phosphatase), respectively, after excision of floxed upstream stop signals. Retinal sections or flatmounts were co-stained for histotypic antigen markers, hPLAP and/or GFP. Cell types were identified by characteristic laminar position, morphology, and marker co-localization. RGCs were clearly distinguished from displaced amacrines by retrograde axon labeling. To assess the heterogeneity in the level of Crx transgene (Cre) expression among progenitors giving rise to different cell types, we conducted dual reporter concordance experiments, as described (Brzezinski et al., 2012). Coexpression of GFP and hPLAP was scored in the outer nuclear (ONL), inner nuclear (INL) and ganglion cell (GCL) layers of adult triple transgenic mice (Crx>Math5 BAC or Crx>Cre BAC; Z/AP; R26floxGFP).

Quantitative assessment of RGCs in transgenic mice

Retinal ganglion cells were counted in Crx>Math5 Tg and non-transgenic littermates at P0 and P22, using Brn3a or retrograde axon labeling to mark RGCs. The fractional contribution of RGCs to the GCL (DAPI nuclei) was determined from 20 sections (400X) representing one eye from n = 2 animals for each P22 genotype, and 12–22 sections (200X) representing n = 4–6 eyes for each P0 genotype.

To evaluate transgenic rescue of RGC development in mutants, Crx>Math5 Tg or Crx>Math5 BAC mice were crossed to Math5 knockout (KO) mice (Atoh7^tm1Gla, (Brown et al., 2001)) for two or more generations. Eyes from informative embryonic, neonatal and adult littermates were immunostained as sections or flatmounts. Retinal cell death was assessed at E16.5 using activated Caspase-3 staining (Gown and Willingham, 2002). RGCs and apoptotic cells were counted in 18 sections (200X) representing n = 6 eyes of each genotype.

EdU pulse-chase and birthdating analysis

To evaluate the overlap between Crx+ and Math5+ cell populations, and compare the timing of Crx and Math5 expression, pregnant dams carrying E13.5 or E15.5 Math5-lacZ/+ embryos were given to a single intraperitoneal injection of EdU (6.7 µg/g body mass). Embryos were harvested after a 4-hr chase and their retinas were stained for Crx, βgal and EdU. For other short-term labeling experiments, a single pulse of EdU or BrdU was given to pregnant dams 1 hr prior to harvest.

To assess alterations in the fate distribution of neurogenic cells exiting mitosis on different days, pregnant dams carrying Crx>Math5 Tg (line 251) and non-transgenic control embryos were given a single injection of BrdU (100 µg/g body mass) on E12.5, E13.5 or E15.5. Retinal sections from the resulting mice were stained for BrdU at P21, and the distribution of strongly BrdU+ cells among GCL, INL and ONL layers was determined. For the E15.5 pulse, we counted 24 sections (200X) from n = 6 eyes of each genotype, representing a total of 816 birthdated cells in Crx>Math5 Tg mice and 1005 birthdated cells in control mice. To evaluate late-stage RGC births, dams carrying Crx>Math5 BAC (line 60) and control embryos were pulsed with EdU at E17.5 and harvested at P22. Likewise, Crx>Math5 Tg (line 251) pups with Math5 KO and heterozygous (het) genotypes, and non-transgenic littermates, were pulsed with BrdU at P1 and harvested at P22.

RGC birthdating curves were generated for rescued and control littermates by giving single EdU pulses at E11 and E12, E13.5, E15.5, or E17.5. The resulting pups (four genotypes) were harvested at P1 and their retinas were stained for Brn3a and EdU as flatmounts. Two 0.05 mm² areas in the central retina were imaged as confocal Z-stacks through the GCL for each flatmount preparation. The density of RGCs (Brn3a+ cells per mm²) and birthdated RGCs (Brn3a+ EdU+ cells per mm²) was determined by direct counting. The normalized RGC birth fraction was calculated by dividing the number of RGCs born at each time point by the sum of RGCs born in all four time points (E11–E12, E13.5, E15.5, E17.5).

Statistics

Comparisons were made using a two-tailed Student’s t-test in Microsoft Excel in cases where equal variance was observed. The Welch t-test was used for comparisons among groups of unequal variance. Errors are reported for biological replicates as SDM (standard deviation of the mean) unless otherwise noted. Jitter plots were generated with Prism software (Graphpad, La Jolla, CA).

Clonal analysis in retinal explants

Retinal explant cultures and retroviral infections were performed as described (Brzezinski et al., 2012) using standard methods (Hatakeyama and Kageyama, 2002; Wang et al., 2002). Briefly, retinas were dissected from E13.5 wild-type embryos, flattened onto Nucleopore polycarbonate membranes (0.4 µm pore size, GE Healthcare, Piscataway, NJ) and transferred to Transwell culture dishes containing neurobasal media (Invitrogen) with B27 and N2 supplements, glutamine (0.4 mM), BDNF (50 ng/mL, Peprotech, Rocky Hill, NJ), CNTF (10 ng/mL, Peprotech), penicillin (50 U/mL), streptomycin (50 µg/mL), and gentamicin (0.5 µg/mL).

The Math5-IRES-GFP retroviral plasmid was constructed by inserting a Math5 cassette in the bicistronic MSCV-IRES-GFP (MIG) retroviral vector (Van Parijs et al., 1999). MSCV-IRES-dnMAML^GFP was generated by replacing the GFP cassette in MIG with dnMAML^GFP. This encodes a fusion protein with residues 12–72 of mouse MAML1 (mastermind-like) at the N-terminus and GFP at the C-terminus (Maillard et al., 2004). Retroviral stocks were prepared in parallel by calcium phosphate transfection of the Phoenix ecotropic packaging cell line (Pear, 2001; Swift et al., 2001) with plasmid vectors. Polybrene (hexadimethrine bromide, 0.8 µg/mL, Sigma Aldrich, St. Louis, MO) was added to filtered media containing infectious particles. These viral stocks were titered on NIH3T3 cells and diluted to ~8 × 10⁵ CFU (colony forming units) per mL. Transductions were performed by pipetting one drop (~25 µL) on top of each fresh explant, to sparsely mark dividing cells (Roe et al., 1993) and their descendants.

Infected explants were cultured for 7 days at the gas-media interface at 37°C under 5% CO₂. Half of the media was replaced with fresh media on days 2, 4 and 6. After one week in culture, explants were fixed in 4% PFA for 30 min and processed for cryosectioning. Serial thick (30 µm) cryosections were immunostained for GFP and Brn3a, and imaged as 3-dimensional confocal Z-stacks. Clones were scored for size (number of GFP+ cells) and composition (number of Brn3a+ RGCs). A clone was defined as an isolated group of directy apposed GFP+ cells, separated by at least 4 cell bodies from other GFP+ cells. We scored 3–4 explants per virus, giving a total number of 70 (IRES-GFP), 60 (Math5-IRES-GFP) and 52 (IRES-dnMAML) clones, respectively.

