Polarization and orientation of retinal ganglion cells in vivo
© Zolessi et al.; licensee BioMed Central Ltd. 2006
Received: 2 August 2006
Accepted: 13 October 2006
Published: 13 October 2006
Skip to main content
© Zolessi et al.; licensee BioMed Central Ltd. 2006
Received: 2 August 2006
Accepted: 13 October 2006
Published: 13 October 2006
In the absence of external cues, neurons in vitro polarize by using intrinsic mechanisms. For example, cultured hippocampal neurons extend arbitrarily oriented neurites and then one of these, usually the one nearest the centrosome, begins to grow more quickly than the others. This neurite becomes the axon as it accumulates molecular components of the apical junctional complex. All the other neurites become dendrites. It is unclear, however, whether neurons in vivo, which differentiate within a polarized epithelium, break symmetry by using similar intrinsic mechanisms. To investigate this, we use four-dimensional microscopy of developing retinal ganglion cells (RGCs) in live zebrafish embryos. We find that the situation is indeed very different in vivo, where axons emerge directly from uniformly polarized cells in the absence of other neurites. In vivo, moreover, components of the apical complex do not localize to the emerging axon, nor does the centrosome predict the site of axon emergence. Mosaic analysis in four dimensions, using mutants in which neuroepithelial polarity is disrupted, indicates that extrinsic factors such as access to the basal lamina are critical for normal axon emergence from RGCs in vivo.
A key step in neuronal morphogenesis is the emergence of correctly oriented axons and dendrites. The cellular and molecular mechanisms that determine how one neurite is selected to become the axon while the others become dendrites have been studied extensively in conditions where this problem is most approachable experimentally, namely in vitro . If hippocampal cells are cultured soon after their final mitotic divisions, multiple neurites emerge simultaneously at seemingly random orientations. From these young multipolar neurons, one neurite then begins to elongate preferentially, marking the beginning of polarization. It becomes the axon , and as it grows it inhibits the other neurites from becoming axons. They become dendrites instead. The inhibitory signal relies on the activities of the small GTPases Rac/Cdc42 and Rho [3, 4] and on the localized inactivation of GSK-3β [5, 6]. Proteins normally associated with the apical junctional complexes of epithelial cells, such as Par-3, Par-6 and atypical protein kinase C (aPKC) have a role in polarization in vitro. Aided by adenomatous polyposis coli and KIF3A (a kinesin superfamily protein), proteins that travel along microtubules, these apical components accumulate at the tips of growing axons. Interference with the activity of any of these proteins compromises polarization [7–9]. The centrosome, acting as a microtubule organizing center, also has a role in axon formation in vitro , and recent evidence suggests that its position determines of the site of axon emergence .
In dissociated cell cultures, neurons develop in the presence of very scarce external cues, and so must perforce break symmetry intrinsically. In vivo, however, neurons are generated within a highly oriented three-dimensional neuroepithelium. In such a situation, differentiating neurons may depend on external cues for polarization. In support of this idea, Rolls and Doe  demonstrated that, in Drosophila mutants lacking the apical junction components Par-3, Par-6 or aPKC, neurons in the central nervous system in vivo are nevertheless appropriately oriented. This study raises two important questions. First, are the dissimilarities in these results due to differences between vertebrates and invertebrates or are they due to differences between the situation in vitro and that in vivo? Second, if extrinsic cues polarize neurons in vivo, how is this done in such a way that neurons become appropriately oriented?
Retinal ganglion cells (RGCs) are an excellent model system with which to study the above questions. Restricted to a layer adjacent to the inner basement membrane, RGCs show 'typical' neuronal polarity, with basally oriented axons and apically oriented dendritic trees. More than a century ago, Ramón y Cajal made observations of the embryonic chick retina and drew differentiating 'RGCs' with bipolar morphologies, including a retracting apical process and an axon extending from the basal side . In line with this, Hinds and Hinds , in their serial electron microscopic studies of the developing mouse retina, suggested that the axons of RGCs arise from the basal process of neuroepithelial-like precursors. By injecting Lucifer yellow into differentiating RGCs in Xenopus embryos, Holt  showed that RGC axons almost always emerge from the basal pole of the cell. Halfter and Schurer  found that disruption of the inner basement membrane of the developing chick retina led to aberrant RGC axon outgrowth. Together, these studies suggest a relationship between axon orientation in differentiating RGCs and the basal surface of the neuroepithelium. However, in all of these studies the images were static and the analysis depended on the cells' having already assumed the beginnings of RGC morphology, including the formation of a primordial axon. To understand what is going on when neurons first polarize, it is essential to be able to follow single cells from their final mitosis to the time when they extend a definitive axon. Only through such studies is it possible to learn, for example, whether differentiating RGCs in vivo go through an early multipolar phase in which they put out several exploratory neurites before they stabilize one as the axon.
