Staggered cell-intrinsic timing of ath5 expression underlies the wave of ganglion cell neurogenesis in the zebrafish retina

The vertebrate central nervous system (CNS) arises during development from specialized neuroepithelial progenitor cells, or neuroblasts. These cells first proliferate to generate the appropriate number of progenitors for each CNS region. Over time, neuroblasts eventually stop dividing and start the process of neurogenesis: that is, the series of developmental programs that causes the cell to exit the cell cycle and differentiate by assuming the phenotype of a mature neuron (Edlund and Jessell, 1999). We will refer to the switch from proliferation to differentiation as the start of neurogenesis, and to the molecular mechanism as `activation of the neurogenic program'. The time at which a given neuroblast activates neurogenesis is carefully regulated, as premature or tardy cell cycle exit can lead to pathological over- or underproduction of neurons (Chenn and Walsh, 2002; Zechner et al., 2003). Moreover, the relative timing of neurogenesis may be a key determinant of cell fate in many regions of the CNS (Edlund and Jessell, 1999; Livesey and Cepko, 2001). Despite the functional importance of neurogenic timing, little is known about how it is specified. Both cell-extrinsic and cell-intrinsic mechanisms have been implicated, but their relative importance is not known (Durand and Raff,2000; Perron and Harris,2000; Zechner et al.,2003; Edlund and Jessell,1999).

In the retina, neurogenesis spreads through the field of progenitors following a precise spatiotemporal pattern, the essential features of which are conserved across vertebrates (reviewed by Vetter and Brown, 2001). Retinal ganglion cells (RGCs), the first cell type to differentiate in the vertebrate retina, initially form a small patch adjacent to the site where the optic stalk attaches to the optic cup. Differentiation then progresses in a manner that, while varying somewhat between species, generally fills first the central retina and then the more peripheral retina in an orderly fashion that resembles an advancing wave front. This feature of retinal differentiation allows the timing of neurogenesis to be predicted from retinal location, an important experimental advantage. In this study we use the zebrafish retina as a model to understand how neuroblasts time their decision to become neurogenically active.

Expression of the proneural gene atonal-homologue 5(ath5; atoh7 – Zebrafish Information Network) is closely associated with the activation of retinal neurogenesis in all vertebrates. The gene encodes a basic helix-loop-helix transcription factor expressed in a wave-like pattern that prefigures the wave of RGC genesis(Masai et al., 2000). Retinoblasts that express ath5 during this wave do so just after their final mitosis; immediately thereafter they begin to differentiate as RGCs (Yang et al., 2003). In the absence of functional Ath5, these cells either ectopically re-enter the cell cycle or fail to exit the cell cycle(Kay et al., 2001), causing both a failure of RGC genesis and an overall delay in the formation of the first retinal neurons (Brown et al.,2001; Kay et al.,2001; Wang et al.,2001). These findings indicate a requirement for ath5 in activating neurogenesis.

As yet, little is known about the control of ath5 expression. A signal from the optic stalk, presumably FGF(Martinez-Morales et al.,2005), appears to induce the first patch of ath5-expressing cells (Masai et al., 2000) (but see Stenkamp and Frey, 2003). How ath5 spreads from that initial patch to cover the rest of the retina is not known, but there is a dominant hypothesis predicting the cellular and molecular mechanisms controlling this process (Amato et al., 2004; Hsiung and Moses, 2002; Jarman, 2000; Kumar, 2001; Malicki, 2004; Neumann, 2001). The hypothesis originated with the observation that vertebrate retinal neurogenesis is reminiscent of the wave-like progression of neurogenesis across the Drosophila eye field. In the fly, the ath5 ortholog atonal (ato) is expressed in a stripe just ahead of the morphogenetic furrow, which contains the differentiating photoreceptors. Both the ato stripe and the furrow advance across the eye imaginal disc due to Hedgehog (Hh) secretion by newborn photoreceptors (reviewed by Kumar, 2001). Hh triggers expression of ato in progenitor cells ahead of the furrow; Ato in turn causes formation of the next group of photoreceptors, which secrete their own Hh, thus forming a self-propagating wave that spreads by sequential induction of new neurons. Loss of ato function blocks photoreceptor formation, thereby removing the cellular source of Hh and bringing the wave to a halt (Jarman et al.,1995).

By analogy with the mechanism in Drosophila, it has been hypothesized that signals derived from newborn RGCs, particularly Sonic hedgehog (Shh), might drive the ath5 wave. In support of this idea, shh is expressed by newborn RGCs, and shh expression spreads across the zebrafish retina in what appears to be a self-propagating wave(Zhang and Yang, 2001; Neumann and Nüsslein-Volhard,2000). Together, these findings have led to a model predicting that Hh molecules, released by RGCs, should drive progression of the ath5 wave and hence the wave of RGC differentiation. We call this, in short, the `sequential-induction' model.

