Live imaging of retinotectal mapping reveals topographic map dynamics and a previously undescribed role for Contactin 2 in map sharpening

Organization of neuronal connections into topographic maps facilitates the efficient transfer of information between brain regions. In the visual system, retinal projections transmit a continuous representation of the external world by maintaining the neighboring relationship of the neurons they originate from in the retina (Cang and Feldheim, 2013; Triplett, 2014). Along the antero-posterior (A/P) axis, retinal ganglion cells (RGCs) in the nasal retina project axons to the posterior optic tectum [or superior colliculus (SC) in mammals], whereas temporal RGCs innervate the anterior tectum. As Sperry first postulated in his chemoaffinity hypothesis (Sperry, 1963), studies in mouse, chick, Xenopus and fish have demonstrated that this precise retinotopic map is first established by specific axon-target interactions, whereby axons with a specific set of receptors interpret guidance cues distributed in a gradient at the target. Subsequently to this guidance process, retinotopic projections are refined by activity-dependent mechanisms triggered by spontaneous retinal activity or visual experience (Kutsarova et al., 2016; Leighton and Lohmann, 2016; Thompson et al., 2017).

It is now well accepted that both guidance cues at the target and patterned retinal activity act together to establish a precise retinotopic map (Benjumeda et al., 2013; Cang et al., 2008; Pfeiffenberger et al., 2006). However, other mechanisms such as repulsive, competitive and stabilizing interactions among axons themselves also participate in initial mapping and refinement (Gosse et al., 2008; Hua et al., 2005; Louail et al., 2020; Rahman et al., 2020; Spead and Poulain, 2020a; Suetterlin and Drescher, 2014; Weth et al., 2014). Yet, our understanding of how and when trans-axonal signaling contributes to retinotopic mapping has remained limited by our inability to selectively manipulate RGCs in a topographic manner. It also remains unclear how the retinotopic map becomes sharper as a whole, as we currently lack the ability to observe the map forming and refining over time in the same embryo in vivo. Retinotopy can be analyzed at early stages by injecting lipophilic dyes or electroporating DNA in specific retinal quadrants. However, both approaches often require fixing specimens for analysis and have some degree of variability, which precludes the study of mapping dynamics and the detection of subtle topographic changes between times or conditions.

Because of its rapid external development and transparency, the zebrafish larvae has become a model of choice for studying retinotopy (Förster et al., 2020; Poulain et al., 2010). Injection of lipophilic dyes in opposite retinal halves has been used extensively to label retinotopic projections during development or regeneration and identify mutants with retinotopic defects (Baier et al., 1996; Harvey et al., 2019; Stuermer, 1988; Trowe et al., 1996; Xiao et al., 2005). Using that approach, early studies have shown that nasal and temporal retinal axons are localized at retinotopic sites as early as 3 days post-fertilization (dpf), with temporal and nasal axons innervating the anterior and posterior tectum, respectively (Stuermer, 1988; Stuermer et al., 1990). Labeling a subset of RGCs in larvae fixed at 4 and 6 dpf has also revealed that projections cover a smaller territory at later stages, suggesting a refinement of the retinotopic map over time (Gnuegge et al., 2001). That reduced coverage is not observed in larvae treated with the voltage-gated sodium channel blocker tetrodotoxin, indicating a role for neuronal activity in retinotopic map maturation. Although these observations highlight similar mechanisms underlying precise retinotopy in zebrafish and other species, we still do not know exactly when and how the retinotopic map sharpens and matures in teleosts. Advances in molecular genetics have allowed the generation of multiple transgenic lines for analyzing the lineage and functions of neuronal populations in zebrafish (Kawakami et al., 2016; Portugues et al., 2013; Robles, 2017), but the lack of enhancers driving transgene expression in specific retinal quadrants has precluded a similar unbiased analysis of retinotopic mapping over time in vivo.

Here, we report the generation of the first genetic model allowing the dynamic and quantitative analysis of retinotopic map formation and refinement directly in vivo. We show that an enhancer located upstream of hmx1 on chromosome 1 drives selective transgene expression in the nasal retina throughout development. We used Cre-mediated recombination of an RGC:colorswitch reporter to specifically label nasal and temporal retinal axons in vivo and image their projection domains at the tectum from 3 to 6 dpf by live confocal microscopy. Our analysis reveals that the A/P retinotopic map forms early but remains dynamic, with nasal and temporal axons expanding their projection domains over time. We further show that nasal projections initially arborize in the anterior half of the tectum but progressively refine and condense their projection domain to the posterior tectum from 4 to 5 dpf. That refinement coincides with a sharpening of the retinotopic map along the A/P axis. We finally demonstrate that the adhesion molecule Contactin 2 (Cntn2) is required for the refinement of nasal projections. Altogether, our study provides the first analysis of a topographic map maturing in real time in a live animal and identifies Cntn2 as a previously undescribed regulator of retinotopic map sharpening in vertebrates.

To identify potential enhancers that would drive specific expression in the nasal or temporal retina in zebrafish, we first assessed genes that were previously reported to be regionally expressed in the retina. Among them, the transcription factor hmx1 has been specifically detected in the nasal retina of several vertebrates (Boisset and Schorderet, 2012; Schulte and Cepko, 2000; Takahashi et al., 2003; Yoshiura et al., 1998). Given its role in retinal patterning, we decided to further analyze and quantify hmx1 expression throughout retinotectal development by in situ hybridization (ISH). At 24 h post-fertilization (hpf), when optic cup morphogenesis is complete (Kwan et al., 2012), hmx1 was strongly expressed in the nasal retina and lens and was also faintly detected in the otic vesicle (Fig. 1A,A′). At 48 hpf, when first retinal axons reach the tectum (Burrill and Easter, 1995; Stuermer, 1988), hmx1 remained strongly expressed in the nasal retina and otic vesicle and could also be detected in the pharyngeal arches (Fig. 1B,B′,E). Interestingly, although hmx1 expression remained stable in the otic vesicle and pharyngeal arches over time, it became restricted to the RGC layer in the nasal retina at 72 hpf (Fig. 1C,C′,F) and 96 hpf (Fig. 1D,D′,G). Hmx1 expression could also be detected at lower levels in the nasal inner nuclear layer at both stages (Fig. 1F,G). We further quantified hmx1 expression levels in the RGC layer along a 360° clockwise trajectory at 48, 72 and 96 hpf (Fig. 1H) and found a sharp gradient of expression along the nasal-temporal axis, with hmx1 being highly expressed in the nasal half of the retina but absent from the temporal half at all three stages (Fig. 1I). As hmx1 and its paralog hmx4 arose from tandem duplication and are tightly linked on the same chromosome in chick and zebrafish (Adamska et al., 2001; Wotton et al., 2009), we also analyzed the expression of hmx4 during retinal development (Fig. S1). As previously described in chick (Deitcher et al., 1994; Schulte and Cepko, 2000), hmx4 had a similar expression to that of hmx1 and was strongly detected in the nasal retina from 24 to 96 hpf.