In the zebrafish, through the use of transgenes that drive fluorescent proteins under the control of an enhancer-promoter from the ath5 gene (atoh7) and four-dimensional (4D) microscopy, it is possible to view the differentiation of RGCs in vivo from their final mitosis at the apical surface, through to the initiation of their axonal and dendritic processes . We took advantage of these innovations to show that RGCs send out axons directly from their basal surface in the absence of other neurites emerging from the cell. Tracing the movements of different apical markers, such as junctional complex or centrosomal proteins, in these transgenic retinas revealed that these components remain in the retracting apical process while the axon extends from the opposite (basal) pole of the differentiating cell body of the RGC.
To look for extrinsic cues in polarizing RGCs in vivo, we used two mutants in which the polarity of the retinal neuroepithelium is disrupted: nagie oko (nok) and heart and soul (has) . Detailed 4D analyses of wild-type RGCs in mutant environments show that the polarization and orientation of RGCs is determined by the local orientation of the neuroepithelium, including factors such as the presence of a basal lamina. RGCs without access to either the inner or outer basal lamina during their differentiation go through a multipolar phase that precedes polarization, as they do in vitro. Our results provide strong evidence for an extrinsic influence in RGC polarization.
Zebrafish were maintained and bred at 26.5°C, and embryos were raised at 28.5°C. The mutant lines used were: nagie oko (nokm 227, a kind gift from Dr Jarema Malicki) and heart and soul (hasm 567, a kind gift from Dr Salim Abdelilah-Seyfried). Both represent null alleles of the respective genes. Two transgenic lines were generated in our laboratory: Tg(pBatoh7:gap43-gfp)cb 1('ath5:gap-gfp') and Tg(pBatoh7:gap43-rfp)cb 2('ath5:gap-rfp'). They express a fluorescent protein (enhanced green fluorescent protein (EGFP) or monomeric red fluorescent protein 1 (mRFP1), respectively) fused to the GAP43 N-terminal palmitoylation signal, under the control of the zebrafish ath5 promoter (comprising 7 kilobases of genomic sequence upstream of the ath5 start codon). For some experiments we used a transgenic line expressing a cytoplasmic form of EGFP under the control of the ath5 promoter ('ath5:gfp', a kind gift from Dr Ichiro Masai) . The ath5:gap-gfp transgenic line was crossed with carriers of both mutations used, to generate an F1 generation from which mutant embryos expressing GAP-EGFP in RGCs could be obtained, namely nokm 227× Tg(pBatoh7:gap43-gfp)cb 1and hasm 567× Tg(pBatoh7:gap43-gfp)cb 1.
Constructs used to inject into living embryos were as follows: ath5:gap-gfp, ath5:gap-rfp, GFP-zcentrin and Par3-GFP . For generating the ath5:gap-gfp and ath5:gap-rfp expression vectors, a fragment containing 7 kilobases from the 5' regulatory region of the zebrafish ath5 gene  has been subcloned upstream to either the GAP-EGFP or GAP-mRFP coding regions . The promoter and coding region were subcloned in the IsceI pBSII SK+ vector, kindly provided by Dr Jochen Wittbrodt . For the GFP-zcentrin construct, pCJW263 is a pCS2+-based plasmid in which zebrafish centrin is joined in-frame, 3' to EGFP. It was created in the following steps. The Sal I site of pCS2P+ was removed by digestion then infilling with Klenow enzyme and religation. This plasmid was cut with Bam HI, the 5' overhangs filled in with Klenow enzyme and then cut again with Xba I. This fragment was ligated with the Afe I-Xba I fragment of pEGFP-C2 (Clontech, Mountain View, CA, USA) that contains the EGFP coding region and multiple cloning sites to create the vector pCS2P+EGFPN. Zebrafish centrin was amplified by PCR using IMAGE clone 5899515 as template and these primers: 5'-TTGGATCCTCATGGCGTCCGGCTTCAGGAAAAGC-3' (forward) and 5'-TTCTCGAGGTACAGATTGGTTTTCTTCATAATCCG-3' (reverse). The PCR product was digested with Bgl II and Xho I and ligated with Bgl II- and Sal I-cut pCS2P+EGFPN. GFP-zcentrin mRNA was transcribed from the Sp6 promoter of pCJW263, linearized with Not I, using the mMessage machine in vitro transcription kit (Ambion, Austin, TX, USA). RNA was purified with the RNeasy RNA purification kit (Qiagen GmbH, Hilden, Germany).