The central prediction of this model, that continuing production of RGCs should require the presence of earlier-born RGCs, has been tested using explant cultures, but results have been conflicting(Masai et al., 2000; McCabe et al., 1999). Here, we devised in-vivo tests of the sequential-induction model, using embryological manipulations and mutant analysis in zebrafish. Our findings reveal that cell-intrinsic factors are sufficient to activate neurogenesis in the zebrafish retina, but also that cell-cell signals may act earlier in development to establish these cell-intrinsic factors or to modulate their activity in order to bring about retinotopic differences in the timing of neurogenesis.

We used wild-type zebrafish of the TL strain. The following mutant and transgenic strains were used: lak^th241; syu^t4 (Tübingen Stock Center); Tg(Pax6DF4:mGFP)^s220(Kay et al., 2004); ath5:GFP (Masai et al.,2003); Tg(Brn3c:mGFP)^s273t(Xiao et al., 2005); Tg(H2AF/Z:GFP)^kca66(Pauls et al., 2001).

In order to accurately assay wave progression, we developed a protocol that ensured developmental synchrony across individuals. Embryos were raised at low density (no more than 50 embryos/100 mm Petri dish or 20 embryos/35 mm dish)at 27°C. All times post-fertilization reported here (aside from those cited in other works) refer to time at 27°C. For some experiments, ath5:GFP carriers were raised at 24°C from 12 hours post-fertilization (hpf). These embryos were staged either by counting somites or by adjusting to hpf at 27°C based on the extent of ath5 wave progression. As a result of raising the embryos at 27°C (rather than 28.5°C), the onset and progression of the ath5 wave in our experiments was slightly later than previously reported(Masai et al., 2000; Masai et al., 2005). In control experiments (not shown), we reared embryos at 28.5°C and found that the timing of the wave was identical to that previously reported.

Embryos were treated with phenylthiourea (PTU; 0.2 mM) to inhibit pigmentation and fixed in 4% paraformaldehyde/1 × PBS overnight at 4°C. Whole-mount immunostains were performed as described(Kay et al., 2001) using the following primary antibodies: Mouse zn5, zn8 and zpr1 (Oregon monoclonal bank); mouse anti-Hu (Molecular Probes); rabbit anti-GFP (Molecular Probes). Secondary antibodies made in goat (Molecular Probes) were conjugated to Alexa488, Alexa546 and Alexa405. Images were collected using a BioRad confocal microscope and were processed with Image J and Adobe Photoshop.

An ath5 antisense DIG-labeled RNA probe of ∼700 bp was made for in-situ hybridization by cloning the ath5 cDNA(Kay et al., 2001) into pCS2,cutting with BamHI, and transcribing with T7 polymerase. This probe also detected the ath5:GFP transgene mRNA due to the inclusion of the ath5 3′ UTR in the transgene construct. To identify lak mutants, ath5 staining was followed by either RFLP mapping (Kay et al., 2001) or by immunofluorescent labeling with the anti-Hu antibody (not shown). Templates for synthesizing patched1 and patched2 riboprobes were a gift of J. Eisen (Oregon). Brightfield images were collected using a CCD camera (Spot).

Embryos from a lak/+ incross were used to generate chimeras by transplantation of blastula cells at the 1000-cell stage, as described(Ho and Kane, 1990; Kay et al., 2004). At 55 hpf,hosts were fixed and stained with zn5 antibody and streptavidin:Alexa 546(Molecular Probes) to reveal donor cells. Donors were genotyped at the lak locus by RFLP (Kay et al.,2001). The absence of zn5 expression identified lakmutant hosts.

Donor embryos hemizygous for ath5:GFP were labeled by injection at the 1-4 cell stage with rhodamine- and biotin-dextran amine (RDA/BDA) in 120 mM KCl (5% w/v). Host embryos (also ath5:GFP/+) were uninjected siblings. Cells were removed from the donor retina using a glass micropipette attached to a microsyringe drive (Stoelting Co.). Transplants were begun when the hosts were ∼24 hpf and continued until ∼26 hpf. In order to ensure that GFP⁺ cells were not transplanted accidentally, cells were never taken from the ventronasal patch itself, but rather from more dorsal regions of nasal retina. We confirmed by in-situ hybridization that the ath5:GFP transgene was not expressed there at this age (not shown). Grafted cells did not begin expressing GFP for several hours, further suggesting (as GFP folds and becomes fluorescent quickly) that GFP mRNA was not present at the time of transplant. After transplantation, embryos were examined periodically for GFP expression until 58 hpf, using a Zeiss Axioskop II microscope and a 20 × air or 40 × water-immersion lens. Further images were collected from live fish using the BioRad confocal microscope. Depending on the exact location into which the donor cells settled, the difference between the predicted differentiation time of the original location of the donor cells and their new location could be quite small. In order to increase this time window, and thereby make more of the transplants informative, we delayed development of the hosts by maintaining them at 24°C starting at 12 hpf. This delayed start of the host wave relative to donors by ∼5 hours.

