Morphogenesis and axis specification occur in parallel during optic cup and optic fissure formation, differentially modulated by BMP and Wnt

2.1. Overall tissue dynamics driving optic cup and optic fissure morphogenesis

To address the tissue dynamics during optic cup formation, we employed in vivo time-lapse imaging of tg(hsp70:kaede) embryos. Kaede is a photo-convertible fluorescent protein. Kaede expression was induced after a heat shock applied at 11 hpf for 15 min. Specific domains of the developing eye were photo-converted from green to red fluorescence (figure 1a). Following the photo-converted cells over time, a gastrulation-like rearrangement of lens-averted cells into the lens-facing domain can be appreciated (figure 1b–f; electronic supplementary material, movie S1). Note the marked flat cells (magenta, figure 1b), located in the lens-averted domain, which over time reach the lens-facing domain of the optic cup and elongate (figure 1c–f). The flow over the distal rims, which has been analysed previously [18], is nicely visible also from a lateral perspective. Importantly, this flow is bilateral, which allows the nasal and temporal domains to extend distally. This can be interpreted as the onset of optic fissure formation, demonstrated in a nuclear- and membrane-labelled optic cup (figure 1g–i).

• Download figure
• Open in new tab
• Download PowerPoint

Notably, the optic vesicle is connected to the optic stalk (figure 1i, arrow). In a transition zone, the stalk is connected to the lens-averted domain of the optic cup in a triangular manner (figure 1j, arrow). This connection can be found on the nasal side and the morphology of this transition zone is changed over time (figure 1j,k).

2.2. Establishment of the temporal optic fissure margin

After this global analysis of tissue dynamics, we focused on the region of the forming optic fissure. At first, we addressed the temporal fissure margin. To visualize the cell dynamics, we made use of a transgenic line with ubiquitous nuclear and membrane labelling (tg(bact2:H2BGFP, bact2:lyntdTomato)). In the lateral perspective, the prospective neuroretinal tissue, located within the lens-averted layer of the temporal ventral domain, is nicely visible (figure 2a,b, between white brackets). In the lateral aspect, these prospective neuroretinal cells are moved over the distal rim into the lens-facing domain of the optic cup (figure 2a,b, white arrows).

• Download figure
• Open in new tab
• Download PowerPoint

Furthermore, it can be appreciated that in the ventral aspect, prospective neuroretinal cells move over a ventral rim via a perpendicular flow into the temporal lens-facing domain of the optic cup (figure 2a,b, yellow arrow; electronic supplementary material, movie S2). This ventral rim corresponds to the temporal fissure margin. After optic cup morphogenesis is complete, the lens-averted layer is made up of flattened cells, the prospective RPE cells (figure 2c, bracket).

To facilitate a better individual cell tracking, we reduced the density of labelled nuclei by injecting RNA coding for H2BGFP into one cell at the 4- to 8-cell staged embryo in the context of a ubiquitous membrane labelling, achieved by injections of RNA coding for lyntdTomato into the zygote. In addition, we increased the resolution in z during time-lapse imaging by performing an exemplary single-plane illumination imaging (SPIM, lightsheet) experiment. Three-dimensional volume analysis over time allowed us to follow individual single cells, artificially labelled with coloured dots, from the lens-averted domain on their way into the optic cup (figure 2d; electronic supplementary material, movie S3). The distal rim (figure 2d, dashed line) and the ventral rim can be appreciated in the max projection of the three-dimensional volume. Individual cells were followed over the distal and distal–ventral rim (magenta and blue) and over the ventral rim (red). This indicates that cells enter the optic cup via a ventral perpendicular flow, which is a direct continuation of the distal flow [18] on the temporal side. The findings are summed up in a scheme (figure 2e).