For transient expression of fluorescent proteins, embryos were injected with either plasmid containing the gene of interest, under the control of a general (cytomegalovirus) promoter or the RGC progenitor-specific (ath5) promoter, or with mRNA transcribed in vitro. DNA injections were made into the cell at the one-cell stage, whereas mRNA injections were done into the yolk at the one-cell to four-cell stage, using a micromanipulator-mounted micropipette and a Picospritzer microinjector. The maximum volumes for injection were 2.5 nl into the cell and 5 nl into the yolk. For the ath5 promoter-driven constructs, the plasmids were injected together with meganuclease I-Sce-I at a concentration of 10 ng/μl into one-cell-stage embryos, as described . For stable transgenesis, embryos expressing the fluorescent protein at the correct location were selected and raised to sexual maturity; transgenic carriers were identified by outcrossing to wild-type fish.
Morpholinos diluted in water were injected into the yolk at the one-cell to eight-cell stage. Morpholinos used were as follows: anti-nok, translation blocking (MORPH1116, Open Biosystems); anti-slit1b, translation blocking (LDHMO2, 5'-GCTCGGTGTCCGGCATCTCCAAAAG-3', designed by L Hutson and C-BC) and anti-slit1a, splice blocking (S1ASDMO1, 5'-GAAATAAACTCACAGCCTCTCGGTG-3', designed by M Hardy and C-BC). The ideal amount to be injected was determined by analyzing a range of concentrations. We found, for the Slit1b morpholino, different responses in different genetic backgrounds, but these were corrected for by adjusting the amount injected, resulting in the same reproducible phenotypes. For the analysis of the data we took into account only embryos that had received the same relative amount of morpholino (namely, 2 to 3 ng for the ath5:gap-gfp line and 6 to 8 ng for the ath5:gap-rfp line) that did not cause obvious effects in cell survival.
Embryo processing and 4D imaging were performed as described previously . Usually, stacks about 100 μm thick, composed of sections separated by 1 μm, were taken every 5 to 10 minutes during an average period of 20 to 24 hours. To avoid damaging the embryos, we maintained the power of the lasers at a minimum (typically 12 to 20%). The 4D data thus obtained were processed and analyzed with Volocity (Improvision, Coventry, UK). Unless stated otherwise in the figure legends, the images shown are maximum-intensity projections of all or most of the confocal stack. Quantifications described in the text were made using Volocity, Openlab (Improvision, Coventry, UK) or ImageJ (National Institutes of Health) software.
Cryosections were made at 10 μm thickness from 4% paraformaldehyde-fixed, OCT-mounted zebrafish embryos. Blocking was for 30 minutes at 20 to 23°C, in 10% heat-inactivated goat serum (HIGS), 1% bovine serum albumin, 0.2% Triton X-100 in PBS. Primary and secondary antibodies were incubated for 1 hour at room temperature, diluted as described below. For whole-mount immunostaining, embryos (grown in 0.003% phenylthiourea) were fixed overnight in 4% paraformaldehyde in PBS, and all subsequent washes were performed in PBS containing 0.2% Triton X-100. Further permeabilization was achieved by incubating the embryos in 0.25% trypsin-EDTA in Hanks balanced salt solution for 15 to 25 minutes at 0°C. Blocking and antibody dilution was as for sections. Antibodies were incubated for at least 36 hours at 4°C, with occasional shaking.