RDA/BDA-labeled ath5:GFP hemizygous embryos were used as donors;age-matched TL (wild-type) embryos were used as hosts. Retinoblasts were removed (as described above) from the central or temporal retina and placed into the head mesenchyme or brain ventricles of a host. Immediately after transplant, and at various times thereafter until 58 hpf, the live hosts were examined for the presence of GFP expression. Importantly, no grafts expressed GFP at the time of transplantation. Donors ranged in age from 20 somites to 29 hpf, depending on the experiment.

Embryos at the 22-24-somite stage, or 35 hpf as a positive control for ath5 expression, were homogenized in Trizol reagent (Gibco). Total RNA was isolated according to the manufacturer's instructions. Two separate RNA samples were prepared for each time point, yielding identical results. First-strand cDNA for ath5 or cdc16 (a ubiquitously expressed positive control) (A. M. Wehman and H.B., unpublished) was synthesized (Promega reverse transcriptase) using the gene-specific primers 5′-TTTCGTAGTGGTAGGAGAAAG for ath5 and 5′-TCCAACACAGAGGACACGAT for cdc16. A 300 bp ath5 PCR product was generated using the `RFLP' primers described(Kay et al., 2001). A ∼200 bp cdc16 PCR product resulted using the primers 5′-CATGGTTTGCTGTTGGATGT (forward) and 5′-GGCCTGGTCATGTTCACTCT(reverse).

The ablation method and equipment are described by Roeser and Baier(Roeser and Baier, 2003). In one set of experiments, animals were at the 22-28 somite stage; in another set they were at 24-26 hpf. A single Pax6DF4:mGFP embryo was transferred to the agarose-coated lid of a 35 mm Petri dish. The weight of the embryo caused it to lie on its side, eye up. The liquid level (embryo medium + 0.02%Tricaine as anesthetic) was adjusted so that the embryo was barely submerged. The ventronasal region of one eye was irradiated with the laser until GFP fluorescence in the targeted area was thoroughly quenched (∼1-2 minutes/embryo). Following staining for ath5, we verified that ablations had indeed killed the ventronasal cells by locating pyknotic cells using DIC optics.

Dechorionated TL embryos were treated with cyclopamine or vehicle (DMSO or methanol), fixed and stained with ath5, ptc1 or ptc2riboprobes. Cyclopamine was obtained from Toronto Research Chemicals and by the generous gift of Dr J. K. Chen (Johns Hopkins/Stanford). We tested dosages between 100-400 μM, but for all experiments shown and quantified here we used cyclopamine at 200 μM, which has been shown by real-time RT-PCR analysis to be optimal for blocking all Hh dependent transcription(Wolff et al., 2003).

The position of the ath5 or RGC wave front was determined by examining stained embryos (labeled with either the ath5 riboprobe or with zn5/anti-GFP antibodies) on a Leica MZ-FLIII dissecting microscope or the Zeiss compound microscope using 10 × or 20 × objectives. Each animal was scored as belonging to one of four categories: (1) wave front in ventronasal retina; (2) wave front in central retina; (3) wave front in temporal retina; or (4) wave over (ath5 expression only in the ciliary margin or RGCs evenly filling the GCL). Because vehicle-treated animals from cyclopamine experiments (both 13 and 25 hpf groups) were indistinguishable from untreated wild-type syu siblings, we pooled the data from these groups into a single `wild-type' category.

The wave of ath5 expression and RGC differentiation in zebrafish occurs during the second day post-fertilization, with ath5 leading RGC differentiation by several hours (Hu and Easter, 1999; Masai et al., 2000). In order to formulate a sensitive assay for progression of the neurogenic wave, we developed a protocol for synchronizing development of zebrafish larvae such that all the wild-type embryos at a given time point showed essentially identical ath5 expression patterns when stained by in-situ hybridization. Staining up to 50 embryos at a time, the position of the ath5 wave front varied by only a few cell diameters across individuals in a given experiment (not shown), and was also very consistent across experiments (see below). Our protocol included lowering the temperature to 27°C to slow down development (see Materials and methods). Under these conditions, ath5 expression commenced in the ventronasal patch consistently at 30 hours post-fertilization (hpf). When the fish were raised at 28.5°C, ath5 was first detectable by in-situ hybridization at 25 hpf, as reported previously(Masai et al., 2000). Unless otherwise noted, all fish ages given here were measured at 27°C.

Fig. 1 shows both the ath5 and RGC differentiation waves, revealed either by staining for ath5 mRNA (Fig. 1A-C);staining for the zn5 antigen (Trevarrow et al., 1990), a specific RGC marker(Fig. 1G-L); or by expression of an ath5:GFP transgene (Masai et al., 2003) that faithfully recapitulates the spread of ath5 mRNA across the retina (Fig. 1D-F; compare with Fig. 1A-C). The ath5 wave starts in the ventronasal retina(Fig. 1A,D) with an initial cluster of cells known as the ventronasal patch(Hu and Easter, 1999), and spreads from there to fill the rest of the nasal retina and then a small patch of the central retina (Fig. 1B,E). Next the wave begins spreading in a central-to-peripheral manner, filling increasingly more peripheral regions of dorsal and temporal retina. Ventrotemporal retina is the last to express ath5 and to make RGCs (Fig. 1C,F)(Hu and Easter, 1999).