2.3. Establishment of the nasal optic fissure margin

We have now established that on the temporal side, the ventral and the distal flow are directly linked (figure 2). On the nasal side, however, the optic stalk is connected to the optic vesicle, probably influencing the morphogenesis of the fissure margin in concert with the distal flow. To address the stalk contribution to fissure margin morphogenesis, we made use of the tg(hsp70:kaede) line again. We first photo-converted in the upper transition zone (figure 3b) to address the onset of stalk contribution. It can be appreciated nicely that the upper domain of the transition zone is moving into the fissure margin over time (figure 3b–f), changing the overall morphology of the transition zone from a triangular shape to a flat shape (figure 3b–f; electronic supplementary material, movie S4), what we termed ‘twist’ of the optic stalk. In order to visualize the further extent of the optic stalk contribution, we specifically labelled a domain within the optic stalk by photo-conversion of Kaede using embryos of the tg(hsp70:kaede) line. The photo-converted domain was then followed in a three-dimensional volume over time (figure 3g–k). It can be appreciated nicely that also this distinctly labelled domain is over time integrated from the optic stalk into the nasal optic fissure margin (figure 3g–k; electronic supplementary material, movie S5). These data are in line with previous observations [24,28]. Notably, in the ventral nasal domain, in which the distal flow and the stalk flow merge, a turbulence can be observed, which we termed ‘vortex’ (figure 3l,m, #; electronic supplementary material, movie S6). The findings are summed up in a diagram (figure 3n). To provide further quantitative data for the stalk contribution, we performed another set of experiments in which we photo-converted parts of the stalk domain early and analysed the contribution of this domain in the optic cup later (figure 3o–q; electronic supplementary material, figure S3 supplement). We found a stalk contribution ranging from 10 to 30%, being higher in the proximal regions of the optic cup and decreasing distally (figure 3o; electronic supplementary material, figure S3 supplement).

• Download figure
• Open in new tab
• Download PowerPoint

2.4. Additional TGFβ-signalling-positive cells enter from the optic stalk

We recently found that TGFβ signalling is essential for optic fissure fusion [19]. In this context, we observed active TGFβ signalling within the optic fissure margins using an in vivo TGFβ-signalling reporter in zebrafish [19]. Here, we addressed whether the TGFβ-signalling domain is extending into the margins by signalling activation within the margin cells, or by secondarily translocation of cells, in which TGFβ signalling was already activated, into the margins. To this end, we performed in vivo time-lapse imaging of embryos of the tg(SBE:GFPcaax), injected with RNA coding for lyntdTomato. We identified a domain in the forebrain and optic stalk, in which TGFβ signalling was activated (figure 4a,e,i, arrowhead). We then followed cells of this domain over time. We found that the active TGFβ-signalling domain was extending through the optic stalk (figure 4; electronic supplementary material, movies S7–S9). We identified TGFβ-signalling-positive cells which were translocated from this domain into the nasal fissure margin secondarily (figure 4; electronic supplementary material, movies S7 and S8). Notably, although the optic stalk was found to be connected predominantly to the nasal fissure margin, we observed TGFβ-positive cells crossing to the temporal fissure margin via the prospective optic nerve head (figure 4c,d,f–h; electronic supplementary material, movies S7 and S8). In distal fissure domains, the reporter signal is lost over time, due to either a continuous flow of these cells into other domains or due to a reduction in signalling activity over time (figure 4j,k).

• Download figure
• Open in new tab
• Download PowerPoint

2.5. BMP signalling induction and inhibition of Wnt secretion affect optic fissure morphogenesis

At this point, we had a framework for optic fissure formation in the context of optic cup morphogenesis at hand. Next, we wanted to address whether this morphogenetic process is hampered in conditions which are known to result in coloboma with a vast cleft, and if so, how. To this end, we made use of two models. On the one hand, we made use of an artificial expression of bmp4. On the other hand, we made use of an inhibition of porcupine, a factor which is crucial for the secretion of active Wnt ligands [29,30].