The primary antibodies, diluted in the blocking solution, were as follows: Zn-5, 1/100 to 1/500 dilution (mAb anti-Ben/DM-GRASP, specific for RGCs in the differentiating neural retina; Zebrafish International Resource Center (ZIRC), Eugene, OR, USA; Zpr-2, 1/100 dilution (mAb specific for retinal pigment epithelium (RPE); ZIRC); anti-laminin 1, 1/60 dilution (poly-clonal antibody (pAb), L9393; Sigma, St Louis, MO, USA); anti-Tau 1, 1/500 dilution (pAb; Dr Itzhak Fischer); anti-aPKC-ζ, 1/250 to 1/500 dilution (pAb; New England Biolabs, Hitchin, UK); and anti-α-catenin, 1/2,000 dilution (pAb, Sigma). Secondary antibodies used were goat anti-mouse IgG and goat anti-rabbit IgG Cy3-conjugated (Chemicon, Temecula, CA, USA), goat anti-mouse IgG and goat anti-rabbit IgG Alexa 488-conjugated (1/1,000 to 1/2,000 dilution; Molecular Probes, Eugene, OR, USA). When necessary, phalloidin-Texas Red (Molecular Probes) was mixed with the secondary antibody. Nuclei were counterstained with 4',6-diamidino-2-phenylindole. For retrograde labeling of RGCs, we micro-injected small amounts of 1,1'-dioctadecyl-3,3,3',3'-tetra-methylindocarbocyanine perchlorate (DiI; Molecular Probes) diluted in chloroform into the right tectum of fixed zebrafish embryos 79 hours after fertilization (hpf). After a variable period of incubation (room temperature or 4°C), embryos were counterstained and processed for confocal imaging.
Photomicrography was performed with either a laser confocal system as described or with Nikon fluorescence microscopes, equipped with cooled charge-coupled device (CCD) Hamamatsu Orca cameras and automated z-drive and fluorescence shutters. Acquisition of z-stacks and deconvolution were performed with Openlab software.
For transmission electron microscopy, embryos were dissected rapidly and fixed for 4 hours at 4°C in 4% glutaraldehyde/0.3% H2O2, in an isotonic phosphate buffer.
After being processed for transmission electron microscopy with the use of standard procedures, ultrathin sections were imaged in an FEI-Philips CM100 system.
Eyes extracted from zebrafish embryos just before or around the onset of RGC differentiation (25 to 28 hpf) were dissociated with trypsin-EDTA at 28.5°C, and cells were seeded at a density of eight eyes per dish in 13 mm coverslip-bottom dishes covered with laminin. After incubation for 1 hour in 200 μl of L15 medium containing 10% FCS, 3 to 4 ml of L15 supplemented with N-2 (Invitrogen, Paisley, Renfrewshire, UK) was added. Cells were then either kept at 28.5°C until needed or used immediately for time-lapse analysis. Cultures from has mutants (and their wild-type controls) were made from 30 to 32 hpf embryos, as cell differentiation seems to be delayed in the mutants. Time-lapse studies of cultured retinal cells were conducted in a Nikon TE300 inverted fluorescent microscope, equipped with a Hamamatsu Orca AG cooled CCD digital camera and automated z-drive and shutters. For data acquisition and analysis we used the Openlab software, taking stacks of images (1 μm steps) every 10 to 20 minutes.
For immunostaining, cells were fixed by adding to the culture medium an equal amount of 4% paraformaldehyde, 15% sucrose in 1 × PBS, for 1 hour at room temperature. After being washed, cells were permeabilized with 0.1% Triton X-100 in PBS, and immunostaining and photomicrography were performed as described for cryosections.
With the aim of generating genetic mosaic embryos, we transplanted 10 to 40 blastomeres from labeled embryos (expressing ath5:gap-gfp transgene and/or injected with dextran-rhodamine or H2B-YFP mRNA (YFP being yellow fluorescent protein) to obtain a general labeling) into the animal poles of unlabeled blastulas. In brief, embryos were embedded in 2% methylcellulose on a coverslip, and usually cells were transferred from one donor to up to six hosts with a glass micropipette as described . Embryos were incubated as usual, keeping the donor apart when necessary to identify the mutants morphologically.
Previous 4D imaging reveals that dividing neuroepithelial cells in the zebrafish retina often maintain a basal process in contact with the inner surface of the neuroepithelium . Electron microscopic studies by Hinds and Hinds  of RGCs differentiating in the mouse retina suggested that the axons of RGCs arise from such basal processes. Without direct time-lapse observations, however, it was not possible for these authors to rule out the possibility that the basal processes of such cells retract before their axons emerge. We therefore took the opportunity offered by 4D imaging to re-examine this question. Until about 40 hpf, growth cones form at the tip of the extended basal process in about half the RGCs examined (Figure 2b,d). In the other half, particularly those adjacent to the inner limiting membrane of the neuroepithelium, the axon emerges without a visible basal process (Figure 2b,b',d; see also Figure 6b below for another example). After 40 hpf, almost all the RGCs form their axons from a basal process (Figure 2d). These results confirm the interpretations of Hinds and Hinds  but show that, at early stages, RGC axons need not emerge from an extended basal process.