The sequential-induction model predicts that spread of the ath5wave should require RGC-derived signals. To test this hypothesis, we used the zebrafish lakritz (lak) mutant, in which a null mutation in the ath5 gene causes complete elimination of RGCs(Kay et al., 2001). We found that the spread of ath5 expression was normal in lak mutants(Fig. 2 and data not shown; n>10 mutants and >20 wild-type siblings for each time point). Thus, neither RGCs nor the ath5 gene itself are essential for driving the ath5 wave.

We next tested whether signals from earlier-born RGCs are essential for driving the RGC differentiation wave. Because the lak/ath5 gene acts cell-autonomously in RGC specification(Fig. 3D-H), we were able to test this prediction by generating lak/wild-type chimeras in which a small number of wild-type cells were situated in an otherwise mutant (and thus RGC-free) retina (see model, Fig. 3C). We generated chimeras at the 1000-cell stage(Ho and Kane, 1990) and assayed RGC differentiation at 55 hpf using the zn5 antibody. Regardless of the host's genotype, retinal clones derived from lak mutant donors never gave rise to zn5⁺ RGCs (n>50 clones; Fig. 3G,H), while clones from wild-type donors always did. Of the wild-type-into-lak chimeras(n=9 eyes with multiple clones per eye), we found three retinae with central or temporal clones that were well isolated from both the optic stalk and other RGC-containing wild-type clones. Contrary to the sequential-induction model's predictions, the isolated clones gave rise to RGCs in all three cases (Fig. 3E,F and data not shown). Indeed, clones in the temporal retina could produce RGCs even when the rest of the retina was completely devoid of RGCs (Fig. 3F). Thus, RGC formation in later-differentiating retinal regions does not depend on prior RGC formation in earlier-differentiating regions.

We next asked whether the ath5-expressing retinoblasts themselves might signal to neighboring retinoblasts, inducing them to express ath5 (Masai et al.,2000). If this version of the sequential-induction model is correct, then removal of the ventronasal patch before ath5 expression should prevent relay of ath5-inducing signals, thereby blocking the wave. To test this prediction, we laser-ablated the neuroblasts of ventronasal retina at times ranging from 22 somites to 26 hpf. To guide the ablations we used a transgenic line that expresses GFP in all retinal neuroblasts(Pax6-DF4:mGFP^s220)(Kay et al., 2004). Photobleaching of GFP protein allowed us to precisely delineate the retinal region targeted in each larva (Fig. 4A-B).

We first tested whether the laser could efficiently kill cells in the targeted ventronasal region and prevent initiation of ath5expression. Only one eye was treated; the other served as an internal control. Fixing and staining for ath5 at 30 hpf, when expression is normally confined to the ventronasal patch, we found that all of the intact eyes showed the expected pattern, whereas none of the ablated eyes showed any ath5 expression (Fig. 4E-F; n=5 for 24-26 hpf ablations and n=6 for 22-somite ablations). Most of the tissue in the laser-targeted region was dying or dead, as evidenced by the pyknotic morphology of the cells(Fig. 4C-D)(Li et al., 2000). To ensure that the ablations eliminated cells before the onset of ath5expression, a control group of age-matched siblings was sacrificed immediately after the ablations. None showed ath5 expression by in-situ hybridization (n>30) or by RT-PCR (age-matched progeny of a different cross; not shown). We conclude that the laser treatment was sufficient to kill most or all of the ventronasal retinoblasts, thereby preventing the normal onset of ath5 expression and formation of the ventronasal patch.

We next tested whether elimination of the ventronasal patch would affect spread of ath5 expression through the rest of the retina. Ablations were performed as before, but this time we assayed for ath5 mRNA expression at 33 hpf. In ablated retinas, the ventronasal portion of the ath5 domain was absent, indicating successful removal of the ventronasal patch. However, outside the laser-targeted region, ath5expression was similar to that seen in unablated retinas(Fig. 4G-H; see Fig. S1 in the supplementary material). In all treated animals (n=27 at 24-26 hpf; n=12 at 22-28 somites), the extent of ath5 wave progression into the central retina was the same in both the ablated and unablated eyes. The ath5 wave evidently can skip over the ablated region, and begin instead in more dorsal regions of the nasal retina, in the central retina, or even in the temporal retina, depending on the size of the ablated domain (see Fig. S1 in the supplementary material). This finding indicates that, if ath5⁺ cells do in fact generate signals that induce ath5 expression in neighboring retinoblasts, such signals are not necessary for propagating the ath5 wave across the retina.