The restricted expression of hmx1 and hmx4 in the nasal retina prompted us to search for potential enhancers regulating their expression. Transcriptional enhancers are cis-regulatory elements containing short DNA sequences bound by specific transcription factors. Their activity has been correlated with the enrichment of specific post-translational modification of histones, allowing the prediction of their position in the genome. Active enhancers are notably associated with the presence of histone H3 lysine 4 monomethylation (H3K4me1) and H3K27 acetylation (H3K27ac), whereas H3K4me3 is predictive of active promoters (Bonn et al., 2012; Heintzman et al., 2007; Rada-Iglesias et al., 2011). We thus analyzed the genomic tracks of H3K4me3, H3K4me1 and H3K27ac modifications previously generated by Bogdanovic et al. (2012) to identify putative distal regulatory elements in the hmx1/4 locus region on chromosome 1 (Fig. 2A). We identified three regions that were characterized by the genomic co-localization of H3K4me1 and H3K27ac marks at 24 (data not shown) and 48 hpf (Fig. 2A). A first 7 kb putative element, hmx1-En1, was located immediately upstream of the hmx1 gene, a second 7 kb putative element, hmx1-En2, was located more distally and a third 1.8 kb putative element, hmx1-En3, was located between the hmx1 and hmx4 genes. We tested the enhancer activity of these sequences by generating stable transgenic lines expressing EGFP targeted to the plasma membrane by the CAAX prenylation motif of Ras (Moriyoshi et al., 1996) under the control of each of these elements. Although hmx1-En3 did not exhibit any enhancer activity, hmx1-En1 and hxm1-En2 drove EGFPCAAX expression in specific and partially overlapping regions at 96 hpf. Both enhancers were active in the pharyngeal arches and lips, but only hmx1-En2 drove EGFP expression in the nasal retina and lens (Fig. 2B-C′). As developmental enhancers can be found in evolutionarily conserved regions (Irimia et al., 2012; Woolfe et al., 2005), we used a multi-species alignment [visualized in the University of California, Santa Cruz (UCSC) genome browser; Kent et al., 2002] to identify conserved domains within hmx1-En2. We delineated a 3 kb region, hmx1-En2s that was conserved across teleosts and amphibians. However, that region did not exhibit any enhancer activity despite its location within hmx1-En2.

As hmx1-En2 was the only enhancer driving expression in the nasal retina at 96 hpf, we analyzed its activity throughout retinal development in more detail (Fig. 3). At 24 hpf, before RGCs differentiate (Hu and Easter, 1999; Laessing and Stuermer, 1996; Schmitt and Dowling, 1999), EGFP expression was strongly detected in the nasal retina and lens of Tg[hmx1-En2:EGFPCAAX] transgenic embryos (Fig. 3A-B′). EGFP remained selectively expressed in the nasal retinal half at 48 (Fig. 3C-D′) and 96 hpf (Fig. 3E-F′). Interestingly, although hmx1 transcripts were only detected in the RGC layer at 72 and 96 hpf (Fig. 1F,G), EGFP remained visible in the entire nasal retina, likely because of its perdurance in vivo. Like hmx1 transcripts, EGFP was also found in other structures and was noticeably detected in the midbrain at 96 hpf (Fig. 3E,E′). To determine whether hmx1-En2 could drive transgene expression in nasal RGCs, we crossed our Tg[hmx1-En2:EGFPCAAX] transgenic line to Tg[isl2b:TagRFP] transgenic fish that express TagRFP under the control of the RGC-specific isl2b promoter (Pittman et al., 2008; Poulain and Chien, 2013). Confocal analysis of double transgenic larvae at 96 hpf revealed that EGFP partially overlapped with TagRFP in the nasal retina (Fig. 3G-H′). Importantly, we could also detect EGFP in nasal retinal axons innervating the posterior half of the tectum (Fig. 3G,G′), indicating that hmx1-En2 is effective at driving selective expression in nasal RGCs at late developmental stages.

As the entire nasal retina and several brain structures beside the tectum were labeled in Tg[hmx1-En2:EGFPCAAX] transgenic larvae, we next sought to generate a stable transgenic line that would allow the direct visualization of the A/P retinotectal map in vivo. The Cre/loxP system has been employed extensively in zebrafish for conditional expression and lineage-tracing analyses (Kawakami, 2007; Mosimann et al., 2011; Yoshikawa et al., 2008). We therefore took advantage of that system to restrict transgene expression to nasal or temporal RGCs. We generated a Tg[hmx1-En2:cre] stable transgenic line that expresses cre in the nasal retina, and a Tg[isl2b:loxP-TagRFPCAAX-loxP-EGFPCAAX] stable reporter line that expresses a switch transgene in RGCs (hereafter referred to as Tg[RGC:colorswitch]). To ensure that the RGC:colorswitch reporter transgene has integrated in an optimal genomic location for Cre-dependent recombination and to eliminate any functional positional effect, we established three independent Tg[RGC:colorswitch] stable lines and tested their responsiveness to Cre by crossing them to Tg[hsp70l:cre] transgenic fish. We selected the Tg[RGC:colorswitch] reporter line, the progeny of which showed complete change of fluorescence in all RGCs following heat shock at 24 hpf (data not shown). We then crossed that line to generate a Tg[hmx1-En2:cre; RGC:colorswitch] double transgenic line (Fig. 4A), and analyzed double transgenic larvae by immunolabeling for EGFP and TagRFP at 4 dpf.

High resolution confocal imaging and 3D-rendering of double transgenic larvae revealed a bi-colored retinotectal map along the A/P axis (Fig. 4; Movie 1). We found that larvae had a bi-colored RGC layer in the retina, with nasal and temporal RGCs expressing EGFP and TagRFP, respectively (Fig. 4B,D-D″). We confirmed by ISH that tagRFP was specifically expressed by temporal and not nasal RGCs at 4 dpf (Fig. 4D‴) and later stages (Fig. S2). We then analyzed the projection domains of nasal and temporal retinal axons at the tectum. After elongating together along both branches of the optic tract (Fig. 4C), TagRFP-positive temporal axons terminated in the anterior tectal half (Fig. 4B″,C′), whereas EGFP-positive nasal axons projected through the anterior tectum to reach the posterior tectum (Fig. 4B‴,C″). The sharp boundary between the nasal and temporal projection domains appeared to split the tectal neuropil into two equivalent halves (Fig. 4B,B′,C). Thus, our observations indicate that hmx1:cre-mediated recombination can be used to drive selective transgene expression in nasal versus temporal RGCs. Our results also establish the Tg[hmx1-En2:cre; RGC:colorswitch] transgenic line as the first genetic model allowing the direct visualization of retinotopic mapping in vivo throughout development.