RESULTS

Crx and Math5 are expressed in comparable neurogenic cell populations

Crx and Math5 progenitor cell populations give rise to 85% (rods, cones and bipolar cells) and 3% of the mouse adult retina (Brzezinski et al., 2012; Furukawa et al., 1997; Jeon et al., 1998), respectively (Fig. 1A). To compare these factors directly, we examined their overlap and onset of expression relative to the terminal S phase, using a lacZ allele (Atoh7^tm1Gla, (Brown et al., 2001)) as a proxy for Math5. We found that Crx and βgal are expressed in distinct, but overlapping, cohorts of cells at both E13.5 and E15.5 (Fig. 1B,C), consistent with the lineage profile of Math5 descendants, over half of which are photoreceptors (Brzezinski et al., 2012). Using EdU or BrdU pulse-chase analysis at these time points, we determined that a small number of Crx and βgal double-positive cells at E13.5 and E15.5 were also labeled with EdU after a 4-hr chase (Fig. 1). At these developmental stages, four hours is sufficient time for some cells progress through S, G2, and M phases and enter G0 (Prasov and Glaser, 2012; Sinitsina, 1971; Young, 1985b). Given that both of these factors are expressed during or after the terminal division (Brzezinski et al., 2012; Muranishi et al., 2011; data not shown), these results suggest that Crx and Math5 can be made in the same cells, simultaneously or sequentially, at the time when cell fate is determined. While Math5 is short-lived, Crx expression persists in a broad population of differentiated photoreceptor and bipolar cells (Brzezinski et al., 2012; Furukawa et al., 2002; Furukawa et al., 1997).

Fig. 1.

Crx and Math5 (βgal) are expressed in overlapping subsets of cells during or shortly after cell cycle exit. (A) Schema comparing the temporal expression patterns and cellular abundance of Crx (red) and Math5 (black) in the retina. (B–C) Sections from E13.5 (B) or E15.5 (C) Math5-lacZ/+ embryos costained for βgal (Math5-lacZ allele), Crx, and EdU following a 4 hour chase in vivo. Many cells coexpress Math5 and Crx (arrowheads), and both factors are expressed in some cells during or shortly after cell cycle exit (EdU+, arrows). Cells initiating expression of Crx or Math5 reflect similar populations of progenitors. RPE, retinal pigmented epithelium; NBL, neuroblastic layer; GCL, ganglion cell layer. Scale bar, 50 µm.

Crx>Math5 conventional and BAC overexpression systems

To critically assess the role of Math5 in biasing progenitor cell fate, we generated transgenic mice with ectopic Math5 expression. Given the overlap of endogenous Crx and Math5 expression, and the similar nature of these neurogenic cells, we chose the Crx promoter to broadly express Math5 in a large population of early post-mitotic precursors. The 2.4 kb Crx promoter has been extensively characterized and is thought to drive specific expression in photoreceptor precursors, and mature rods, cones and bipolar cells (Furukawa et al., 2002). However, to limit position effects and ensure faithful Crx expression, we also built a BAC transgene containing 134 kb regulatory DNA upstream of the Crx start site (exon 1), and 69 kb downstream of the polyadenylation signal (exon 3). We then generated conventional and BAC transgenic mice, termed Crx>Math5 Tg and Crx>Math5 BAC, respectively (Fig. 2A). The transgenes express Math5 and Cre from bicistronic transcripts. The Cre recombinase allowed us to trace the fate of cells expressing transgene-derived Math5 using the R26floxGFP reporter (Mao et al., 2001), which makes GFP after excision of a loxP-flanked stop signal. In parallel, we generated matched Crx>Cre Tg and Crx>Cre BAC control transgenic mice to confirm the Crx lineage, and isolate Math5 effects.

Fig. 2.

Characterization of the Crx>Math5-IRES-Cre and Crx>Cre conventional and BAC transgenic mice. (A) Map of conventional and BAC transgenes. In conventional transgenes, Math5-IRES-Cre pA and control Cre pA cassettes are positioned downstream of the 2.4 kb Crx promoter fragment, as Crx>Math5 Tg and Crx>Cre Tg, respectively. In BAC transgenes, equivalent cassettes were precisely inserted in BAC clone RP23-81H17, at the first ATG of the Crx gene, in exon 2, as Crx>Math5 BAC and Crx>Cre BAC. (B) Expression patterns of Crx>Math5 BAC, Crx>Cre BAC, Crx>Math5 Tg, and Crx>Cre Tg mice carrying R26floxGFP reporters. In adult and embryonic retinas for each transgene (right, far right), Crx and Cre are coextensive, suggesting faithful recapitulation of the endogenous Crx pattern. In Crx>Math5 BAC and Crx>Cre BAC retinas, reporter expression (R26floxGFP) is largely confined to the photoreceptor and bipolar precursors in the neural retina. In Crx>Math5 Tg and Crx>Cre Tg, the GFP reporter is widespread throughout the neural retina, and is present in the RPE and ciliary body (Suppl Fig. 2). (C) RT-PCR of RNA from P21 Crx>Math5 Tg tissues, P21 non-transgenic control (cnt) retinas, and E14.5 control eyes. The products show total, endogenous (endo), and transgenic (Tg) Math5, and βactin mRNA expression. Endogenous Math5 is expressed in the embyonic retina, and at low levels in the brain, but is not detected in other tissues (including the adult retina). In contrast, transgenic Math5 is strongly expressed in P21 retinas, and is restricted to eye tissue. No PCR products are detected in the absence of reverse transcriptase (− RT). (D) Taqman qRT-PCR comparing the relative abundance of Math5 and Crx transcripts in P21 Crx>Math5 BAC (BAC 60), Crx>Math5 Tg (Tg251), and control retinas. Math5 (a) and Crx (b) expression levels are normalized to Gapdh, with expression reported relative to Crx>Math5 BAC (left). Math5 expression is also reported relative to endogenous Crx levels (right), calculated using triplex competitive PCR data (k) from Crx>Math5 BAC to normalize expression levels (Suppl Fig. 3). Math5 expression in Crx>Math5 Tg retinas is 16-fold higher than Crx>Math5 BAC retinas, and 8-fold higher than endogenous Crx. Ex, exon; IRES, internal ribosome entry site; FRT, Flipase recognition target; on, optic nerve; cb, ciliary body; IS, inner segment; ONL, outer nuclear layer; INL, inner nuclear layer. Scale bar, 50 µm.

Multiple independent insertions (≥ 4) of each transgene were analyzed for GFP and Cre expression, after crossing to R26floxGFP reporter mice (Fig. 2B). These patterns were consistent, indicating that position effects were minimal in the vast majority of the transgenic lines (Suppl Fig. 1 and data not shown). For subsequent analysis, we chose a single representative line for each of transgene.