The results of our experiments so far suggested that cell-cell signaling is not as essential for triggering ath5 expression as had previously been assumed. We therefore wondered whether retinoblasts would express ath5 even when removed entirely from the normal retinal signaling milieu. To test this possibility, we devised a method for transplanting retinoblasts from a labeled donor retina into non-retinal tissues of an unlabeled host (note that this experiment is quite different from the lak/wild-type blastula transplants; see Materials and methods). If retinal signals are required to trigger ath5 expression,heterotopically transplanted retinoblasts should fail to express ath5. Donor cells (labeled with RDA and carrying the ath5:GFP transgene) were removed from the temporal retina at various times between the 20-somite stage and 29 hpf. This latest time point is still∼5 hours before the first temporal retinal cell expresses mRNA for the ath5:GFP transgene (data not shown). Donor cells were placed in the telencephalic or mesencephalic ventricular space of a wild-type,non-transgenic host brain; some transplants were also placed in the head mesenchyme. By 50 hpf, grafted cells in both the ventricular space(Fig. 5A-B) and the head mesenchyme (not shown) expressed GFP. Some of these cells had clearly differentiated into RGC-like neurons, as they possessed long axons tipped by growth cones (Fig. 5B). To ensure that the transplants were done before the onset of ath5expression, we used donors at the 20-24-somite stage. RT-PCR experiments showed that ath5 is not yet expressed at 24 somites (not shown). Three of the seven grafts from 20-24-somite donors expressed ath5:GFPby 50 hpf. These results demonstrate that retinoblasts can express ath5, and possibly even assume the RGC fate, even when stripped out of the retinal neuroepithelium and placed in a variety of ectopic locations.

We next tested whether the transplanted retinoblasts could undergo neuronal differentiation. We found that heterotopically grafted retinoblasts in the brain, brain ventricles or head mesenchyme, as well as homotopically grafted cells in the retina, could differentiate as both RGCs and photoreceptors, as judged by expression of cell-type-specific markers (see Figs S2, S3 in the supplementary material). Not every grafted cell was able to differentiate successfully. For example, only ∼15-35% of cells transplanted to the host retinal ganglion cell layer (GCL) successfully expressed RGC markers by 72 hpf(see Fig. S2 in the supplementary material). This observation suggests that certain cell-cell signals, perhaps those provided by being part of an intact neuroepithelium, may be permissive for the activation of neurogenesis and/or for differentiation. Nevertheless, intra-retinal signals do not appear essential for activation or execution of the retinal neurogenic program, as both ath5 expression and cell-type-specific marker expression can occur in the absence of such signals.

Signaling between retinoblasts, while not necessary for ath5expression, might be required to produce accurate timing of the ath5wave. Alternatively, each retinoblast might be intrinsically programmed to activate neurogenesis on a staggered schedule, depending on its retinal position. To distinguish between these possibilities, we performed two retinoblast transplant experiments.

We first took nasal retinoblasts from RDA-labeled ath5:GFPcarriers at 24-26 hpf and placed them into the temporal region of a host ath5:GFP retina. GFP expression was then monitored in live fish at regular intervals (generally ∼1-2 hours), until well after the wave was complete (58 hpf). We found that these nasal-derived cells in temporal retina either began expressing ath5 while the host wave was still confined to nasal retina (Fig. 5C,D;31%; n=4/13), or failed to express ath5 at all by 58 hpf(n=9/13). None of the grafted nasal cells expressed ath5 on a temporal schedule. This result supports the notion that intrinsic factors,not signals from the advancing wave front, determine the timing of ath5 expression. Although only 30% of nasal-into-temporal grafts succeeded in expressing ath5:GFP by 58 hpf, this rate of differentiation was typical for grafts into the GCL that were made without regard to the retinal sector from which the cells were taken from or where they were placed (see Fig. S2 in the supplementary material). It is therefore unlikely that the temporal environment inhibited nasal donor cells from differentiating.

We next asked whether relative timing differences would be maintained even in the absence of retinal signals. Using RDA-labeled ath5:GFPcarriers as donors, retinoblasts were removed from either the central or temporal retina at 26-29 hpf and transplanted into the head of wild-type(non-transgenic) hosts. This time point was several hours prior to the onset of transgene mRNA expression in the central or temporal retina (not shown). Each donor gave rise to two hosts, one carrying the donor's central retinal cells and one carrying its temporal cells. These live hosts were then checked periodically for ath5:GFP expression in the grafted cells. In each pair of hosts, the one carrying the central graft always expressed ath5:GFP first. In fact, all the central grafts had begun expressing ath5 before the first temporal graft did so(Fig. 5B; n=5 central, n=5 temporal). Only one graft failed to express ath5 at all(n=11 hosts). Thus, progenitors do not require retinal signals after 26 hpf (27°C) to maintain the neurogenic timing conferred by their original retinal position. These experiments suggest that the spatiotemporal pattern of the RGC differentiation wave might arise because retinoblasts are intrinsically programmed, before the wave, to begin neurogenesis at different times.