We next examined retinotopic mapping in living larvae from 3 to 6 dpf (Fig. 5) by establishing a consistent imaging and quantification method across larvae for reproducible and unbiased analyses (Fig. S3). Confocal stacks of the retinotectal system were consistently rotated along the x, y, and z-axes to orient all larvae in a similar and comparable manner (Fig. 5A,A′; Fig. S3A,A′). We then used maximal projections of rotated stacks to delineate several landmarks at the tectum and analyze the projection domains of nasal and temporal retinal axons (Fig. 5A″; Fig. S3B-B″).

At 3 dpf, when the retinotectal map can first be visualized in fixed embryos (Burrill and Easter, 1995; Stuermer, 1988), we found that EGFP-positive nasal axons had already elongated through the anterior half of the tectum to innervate the posterior half (Fig. 5B). Some nasal axons also seemed to arborize in the anterior tectal half, just rostral to the equator. On the other hand, TagRFP-positive temporal axons projected specifically to the anterior tectum and were not observed in the posterior half (Fig. 5B′). From 4 to 6 dpf, the projection domain of temporal axons expanded within the anterior tectal half, thereby pushing the TagRFP boundary towards the equator (Fig. 5C′-E″). Conversely, the projection domain of EGFP-positive nasal axons appeared denser and progressively more restricted to the posterior tectal half (Fig. 5C-E). To better analyze the dynamics of retinotopic map formation, we quantified the tectal area covered by nasal and temporal axons over time (Fig. 5F-J; Fig. S3B″). Overall, the total tectal coverage (area of the tectum covered by nasal and temporal axons) significantly increased from 3 to 6 dpf (P<0.001), indicating a continuous growth and innervation of the tectum (Fig. 5F). The area covered by temporal axons in the anterior half of tectum steadily and significantly increased as well (Fig. 5G). The arborization field of temporal axons (area covered by temporal axons over the entire tectum) also progressively expanded from 3 to 6 dpf (Fig. 5H), suggesting that the increasing innervation by temporal axons significantly contributes to the tectum growth. Finally, the area covered by EGFP-positive nasal axons in the posterior half of the tectum also steadily and significantly increased from 3 to 5 dpf (P<0.001), but then remained stable from 5 to 6 dpf (Fig. 5I). Thus, our results demonstrate that the A/P retinotopic map is formed early but remains dynamic, with both nasal and temporal projection domains expanding over time.

In contrast to temporal axons that reach the anterior tectum immediately, nasal retinal axons must navigate through the anterior half of the tectum to reach their correct target in the posterior half. Interestingly, we noticed that some nasal axons appeared to arborize in the anterior tectal half just rostral to the equator at 3 and 4 dpf (Fig. 5B,C). However, these arborizations were not as clearly observed at 5 dpf, as shown by the apparent decrease in EGFP intensity between 4 and 5 dpf (arrows in Fig. 5C,D). We analyzed in more detail the tectal coverage of nasal axons in the anterior half of the tectum. Strikingly, the area covered by nasal axons in the anterior tectum significantly decreased between 3 and 5 dpf but remained stable between 5 and 6 dpf (Fig. 5J). Although the values obtained at 6 dpf likely represent fluorescence from the nasal axonal bundles that have extended through the anterior tectum, the significant decrease in tectal coverage observed from 4 to 5 dpf suggests that nasal retinal projections in the anterior tectal half might refine during that period. We thus calculated a nasal axon mistargeting index (NAMI) as the ratio between the anterior and posterior tectal areas covered by nasal axons (Fig. 5K). That index significantly decreased between 3 to 4 dpf and 4 to 5 dpf, but remained stable between 5 and 6 dpf. We also established a refinement index corresponding to the change in the NAMI between two consecutive days (Fig. 5L). The refinement index was greater than 1 between 3 and 4 dpf, and 4 and 5 dpf, indicating a refinement of nasal retinal projections between these stages. That refinement was not due to an absence of Cre recombination in later-born axons, as no tagRFP transcripts could be detected in nasal RGCs at 4, 5 or 6 dpf (Fig. S2). In contrast, the refinement index averaged a value of 1 between 5 and 6 dpf (0.97±0.04; mean±s.e.m.), indicating that no detectable refinement occurred over that period. Thus, our results indicate that nasal retinal axons refine and condense their projection domain to the posterior tectum from 3 to 5 dpf.

To determine the effect of that refinement on the retinotopic map, we decided to analyze the sharpness of the map at 4 and 5 dpf. We used sum projections of rotated stacks to measure the mean fluorescence intensity of EGFP and TagRFP in bins of equal height distributed along the A/P axis of the tectum (Fig. S3C-C″). Interestingly, the mean EGFP intensity significantly decreased between 4 and 5 dpf in anterior bins 3 and 4, whereas it significantly increased in posterior bins 6-9 (Fig. 5M). We further analyzed the sharpness of the EGFP-TagRFP boundary by normalizing fluorescence intensities to their maximum values at 4 and 5 dpf and plotting them along the A/P axis (Fig. 5N,O). We then determined the distance from the anterior tectal boundary at which EGFP and TagRFP intensities reached 50% of their maximal value (dashed lines in Fig. 5N,O). Interestingly, EGFP_50% shifted from an averaged position of 40.73±13.16 µm at 4 dpf (anterior region of bin 3) to 53.45±9.92 µm (bin 4) at 5 dpf. In contrast, TagRFP_50% kept a similar location between bins 4 and 5 from 4 to 5 dpf. We next calculated a sharpness index corresponding to the absolute value of the distance between EGFP_50% and TagRFP_50% (double arrows in Fig. 5N,O). That index significantly decreased between 4 and 5 dpf (Fig. 5P), demonstrating that the boundary between nasal and temporal projection domains sharpens during that interval. Thus, our results demonstrate that the zebrafish retinotectal map sharpens and becomes more precise as nasal projections refine and disappear from the anterior tectum.