2.5.1. BMP induction

First, we addressed the effect of induced BMP signalling. It was previously shown that the bilateral distal flow could be precociously arrested by induced BMP expression [18], the likely reason for this being an oversaturation of the BMP antagonist follistatin a (fsta), which was found expressed in the nasal and temporal domain of the forming optic cup [18]. We now addressed the expression of fsta also in the forming ventral aspect of the optic cup. Notably, we find it expressed in the transition zone from the stalk to the lens-averted domain (figure 5a–c). We then crossed fish from tg(hsp70:bmp4, cmlc2:GFP) with fish from tg(SBE::GFPcaax), both described previously [19], to produce double transgenic embryos. Heat shock inductions of bmp4 were performed at 17 hpf. Subsequently, the embryos were subjected to in vivo time-lapse imaging (figure 5d). bmp4 inductions starting at 17 hpf resulted in overt morphogenetic defects of the optic cup (figure 5e–h). The optic stalk and the TGFβ-signalling active domain within were found enlarged (figure 5e–h), compared to controls (figure 4e–h) and the connection of the optic stalk to the optic cup was found broadened. A clear symmetry breaking connection of the stalk to a future nasal domain including a twist of the optic stalk could not be identified. Importantly, the optic stalk was found stuck and not integrated into the optic cup (electronic supplementary material, movie S10). Nevertheless, under these conditions without proper tissue dynamics, TGFβ-signalling active domains were later found within the optic cup but also within lens-averted domains (figure 5f–h). Notably, TGFβ-signalling activity within the optic cup can be observed during retinal differentiation (not shown). The perpendicular flow on the temporal domain was also largely absent (electronic supplementary material, movie S10). Notably, however, the distal flow in the ventral domains was not abrogated, explaining that in distal domains, an optic fissure was forming (electronic supplementary material, S1A–D and movie S11). This was at odds with findings from embryos in which bmp4 was driven from the rx2 cis-regulatory element, in which especially the distal flow on the temporal side was arrested [18] (electronic supplementary material, figure S5 supplement 1, E–N, see O–X as control and movies S13 (proximal), S14 (distal), see S15 (proximal) and S16 (distal) as control). Notably, in this scenario, the optic stalk is also stuck and the perpendicular flow on the temporal side is largely absent (electronic supplementary material, figure S5 supplement 1, E–I and J–M). Notably, to a certain extent, a pathologically formed fissure can be found in distal domains [18] (electronic supplementary material, figure 5 supplement 1, N). We next performed heat shock inductions of bmp4 at 13 hpf. Subsequently, the embryos were subjected to in vivo time-lapse imaging (figure 5, timeline). Bmp4 inductions starting at 13 hpf resulted in overt and even more pronounced morphogenetic defects of the optic cup (figure 5i–l; electronic supplementary material, figure S5 supplement 3, showing the overall morphology post-heatshock at 13 hpf and 17 hpf). Large ectopic putative neuroretinal domains can be appreciated (figure 5i–l; electronic supplementary material, movie S12). Importantly, also the stalk-derived flow as well as the perpendicular flow on the temporal side were found absent resulting in an absence of the optic fissure (figure 5i–l; electronic supplementary material, figure S5 supplement 2, A–D). Accordingly, the optic fissure is not only an important entry route for the POM and blood vessels, which enter through the fissure, but also for prospective neuroretinal tissue which is integrated into the ventral optic cup via the optic fissure margins. Thus, it is comprehensible that large domains of the ventral optic cup must be missing. We therefore addressed the ocular phenotype resulting from a bmp4 induction via heat shock 13 hpf at a later time point. Notably, at 40 hpf, the lens is protruding ventrally, explainable by missing ventral optic cup domains (electronic supplementary material, figure S5 supplement 2, E–H). So, even though no optic fissure was forming, at later developmental stages, a ventral gap is visible, not corresponding to the normal optic fissure. Since the overall flow movements were arrested drastically, the shape of the optic cup including the ‘abnormal ventral fissure’ must be resulting predominantly from a bending of the cup, probably caused by ojoplano-mediated basal constriction [20,21].

• Download figure
• Open in new tab
• Download PowerPoint

2.5.2. Wnt inhibition

We next addressed the role of Wnt signalling during optic cup and fissure formation. In a mouse knockout, the loss of porcupine has been shown to result in a coloboma with a vast cleft and ectopic neuroretina [11]. We inhibited porcupine with a small compound inhibitor (LGK-974, Hycultec GmbH, Beutelsbach) mimicking the coloboma phenotype observed in mouse. We hypothesized that the vast coloboma could well result from a hampered optic fissure morphogenesis, even though the interpretation of the ocular phenotype found in the porcupine KO mouse was a different one [11]. We made use of in vivo time-lapse imaging using embryos with the TGFβ-signalling reporter transgene as a stalk marker, in which we injected with lyntdTomato RNA at the zygotic stage. The embryos were transferred to fish-medium containing the compound inhibitor LGK-974, an inhibitor for porcupine, at 13 hpf. To test the efficacy of the inhibitor, we performed a dose–response analysis for the inhibitor (electronic supplementary material, figure S6 supplement 2) and also made use of an established Wnt reporter in zebrafish, tg(7xTCF-Xla.Siam:GFP) (electronic supplementary material, figure S6 supplement). A reduced Wnt-signalling activity can be appreciated nicely. Strikingly, and according to our hypothesis, the inhibition of porcupine was also resulting in a defect of the flow movements during optic cup morphogenesis (figure 6; electronic supplementary material, movies S17 and S18). This can nicely be appreciated in the dorsal domain of the optic cup, where ectopic domains of the prospective future neuroretina can be observed on the nasal and the temporal side (figure 6b,c, arrows). However, even more strikingly, the domain of TGFβ-signalling positive cells was found largely stuck within the optic stalk (figure 6c,d, arrowhead). In addition, this also negatively affected the ventral perpendicular flow on the temporal side (electronic supplementary material, movie S18). Moreover, we quantified distinct parameters (figure 6k). While the width of the optic cup was unchanged, its length was reduced. We also noted an increased width of the optic fissure, while its length was reduced. Overall, our data demonstrate that Wnt ligands are important for optic cup and optic fissure morphogenesis. The data furthermore indicate that the coloboma phenotype resulting from a hampered Wnt signalling is a secondary, morphogenetic coloboma.