To evaluate transgene fidelity, we co-immunostained sections from adult and embryonic mice for GFP, Cre, and Crx antigens (Fig. 2B). In both Crx>Math5 Tg mice (line 251) and Crx>Cre Tg mice (line 352), we observed cumulative expression (GFP) throughout the neural retina, RPE (retinal pigmented epithelium), and ciliary body, but not in lens or scleral tissue (Fig. 2B, Suppl Fig. 2). GFP expression was also evident in the pineal gland (Suppl Fig. 2), and in embryonic forebrain regions (data not shown). Within the retina, all cell layers, and the vast majority of cells, were labeled with GFP by both conventional transgenes. In contrast, the patterns of Cre and Crx immunostaining were much more restricted. Cre antigen was completely co-extensive with Crx in adult photoreceptors and bipolar cells, although the relative expression levels varied among INL cells. Likewise, in the embryonic retina, all Cre+ cells were Crx+ (Fig. 2B and data not shown). These results indicate that the Crx 2.4 kb promoter drives strong expression in photoreceptor and bipolar precursors, but is also active in multipotent progenitors or in other post-mitotic cells. Together, these patterns are most consistent with high-level expression in Crx domains (Chen et al., 1997; Furukawa et al., 1997), and leaky expression in the domains of closely related homeodomain factor Otx2 (Bovolenta et al., 1997; Simeone et al., 1993). In Crx>Cre BAC mice (line 60) and Crx>Math5 BAC mice (line 764), cumulative transgene expression (GFP) was confined to the retina (Fig. 2B), and was not observed in the RPE or ciliary body as noted with the conventional transgenes (Suppl Fig. 2). Again, Cre+ cells were co-extensive with the Crx population of cells at both adult and embryonic time points (Fig. 2B). Within the adult retina, only a few scattered cells were labeled with GFP in the inner INL and GCL by the Crx>Cre BAC and Crx>Math5 BAC transgenes, as expected. Overall, the patterns of BAC and conventional transgene expression were very different, suggesting that the Crx 2.4 kb promoter is active beyond the endogenous Crx domain.

To further characterize the BAC and conventional transgenes, we assessed the distribution and level of Math5 mRNA expression by RT-PCR. In the Crx>Math5 Tg, transcripts were evaluated in various tissues using primers specific to transgenic (Tg) or endogenous Math5 (Fig. 2C). Both species were confined largely to the eye, with a low level of endogenous Math5 mRNA in the brain (Saul et al., 2008). Only transgenic Math5 was detected in the adult retina, consistent with the known patterns of Math5 and Crx expression (Brown et al., 1998; Brzezinski et al., 2012; Chen et al., 1997; Furukawa et al., 1997). To quantitatively compare the levels of BAC and conventional Crx>Math5 transgene expression, we first determined the ratio of Crx>Math5 BAC and endogenous Crx transcripts using a triplex competitive RT-PCR assay (Prasov et al., 2010) with a common end-labeled forward primer located in Crx exon 1 (Suppl Fig. 3). We found that BAC-derived Math5 transcripts were present at 53 ± 3% (k) the level of endogenous Crx transcripts in the adult retina, or approximately single-copy expression levels. We next measured Math5 (a) and Crx (b) RNAs relative to Gadph in Crx>Math5 BAC, Crx>Math5 Tg and control adult retinas by TaqMan quantitative PCR. We found that Math5 levels in Crx>Math5 Tg retinas were 16-fold higher than those in BAC transgenic retinas. As expected, Math5 was not detected in controls and the level of Crx mRNA did not vary between genotypes. In the adult Crx>Math5 Tg retinas, the level of Math5 expression was thus 8-fold higher than endogenous Crx (a × k/b). The differences in reporter expression observed between BAC and conventional transgenic mice (Fig. 2B) may thus reflect differences in the pattern and/or level of expression.

Lineage analysis of Crx>Math5 transgenes

To determine whether ectopic Math5 biases progenitors towards particular cell fates, we evaluated the distribution of cell types in Crx>Math5 BAC and Crx>Math5 Tg retinas compared to control Crx>Cre BAC and Crx>Cre Tg retinas, by examining GFP staining pattern of double transgenic mice carrying the R26floxGFP reporter. As expected, each transgene labeled all photoreceptors and bipolar cells. Persistent Math5 expression did not grossly alter the subtype distribution of cones or bipolar cells (Suppl Fig. 4).

In addition to rods, cones and bipolar cells, the Crx>Math5 BAC and Crx>Cre BAC control transgenes marked a small number of ganglion, horizontal and amacrine neurons, with frequency that varied from region to region in the retina and among cell types (Fig. 3). Early cell types, particularly horizontal neurons, were rarely marked by Crx>Cre BAC or Crx>Math5 BAC transgenes (Fig. 3E,F), consistent with the onset of Crx expression at E12.5 (Chen et al., 1997; Furukawa et al., 1997). Likewise, in conventional Crx>Math5 Tg and Crx>Cre Tg mice, the vast majority of retinal cells were marked with GFP, including all major cell types. As noted with the BAC transgenes, fewer RGCs and horizontal cells were marked compared to other cell types (Suppl Fig. 5 and data not shown).

Fig. 3.

Widespread Crx>Math5 expression has little effect on cell fate decisions in the retina. (A–F) Sections from adult Crx>Math5-IRES-Cre BAC (shortened as Crx>Math5 BAC) or matched Crx>Cre BAC control transgenic mice carrying R26floxGFP reporters were coimmunostained with GFP and informative markers. GFP and marker costaining (bottom) and GFP alone (top) are indicated (arrows). Müller glia were identified by Sox9 (A,B), amacrines by AP2α (C,D) and horizontal neurons by calbindin (E,F). Arrowheads in E and F mark calbindin+ horizontal cells that are GFP−. (G–H) Flatmounts of adult Crx>Math5 BAC and Crx>Cre BAC retinas labeled with retrograde rhodamine dextran and imaged through the GCL. Some lineage-marked (GFP+) rhodamine dextran (rhod dext) [+] RGCs (arrows) and rhod dext [−] displaced amacrine cells (arrowheads) are indicated (insets). While both transgenes mark essentially all photoreceptors and bipolar cells (Suppl Fig. 2), a small fraction of other cell types was also labeled in each case. The fate spectra are similar, with or without Math5. Rare horizontal cells were labeled in Crx>Math5 BAC Tg mice, but not in Crx>Cre BAC Tg controls, most likely due to differences in transgene expression level. Scale bar, 50 µm.

In principle, the labeled RGCs, horizontal and amacrine cells in BAC transgenic mice could represent one of three classes: [1] lineal descendents of proliferating progenitors that expressed low or high levels of the transgene, [2] rare Crx+ precursors that adopted non-photoreceptor or bipolar fates, [3] cells whose fate was shifted due to the action of Math5. To distinguish these mechanisms, we used two approaches. First, we assessed the heterogeneity of Cre expression levels among cells using a dual-reporter concordance paradigm (Brzezinski et al., 2012). Adult triple transgenic Crx>Math5 BAC or Crx>Cre BAC mice, carrying R26floxGFP and Z/AP (Lobe et al., 1999) reporters, were immunostained for GFP and hPLAP (Suppl Fig. 6). Among photoreceptors in the ONL and bipolar cells in the outer INL, concordance for GFP and hPLAP was uniformly high for each transgene (nearly 100%). However, among neurons in the inner INL and GCL, concordance was low (<40%), indicating that these cell types expressed a low level of Cre, and thus recombined stochastically at only one reporter locus. Second, we analyzed the cycle kinetics and spatial distribution of informative cells in E15.5 embryos carrying Crx>Math5 BAC and R26floxGFP. After a 1 hr EdU pulse, no Cre+ EdU+ cells were detected (Fig. 4A). However, many EdU+ cells were observed in GFP+ vertical stripes (Fig. 4B), where most labeled Brn3b+ RGCs were localized (Fig. 4C). These findings are consistent with the general clustering of GFP+ cells in the adult retinas, and suggest a clonal origin (Reese et al., 1999; Turner and Cepko, 1987). Similar patterns were observed for the Crx>Math5 Tg embryos, but the relative abundance of GFP+ progenitors was much greater (data not shown).