Our unique ability to analyze subtle dynamic changes in retinotopic mapping over time prompted us to test whether we could use our transgenic line to discover as yet unreported gene functions. Cntn2 is an adhesion molecule known to promote axon growth and fasciculation in several brain circuits (Mohebiany et al., 2014). It has been specifically detected in the nasal retina in zebrafish (Gurung et al., 2018; Lang et al., 2001; Warren et al., 1999), but its role in the retinotectal system has remained surprisingly uncharacterized. We therefore decided to test whether Cntn2 could regulate retinotectal mapping along the A/P axis. We first confirmed that cntn2 was strongly and selectively expressed in the nasal RGC layer from 3 to 6 dpf (Fig. 6A-B), suggesting it might regulate the targeting and/or maturation of nasal projections at the tectum. We then took advantage of the two-RNA component (crRNA:tracrRNA) version of the CRISPR system (Jacobi et al., 2017) to induce cntn2 mutations in transgenic embryos. We designed two independent crRNAs targeting sequences in exons 3 and 5 that encode the first and second N-terminal Ig-like domains of Cntn2, respectively. As previously described (Hoshijima et al., 2019), crRNA:tracrRNA:Cas9 ribonucleoprotein (RNP) complexes were highly effective at inducing mutations directly in injected ‘crispant’ embryos (Fig. S4). Crispant larvae injected with either RNP complex (gRNA1 or gRNA2) or a combination of both gRNAs were undistinguishable from Cas9-injected controls and did not exhibit any obvious morphological or developmental defects.

We then analyzed retinotopic mapping in cntn2 crispants at 4 and 5 dpf. Both crispants and controls had a bi-colored A/P retinotectal map at both stages (Fig. 6A-F′), with nasal axons innervating the posterior tectal half and temporal axons, the anterior half. Quantifications showed a similar increase in the total tectal coverage from 4 to 5 dpf in crispants and controls, indicating a normal tectal growth (Fig. 6G). We did not observe any difference in the temporal arborization field between crispants and controls (Fig. 6I), and the anterior tectal coverage of temporal axons increased similarly among groups from 4 to 5 dpf (Fig. 6H), suggesting that cntn2 mutations did not overly affect temporal projections. We noticed, however, that nasal projections did not appear to disappear from the anterior tectum at 5 dpf in crispants (Fig. 6C,D) compared with controls (Fig. 6E,F). Although the posterior tectal coverage of nasal axons increased in both crispants and controls (Fig. 6J), we observed significant differences in the anterior tectal coverage, which decreased in controls but remained constant or slightly increased in cntn2 crispants (Fig. 6K). Likewise, the NAMI significantly decreased in controls but not in crispants (Fig. 6L), demonstrating that the refinement of nasal projections does not occur when cntn2 is mutated. Interestingly, both anterior tectal coverage and NAMI remained remarkably constant from 5 to 6 dpf in crispants, further indicating that the nasal projection domain does not refine even at a later stage in the absence of Cntn2 (Fig. S5B,C). Lack of refinement was further confirmed by the refinement index that averaged 1 from 4 to 6 dpf in crispants (Fig. 6N; Fig. S5D). As the refinement of nasal projections correlates with a sharpening of the retinotopic map (Fig. 5), we next examined how the refinement defects observed in cntn2 crispants might affect topographic mapping. Analysis of the EGFP-TagRFP boundary sharpness at 4 and 5 dpf revealed that the distance between EGFP_50% and TagRFP_50% did not decrease in crispants compared with controls (Fig. S6). Moreover, the sharpness index remained stable in crispants instead of decreasing like in controls (Fig. 6M), demonstrating that cntn2 is necessary for retinotectal map sharpening. As cntn2 is known to modulate neurogenesis at early stages of nervous system development (Ma et al., 2008), we next tested whether the loss of cntn2 could affect retinal patterning along the A/P axis. Analysis and quantification of hmx1 expression throughout retinotectal development revealed a normal retinal patterning in cntn2 crispants (Fig. S7), indicating that the absence of tectal refinement is not due to a loss of RGC positional identity in the retina. Thus, by detecting subtle retinotectal targeting phenotypes in cntn2 crispants, our study unraveled a novel function for Cntn2 in refining and sharpening retinal projections during visual system maturation.

Our understanding of topographic map development and maturation has so far been limited by a lack of genetic models allowing the direct observation of maps over time. Here, we report the generation of a novel zebrafish transgenic line that, for the first time, enables the unbiased and quantitative analysis of retinotopic map formation and refinement directly in vivo. Using live confocal imaging of transgenic larvae from 3 to 6 dpf, we show that the A/P retinotopic map is formed at early stages but remains dynamic as both retina and tectum grow, with the projection domains of nasal and temporal axons expanding over time. We further demonstrate that nasal retinal projections initially arborize in the anterior tectal half but progressively refine and condense their projection domain to the posterior tectum from 4 to 5 dpf, leading to the sharpening of the A/P retinotopic map. Finally, we demonstrate that Cntn2, an adhesion molecule expressed in nasal RGCs, is required for the refinement of nasal retinal axons and retinotectal map sharpening.

In agreement with previous studies in zebrafish and other species (Boisset and Schorderet, 2012; Deitcher et al., 1994; Schulte and Cepko, 2000; Stadler and Solursh, 1994; Wang et al., 2000; Yoshiura et al., 1998), our data reveal that the homeobox transcription factors hmx1 and hmx4 are expressed in a sharp nasal-high to temporal-low gradient in the retina throughout development. The detection of hmx1 and hmx4 throughout the nasal retinal neuroepithelium at early stages indicates that both genes are expressed in proliferating neuroblasts and might regulate their positional identity and differentiation. Supporting that hypothesis, reduced hmx1 expression has been shown to block retinal cell differentiation in zebrafish (Boisset and Schorderet, 2012; Schorderet et al., 2008) and cause microphthalmia in mouse (Munroe et al., 2009) and human (Gillespie et al., 2015; Schorderet et al., 2008; Vaclavik et al., 2011). Although zebrafish embryos lacking functional hmx1 do not exhibit any eye patterning defect (Boisset and Schorderet, 2012), misexpression of hmx1 or hmx4 does alter the regional specification of the retina along the nasal-temporal axis in chick (Schulte and Cepko, 2000; Takahashi et al., 2009), suggesting that hmx1 and hmx4 have redundant functions in teleosts. Interestingly, our analysis shows that the expression of both hmx1 and hmx4 becomes restricted to the nasal RGC and inner nuclear layers at later stages. We have also identified a distal regulatory element upstream of hmx1 and hmx4 genes that drives expression in nasal RGCs at 4 dpf and later. Altogether, these data suggest that hmx1 and hmx4 are expressed by mature RGCs themselves at late stages of development. Supporting that observation, hmx1 transcripts have been detected in RGCs, horizontal cells and Müller glia of the adult human retina by single cell profiling (Cowan et al., 2020; Lukowski et al., 2019; Voigt et al., 2019). Examining the effects of manipulating hmx1 expression in nasal or temporal RGCs will thus be of great interest to better understand the progression of the oculoauricular syndrome caused by hmx1 mutations in human (Gillespie et al., 2015; Schorderet et al., 2008; Vaclavik et al., 2011).