The effect of the Slit1b morpholino on polarization seems temporary. The RGC-specific antibody Zn-5 revealed that in the morphants, RGCs eventually do retract their apical processes and form a normal ganglion cell layer, although they do so much later than their wild-type counterparts (Figure 4d–f).
The results described above show that there is no obligatory temporal relationship between the retraction of the apical process and the onset of axon outgrowth and that the axon, even of wild-type RGCs, almost always forms before the apical process completely retracts. This retracting apical process is a remnant of the apical compartment of the neuroepithelial progenitor. In vitro studies have shown that molecules normally confined to this apical compartment begin to accumulate in the emerging axon . Thus it might be that in vivo, although the morphology shows an apical process at the time of axon emergence, molecularly the situation is similar to what happens in vitro. If so, we should be able to test this idea by following apical proteins in RGCs during this transition period.
We decided to investigate this issue by using the fusion protein ASIP/PAR-3-EGFP (Par3-GFP) as an in vivo apical marker. This protein was previously shown to accumulate at the apical side of the neural tube epithelium in the zebrafish . Of the three splice isoforms of zebrafish Par-3 described so far , this one is the most similar to the 150 kDa Pard-3a, which does not cause retinal patterning defects when overexpressed . Consistent with all these observations, we found that the overexpression of Par3-GFP does not affect retinal lamination or differentiation and that it accumulates at the apical border of the retinal neuroepithelium at early stages. Interestingly, just at the stage at which RGCs start to differentiate, granules containing Par3-GFP move from the apical towards the basal side of the retina (Figure 6a and Additional file 4). They do not, however, travel all the way to the basal surface but seem instead to accumulate throughout the developing neuroepithelium, particularly around the central region, at the stages examined (Additional file 4). Do these Par3-GFP granules, we wondered, remain in the apical processes of differentiating cells? Indeed, 4D analysis of embryos double-labeled with this construct and ath5:gap-rfp reveals that these granules dynamically co-localize with the tips of retracting apical processes of RGCs (Figure 6b and Additional file 5). In our experimental conditions, we failed to see an accumulation of this apical marker at the tip of the growing axons (see Figure 6b and Additional file 5). Moreover, antibody staining of differentiating RGCs revealed that other apical markers, such as aPKC, α-catenin and F-actin, remain at the tips of RGCs retracting apical processes (Figure 6c–e). In the Slit1b morphants, apical process retraction is inhibited, and the Par3-GFP signal in RGCs remains at the apical border of the retina even after the formation of the axon (Figure 6f).
Another structure associated with the apical compartment of many epithelia, including the retinal neuroepithelium, is the centrosome [14, 29]. To visualize the localization of the centrosome in differentiating RGCs in vivo, we generated a fusion protein containing the full zebrafish sequence of the pericentriolar protein centrin , attached to EGFP ('GFP-zcentrin'), and used it to follow the subcellular localization of the centrosome in ath5:gap-rfp-labeled differentiating RGCs.
We find that GFP-zcentrin labels small dots located at the apical side of the undifferentiated retinal neuroepithelium. Some of these centrosomes are clearly in the apical tips of ath5:gap-rfp-positive cells. When these cells enter into the differentiation process that will lead to an RGC, the apical process is retracted as we have described, and the centrosome remains associated with the tip of the retracting process (20 out of 20 cells in five different embryos), even when the cell is extending its axon on the opposite side (Figure 6g and Additional file 6). In all the cases studied, the centrosome approaches the nucleus (on its apical side) only just before the completion of apical retraction. By analogy with what we found for Par3-GFP, we failed to see a basal localization of the centrosome in the differentiating RGCs. Two out of the 20 cells that were followed showed a lateral localization of the GFP-zcentrin-positive centrosome, starting at about 1.5 hours after the apical process had completed the retraction; the rest remained clearly apical.
Taken together, these observations show that, during their differentiation, RGCs undergo a transition phase in which they retain some characteristics of a neuroepithelial cell, such as the apical localization of junctional molecules and the centrosome, while forming the axon. They also show that, at least for RGCs differentiating in vivo, the axon emerges basally, whereas apical proteins and the centrosome remain apical.