Fig. 4.

The Crx>Math5 BAC transgene is expressed at low levels in proliferative retinal progenitors. E15.5 Crx>Math5 BAC; R26floxGFP embryos were pulsed with EdU for 1 hr and costained for the indicated markers. (A) Projection images of 10 µm optical sections show that all Cre+ cells were EdU− indicating that Cre is not expressed at high levels during S-phase. (B) In contrast, many GFP+ stripes contain EdU+ cells (arrows, inset), consistent with stochastic Cre expression in proliferating RPCs. (C) A small number of Brn3b+ RGCs were identified among clustered GFP+ cells (arrow, inset). Scale bar, 50 µm.

Crx>Math5 expression does not stimulate RGC genesis

Given the limitations of using the Cre lineage reporter to assess cell fate, we employed other metrics to assess the effects of broad Math5 overexpression on the overall fate distribution. We focused primarily on ganglion cells for these experiments, because Math5 is necessary for RGC development, and we used Crx>Math5 Tg, because this transgene is expressed at much higher levels than the BAC counterpart (Fig. 2D). We observed a similar abundance of Brn3a+ or rhodamine dextran-labeled RGCs in Crx>Math5 Tg and control mice throughout development (Fig. 5A–C). At P0, prior to the neonatal culling of RGCs (Erkman et al., 2000; Farah and Easter, 2005; Galli-Resta and Ensini, 1996; Young, 1984), 56 ± 5% SD of GCL neurons in Crx>Math5 Tg retinas were RGCs (Brn3a+), similar to controls (59 ± 5% SD, P= 0.4, Fig. 5D). Likewise, at P22, no significant difference was observed in the fraction of RGCs (rhodamine dextran-labeled) in the GCL between the two genotypes (41.1 ± 0.1% SD for wild-type, 43 ± 8% SD for Crx>Math5 Tg, P= 0.8 Fig. 5D) or previous reports (41 ± 4% SD (Jeon et al., 1998)). We conclude that broad ectopic Math5 expression does not promote RGC fate, alter the survival of RGCs, or drive Brn3a expression in a Math5 wild-type retina.

Fig. 5.

Widespread Crx>Math5 expression does not alter RGC abundance or retinal histology. (A–C) Sections from E13.5 (A), P0 (B) and P22 (C) non-transgenic (non-Tg) or Crx>Math5-IRES2-Cre (Crx>Math5 Tg) retinas stained for Brn3a (A,B) or rhodamine dextran (C) to mark RGCs, and counterstained with DAPI to mark nuclei. There is non-specific staining in the RPE and vitreous due cross-reactivity of anti-mouse secondary antibodies with mouse IgG. (D) RGC fraction among GCL neurons at P0 and P22. There is no significant difference in the RGC fraction between transgenic (Tg) and control retinas. (E) Low (top) and high (bottom) magnification views of basic fuchsin-and methylene blue-stained plastic sections. The retinal histology is similar in Crx>Math5 Tg and control mice. Scale bar: 50 µm in A–C; 100 µm in E.

Crx>Math5 expression alters the distribution of early born cell types

Given the persistent high level of expression of the Crx>Math5 Tg in rods, cones and bipolar cells, we assessed overall retinal histology in plastic sections (Fig. 5E). The morphological features of photoreceptor nuclei, inner and outer segments, and other layers were not affected at this level. However, subtle fate shifts might occur during development, which are counterbalanced by homeostatic feedback mechanisms. To assess these effects, we used a birthdating approach. Embryos were exposed to single BrdU pulses at E12.5, E13.5, or E15.5 and their retinas were analyzed at P22. At each time point, fewer birthdated nuclei were observed in the ONL of Crx>Math5 Tg mice than controls (Fig. 6A, Suppl Fig. 7). For quantitative analysis, we focused on the E15.5 time point, as these litters contained a sufficient number of animals of each genotype for statistical comparisons. We counted the number of strongly BrdU+ nuclei in ONL, INL and GCL layers. We observed a 2-fold decrease in the fraction of birthdated photoreceptors (ONL cells) in Crx>Math5 Tg retinas (21 ± 3% SD) compared to controls (43 ± 2% SD, t-test P < 10⁻³), with a corresponding increase in the INL and to lesser extent GCL (Fig. 6A).

Fig. 6.

Widespread Crx>Math5 expression does not extend the profile of RGC births, but decreases the numbers of early-born photoreceptors. (A) Birthdating of Crx>Math5-IRES-Cre transgenic (Crx>Math5 Tg) and non-Tg littermate retinas. Embryos were exposed to BrdU at E15.5 and analyzed at P22. There were 2-fold fewer E15.5 birthdated cells in the ONL, representing rod (arrow) and cone (arrowhead) photoreceptors in Crx>Math5 transgenic animals compared to controls, and corresponding increases in the INL and GCL. (B) Crx>Math5 Tg pups were similarly exposed to BrdU at P1 and harvested at P22. Few, if any, GCL cells were labeled with BrdU in transgenic or control mice. (C) Crx>Math5-IRES-Cre BAC (BAC) embryos were exposed to EdU at E17.5, and their RGCs were labeled with rhodamine dextran at P22. No EdU+ RGCs were detected in central flatmounts of BAC or control retinas. Prolonged transgenic expression of Math5 does not extend the RGC birthdating profile. Scale bar: 100 µm in A–B; 50 µm in C.

In principle, the loss of early-born photoreceptors could be due to cell death from persistent Crx>Math5 Tg expression or a small, bona fide fate shift. To distinguish these mechanisms, we evaluated Crx>Math5 Tg and control retinas for apoptosis by activated Caspase-3 immunostaining. At multiple time points between E13.5 and P0, we observed 1–2 apoptotic cells per field (200X) in both Crx>Math5 Tg and control retinas (Suppl Fig. 8), consistent with previous studies of cell death in the embryonic retina (Vecino et al., 2004). Crx>Math5 Tg is thus unlikely to induce cell death in photoreceptor precursors. Instead, ectopic Math5 appears to shift the fates of some early rod and cone photoreceptors.

Crx>Math5 expression does not extend the temporal profile of RGC births

In Crx>Math5 Tg mice, the expression of Math5 is extended through postnatal development, whereas endogenous Math5 mRNA is downregulated by P0 (Brown et al., 2001; Brzezinski et al., 2012). To test whether prolonged expression in neurogenic cells extends the profile of RGC births, we pulsed Crx>Math5 Tg mice with BrdU at P1 and harvested eyes at P22. In the central two-thirds of the retina, no BrdU+ cells were detected in the GCL of either genotype (Fig. 6B), consistent with the completion of displaced amacrine and RGC genesis in these areas by P1 (Farah and Easter, 2005; LaVail et al., 1991; Reese and Colello, 1992; Voinescu et al., 2009; Young, 1985a). Similarly, in flatmounts of adult Crx>Math5 BAC retinas that were exposed to an EdU pulse at E17.5, we observed very few, if any, EdU+ RGCs (Fig. 6C). Therefore, the envelope of RGCs births is not extended by prolonged Math5 expression.