Using hmx1:cre-mediated recombination of an RGC:colorswitch reporter, we have generated a novel transgenic line that enables the unprecedented visualization of the A/P retinotopic map in vivo. We were able to image and analyze for the first time retinotectal map development within the same larvae over successive days. This unparalleled temporal resolution allowed us to observe dynamic changes in retinotopic mapping that could not be seen previously in fixed embryos. We found that, although the A/P retinotopic map is established early on, it progressively shifts caudally as temporal axons expand their innervation of the anterior tectum. This caudal shift is accompanied by the progressive refinement of nasal projections that condense their projection domain to the posterior tectum. Interestingly, the parameters of retinotopic mapping we measured did not show much variability across larvae, indicating that retinotopic mapping is a robust and highly stereotyped process. That property allowed us to characterize in detail the dynamic changes underlying retinotopic map maturation. Notably, we found that nasal projections initially covering the caudal part of the anterior tectum refine and progressively condense their domain to the posterior half between 3 and 5 dpf. This refinement is unlikely to be caused by cell death in the retina, as apoptosis in the RGC layer peaks from 1.5 to 3 dpf before sharply decreasing from 4 to 6 dpf (Biehlmaier et al., 2001; Cole and Ross, 2001). Instead, it is likely driven by the dynamic rearrangement of axonal branching pattern as nasal axons extend caudally and arborize in their final zone in the posterior tectum. In Xenopus, nasal axons retract their branches from the anterior tectum after initiating branches in both anterior and posterior tectal halves (O'Rourke and Fraser, 1990). Although retinal axons in zebrafish were initially thought to elongate along straight trajectories and only arborize after reaching their target area (Kaethner and Stuermer, 1992), more recent studies using high-resolution time-lapse imaging have instead revealed that axons continuously extend and retract branches and navigate by selective branch stabilization (Kita et al., 2015; Simpson et al., 2013). Once in their termination area, axonal arbors remain highly dynamic, with only a small fraction of them being maintained in the mature arbor (Alsina et al., 2001; Ben Fredj et al., 2010; Campbell et al., 2007; Meyer and Smith, 2006; Munz et al., 2014; Ruthazer et al., 2006). A retraction of proximal branches that have extended in inappropriate areas coupled with a constant remodeling of arborizing axons might thus redistribute the position of branches and cause arbors to shift caudally, leading to the progressive refinement of nasal projections we observed.

Using a CRISPR-mediated functional approach in our retinotopic transgenic line, we found that the refinement of nasal projections requires Cntn2, a glycosylphosphatidylinositol (GPI)-anchored adhesion molecule selectively expressed in the nasal retina throughout development. Although previous studies had described the selective expression of cntn2 in the nasal retina in zebrafish (Gurung et al., 2018; Lang et al., 2001; Warren et al., 1999), none had reported a function for Cntn2 in retinotopic mapping, likely because the labeling methods available then did not allow for the visualization of nasal projections refining over time in a living animal. Our detection and unbiased analysis of retinotectal maturation in vivo revealed that retinotopic map sharpening is prevented in cntn2 crispants, indicating that the remodeling of nasal retinal projections strongly contributes to the increasing precision of the retinotopic map. Interestingly, we did not observe any defects in the initial tectal coverage of nasal or temporal axons in cntn2 crispants, suggesting that retinal axon elongation and guidance are not affected in the absence of Cntn2. The lack of an earlier phenotype is unlikely to be caused by an incomplete knockdown, as we did not detect any additional phenotypes or worsened refinement defects in crispants injected with a combination of gRNAs (Fig. S8). Our results rather highlight a specific, later role for Cntn2 in the visual system that contrasts with its known functions at earlier developmental stages in other circuits (Masuda, 2017). Indeed, Cntn2 is known to regulate axon growth, guidance and fasciculation in several systems (Mohebiany et al., 2014). It is required for the proper guidance of spinal commissural axons and cerebellar granule cell axons in chick (Baeriswyl and Stoeckli, 2008; Fitzli et al., 2000; Stoeckli and Landmesser, 1995). In mice, it regulates the fasciculation of dorsal root ganglion (DRG) axons (Law et al., 2008) and controls the guidance and fasciculation of motor axons, preventing them from entering DRGs (Suter et al., 2020). In zebrafish, Cntn2 promotes the growth and fasciculation of axons extending from the nucleus of the medial longitudinal fascicle (Gurung et al., 2018; Wolman et al., 2008). Cntn2 is the only member of the contactin family known to participate in homophilic interactions (Mohebiany et al., 2014). It also engages in heterophilic interactions with other adhesion molecules (Buchstaller et al., 1996; Fitzli et al., 2000; Kuhn et al., 1991; Kunz et al., 1998; Suter et al., 1995), raising the possibility that the lack of refinement observed in cntn2 crispants might result from a fasciculation defect among nasal axons that would delay nasal axon extension to the posterior tectum. However, we did not observe any fasciculation defects of nasal axons as they navigate along the optic chiasm and optic tract. The lack of tectal refinement persisting at late stages in cntn2 crispants also argues against a delayed phenotype. Instead, Cntn2 might regulate the selective retraction of proximal nasal axonal branches that have inappropriately extended in the anterior tectum. Cntn2 might, for example, modulate the responsiveness of axonal branches to a repellant cue enriched in the anterior tectal region, just as it regulates the sensitivity of sensory axons to semaphorins (Law et al., 2008; Dang et al., 2012).