The programming of neurogenic timing could be accomplished by some of the same patterning mechanisms that create anteroposterior (AP) or dorsoventral(DV) differences in gene expression across the retina(McLaughlin et al., 2003). To test this idea, we perturbed DV patterning of the diencephalon by disrupting Hh signaling. Hh-family molecules are expressed in the ventral midline of the diencephalon and have a well-established role in patterning the eye (reviewed by Amato et al., 2004). Shh is also expressed within the retina during RGC neurogenesis, and this retina-derived Hh has been predicted, as part of the sequential-induction model, to regulate ath5 expression (reviewed by Amato et al., 2004; Neumann, 2001). Thus, it was important for us to separate the effects of early, midline Hh signaling from later, intra-retinal Hh signaling.

We first examined ath5 expression in the shh null mutant syu^t4 (Schauerte et al., 1998). We found that the spatial pattern of the ath5wave was unaffected in these mutants, but the timing of wave progression was dramatically altered (Figs 6, 7). Initiation of the wave in ventronasal retina was normal at 30 hpf(Fig. 6A,E; Fig. 7A). Subsequently, the ath5 wave spread to the central and temporal retina as in wild type,but it did so on a delayed schedule. For example, whereas the wild-type wave reached the temporal retina at 41 hpf, in syu mutants the ath5 wave was confined to the central retina at 41 hpf(Fig. 6C,G; Fig. 7C). The wave did not reach the temporal retina until 50 hpf in the mutants, indicating that temporal retinoblasts were delayed in expressing ath5 by almost 10 hours (Fig. 6D,H; Fig. 7D). Loss of shhfunction thus did not block the ath5 wave, but it did have a substantial effect on the time of neurogenic activation.

In order to determine whether this phenotype results from loss of midline-or retina-derived Shh, we defined the time window during which shh is necessary for timely wave progression. To do this, we employed the drug cyclopamine, a small molecule inhibitor of the Hh receptor Smoothened(Chen et al., 2002). Cyclopamine treatment fully inactivated the Hh signaling pathway at the dose we used (200 μM) (see Wolff et al.,2003), as expression of the Hh target genes patched1 and patched2 was severely reduced or eliminated in all tissues(Fig. 8A-B and data not shown; n=20 for each probe). First, we specifically blocked retina-derived Hh signaling by starting cyclopamine treatment at 25 hpf, after specification of the retinal DV axis but before ath5 expression. We saw no effect on the progression of the ath5 wave(Fig. 7A-D; Fig. 8C,E; n=∼20 treated embryos/time point). However, when we gave cyclopamine at 13 hpf,during patterning of the DV axis but before shh was expressed in the retina, we observed a delay in ath5 expression similar to that seen in syu mutants (Fig. 8D; Fig. 7D-E; n=9). Together, these results suggest that early, midline-derived shh signaling is required to ensure the subsequent timely expression of ath5 by retinoblasts.

To determine whether the altered timing of the ath5 wave also changes the timing of RGC differentiation, we analyzed the expression of two different RGC markers following 13 and 25 hpf cyclopamine treatment. Indeed,we found that treatment at 13 hpf, but not 25 hpf, delays expression of RGC-specific markers (Fig. 9). Thus, while loss of Hh signaling during the ath5 wave has no effect on ath5 expression or RGC differentiation, blockade of early,midline-derived Hh signals causes a delay in both the ath5 and RGC differentiation waves. This finding raises the possibility that axial patterning mechanisms, such as midline Hh signaling, might be involved in creating spatial differences in the timing of neurogenic activation.

In this study we have investigated how neuroblasts in the vertebrate retina activate neurogenesis – how they switch at the appropriate time from a proliferative to a neurogenic state. Because the ath5 proneural gene is required for this switch in the earliest-differentiating retinoblasts(Kay et al., 2001; Yang et al., 2003), we looked into the cellular and molecular factors that determine the timing of ath5 expression. We found that retinoblasts do not require signals from the retinal environment in order to correctly time ath5expression. We eliminated RGC-derived signals; signals derived from ath5-expressing cells; and finally signals from all retinal cells. In each case at least the relative timing of ath5 expression was normal. These experiments indicate that retinoblasts may possess a cell-intrinsic program that activates neurogenesis. The intrinsic program appears to be at least partially established by the patterning activity of midline-derived Shh. Our findings may explain the origins of the wave of RGC differentiation that sweeps across the retina: Because the timing of neurogenesis is pre-specified and staggered according to retinal position, the wave may emerge through the collective timing decisions of individual neuroblasts. In this model,progressive cell-cell signaling is not required to drive the ath5wave, although such a mechanism may still influence retinoblast differentiation and/or cell fate selection.

We show that central retinoblasts are already competent to express ath5 independently of signals from ventronasal retina by 22 somites(∼20 hpf at 28.5°C), and temporal retinoblasts are independent of retinal signals by 20 somites (∼19 hpf at 28.5°C). When the retina is removed at 18 somites and the nasal and temporal halves are cultured separately, the temporal explant is delayed or fails to express ath5(Masai et al., 2000). This result, together with ours, may indicate that an important intra-retinal signaling event occurs between the 18- and 20- somite stages (a time window of about 1 hour). Alternatively, explanting the retina may remove a source of extra-retinal signals, such as Shh, that are required for timely differentiation of temporal retinoblasts. Regardless of the precise cellular mechanism through which the timing of ath5 expression is set, our results are significant for showing that the time window during which signals act to influence neurogenic activation is substantially earlier than previously suspected.