• Download figure
• Open in new tab
• Download PowerPoint

2.6. Dorsal ventral axis specification during optic cup formation and the role of BMP and WNT-signalling

A hampered optic fissure morphogenesis was observed resulting from both induced bmp4 expression and porcupine inhibition. However, next to affecting morphogenetic tissue dynamics, BMP and Wnt-signalling both could affect the dorsal ventral axis within the optic cup [31]. We thus next addressed if bmp4-induced signalling and Wnt-signalling inhibition was also affecting the identity of cells within the optic cup. We addressed the expression of several markers, known to be expressed in distinct domains of the optic cup, namely bambi [18,32] in the dorsal optic cup, vax2 [32,33] in the ventral optic cup and pax2a [33] in the optic stalk and ventral most optic cup. bmp4 induction, performed at 17 hpf, resulted in a dramatic increase in the bambia expression domain. While bambia can be found expressed in the dorsal domain of the optic cup in controls (figure 7a–c), bmp4 induction increased the expression domain to the entire optic cup, the optic stalk and large domains of the forebrain (figure 7d–f). The expression of vax2, which in controls can be found in the ventral domain of the optic cup and optic stalk (figure 7g–i), was largely absent after bmp4 induction (figure 7j–l). Noteworthy, however, was a remaining expression of vax2 in the lens-averted domain on the temporal side (figure 7j). This domain is corresponding to the domain which was stuck during morphogenesis and was not correctly integrated into the lens-facing side of the optic cup (figure 7j). The induction of bmp4 furthermore resulted in a reduced pax2a expression domain within the optic cup (figure 7, m–o controls, p–n bmp4 over-expression). The pax2a expression domain in the optic stalk, however, was found enlarged (figure 7p).

• Download figure
• Open in new tab
• Download PowerPoint

The inhibition of porcupine resulted in a reduced expression intensity of bambia, though the expression domain seemed largely unaffected compared to controls (figure 8a–c controls, d–f treatment). Notably, we found bambia expressed to a similar extent at 17 hpf in controls and porcupine-inhibited embryos (figure 9a–f), suggesting an onset of repression in between 17 hpf and 24 hpf. Vax2 was found expressed in the ventral optic cup and optic stalk also after porcupine inhibition (figure 8j–l), in domains where it can also be found in controls (figure 8g–i). Importantly, vax2 could also be found in lens-averted domains of the optic cup (figure 8j–l). These domains corresponded to the domains which failed to be integrated into the lens-facing domain of the optic cup correctly during morphogenesis. This is indicating that the domain of vax2, often used as a ventral marker for the optic cup, is largely originating from the lens-averted part of the early optic cup. We further analysed this by addressing the vax2 expression domain at 17 hpf (figure 9g–l). It can be appreciated nicely that in controls, but also in porcupine-inhibited embryos, the vax2-positive domains can be found in the lens-averted domain of the early optic cup (figure 9g–l). The domain of pax2a, expressed in the ventralmost optic cup and the optic stalk in controls, was found mildly reduced after inhibition of porcupine (figure 8m–r). Notably, after porcupine inhibition, pax2a expressing cells could be found in the lens-averted domain on the nasal side (figure 8p–r). An extension of the pax2a domain from the optic stalk into the nasal lens-averted domain resulting from porcupine inhibition could be seen already at 17 hpf (figure 9p–r). See figure 9m–o as control.