Alternatively, Cntn2 might promote retinotectal refinement by modulating the activity of RGCs. Spatiotemporal patterns of retinal activity are indeed known to drive the sharpening of visual circuits in vertebrates (Burbridge et al., 2014; Cang et al., 2005; Chandrasekaran et al., 2005; Dhande et al., 2011; Hiramoto and Cline, 2014; Munz et al., 2014; Rahman et al., 2020; Ruthazer et al., 2003; Stellwagen and Shatz, 2002; Xu et al., 2015, 2016; Zhang et al., 2011). Activity is notably required for the elimination of branches with a firing pattern that does not match that of their neighbors. Consequently, blocking neuronal activity with tetrodotoxin causes enlarged axonal arbors in frog (Cohen-Cory, 1999; Reh and Constantine-Paton, 1985) and prevents the refinement of retinal fibers that overshoot their termination zone in chick (Kobayashi et al., 1990). Similarly in mouse, axons occupy a larger area in the SC after RGCs have been silenced (Benjumeda et al., 2013). In zebrafish, altering RGC activity also modifies the size and morphology of terminal arbors (Ben Fredj et al., 2010; Gnuegge et al., 2001; Hua et al., 2005; Kaethner and Stuermer, 1994; Schmidt et al., 2000; Smear et al., 2007; Stuermer et al., 1990), and the projection field of silenced retinal axons appears to be more diffuse (Ben Fredj et al., 2010; Gnuegge et al., 2001). Interestingly, we observed a lack of nasal axon refinement and map sharpening between 4 and 5 dpf in zebrafish maco mutants (Spead and Poulain, 2020b preprint), in which a downregulation of voltage-gated sodium channels causes a lack of neural activity in RGCs and peripheral sensory neurons (Gnuegge et al., 2001; Granato et al., 1996; Ribera and Nüsslein-Volhard, 1998). Maco mutants harbor a mutation in pigk, a component of the transamidase complex responsible for GPI anchor synthesis and attachment to nascent proteins (Carmean et al., 2015; Ohishi et al., 2000). Although it remains unknown which GPI-anchored proteins are impaired in maco, Cntn2 appears as a good candidate considering the shared phenotype between maco mutants and cntn2 crispants. Moreover, Cntn1, another member of the contactin family, is known to interact with several voltage-gated sodium channels via its fibronectin type III-like (Fn-III) domains and to increase their density at the plasma membrane (Chen et al., 2004; Kazarinova-Noyes et al., 2001; Liu et al., 2001; McEwen and Isom, 2004; McEwen et al., 2004; Rush et al., 2005; Shah et al., 2004). Considering the high conservation between the Fn-III domains of Cntn1 and Cntn2, Cntn2 might be able to interact with sodium channels as well. An absence of functional Cntn2 would then alter the distribution of sodium channels in nasal retinal axons, leading to a lack of activity-dependent refinement and map sharpening at the tectum. Conversely, neuronal activity might regulate the targeting of Cntn2 to the plasma membrane, just as it modulates that of Cntn1 in hypothalamic axons (Pierre et al., 2001). Cntn2 would then act downstream of neuronal activity for mediating activity-dependent morphological remodeling. By enabling the visualization of map sharpening over time as well as the selective manipulation of nasal and temporal RGCs, our new genetic model will provide new strategies for analyzing the molecular mechanisms by which Cntn2 and activity cooperate for precise retinotopic mapping in vivo.

All experiments and procedures were approved by the Institutional Animal Care and Use Committee of the University of South Carolina. Zebrafish wild-type (WT) and transgenic embryos were obtained from natural matings, raised at 28.5°C in E3 medium (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl₂ and 0.33 mM MgSO₄) in the presence of 150 mM of 1-phenyl-2-thiourea (PTU; Millipore Sigma) to prevent pigment formation, and staged by age and morphology (Kimmel et al., 1995). WT embryos were from the Tübingen or AB strains. Embryos were anesthetized in tricaine-S (Western Chemicals) before fixation or imaging. Zebrafish larvae and young fish were nurtured using rotifer suspension and dry food (Gemma 75 and 150, Skretting USA). Adult fish were fed with dry food (Gemma 300, Skretting USA).

For cloning hmx1, hmx4 and cntn2 cDNAs, zebrafish mRNA was isolated from embryos at 24, 48 and 96 hpf using Trizol and the RNeasy mini kit (Qiagen), and cDNA was prepared from RNA using the SuperScript III First-Strand Synthesis system (Invitrogen). Hmx1, hmx4 and cntn2 cDNAs were amplified using the following full length primers: hmx1-fw, 5′-ATGCATGAAAAAAGCCAGCAACAGC-3′; hmx1-rv. 5′-TCAGACAAGGCCTGTCATCTGC-3′; hmx4-fw, 5′-ATCTAACGGAGAATATGAGCAAGGAG-3′; hmx4-rv, 5′-TCATATATCTCCATCAAACAGGCTGAAATAC-3′; cntn2-fw, 5′-ATGAGGATTCTGTTGTGTCTG-3′; cntn2-rv, 5′-TCACAGTCCTGATGAGCCAA-3′. Amplicons were subcloned into PCRII-TOPO (Invitrogen) and sequenced to verify gene identity and confirm sequence orientation (sequencing by Eton Bioscience).

Hmx1 putative enhancer elements were amplified by PCR from total genomic DNA using the LA Taq PCR kit v2.1 (TaKaRa) and the following primers: hmx1-En1-fw, 5′-ACCGCACCACTAAAGAGTCACAG-3′; hmx1-En1-rv, 5′-GGGTGATACGTGAATACCTCTAAGCA-3′; hmx1-En2-fw, 5′-GAGGGTGCCAGATGGAGATACAC-3′; hmx1-En2-rv, 5′-ACTGGCTCTGCTATGCTTCTGTTTC-3′; hmx1-En2s-fw, 5′-GAACGGTACCGAACCGTCTATTAAAAGATTACACTAC-3′ (KpnI restriction site added in primer); hmx1-En2s-rv, 5′-GAACGGATCCAATAAACAAGGGACTAATAATTCAAGG-3′ (BamHI restriction site added in primer); hmx1-En3-fw, 5′-GAACGGTACCTCTTTGGAGACTGGCTGAACTGAC-3′ (KpnI restriction site added in primer); hmx1-En3-rv, 5′-GAACGGATCCATTCTCCGTTAGATGCGGGTCC-3′ (BamHI restriction site added in primer). Amplicons were purified on gel using the Qiaquick gel extraction kit (Qiagen), subcloned into PCRII-TOPO (Invitrogen) and sequenced before being digested and ligated into the Gateway p5E-MCS entry vector (Kwan et al., 2007) using the following restriction endonucleases: XhoI/BamHI (for hmx1-En1), DraII (for hmx1-En2) and KpnI/BamHI (for hmx1-En2s and hmx1-En3).