The above experiments show that RGCs polarize in harmony with the neuroepithelium in which they arise. This suggests that the local environment may influence the site of axon outgrowth in differentiating RGCs. If so, axon emergence in RGCs may be affected if the polarity of the retinal neuroepithelium is disrupted, as it is in the zebrafish mutants nok and has. The nok mutant is defective in Pals-1 (also known as Stardust or MPP5), a MAGUK (membrane-associated guanylate kinase) protein that associates with the apical junctional complex. The atypical protein kinase C aPKC-λ, the protein affected in has mutants, is a core component of this complex. To investigate how RGC axons emerge in these disrupted environments, we injected the ath5:gap-gfp construct or crossed ath5:gap-gfp transgenic fish onto these mutant backgrounds.
Are these nok mutant RGCs polarized in the wrong direction from the beginning of their differentiation process? Previous studies  have shown that filamentous actin is normally concentrated at the apical side of the neuroepithelium, but in random central positions in nok mutant retinas. Figure 7g shows one of these F-actin accumulations in the proximity of the non-axonal process of an ectopic RGC, suggesting that its polarity is completely inverted; that is, its basally directed process is the equivalent of the apical process in a normal RGC. Consistent with this was our observation, by 4D microscopy, of 12 ectopically differentiating RGCs from 8 different nok embryos, which clearly retract this 'apical process' while forming the axon on the retinal outer surface (Figure 7j and Additional file 8). We also looked at the movements of Par3-GFP granules in ath5:gap-rfp-positive cells in these mutant retinas (Figure 7i). Figure 7h shows a Par3-GFP granule at the tip of a basally directed 'apical' process in an ectopic RGC, which is extending an axon on the outer retinal surface. However, we never found any accumulation of Par3-GFP at the axon growth cone.
The RPE-specific antibody Zpr-2 made it possible to compare the distribution of these cells with that of ectopic neurons in the different mutant backgrounds. In nok mutants, the ectopic ath5:gap-gfp-expressing cells and axons seem to be concentrated in areas of the retina devoid of RPE (Figure 9b,f). In has embryos, however, ath5:gap-gfp-positive cells seem to be more evenly distributed across the width of the retina (Figure 9f). To quantify this we measured the profiles of ath5:gap-gfp fluorescence intensity on confocal sections of these double-labeled retinas. The results, presented in Figure 9e, show that the accumulation of label at the inner side of the retina in the wild-type retina is converted into an accumulation of signal at both retinal surfaces in the nok mutant retina. In nok morphants (which show a milder nok-like phenotype) there is clearly a small peak of apical ath5:gap-gfp signal in areas where the RPE is absent. These data are consistent with the RGCs being close to the surface on which they are growing axons. This may be so, because once the axon forms, it is likely to exert enough tension on the cell body to tow it towards the relevant surface. In has mutants, although the RPE covers most of the retinal apical surface, there are only few areas devoid of RPE. In these areas we also found a slightly higher apical ath5:gap-gfp signal.
The graphs in Figure 10g–h show a quantitative comparison of directed outgrowth of axons from ectopic RGCs in nok and has mutants, compared with that of ectopic transplanted RGCs in mosaics. The data support the hypothesis that the inversion of ectopic RGC orientation in nok mutants is dependent on the absence of the RPE. To test this further, we used low doses of a translation-blocking nok morpholino (0.17 to 0.20 pmoles per embryo; Figure 9a), which, although able to produce ectopic RGCs, are not efficient at removing the RPE. Thus, ectopic RGCs in these embryos are usually apposed to the RPE rather than Bruch's membrane. According to the hypothesis being tested, these cells should also have difficulty in polarizing. Indeed, these ectopic RGCs do not generally orient in either the normal or reverse direction, and the axons of these cells, when present, tend to grow aberrantly, often towards the retinal periphery (Additional file 11). A summary of the observed behaviors of RGCs in the analyzed conditions is presented in Figure 10i.
Intrinsic mechanisms are used predominantly to break symmetry in neurons that develop in a two-dimensional symmetric in vitro environment, but in vivo the polarized three-dimensional environment provides extrinsic signals that orient differentiating neurons. To approach the mechanisms that drive neuronal polarization in vivo, we have used 4D microscopy to examine how RGCs in zebrafish make the morphological transition from post-mitotic neuroepithelial-shaped cells to neurons with basally oriented axons. Comparison of the polarization of mammalian hippocampal neurons, or indeed zebrafish RGCs in vitro, shows that there are crucial differences between RGC polarization in culture and in the living tissue. These differences may be explained on the basis of environmental influences. Cultured cells are in an almost completely artificial environment. They are in contact with a flat substrate (often rich in laminin) and rarely come into contact with other cells. In vivo, however, differentiating neurons are naturally almost completely surrounded by cells with which they interact extensively. In addition, the native environment provides heterogeneously distributed positional cues that are difficult to emulate in vitro.