• Download figure
• Open in new tab
• Download PowerPoint

• Download figure
• Open in new tab
• Download PowerPoint

2.6.1. Summary, discussion and conclusion

Even though the role of tissue dynamics is more and more appreciated for optic cup formation [18,22–25], only very little was so far known about optic fissure morphogenesis and whether or not this also is affected by tissue dynamics. A still prevailing view suggests a mere ventral bending of the nasal and temporal domains to drive fissure formation [4]. We show, however, with in vivo time-lapse analyses of zebrafish embryos that tissue dynamics play a major role for optic fissure formation (see figure 10 for a summary of our findings). Based on cell tracking, we find the ‘bilateral neuroretinal flow’ over the distal rim [18] in direct continuation with a ventral perpendicular flow on the temporal side. On the nasal side, however, a tissue flow derived from the optic stalk is shaping the fissure margin, especially in the proximal domain. A certain contribution of stalk-derived cells during optic cup development was also found in Xenopus [28] and a similar finding was described as a late evagination in zebrafish [24]. We further uncovered a cellular vortex in the region of the nasal margin, probably resulting from converging streams of cells from the distal flow [18] and the stalk flow (this study) during which the optic stalk undergoes a twist movement. The surface ectoderm including the forming lens and the forming optic cup are suggested to interdepend due to the exchange of inductive signals [5]. Prospective neuroretinal cells, which are only secondarily integrated into the lens-facing domain of the optic cup over the distal rims, could therefore be shielded from such inductive signals, a concept potentially important for the establishment of the stem cell containing ciliary marginal zone, recently discussed [18]. Here, we show that the optic fissure margins serve as ventral rims, over which, next to the distal rims, cells are secondarily integrated into the optic cup. These cells will also be shielded from inductive signals before they enter the lens-facing domain of the optic cup. This affects the population of cells derived from the lens-averted domain of the optic vesicle but also the stalk-derived cells. Notably, we identified a translocation of a specific set of cells, which are moved from the optic stalk into the optic cup. In these cells, TGFβ signalling was activated. Even though we found the optic stalk mainly connected to the nasal fissure margin, especially in the distal aspect, the TGFβ-signalling-positive cells moved through the nasal fissure and via the presumptive optic nerve head also to the temporal fissure margin. Notably, TGFβ signalling was shown to be important for optic fissure fusion [19], a process following the orderly formation of the optic fissure margins. The translocation of the TGFβ-signalling active cells nicely shows that the secondarily integrated cells might not just be potentially shielded from inductive signals from the surface ectoderm but also might be activated or primed in proximal domains before they are sent to their destination inside the optic cup. It is also mentionable that this population is sent to both fissure margins, meaning that it is not respecting the distinct nasal and temporal domains [5,34]. The finding of a bilateral flow per se suggests a static dorsal and ventral pole within the forming optic cup [18]. The dorsal pole can be nicely appreciated, because it is causing the emerging ‘dorsal fissure’ [35]. The region of the ventral pole must be corresponding to the prospective optic nerve head, marking the proximal end of the optic fissure. The flow of tissue entering from the optic stalk could be a contributing factor to the proximal initiation of a split flow. Overall, such detailed knowledge about optic fissure morphogenesis is crucial to understand the origin of diverse coloboma phenotypes. Even though a defective fusion of the optic fissure margins may be the most appreciated cause for coloboma, a fusion defect is very unlikely to be the reason for coloboma in which the cleft is so big that the margins cannot touch. Such coloboma are more likely to be ‘indirect’ and rather result from optic fissure formation defects [36].

• Download figure
• Open in new tab
• Download PowerPoint

We addressed two coloboma models presenting a vast cleft as a phenotype [11,18]. We used bmp4 induction [18,19] and small compound-mediated inhibition of porcupine. The latter is mimicking the phenotype observed in porcupine KO mice [11]. Porcupine is crucial for the secretion of active Wnt ligands [29,30]. Mutation of β-catenin resulted in a similar coloboma phenotype [10]. This is suggesting that the vast coloboma phenotypes observed in β-catenin mutants and in porcupine mutants are resulting from a hampered canonical Wnt signalling.