All transgenesis constructs were generated using the Tol2kit Gateway cloning system (Kwan et al., 2007). To generate hmx1 reporter constructs, p5E-hmx1-En1, p5E-hmx1-En2, p5E-hmx1-En2s, p5E-hmx1-En3, pME-EGFPCAAX and p3E-polyAv2 were recombined into the pDestTol2pA2 destination vector using a Gateway Multisite LR reaction (Kwan et al., 2007; Invitrogen). p5E-hmx1-En2, pME-iCre and p3E-polyAv2 were recombined into pDestTol2CR3 (pDestTol2pA3 with myl7:TagRFP transgenesis marker) to generate the hmx1-En2:iCre construct. The sequence encoding loxP-TagRFPCAAX-polyA-loxP was amplified by PCR, purified on gel using the Qiaquick gel extraction kit (Qiagen) and recombined into the pDONR221 destination vector to generate the pME-loxP-TagRFPCAAX-polyA-loxP entry vector. A modified p5E-isl2b-gata2a entry clone encoding a 7.6 kb genomic fragment upstream of the isl2b start codon fused to the 1 kb promoter of gata2a (Ben Fredj et al., 2010; Pittman et al., 2008), pME-loxP-TagRFPCAAX-polyA-loxP, and p3E-EGFPCAAX-polyA were recombined into the pDestTol2pA2 destination vector to generate the isl2b:loxP-TagRFPCAAX-loxP-EGFPCAAX (RGC:colorswitch) reporter construct.

Stable transgenic lines were generated using the Tol2 transposon method as described previously (Kawakami et al., 2000). We co-injected 10 to 40 pg of purified DNA (pTol2pA2-hmx1-En1:EGFPCAAX, pTol2pA2-hmx1-En2:EGFPCAAX, pTol2pCR3-hmx1-En2:iCre, RGC:colorswitch) with 25 pg of synthetic mRNA encoding Tol2 transposase at the one-cell stage, and injected embryos with transient expression of transgenes were raised up to adulthood as F0 generation. F0 fish were then out-crossed to WT to screen for positive F1 embryos expressing the transgenes. Expression of TagRFP driven in the heart by the myl7 promoter was used to identify hmx1-En2:iCre carriers. Transgenic F1 carriers were subsequently out-crossed to WT to generate stable lines with a single-copy insertion.

Potential gRNA target sites were identified using the proprietary Alt-R^® CRISPR-Cas9 guide RNA design platform developed by Integrated DNA Technologies (IDT). Two independent target sequences in two different exons were chosen based on the high predicted editing performance score (above 65) of the corresponding gRNA and the lack of potential off-target sites with fewer than 2 bp mismatches in the GRCz11 genome. The cntn2 target sequences (with PAMs in brackets) used in this study were: cntn2 gRNA1, TGAAGAGTCGCACTACACAC[AGG] (targeting exon 4 of ENSDART00000000486.9 Ensembl transcript); and cntn2 gRNA2, GGAACTCGTTGATAAACCAG[CGG] (targeting exon 6). Target-specific Alt-R^® crRNA and universal Alt-R^® tracrRNA were synthesized by IDT. Each RNA was dissolved in nuclease-free duplex buffer (IDT) as a 100 mM stock solution that was stored at −80°C. To prepare the crRNA:tracrRNA duplex, equal volumes of 100 mM Alt-R^® crRNA and tracrRNA stock solutions were mixed together, heated for 5 min at 95°C and annealed by gradual cooling at room temperature. Cas9 protein (Alt-R^® S.p. Cas9 nuclease V3, IDT) was diluted to 1 μg/μl in 30 mM HEPES (pH 7.5), 100 mM potassium acetate, aliquoted and stored at −20°C. CrRNA:tracrRNA:Cas9 ribonucleoprotein (RNP) complexes were assembled by gently mixing 3 μl of crRNA:tracrRNA duplex and 3 μl of Cas9 stock protein. The RNP solution was heated for 10 min at 37°C and allowed to cool at room temperature. Then 1 μl of 0.25% phenol red solution was added to the RNP complex solution before microinjection. Approximately 2 nl of 5 μM RNP complex solution was injected into the cytoplasm of one-cell-stage embryos. Control embryos were injected similarly with Cas9 only.

After final imaging of the retinotectal map, genomic DNA was extracted from Cas9 control and cntn2 gRNA-injected larvae for analyzing gRNA efficiency. Individual larvae were incubated in 20 μl of 50 mM NaOH at 95°C for 25 min. After cooling to 4°C, 2 μl of 1 M Tris-HCl (pH 8.0) was added for neutralization. The presence of mutations at the target sites in exons 4 and 6 was assessed by HRMA in all larvae analyzed, as previously described (Dahlem et al., 2012; Parant et al., 2009). Amplicons including the entire genomic target site were generated using the following primers: gRNA1-HRMA-fw, 5′-ATGGCACTGACATATCATTTG-3′; gRNA1-HRMA-rv, 5′-CCATTCTGAGGGTTGTTGA-3′; gRNA2-HRMA-fw, 5′-CTGTTTGTTAGCTTTGTCCTAC-3′; gRNA2-HRMA-rv, 5′-GAAACAAACCACCTGCCC-3′. To confirm HRMA results, larger genomic sequences including the target sites were amplified from a subset of larvae and analyzed by sequencing. Amplicons to be sequenced were generated using the following primers: gRNA1-seq-fw, 5′-GTGTGTGATTTCCTCCCAG-3′; gRNA1-seq-rv, 5′-CCGAACTTGAGATTAGCGAC-3′; gRNA2-seq-fw, 5′-GAACATTTCTTGCCTGCCAG-3′; gRNA2-seq-rv, 5′-GCTGAAGATGCTCTTGGTGC-3′.

For making ISH probes, cDNA templates cloned into pCRII-TOPO were amplified by PCR using M13fw and M13rv primers and purified on gel. In vitro transcription of digoxigenin-labeled probes was performed using the DIG RNA Labeling Kit (Millipore Sigma) according to the manufacturer's instructions. Embryos were dechorionated at the appropriate developmental stages and fixed in 4% paraformaldehyde in phosphate buffered saline (PBS; pH 7.4) for 2 h at room temperature and overnight at 4°C. Whole-mount ISH was performed as previously described (Thisse and Thisse, 2008). After staining, embryos were cleared in 80% glycerol for imaging. Sense probes were used as controls and did not reveal any staining. Images were acquired using an Olympus SZX16 stereomicroscope equipped with an Olympus DP80 dual color camera and Cellsens standard software. Digital images were cropped and aligned using Adobe Illustrator.

Quantification of gene expression in the retina was carried out according to Picker and Brand (2005) with the following modifications: eyes were dissected from embryos stained by ISH using a sharpened tungsten needle and imaged in 80% glycerol in a lateral view as described above. Images were imported into Fiji ImageJ analysis software (Schindelin et al., 2012; Schneider et al., 2012), transformed to 8-bit grayscale images and inverted. An oval selection was applied half-way between the lens and the RGC layer periphery, and signal intensity was measured along a 360° trajectory using the ‘Oval Profile Plot’ plugin. Values were exported and analyzed in Microsoft Excel.