Retroviral Math5 does not induce RGC fate or cell cycle exit in retinal explants

We also tested whether Math5 can bias proliferating progenitors towards RGC fate, using a retroviral vector to transduce cultured retinal explants. E13.5 retinas were infected at low density ex vivo with MIG vectors. The resulting single-copy proviruses express Math5 and GFP, or GFP alone from the potent MSCV LTR promoter (Hawley, 1994) (Fig. 7A). After 7 days in culture, we counted the number of Brn3a+ RGCs (Fig. 7B–C). Among 70 clones infected with the IRES-GFP retrovirus, 4/272 GFP+ cells were Brn3a+ (1.5 ± 0.7% binomial SD, Fig. 7D), in accord the fraction of RGCs produced by in vivo clonal analysis (Turner et al., 1990). Among 60 clones infected with Math5-IRES-GFP retrovirus, 6/262 GFP+ cells were Brn3a+, which is not different from those transduced with GFP alone (2.2 ± 0.9% binomial SD, Fisher’s exact P = 0.34). Because overexpression of bHLH factors can promote cell cycle exit and differentiation (Farah et al., 2000), we also evaluated clone size. We found that the size distribution did not vary significantly between explants infected with these viruses (Fig. 7E, χ² test P = 0.6 for df = 4), indicating that high-level single-copy expression of Math5 does not significantly promote cell cycle exit. As a positive control, we also analyzed explants infected with an MSCV-IRES-dnMAML^GFP retrovirus. The dnMAML^GFP fusion protein autonomously blocks Notch signaling by interfering with the NICD-CSL transcriptional complex (Maillard et al., 2004). Inhibition of Notch activity is associated with premature cell cycle exit and stimulation of RGC fate among early progenitors (Austin et al., 1995; Nelson et al., 2007). Among 52 clones, we detected a modest increase in the fraction of Brn3a+ RGCs (Fig. 7D, 3/60 GFP+ cells, 5 ± 3%, P = 0.11). We also observed a significant reduction in clone size, with all cells deriving from one- or two-cell clones (Fig. 7E, χ² test P < 10⁻⁸ for df = 4). These results confirm that blockade of Notch signaling by dnMAML^GFP drives cell cycle exit, and that our explant system is sufficiently robust to detect this effect. Ectopic Math5 expression in progenitors does not stimulate cell cycle exit or significantly promote RGC fate in this system, consistent with our transgenic overexpression findings (Fig. 5).

Fig. 7.

Retroviral Math5 overexpression does not stimulate RGC fate or cell cycle exit in retinal explant cultures. (A) Experimental design. Retinas were explanted from E13.5 embryos, flattened on polycarbonate membranes, infected at low density with the indicated MSCV retrovirus, and cultured for 7 days in vitro (DIV). Isolated GFP+ clones were scored for RGC number by Brn3a immunoreactivity and clone size. (B–C) Example clones from explants infected with IRES-GFP (B) or Math5-IRES-GFP (C) retroviruses. (D) Plot showing the fraction of GFP+ cells that developed as Brn3a+ RGCs in transduced explants. There was no significant difference in the RGC fraction of explants transduced with Math5-IRES-GFP or IRES-GFP control. A modest increase in RGCs was observed when Notch signaling was autonomously blocked in clones with the IRES-dnMAML virus. Error bars show binomial standard deviation. (E) Clone size distribution. There was no difference between Math5-IRES-GFP and IRES-GFP explants, but clone size was significantly reduced in explants infected with IRES-dnMAM. Scale bar, 50 µm.

Crx>Math5 expression partially rescues the RGC deficiency in Math5 KO mice

Although Crx>Math5 Tg does not stimulate RGC genesis in the wild-type environment, cells expressing ectopic Math5 may be prevented from adopting the RGC fate by strong negative feedback from nascent RGCs (Austin et al., 1995; Belliveau and Cepko, 1999; Waid and McLoon, 1998; Wang et al., 2005; Zhang and Yang, 2001). Indeed, we have observed that many Brn3a+ cells are generated in E13.5 Math5 KO retinal explants transfected with human ATOH7 (Prasov et al., 2012). We therefore crossed Crx>Math5 Tg and Crx>Math5 BAC transgenes onto the Math5 KO background to examine the potential of Crx-driven Math5 to stimulate RGC genesis, heterochronically and heterotopically (Fig. 1A), in a deficient environment.

We evaluated retinal flatmounts for the density of mature RGC cell bodies, axons and fascicles (Fig. 8A–C). In Math5 heterozygous mice, the vast majority of axons made radial projections to the optic disc and were fasciculated (Fig. 8A,B), and the GCL contained many Brn3a+ RGCs (Fig. 8C). In contrast, Math5 KO retinas had vastly reduced axon density, in accord with previous estimates (Lin et al., 2004). The residual axons were largely unfasciculated and exhibited pathfinding defects similar to those in Brn3b −/− mice (Badea et al., 2009; Gan et al., 1999) (Fig. 8B and Suppl Fig. 9). Among 10 flatmounts examined from Math5 KO eyes, no Brn3a+ cells were observed in any area (Fig. 8C). The Crx>Math5 Tg transgene, and to a lesser extent the Crx>Math5 BAC, were variably capable of rescuing axons and preventing fasciculation defects in Math5 KO retinas. However, rescue was less pronounced with successive generations of mice, ostensibly due to epigenetic reductions in transgene expression levels (Garrick et al., 1998). The Crx>Math5 Tg, but not the Crx>Math5 BAC, transgene was capable of restoring Brn3a expression among some adult ganglion cells. As expected, all rescued Brn3a+ ganglion cells were derived from progenitors that expressed the Crx>Math5 Tg transgene (GFP+), whereas only ~40% of Brn3a+ cells were GFP+ in the wild-type (Fig. 8D). Finally, we observed small optic nerves in rescued mice carrying the Crx>Math5 Tg (Fig. 8E), but not in Math5 KO controls. Together, these results suggest that the Crx>Math5 Tg transgene, which expresses high levels of Math5 in early post-mitotic precursors and low levels in progenitors, can partially rescue RGC genesis and axonal guidance defects in Math5 KO mice.

Fig. 8.

Crx>Math5 expression partially rescues RGC fate specification and optic nerve development in Math5 knockout (KO) mice. (A–C) Flatmounts of adult retinas from the indicated genotypes stained for TuJ1 (A, B) to mark axons, and Brn3a (C) to mark cell bodies of mature RGCs. In Math5 heterozygous mice (het), RGC axons fasciculate and project to the optic disc (OD), and their cell bodies are reactive for Brn3a. In Math5 KO mice, RGC axons are very sparse, meander and do not fasciculate, and their cell bodies not express Brn3a; and the presumptive optic disc (POD) is not fully developed. The Crx>Math5 Tg – and to a much lesser extent Crx>Math5 BAC – transgene is able to partially rescue RGC density (Brn3a+), pathfinding and fasciculation defects in Math5 KO mice. (D) Sections from adult retinas of the indicated genotypes, stained for GFP and Brn3a. In heterozygous mice, the Crx>Math5 Tg marks ~40% of the Brn3a+ RGC population. In contrast, rescued Brn3a+ RGCs derive exclusively from the Crx>Math5 Tg lineage (arrows). (E) Optic nerves of adult Math5 het and Crx>Math5 Tg; Math5 KO mice, stained with H+E. The rescued nerve is much thinner than the control. (F) Birthdating transgenic and control Math5 KO retinas. Pups were exposed to a pulse of BrdU at P1 and harvested at P22. Few, if any, GCL cells were labeled with BrdU in either genotype, indicating that rescued RGCs are born prior to P1. P, proximal; D, distal. Scale bar, 50 µm.