In both our inducible coloboma models, we found a drastic effect on the ventral invagination (summed up in figure 10). The flow derived from the optic stalk was arrested, resulting in a failure of the proximal optic fissure to form. This suggests that the optic stalk is indeed contributing to the initiation of optic fissure formation. Since bmp4 induction is also affecting the distal flow in the ventral domains, it is understandable that the phenotype is ultimately resulting in an almost complete absence of the optic fissure. Although both coloboma models share a ventral flow defect, they showed also marked differences in other aspects. While the inhibition of porcupine resulted in a strong morphogenetic defect affecting mostly the tissue dynamics, heat shock-induced induction of bmp4 affected the cells also in other ways. Mentionable is the ectopic activation of TGFβ signalling found during in vivo time-lapse imaging as well as the ectopic induction of bambia expression, extending through the entire optic cup and the almost completely missing vax2 expression, which could only faintly be seen in the lens-averted domain. These effects can probably be interpreted as a shift within the dorsal ventral axis of the optic cup [26,27,32] in line with a dorsalization phenotype. The remaining, though faint, expression domain of vax2 in the lens-averted domain, nicely visible on the temporal ectopic domain, however, underlines the failure of tissue dynamics. Our expression domain analyses of the porcupine-inhibited embryos are also indicating a morphogenetic defect. Our data clearly show that the expression domain of vax2 is following the tissue flow. In control embryos, analysed at 17 hpf, vax2 was found expressed largely in lens-averted domains and the optic stalk. Only later, at 24 hpf, could the classical ventral domain be seen. The inhibition of porcupine, blocking the morphogenetic tissue movements, however, resulted in the vax2 domain largely being stuck in the lens-averted domain of the optic cup. These findings show the importance of combining time-lapse imaging analyses with expression analyses. If the tissue dynamics would have been unknown or ignored, the altered localization of vax2 expressing cells could easily have been mistaken for a trans-differentiation of RPE precursors. The relatively normal sized, though mislocalized, vax2 expression domain indicates a normal ventral identity of the cells making up the domain. However, the expression intensity of bambia in the dorsal domain was found reduced. The reason for this likely is a hampered maintenance of the dorsal cell identity found in the absence of Wnt-signalling in line with previous findings [31]. The domain of pax2a was found slightly reduced in the optic cup and mislocalized to the lens-averted domain and stalk. An extension from the optic stalk into this domain was already apparent at 17 hpf.

Overall, we provide a solid framework of data for optic fissure morphogenesis. This framework is the basis for the analyses of coloboma having in common a vast cleft, which are unlikely resulting from a mere defect of optic fissure fusion. As a proof of concept, we applied our framework and analysed two known vast coloboma phenotypes. It is very likely that this framework will be applicable also for many other colobomas with a vast and extended cleft. Furthermore, the concept of a defective morphogenetic process as a basis for vast cleft phenotypes will probably also be useful for other tissues (e.g. palate and neural tube).

3.1. Zebrafish

Husbandry. Zebrafish (Danio rerio) were kept in closed stocks in accordance with local animal welfare law. The fish facility is under supervision of the local representative of the animal welfare agency. Fish were maintained in a constant recirculating system at 28°C on a 12 L : 12 D cycle. Age of zebrafish embryos was determined according to Kimmel et al. [37].

Transgenic zebrafish. TGFβ reporter: (tg(SBE:GFPcaax) were generated previously, as described by Knickmeyer et al. [19], rx2 reporter: tg(Ola.rx2:eGFP-caax) were described and used previously [18]. Ubiquitous nuclear and membrane reporter: The double transgenic line containing, tg(actb2:H2BGFP) tg(actb2:lyntdTomato) was created in the laboratory of Joachim Wittbrodt and kept in a AB/Casper background. Kaede reporter: tg(hsp70:kaede;cmlc2:eGFP) was created in the laboratory of Jochen Wittbrodt and kept in a AB/WIK background. The two lines were kindly provided by the laboratory of Jochen Wittbrodt. Wnt reporter: The tg(7xTCF-Xla.Siam:GFP) transgenic line was kindly provided by Mathias Carl. HS-bmp4: tg(hsp70:bmp4; cmlc2:eGFP) were described as used previously [19]. Rx2-driven bmp4 overexpression: tg(rx2:bmp4; cmlc2:eGFP) was described as used previously [18].

Transient labelling of zebrafish. Where indicated, H2BGFP (nuclear localized GFP) (50 ng µl⁻¹), lyntdTomato (75–150 ng µl⁻¹) was injected into 1- to 8-cell staged zebrafish embryos enabling four-dimensional imaging of mosaic or ubiquitous labelled zebrafish.