Larvae were fixed in 4% paraformaldehyde in PBS for 2 h at room temperature and then overnight at 4°C. Larvae were washed three times in PBT (PBS+0.5% Triton X-100). Antigen retrieval was carried out in 150 mM Tris (pH 9) for 5 min at room temperature followed by 20 min at 70°C. Larvae were then permeabilized at room temperature for 15 min in water first, and then for 30 min in PBS with 1% Triton and 0.1% collagenase. Larvae were blocked for 1 h at room temperature with blocking buffer [PBS with 0.5% Triton, 1% DMSO, 1% bovine serum albumin (BSA), and 2% normal goat serum]. Primary anti-EGFP (ab290, Abcam) and anti-TagRFP (M204-3, MBL International) antibodies were applied at 1:500 dilution in PBT supplemented with 1% DMSO and 1% BSA overnight for 4°C. Larvae were washed three times in PBT. Secondary Alexa Fluor 488 goat anti-rabbit (111-545-003, Jackson ImmunoResearch) and Alexa Fluor 594 donkey anti-mouse (715-585-150, Jackson ImmunoResearch) antibodies were diluted 1:500 in PBT with 1% DMSO and 1% BSA and applied together with TO-PRO-3 (T3605, Thermo Fisher Scientific) diluted 1:1000 overnight at 4°C. Larvae were washed five times in PBT and mounted in 1% ultrapure low melting point (LMP) agarose (16520050, Thermo Fisher Scientific) for confocal imaging.

For time-course imaging of live larvae from 3 to 6 dpf, larvae were anesthetized in 0.015% tricaine-S and embedded dorsally in 1% LMP agarose in E3 medium+PTU in a lumox membrane-bottomed dish (Greiner Bio-one). Images were acquired on a Leica TCS SP8X laser-scanning confocal microscope equipped with LASX software, HyD detectors and a 20× objective. z-series of the entire retinotectal system were acquired at 512×512 pixel resolution with a zoom of 1 and z-intervals of 1.5 µm. After imaging, larvae were kept individually in a 12-well dish and allowed to recover for 24 h before being re-anesthetized and mounted dorsally for the next day of imaging. Maximal and sum intensity projections were compiled at each time point in ImageJ software.

For high resolution imaging of the retinotectal system at 4 dpf, immunolabeled larvae were mounted either laterally after removing the contralateral eye or dorsally in 1% LMP agarose and imaged as described above with a z-interval of 1 µm. 3D reconstructions of the retinotectal system were generated using FluoRender (Wan et al., 2012, 2017).

All quantitative analyses of topographic mapping were conducted using ImageJ software. For unbiased analysis, dorsal view z-series were consistently rotated along the x, y and z-axes using the TransformJ Plugin (Meijering et al., 2001), so that the left and right tecta were aligned horizontally, both optic tracts intersected at an angle of 60° and the roundness of the right tectal neuropil was equal to 1. To analyze the area of coverage, rotated images were maximally projected and binarized using a threshold of 80 for TagRFP and of 75 for EGFP. Thresholds were chosen to best represent the raw images acquired. Using the TagRFP channel, we delineated the anterior tectal boundary as the anterior line where retinal axons enter the tectum, and the TagRFP posterior boundary. Using the EGFP channel, we delineated the posterior tectal boundary as the caudal-most border where retinal axons arborize at the tectum. We defined the equator as half the length of the tectum measured along the A/P axis between the anterior and posterior tectal boundaries. We used the ‘Analyze Particles’ tool in ImageJ to measure the tectal coverage (area) of the TagRFP-positive temporal axons, of the EGFP-positive nasal axons in the anterior half of the tectum (rostral to the equator) and of the EGFP-positive nasal axons in the posterior half of the tectum (caudal to the equator). We calculated the total tectal coverage as the sum of the TagRFP and posterior-EGFP axonal coverages. To further analyze retinotopic mapping along the A/P axis, we established NAMI as the ratio between the EGFP area of coverage in the anterior half of the tectum and the EGFP area of coverage in the posterior half of the tectum. We calculated a refinement index as the ratio change of the NAMI between two consecutive days.

To quantify the sharpness of the boundary between the TagRFP and EGFP projection domains, rotated images were sum-projected for measuring the mean fluorescence intensity of the TagRFP and EGFP signals. We divided the A/P length of the tectum into 10 bins of equal height on merged images using the ‘Polygon Selection’ tool in ImageJ, and measured the mean fluorescence intensity of each channel in each bin using the ‘Measurement’ function in ImageJ. Bin 1 was defined as the anterior-most tectal bin and bin10 as the posterior-most tectal bin. Intensity values were normalized to the maximum value for each channel and plotted along the antero-posterior axis. We determined the point at which fluorescence intensity reached 50% of its normalized maximum value for each channel, and defined the sharpness index as the absolute distance between the EGFP_50% and TagRFP_50% positions at 4 and 5 dpf (see Fig. S3 for a detailed illustration of our quantifications).

All statistical analyses were performed using GraphPad Prism 9 software. We define biological replicates as individual larvae from a mixed clutch born from pairings of at least two males and two females. Each experiment was repeated under similar experimental conditions. Sample size was decided based on the low variability detected in pilot studies. Data are presented as mean±s.e.m. For better communicating variability across samples and experimental reproducibility, graphs are represented as ‘SuperPlots’ (Lord et al., 2020), in which biological replicates representing independent experiments are color-coded, with circles representing individual larvae tested and triangles representing averages. We compared groups with repeated measures over time (Fig. 5F-K) using repeated measures one-way ANOVA with Tukey's multiple comparisons post-hoc test. We analyzed the refinement index using a two-tailed, paired t-test compared with 1 (1 representing no change). We analyzed the sharpness index in Fig. 5P using a two-tailed, paired t-test, and the EGFP mean distribution in Fig. 5M using a two-tailed, paired t-test within each bin. We compared data with both ‘within-subjects’ and ‘between-subjects’ factors (Fig. 6G-M) using mixed effects one-way ANOVA followed by Tukey's multiple comparisons post-hoc test. The number of larvae analyzed are described for each experiment in the figure legends. In all figures, *P<0.05, **P<0.01 and ***P<0.001.

We thank the expert help of Dr Edsel Pena (University of South Carolina) for assistance with statistical analyses, and Quill Thomas for technical assistance and fish husbandry.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.199584.