Crx>Math5 expression alters the RGCs birth profile in Math5 KO mice

Given the differences in the timing of Crx and endogenous Math5 expression, we tested whether the rescued RGCs were born within the same temporal envelope as native RGCs, which are only generated prenatally. We first exposed P1 pups to a single pulse of BrdU, and stained their retinas at P21. Few, if any, GCL cells were birthdated at P1 in Math5 KO mice carrying Crx>Math5 Tg (Fig. 8F).

To fully explore the temporal profile of RGC births and the extent of RGC genesis in rescued mice, we generated partial birthdating curves (Young, 1985a) for Crx>Math5 Tg; Math5 KO mice and littermate controls (Fig. 9). Crx>Math5 Tg; Math5 KO mice were crossed to Math5 heterozygotes and the pregnant dams were given two injections of EdU at E11 and E12 to mark the earliest born RGCs (Fig. 9A), or a single injection of EdU at E13.5, E15.5, or E17.5 (Fig. 9B–D). Retinas were harvested from the resulting pups and stained as flatmounts for Brn3a and EdU at P1, a time point before the neonatal RGC culling period (Farah and Easter, 2005), but after all RGCs have been generated (Figs. 6B and 8F). In Math5 KO and Crx>Math5 Tg; Math5 KO mice, very few Brn3a+ RGCs were born during early development (Fig. 9A). In all 4 genotypes, the majority of RGCs were born between E13.5 and E15.5 (Fig. 9B,C), consistent with RGC birthdating curves for wild-type retinas (Drager, 1985; Farah and Easter, 2005; Young, 1985a). Furthermore, no E17.5 birthdated Brn3a+ cells were detected in any of these mice, after careful examination of confocal Z-stacks through the GCL (Fig. 9D).

Fig. 9.

RGC birthdates in transgenic, Math5 KO and rescued animals. Embryos with the four indicated genotypes were exposed to a single pulse of EdU at E11 and E12, E13.5, E15.5 or E17.5, and are compared as littermates. (A–D) GCL confocal views of P1 birthdated retinal flatmounts, stained for Brn3a (green) and EdU (magenta). (E) RGC birthdating curves for each genotype. No Brn3a+ RGC births were detected in any genotype after E17. The overall number of RGC births is substantially reduced in Math5 KO and Crx>Math5 Tg (Tg) rescued mice. (F) Normalized RGC birthdating curves for each genotype. Because virtually all RGCs are born between E11 and E17, the total number of RGCs was summed across the 4 time points for each genotype, and the RGC birth fraction at each time is plotted relative to this total. The normalized curves are quite similar. However, very few, if any, Brn3a+ RGCs were born during early neurogenesis (E11–E12) in Math5 knockout (KO) or rescued (Tg; Math5 KO) mice. In the rescued mice, a larger fraction of RGCs were born during mid-gestation (E15.5). (G) Brn3a+ RGC density jitter plots for each genotype. Each data point represents a single eye. The Crx>Math5 transgene (Tg) partially rescues the RGC deficiency in Math5 KO mice, which have significantly more Brn3a+ RGCs than Math5 KO controls, although this effect is variable. The Crx>Math5 Tg does not significantly affect RGC density in heterozygotes. Scale bar, 50 µm.

The birthdating data and RGC counts reveal four important trends (Fig. 9E,F). First, the curves for Math5 heterozygous (het) and Crx>Math5 Tg; Math5 het mice were nearly identical, suggesting that Crx>Math5 Tg does not alter the profile of RGC births (χ² test P = 0.26 for df = 2). Second, the normalized distribution of RGC births for Crx>Math5 Tg; Math5 KO and Math5 het mice differed significantly (χ² test P = 0.01 for df = 2). In particular, very few RGCs were generated in the rescued mice (Crx>Math5 Tg; Math5 KO) at the earliest developmental times, and a larger number of RGCs were born at E15.5. Third, RGCs were generated within the same temporal envelope in all 4 genotypes, suggesting that this time window is fixed. It does not depend on Math5, and cannot be shifted by protracted or elevated Math5 expression.

Fourth, the absolute number of rescued RGCs at P1 in Crx>Math5 Tg; Math5 KO mice was increased 2.3-fold in comparison to control Math5 KO mice (Welch t-test P = 0.02, Fig. 9G), but this effect was variable, and the number of RGCs was low in comparison to Math5 hetetozygotes (12%, 1160 ± 250 Brn3a+ cells per mm², P < 10⁻⁸). The Crx>Math5 transgene did not increase the RGC density in Math5 hets (10,400 ± 400 Brn3a+ cells per mm² vs. 10,800 ± 600, Student’s t-test P = 0.54), consistent with the results above (Fig. 5). Likewise, the number of RGCs was significantly reduced in control Math5 KO mice compared to heterozygotes (5%, 500 ± 50 Brn3a+ cells per mm², P < 10⁻⁸), as previously reported (Lin et al., 2004).

Increased cell death and optic nerve defects in rescued Crx>Math5 Tg mice

The variability and incomplete rescue of RGCs in Crx>Math5 Tg mice on the Math5 KO background was surprising. To explore the mechanism underlying this variability, we analyzed optic nerve development during early embryogenesis, using TuJ1 to identify RGC axons. At E15.5 and E17, we observed coalescence of TuJ1+ fibers and formation of optic nerves in Crx>Math5 Tg; Math5 KO, but not in Math5 KO retinas (Fig. 10A,B). These rescued optic nerves were much thinner than those in wild-type controls. In some cases, these nerves exhibited severe pathfinding defects as they exited the retina, and formed large ‘knot’ structures (Fig. 10A, arrowhead). Because RGCs that fail to properly establish connections in the CNS are eliminated (O'Leary et al., 1986), we evaluated cell death by activated Caspase-3 immunostaining at E16.5, near the end of RGC genesis. We observed a significant increase in cell death in Math5 KO mice compared heterozygous controls (Fig. 10C–D, 9 ± 1 vs. 1.1 ± 0.2 Casp3+ cells per field, Welch t-test P < 0.001), consistent with previous reports (Feng et al., 2010). In Crx>Math5 Tg; Math5 KO mice, there was a further increase in cell death compared to Math5 KO retinas (12.9 ± 0.6 Casp3+ cells per field, P = 0.016). The vast majority of dying cells were located in the GCL (Fig. 10C), suggesting that aberrant RGCs are eliminated. We also observed knots of RGC axons in Crx>Math5 BAC; Math5 KO mice, consistent with the small degree of rescue by this transgene (Fig. 10E).

Fig. 10.