3.2. Heat shock

Heat shocks for the tg(hsp70:kaede; cmlc2:GFP) and tg(hsp70:bmp4; cmlc2:GFP) lines were applied in a heating block at 37°C for different times. Early heat shocks up to 13 hpf were applied for 15 min, later they were applied for 1 h.

For tg(hsp70:bmp4, cmlc2:eGFP) embryos, heat shocked wild-type siblings from the same clutch of eggs were used as controls.

3.3. Inhibitor treatment

Porcupine inhibition: LGK-974 (Hycultec GmbH Beutelsbach) was solved in DMSO (10 mM). Embryos were treated with 20 µM LGK-974 from 13 hpf on until end of imaging. For the dose–response curve, embryos were treated with 1, 10, 20 and 50 µM of inhibitor. Control embryos were treated with an equal amount of DMSO without inhibitor.

3.4. Whole-mount in situ hybridization

Whole-mount in situ hybridization (WMISH) was performed according to Quiring et al. [38] and Heermann et al. [18]. WMISHs for confocal imaging were stained with FastRed Naphthol (Sigma-Aldrich). The nuclei were stained with DAPI (4 μg ml⁻¹). For each in situ 8–10 embryos were stained, 4–6 were confocally imaged.

3.5. Microscopy

Whole-mount in situs were imaged with a Leica TCS SP8 confocal microscope with a z-spacing of 3 µm.

3.6. Time-lapse imaging

Time-lapse imaging was performed on inverted Leica TCS SP8 set-ups with two internal hybrid detectors, and a 40× long distance objective (water immersion), using Immersol (Carl Zeiss) as immersion medium. For time-lapse imaging, embryos at appropriate stages were embedded in 1% low melting agarose in glass bottom dishes (MatTek, Ashland, MA, USA) and covered with zebrafish medium, including tricaine for anaesthesia. Confocal stacks were taken every 10 min with a resolution of 1024 × 1024 and 3 µm z-spacing. The left and right eyes were used and oriented to fit the standard lateral view.

3.7. Processing of time-lapse imaging data

The stacks were evaluated with FIJI software [39]. For denoising of the movies, the PureDenoise plugin [40] was used with four cycle-spins and three frames multiframe.

3.8. Single plane illumination (lightsheet) imaging

Time-lapse SPIM imaging was performed with a Leica TCS SP8 set-up upgraded with a DLS (digital lightsheet). A 5× illumination objective, a 25× dipping lens together with 2.5 mm mirror caps were used to obtain a lightsheet. A stack was taken every 5 min. For imaging, glass-bottom-dishes (MatTek) were coated with 2% Agarose. Embryos were embedded in 1% low melting agarose and placed on the coating. Two notches were cut in the agarose with razorblades, so that the embryo was placed on a 2 mm wide stripe that could fit between the mirror caps. The embryo was covered with zebrafish medium, including tricaine for anaesthesia. Data from left and right eyes were used and oriented to fit the standard lateral view.

3.9. Tracking and visualization

Tracking of cells was performed with MTrackJ [41]. Tracks were visualized using custom-made ImageJ plugins as in [18]. The overlay images of tracks and raw data were used to produce maximum intensity projections for figure 2 and electronic supplementary material, movie S4.

3.10. Quantitative analysis of the length of the optic fissure

To analyse the length of the optic fissure, DAPI and anti-β-catenin-stained embryos were embedded upside down in low melting agarose. z-stacks with 3 µm z-spacing were taken. The length of the fissure was measured on an optical section below the lens, in a straight line from the embryonic retina to the surface ectoderm. The width of the OF and the eye was measured, at 50% length of the OF. The width of the neuroretina was measured in extension of the length of the OF. The length of the eye is the addition of the retina and the length of the OF. For each condition, seven eyes were measured. To determine significance, an unpaired t-test with Welch's correction was used, with p < 0.05.

3.11. Quantitative analysis of the contribution from the optic stalk to the neuroretina

Kaede in the optic stalk was photo-converted in a way that only the stalk and as little as possible from the optic vesicle was red. The embryo was then imaged until the flow movements were over. From every third optical section (section thickness 4 µm), the area of the optic cup was measured using the measure tool in Fiji. In the same sections, the red area was also measured and a percentage of red area in the optic cup after fissure development was calculated. This was done in three animals.