If retinoblasts have an intrinsic tendency to activate neurogenesis at a particular time, there must be some mechanism that establishes the neurogenic timing for each cell. What is this mechanism, and when does it act? We find that retinal location is a key variable, implying that an asymmetric spatial signal might set the neurogenic timer in order to impart location-specific timing information. A good candidate for such a signal is Shh derived from the ventral midline of the diencephalon. Previously, midline Hh signals were reported to be required for ath5 expression(Stenkamp and Frey, 2003). Here we extend this important finding by showing that Shh acts between 13 and 25 hpf not as an absolute prerequisite for ath5 expression, but rather to ensure timely expression of ath5 during the wave, hours later. Before 25 hpf, shh and its relative, tiggy-winkle hedgehog (twhh), are expressed in the diencephalic ventral midline but not in the retina (Ekker et al., 1995). In fact, shh expression is not detectable in the retina before the ath5 wave has already reached the temporal retina (Masai et al., 2005). These combined results suggest that the midline source of Hh signals, in addition to patterning the DV axis of the eye(Amato et al., 2004; McLaughlin et al., 2003), also seems to have a role in patterning the timing of retinal neurogenesis.

In a recent study, Masai et al. (Masai et al., 2005) narrowed the time window of Hh action even further by using forskolin, a potent activator of protein kinase A (PKA), which in turn antagonizes Hh signaling. In keeping with our results, these authors show that the ath5 wave can only be blocked when forskolin treatment occurs before, but not during, wave progression (21-25 hpf at 28.5°C)(Masai et al., 2005). Because forskolin causes a more severe ath5 phenotype than either cyclopamine or various genetic manipulations that block Hh signal transduction, it appears that multiple signaling pathways impinge on PKA to regulate the subsequent timing of ath5 expression.

Loss of early Hh signaling delays both the ath5 and RGC differentiation waves. By contrast, retina-derived Hh appears not to be required for ath5 expression or RGC differentiation. In Xenopus, chick and mouse, reducing retinal Hh signaling has effects on RGC genesis similar to those we describe(Dakubo et al., 2003; Perron et al., 2003; Zhang and Yang, 2001). By contrast, it had previously been reported that retinal Hh signaling is essential for RGC formation in zebrafish(Neumann and Nüsslein-Volhard,2000), a finding that we could not confirm. It is not clear why the results of the two studies differ. In both experiments, cyclopamine was used to block Hh signaling specifically during the period of RGC genesis(starting at ∼25 hpf). We show, using ptc1/2 staining on larvae treated in parallel with those tested for RGC formation, that our drug was effective at eliminating Hh signaling. To control for drug quality, we obtained cyclopamine from two independent sources and dissolved it in two different solvents – each permutation of the experiment gave identical results.

Another line of experimentation in zebrafish recently arrived at the conclusion that short-range Hh signaling is required within the retina for progression of the ath5 wave(Masai et al., 2005). These authors injected embryos with antisense morpholino oligonucleotides that abrogate Shh and Twhh function (Hh-MO). Cells from Hh-MO embryos were then transplanted at the blastula stage into wild-type hosts. Hh-MO cells did not express ath5 at 33 hpf (28.5°C), the latest time point studied(equivalent to 40 hpf at 27°C), despite being surrounded by wild-type cells from the host. Host cells adjacent to Hh-MO clones also occasionally failed to form RGCs. While this finding is consistent with a role for short-range Hh signaling in the progression of the ath5 wave, other interpretations are possible. In particular, it is unclear if analysis of donor-derived clones in the eye would allow a strict test of whether Hh signaling acts in the eyecup to regulate ath5. As the Hh-MO clones lack Shh/Twhh activity throughout their developmental history, the effects on ath5 expression could reflect a requirement for Hh signaling before eye evagination. Even cell-non-autonomous effects on neighboring cells are plausible given the close juxtaposition of cells in neural plate that will later give rise to the eye (Varga et al.,1999). Furthermore, if Hh signaling propagates ath5expression between cells, then some of the Hh-MO donor cells near Hh-expressing host cells should have expressed ath5, which apparently was not seen (Masai et al.,2005). While it is possible that Hh signals act within the eye to affect ath5 expression (perhaps by an autocrine mechanism), further studies will be required to resolve this question definitively. Retina-derived Hh clearly has important functions in axon guidance (see Fig. 9) and in the differentiation of later-born cell types, such as bipolar cells and photoreceptors (Shkumatava et al.,2004; Stenkamp and Frey,2003; Stenkamp et al.,2000). In any case, our studies suggest that Hh is unlikely to be required as a sequential inducer of ath5 expression.