Survival and generation of late-born RGCs are inhibited in rescued animals. (A–B) βIII-tubulin (TuJ1) staining of rescued (Crx>Math5 Tg; Math5 KO) retinas at E15.5 and E17, compared to Math5 heterozygous littermates (het, A,C) or Math5 wild-type (WT) control (E17.5, B). Math5 KO (left) retinas form a nerve fiber layer, but there is no clear coalescence of axons into an optic nerve. Crx>Math5 Tg rescued retinas have thin optic nerves, and these occasionally form axon knots in the optic stalk (arrowhead, inset). (C–D) Rescued and control sections stained for cleaved Caspase-3 (arrows) at E16.5 to mark apoptotic cells. Dying cells primarily reside in the forming ganglion cell layer. Crx>Math5 Tg rescued animals (Tg; KO) exhibit increased levels of apoptosis as compared to Math5 KO mice, which have much higher levels of cell death than heterozygous animals. (E) TuJ1 staining of E15.5 Crx>Math5 BAC; Math5 KO mice, shows partial rescue of optic nerve development and the appearance of a similar RGC axon knot (arrowhead). Scale bar, 50 µm.

DISCUSSION

The patterns of Crx>Cre BAC and Crx 2.4 kb transgene expression

The Crx 2.4 kb promoter has been used to drive Cre, lacZ, Nrl or Nr2e3 expression (Cheng et al., 2006; Furukawa et al., 2002; Koike et al., 2005; Nishida et al., 2003; Oh et al., 2007), and Crx BACs have been used to drive GFP, hPLAP or lacZ expression (Muranishi et al., 2010; Samson et al., 2009). From these studies, it is clear that both types of transgenes are expressed at high levels in photoreceptors and bipolar cells, and their precursors. However, thorough lineage data reflecting cumulative transgene expression are lacking. In this study, as control for Math5 overexpression, we characterized the cumulative expression of Crx 2.4 kb promoter and Crx>Cre BAC transgenes using R26floxGFP and Z/AP reporters. At the level of Cre immunodetection (Fig. 2), both transgene formats recapitulated endogenous Crx expression throughout development, and were restricted to known Crx domains in the eye and pineal gland (Fig. 2, Suppl Fig. 2) (Chen et al., 1997; Furukawa et al., 1997). However, when Cre activity was assessed using highly sensitive reporters, the Crx 2.4 kb promoter and, to a much lesser extent the Crx>Cre BAC, were found to mark all major cell types in the retina, contrary to previous reports (Koike et al., 2005; Nishida et al., 2003). Multiple independent insertions gave similar results (Suppl Fig. 1 and data not shown), so the ectopic patterns cannot be attributed to chromatin position effects.

There were notable differences between conventional and BAC transgenes. First, the GFP reporter was detected in the RPE, ciliary body and embryonic brain with conventional, but not with BAC transgenes (Suppl Fig. 2 and data not shown), and GFP was more broadly expressed within the neural retina. These differences may reflect leaky expression of the Crx 2.4 kb promoter in Otx2 domains. Otx2 and Crx are evolutionary paralogs (Plouhinec et al., 2003), and Otx2 is expressed in the RPE, ciliary body, retinal progenitors, and embryonic forebrain (Bovolenta et al., 1997; Brzezinski et al., 2010; Muranishi et al., 2011; Simeone et al., 1993). Given their structural and functional similarity, common evolutionary origin, and overlapping expression in photoreceptors and the pineal, at least some transcription factors are likely to regulate both genes, and bind within this 2.4 kb sequence. Indeed, the zebrafish Crx ortholog is expressed in proliferating retinal progenitors (Shen and Raymond, 2004), and it has been suggested that bovine Crx, along with Otx2, regulate gene expression in the RPE (Esumi et al., 2009). Second, the level of expression was significantly higher (16-fold) from the conventional transgene (Fig. 2 and Suppl Fig. 3). Together, the qualitative and quantitative differences we observed can explain the greater abundance of GFP+ cells in Crx>Cre Tg mice compared to Crx>Cre BAC mice.

We believe the patterns of cumulative transgene expression reflect dichotomous low-level or leaky activity in proliferating RPCs, and high-level activity in photoreceptor and bipolar precursors, for four reasons. First, Cre immunoreactivity closely matched endogenous Crx expression, in adult and embryonic retinas (Fig. 2). Second, high-level Crx or Cre expression was observed only in post-mitotic cells (Fig. 4 and data not shown) (Muranishi et al., 2011). Third, GFP reporter expression was distributed in radial stripes in BAC transgenic mice (Figs. 2–4 and Suppl Fig. 1), suggesting a clonal origin. Fourth, low concordance (<40%) was observed between R26floxGFP and Z/AP reporters in the GCL and inner INL of Crx>Math5 BAC and Crx>Cre BAC mice, but high concordance (~100%) was observed in photoreceptor and bipolar neurons. This dichotomy contrasts starkly with the uniformly high concordance observed for a Math5>Cre BAC transgene (Brzezinski et al., 2012), and demonstrates the utility of this approach to distinguish populations with heterogeneous levels of Cre.

Our analysis suggests that some previous results obtained using conventional Crx>Cre transgenes should be reinterpreted. Notably, Nishida et al. (Nishida et al., 2003) ablated Otx2 using a Cre transgene driven by a 12 kb Crx promoter segment. Paradoxically, this resulted in complete loss of Crx mRNA at E18.5 (cf. Fig. 4C,D) despite the inherent time delay required for Cre protein expression, excision of Otx2 genomic sequences, and decay of existing pools of Otx2 and Crx mRNA and protein (Nagy, 2000). Thus, the ablation of Otx2 in this experiment must have occurred earlier, in retinal progenitors, well before the onset of Crx trancription. Likewise, Koike et al. (Koike et al., 2005) conditionally ablated atypical protein kinase C (aPKC) and observed a major disruption in retinal organization. On this basis, they concluded that photoreceptors were critical for proper lamination of the retina. Instead, we believe a more parsimonious explanation is that aPKC function in RPCs is critical for lamination and epithelial polarity, as has been demonstrated for other neural progenitors (Cui et al., 2007; Wodarz et al., 2000).

Ectopic expression of Math5 does not stimulate RGC fate

After establishing the patterns of Crx BAC and Crx 2.4 kb transgenic expression, we were able to test the effects of massive Math5 overexpression on the fate trajectory of retinal cells. Our findings show that Math5 overexpression does not significantly bias RGC fate in the wild-type environment for three reasons. First, the fraction of RGCs within the GCL was not increased in Crx>Math5 Tg mice at any point during development (Fig. 5). It is likely that negative feedback from nascent RGCs (Austin et al., 1995; Belliveau and Cepko, 1999; Waid and McLoon, 1998; Wang et al., 2005; Zhang and Yang, 2001) restricts Math5 from inducing supernumerary RGCs. Second, the temporal profile of RGC births was not extended by Crx>Math5 Tg expression (Figs. 6 and 9), despite abundant Crx expression in early post-mitotic precursors during the post-natal period.