These differences are likely to influence the initial stages of differentiation. In vitro, the cell is rounded and extends pseudopodia and filopodia all around its free surface, whereas in vivo the newborn RGC first extends from the apical to the basal surface of the neuroepithelium assuming a spindle shape, typical of neuroepithelial cells, and then starts retracting its apical process. RGCs in vivo show filopodial activity at the basal side for 1 or 2 hours before axonogenesis, whereas RGCs differentiating in culture extend short neurites at several points of their cell bodies, before one of them starts to elongate in an axon-like manner. This seems very comparable to stage 2 of rat hippocampal neurons differentiating in vitro, although we found that RGCs, in our culture conditions, do not usually present a multipolar morphology, but just an alternative growth and retraction of neurites at different points. Another difference from rat hippocampal cells is that zebrafish RGCs seem to form their dendrites de novo in the transition between stages 3 and 4. The second stage, which can last up to several hours in cultured RGCs, is not seen in normal differentiating RGCs in vivo, because these cells seem to pass directly from a neuroepithelial-like spindle-shaped cell to one in which the RGC has only a single fast-extending neurite, which invariably becomes the axon. This neurite is always formed at the basal pole of the cell, opposite to a retracting apical process. Signals distributed heterogeneously in the retinal neuroepithelium could be responsible for restricting axon formation to the basal side . A similarly restricted cellular behavior has very recently been described for HSN motorneurons differentiating in living C. elegans larvae .
The seminal work of Hinds and Hinds  in the developing mouse retina, following the much earlier observations of Ramón y Cajal , suggested that in RGCs the axon and the basal process were the same thing, or that the axon was formed from the basal process. Recent time-lapse observations of bipolar cells in vivo showed that the neurites extending into the outer plexiform layer and the inner plexiform layer develop respectively from the unretracted apical and basal processes of the migrating precursor . In the present study, 4D imaging allowed us to observe the dynamics of axonogenesis in RGCs in vivo and to see that in many cases, especially at early stages, the axon emerges directly from the cell body after it reaches the basal surface of the neuroepithelium and is in apposition to it. It is interesting that the filopodial activity at the basal side of such differentiating RGCs starts before the extension of the axon, similar to what was previously described for differentiating rat RGCs . At later stages, however, most RGC axons do seem to grow from a basal process. The reason why basal process-free differentiating RGCs are only seen at relatively early stages of retinal differentiation may reflect the fact that the first RGCs are able to translocate their somas all the way to the basalmost position, whereas later-differentiating somas are blocked from reaching the basal surface by a layer of tightly packed RGCs and must form axons from a distance (see Figure 2a).
We wondered whether the retraction of the apical process or components that are normally localized to this process provide intrinsic information that tells the differentiating RGC when and where to form the axon. In the retina, post-mitotic RGC precursors transiently acquire a spindle-like, neuroepithelial morphology. They begin to lose this neuroepithelial morphology with the retraction of the apical process, and the axon is usually formed after this retraction has started but has not yet completed. However, our observations also show that axons can emerge from RGCs before retraction of the apical process begins. In Slit1b morphants, most RGCs form axons before apical process retraction begins. Our results with Slit1b are consistent with previous work showing a role for Slit proteins in neuronal migration , and it will be very interesting to investigate how Slit1b is involved in apical retraction, although that would have to be the subject of another study. All these observations indicate that axon extension occurs independently of the timing of apical retraction.
It has been confirmed by different groups that the Par-3-Par-6-aPKC complex accumulates at the tip of the growing axon of cultured hippocampal neurons, where it has a fundamental role in the determination of axon identity [7, 9]. We did not find detectable accumulation of Par3-GFP in the RGC axons forming in vivo. Rather, this protein, although present in differentiating RGCs, remains apical. This finding supports the apparent absence of a role for apical complex components in mushroom body neurons during axonogenesis (or dendritogenesis) in living Drosophila larvae . We also show that aPKC-λ-deficient (has-/-) RGCs are able to polarize efficiently and extend axons in vivo when these cells are in contact with the inner surface of the retina and that they can also polarize in vitro. In this case it may be that because the has mutation affects aPKC-λ and not aPKC-ζ, the latter compensates for the absence of the former and that double mutants may be needed to reveal whether aPKCs are necessary for oriented RGC axon outgrowth in vivo and intrinsic polarization in vitro.