Our results naturally raise the question of how retinoblasts manage to translate extracellular signals into timing information that will be used hours later. One possibility is that Shh and/or other time-specifying signals might regulate cyclin-dependent kinase inhibitors (CDKIs). In several types of neural progenitor cells, including retinoblasts, gradual accumulation of CDKIs, particularly those of the Kip family, appears to be a mechanism for measuring proliferative time (reviewed by Durand and Raff, 2000; Edlund and Jessell, 1999; Ohnuma and Harris, 2003). Signaling pathways could therefore modulate the subsequent timing of neurogenesis by influencing the initial levels or activities of CDKIs. There is ample evidence for genetic interactions between basic helix-loop-helix proteins (including Ath5) and Kip CDKIs(Farah et al., 2000; Kitzmann and Fernandez, 2001; Ohnuma et al., 2002; Vernon et al., 2003). Thus,CDKIs could provide the link between early timing specification signals and subsequent ath5 expression. Although the ability of Hh to influence CDKI expression is yet untested, PKA activity can influence expression of the Kip gene p27 in the zebrafish retina(Masai et al., 2005). As Hh derived from the ventral diencephalic midline can act over long distances to regulate other cell cycle genes (Ishibashi and McMahon, 2002), it is plausible that midline Hh signaling might influence retinal CDKI expression. It will be interesting to see whether this mechanism influences the timing of neurogenesis in the retina.

To date, there have been few attempts to determine how specific neuroblast populations integrate cell-intrinsic and cell-extrinsic information for neurogenic activation. For oligodendrocyte precursors, it has been shown that an intrinsic timer operates, which is licensed by permissive extrinsic signals(Durand and Raff, 2000). Retinoblasts may use a similar cell-intrinsic mechanism to decide when to differentiate (Cayouette et al.,2003). While this work in cell culture, together with our experiments in vivo, highlights the importance of cell-intrinsic factors, RGC neurogenesis is certainly not completely cell-autonomous. First, as in oligodendrocyte precursors, there may be licensing signals – the fact that the majority of transplanted retinoblasts fail to differentiate in our experiments implies the existence of such signals. Second, cell-cell signals may affect steps of RGC differentiation downstream of the cell-intrinsic trigger that activates the neurogenic program (and thus presumably downstream of ath5 expression).

Third, cell-cell signaling may be necessary to counteract the effects of the intrinsic timer. Our results reveal the existence of an intrinsic tendency toward neurogenesis, but there must be some signal that opposes this tendency– otherwise no retinoblasts would be reserved to make later-born cell types. Similarly, to ensure that the correct number of progenitors become RGCs, there may also be signals that promote neurogenesis over and above the basal level provided by the intrinsic timer. There is strong evidence that Notch, perhaps in concert with other signals, plays both these roles, acting in one context to promote cell cycle exit during the RGC wave(Ohnuma et al., 2002), and in another context to terminate RGC genesis behind the wave front(Silva et al., 2003). However,Notch probably accomplishes both these roles without affecting the spatiotemporal pattern of differentiation(Ohnuma et al., 2002; Scheer et al., 2001; Silva et al., 2003). Thus, it appears that extrinsic signals balance but do not fundamentally alter the intrinsic program underlying RGC genesis.

In the developing Drosophila retina, a progressive cell-cell signaling loop, centered on the ato and hh genes, initiates neurogenesis (Kumar, 2001). The discovery that close homologs of these genes, ath5 and shh, play a role in differentiation of the first-born neurons of the vertebrate retina led to the hypothesis that the mechanisms for initiating retinal neurogenesis are conserved between flies and vertebrates (reviewed by Kumar, 2001). Despite the appeal of this hypothesis, there have been few attempts to test whether neurogenesis spreads across the vertebrate retina by a fly-like sequential mechanism. In one such study, the peripheral portion of a chick retinal explant, dissected away from the RGC wave front and cultured separately, was still competent to generate RGCs, implying that differentiation does not require a progressive, wave-like mechanism(McCabe et al., 1999). However, a similar experiment in zebrafish yielded the opposite result(Masai et al., 2000). Although there are experimental design differences that might explain the differing results of these two studies, neither has settled the question of whether sequential induction triggers retinal neurogenesis in vertebrates. Here we have devised in-vivo tests of the fly-inspired sequential-induction model, and we have found very little evidence to support it. Our findings indicate that while hh and ato family genes may have a broadly conserved function in RGC genesis, the cellular context in which they operate differs significantly between vertebrates and insects.

We thank James K. Chen (Stanford) for providing cyclopamine and Judith Eisen (Oregon) and Ichiro Masai (RIKEN) for providing plasmids and fish strains. We also thank E. Ober, C. J. Neumann, and J. K. Chen for helpful comments. The work was supported by an NSF predoctoral fellowship (J.N.K.), by the NIH (EY13855 to H.B. and EY014167 to B.L.) and by a David and Lucille Packard Fellowship (H.B.).