Third, retroviral expression of Math5 did not significantly promote RGC fate or cell cycle exit in cultured embryonic retinal explants. These results differ from previous studies in chick and frog, in which overexpression of Math5 orthologs favored RGC fate (Kanekar et al., 1997; Liu et al., 2001). In these studies, effects on cell cycle dynamics and cell fate could not be completely isolated. By itself, any experimental manipulation that forces cell cycle exit during early neurogenesis can cause RPCs to adopt early fates, by effectively stopping progression of the ‘histogenetic clock’ (Ohnuma et al., 2002). Indeed, high-level expression of proneural bHLH factors induces cell cycle exit in vitro (Farah et al., 2000), and Xath5 overexpression reduces clone size in vivo (Moore et al., 2002). Furthermore, misexpression of other bHLH factors, such as Neurod1, during early frog development promotes RGC fate, whereas overexpression of Xath5 during late developmental stages favors non-RGC fates (Moore et al., 2002). In mice, Math5 is unlikely to be a major determinant of cell cycle exit, because it is variably expressed, during or after the terminal division, and lineage-marked cells do not re-enter the cell cycle in Math5 KO mice (Brzezinski et al., 2012). These disparate gain-of-function results may reflect differences between species in the timing, level or unique post-translational regulation of the endogenous Math5 ortholog, or the effective dose of bHLH protein delivered in these experiments.

Our transgenic and retroviral clone analyses separate cell cycle and fate effects. In the explant experiments, the level of proviral Math5 expression in transduced RPCs was not sufficient to promote cell cycle exit (Fig. 7E) and RGC fate was not favored (Fig. 7D). In contrast, the dnMAML retrovirus, which blocks Notch signaling, drove RPCs out of the cell cycle and increased RGC abundance, consistent with previous results (Austin et al., 1995; Nelson et al., 2007; Ohnuma et al., 2002). Some RGC fate effects attributed to Ath5 orthologs may be explained by premature cell cycle exit. When expressed at eight times the level of endogenous Crx (Fig. 2D), ectopic Math5 does not significantly bias progenitors or post-mitotic precursors towards the RGC fate in a wild-type environment, supporting its role as a competence factor. However, ectopic Math5 does favor RGC development in a deficient environment (Math5 KO). We observed partial rescue of RGC and optic nerve formation in adults and embryos (Figs. 8–10). These findings are generally consistent with the stimulation of ganglion cell fate by Math5 transfection in neurosphere cultures (Yao et al., 2007) or by electroporation of human ATOH7 in Math5 KO retinal explants (Prasov et al., 2012).

In the wild-type environment, we did observe one relatively minor fate effect of transgenic Math5. The number of early-born photoreceptors was significantly reduced (2-fold at E15.5) in Crx>Math5 Tg mice. However, adult retinal histology and photoreceptor morphology were grossly unaffected (Fig. 5E). Furthermore, widespread Math5 expression is not sufficient to stimulate ectopic Brn3b immunoreactivity (Suppl Fig. 10), a known downstream target (Liu et al., 2001; Mu et al., 2005). Finally, the co-expression of Crx and Math5-lacZ in some progenitors (Fig. 1) and the persistent expression of transgenic Math5 in mature adult photoreceptors (Fig. 2) show that downregulation of Math5 is not an essential step in the specification of these cell types.

Math5 is not the sole determinant of RGC competence

Our birthdating analyses (Figs. 6–9) provide novel insights into the window of RGC competence and the role of Math5. The timing of Math5 expression closely mirrors the RGC birthdating curve and Math5 is required for RGC competence (Brown et al., 1998; Brown et al., 2001; Brzezinski et al., 2012; Wang et al., 2001). Thus, in principle, the pattern of Math5 expression may be the sole factor temporally restricting RGC specification. Indeed, in Gdf11 mutant mice, an overproduction of RGCs during development is correlated with a spatiotemporal increase in Math5 expression (Kim et al., 2005). Our results, however, suggest that prolonged expression of Math5 in Crx>Math5 Tg mice does not extend the profile of RGC births, even when few nascent RGCs are present, in the Math5 KO rescue. In the rescued mice, the peak of RGC birthdates was shifted by approximately two days (Fig. 9), consistent with the later onset of Crx expression, but this modest heterochronic effect occured within the normal envelope for RGC genesis. Furthermore, the profile of residual RGC births in Math5 KO mice closely matches wild-type birthdating curves, except for the extremely low RGC abundance (Fig. 9). Together, these findings suggest that the pattern of Math5 expression is not the sole factor restricting RGC competence. Instead, a complex network of interactions, including Math5, is likely to determine the spatiotemporal pattern of RGC genesis. The envelope of RGC genesis also appears to be shaped by other transcription factors, Notch signaling, and microRNAs (Elliott et al., 2008; Georgi and Reh, 2010; Silva et al., 2003). These factors may function upstream or together with Math5 to permit RGC differentiation.

A pioneering model for RGC fate specification

In Crx>Math5 Tg; Math5 KO and Math5 KO mice, the loss of early-born RGCs (Fig. 9) is correlated with pathfinding defects in remaining RGCs (Fig. 10 and Suppl Fig. 9). These observations can be explained in two ways. First, the residual and rescued RGCs may follow an aberrant RGC differentiation pathway. In adult Math5 KO mice, residual RGCs form dendritic arbors with normal size and spacing (Lin et al., 2004), but these cells fail to express Brn3b and Brn3a (Fig. 8 and data not shown), which are critical for axon pathfinding, dendritic stratification, and cytodifferentiation (Badea et al., 2009; Gan et al., 1999). It is thus possible that Crx>Math5 Tg derived RGCs are intrinsically defective and express an aberrant set of RNA transcripts. Alternatively, the pathfinding defects in these RGCs may result indirectly, from a deficiency of early-born ganglion cells, which may limit the extent of rescue overall. Nascent RGCs are known to elaborate signals, such as sonic hedgehog (Shh), which promote intraretinal axon pathfinding generally (Erskine and Herrera, 2007; Oster et al., 2004). In zebrafish, the establishment of early RGC axons is necessary and sufficient for pathfinding and survival of later RGCs (Pittman et al., 2008), and this community effect may be widespread in the nervous system (Raper and Mason, 2010). In Math5 KO mice, residual RGC axons are poorly fasciculated, and often branched (Fig. 8), and do not extend radially toward the central retina (Suppl Fig. 9). In Crx>Math5 Tg; Math5 KO animals the extent of fasciculation is roughly correlated with the number of surviving RGCs (data not shown). Thus, isotypic interactions are likely to be critical for proper pathfinding and fasciculation of the transgene-rescued RGCs. These pioneering effects are not limited to intraretinal pathfinding, as knots of tangled fibers were apparent behind the retinas of rescued Crx>Math5 Tg and BAC animals (Fig. 10). These defects appear to be resolved by apoptosis, although a small number of ganglion cells survive, and are likely to make synaptic connections in the brain (Triplett et al., 2011). Our results highlight the robust pathfinding mechanisms that operate in the retina, and the strong homeostatic mechanisms that balance the ratio of diverse cell types during development.

• Gain-of-function test of Math5 bHLH factor action in mice

• BAC and conventional Crx>Math5-ires-Cre transgenes, with Crx>Cre controls

• Temporal shift in retinal ganglion cell (RGC) birthdates but no change in overall time envelope

• Math5 is not the sole determinant of RGC fate competence

• Heterochronic rescue of RGC agenesis phenotype in Math5 KO mice, with axonal pathfinding defects