It was reported recently that the localization of the centrosome is involved in establishing the site of axon emergence from cultured rat hippocampal cells . The authors of that study suggested that during telophase of a neurogenic apicobasal cell division [37–39], the centrosome of the basal post-mitotic daughter cell, being located opposite to the apical surface, would determine the emergence of the axon from the basal side. However, this could not work in our case because there are no apicobasal cell divisions in the differentiating zebrafish retina, and RGCs originate from planar oriented divisions [17, 21]. The absence of apicobasal divisions in the zebrafish retina is also consistent with our observation that RGC precursors inherit components such as Par3-GFP, aPKC, α-catenin and F-actin of the apical adhesion complexes in the tips of their retracting apical processes. Surprisingly, through the dynamic analysis of RGC differentiation by 4D microscopy, we also found that the centrosome does not move from its apical position during the whole process of axonogenesis. This observation does not exclude the possibility of an essential role of this organelle in axon determination or axon growth but shows that its localization close to the site of axon formation is not necessary for the normal polarization of zebrafish RGCs in vivo. In addition, our data support previous data obtained from the ultrastructural analysis of the differentiating mouse retina  and other systems, including hippocampal pyramidal cells [40, 41]. Many studies have shown that the centrosome, which is essential for nuclear translocation, is located on the leading edge side of the nucleus in migrating neurons . In our system, the centrosome localizes to the trailing edge as the nucleus translocates towards the basal surface. This difference could be because RGCs do not actually migrate but just elongate and translocate their nuclei.
We have used two mutations, nok and has, to analyze the role of neuroepithelial polarity in the orientation of RGC polarity in vivo. We found in nok mutants that ectopic RGCs are able to polarize properly (that is, to form only one axon) but that their orientation is often inverted (that is, the axon is directed towards the outer retina). In accordance with previous findings that apical markers, including F-actin and Par-3, are accumulated in random positions inside the retina in these mutants [28, 31], we found that basally directed processes of these ectopic differentiating RGCs express Par3-GFP in the proximity of F-actin staining. In the nok mutant retinas there are large gaps where the RPE is not present and where Bruch's membrane, the basal lamina of the RPE, becomes apposed to the surface of the retinal neuroepithelium. In these regions the neuroepithelial polarity seems completely inverted (that is, the apical complex is basally located with respect to a basal lamina present at the apical surface of the neuroepithelium).
We also unexpectedly found that there were fewer ectopic RGCs concentrated on the apical side of the retina, as well as many fewer ectopic axons in has mutants than in nok mutants. The RPE is more intact in has mutants. This observation suggests that the RPE may normally inhibit reverse polarization, either by generating an inhibitory signal or by blocking access to permissive signals on Bruch's membrane. It has previously been shown in chick and quail that a small proportion of axons naturally escape from the optic nerve layer and grow between the cells of the RPE and Bruch's membrane, and that the Bruch's basal lamina is a good substrate in vitro for RGC axon growth . This favors the plausibility of the second hypothesis. In addition, inversion of RGCs has been observed in organ-cultured retinas in which the vitreal surface was exposed to chondroitin sulfate . In this case, too, ectopic RGCs extend axons on the outer surface of the retina. Many previous studies have proposed that the RPE has an essential role in retinal lamination [31, 45–47], and it is tempting to speculate that our observations could also help explain their results.
The polarity of the neuroepithelium seems to be a major determinant in the site of axon outgrowth from RGCs. Our work suggests that there is an intrinsic tendency for RGCs to polarize (to form only one axon) but that RGCs differentiating in vivo use signals, like those normally found at the inner limiting membrane of the retina, to define the orientation of this polarization and the position of axon emergence.
= atypical protein kinase C
= charge-coupled device
= enhanced green fluorescent protein
= heart and soul
= hours after fertilization
= monoclonal antibody
= monomeric red fluorescent protein
= nagie oko
= polyclonal antibody
= phosphate-buffered saline
= polymerase chain reaction
= retinal ganglion cell
= retinal pigment epithelium.
We thank A Reugels and J Wittbrodt for sharing plasmids used in this work; S Abdelilah-Seyfred, J Malicki and I Masai for zebrafish lines; I Fischer for antibodies; M Bate, A Brand, C Holt and R Wong for critically reading the manuscript; and I Pradel for technical assistance in molecular cloning.
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.