Visual system development of the spotted unicornfish, Naso brevirostris (Acanthuridae)

Many animals use vision to perform important behavioural tasks such as feeding, mating, avoiding predators and finding a suitable home (Cronin et al., 2014). At the core of the vertebrate visual system is the retina, an extrusion of the brain which is subdivided into various functional layers, two of which are at the centre of this study, the photoreceptor layer and the ganglion cell layer.

The photoreceptor layer is the first stage of visual processing and is composed of morphologically diverse cone and rod photoreceptor cells which absorb light, transform it into an electrical signal, and send the information downstream to various neural cells via the phototransduction cascade. Cones mediate vision in bright-light conditions and colour vision, while rods mediate vision in dim-light conditions (Walls, 1942). Cones can further be classified into different types depending on their morphology and/or the type of photopigment (an opsin protein covalently bound to a light-absorbing chromophore) they possess (Hunt et al., 2014). Morphologically, cones can be classified as single, double, triple or quadruple, although only the first two configurations are common and are often arranged in regular and specific patterns or mosaics (Bowmaker and Kunz, 1987; Peichl et al., 2004). Molecularly, cones are also classified into four types based on the opsin genes they express, which encode different protein classes that are sensitive to different parts of the visible light spectrum: a short-wavelength protein class 1 opsin (SWS1) maximally sensitive to UV–violet wavelengths (355–450 nm peak spectral sensitivity, λ_max), a second short-wavelength class opsin (SWS2) maximally sensitive to the violet–blue part of the spectrum (410–490 nm λ_max), a middle-wavelength class 2 opsin (RH2) maximally sensitive to blue–green wavelengths (435–535 nm λ_max) and a long-wavelength class opsin (LWS) maximally sensitive to the green–red part of the light spectrum (490–625 nm λ_max) (Bowmaker, 2008; de Busserolles et al., 2017; Torres-Dowdall et al., 2017). In percomorph fishes, SWS proteins are found in single cones, whereas RH2 and LWS opsins occur in double cones (Hunt et al., 2014). Most vertebrates possess a single type of rod photoreceptor expressing the rod opsin protein (RH1; 460–530 nm λ_max; Bowmaker, 2008; but see Musilova et al., 2019, for deep-sea fishes with multiple RH1s that extend their sensitivities to ∼440 nm λ_max).

The ganglion cell layer is the last stage of visual processing in the retina and is composed of ganglion cells that possess axons that reach to the inner surface of the retina and converge into the optic nerve to send the information into the central nervous system (Walls, 1942). Therefore, the arrangement and spacing between ganglion cells is one of the determining factors of visual acuity (or resolution) (Fernald, 1988).

In order to perform at its optimum, the visual system of a particular species is adapted to the type of habitat they live in and to the prevailing surrounding light conditions (Lythgoe, 1979). In general, vertebrates range from cone monochromats with a single spectral class of cone photoreceptor (e.g. sharks and many rays), over dichromats and trichromats (e.g. most mammals and many marine fishes), to tetrachromats (e.g. most birds and many freshwater fishes; Bowmaker, 2008). Cone photoreceptors and their respective opsin repertoires are particularly diverse in teleost fishes (e.g. Lin et al., 2017; Musilova et al., 2019). This is thought to primarily be due to the different light environments that fish inhabit (Cronin et al., 2014; Lythgoe, 1979), but in some instances may also be driven by sexual selection (Endler, 1990) and/or the feeding habits of species. For example, UV photoreception increases feeding efficiency in some fishes eating UV-absorbing or scattering zooplankton (Novales-Flamarique, 2016; Loew et al., 1993; Browman et al., 1994), while herbivorous fishes may profit from visual systems tuned to longer wavelengths as a result of the red-reflecting properties of chlorophyll (Cortesi et al., 2018 preprint; Marshall et al., 2003; Stieb et al., 2017).

Moreover, the density of photoreceptors and ganglion cells can also vary not only between species but also within an individual's retina over its lifespan (Shand et al., 1999, 2000). The study of their distribution using the wholemount technique (Coimbra et al., 2006; Stone and Johnston, 1981; Ullmann et al., 2012) provides useful information on the visual ecology of a species, which usually reflects its habitat and behavioural ecology (Bozzano and Collin, 2000; Collin and Pettigrew, 1988a,b, 1989; Hughes, 1977). Two main specializations can be found in vertebrates: area and streaks (Collin and Pettigrew, 1988a,b). Both specializations have higher densities of cells compared with the rest of the retina, resulting in regions of acute vision in the corresponding field of view. An area is a concentric increase in cell density in a particular region of the retina; in some vertebrates, it is termed a fovea because of other structural adaptations (Walls, 1942). In teleost fishes, areas are often located temporally (i.e. area temporalis) and found in species that live in enclosed environments such as caves or coral structures, and/or coral overhangs (Collin and Pettigrew, 1989; Collin and Shand, 2003). The temporal area receives the visual information from the frontal field of view, corresponding to the natural swimming direction of fish. Nevertheless, multiple area centralis can also be found in a single retina (Collin and Pettigrew, 1989). For example, triggerfishes (Balistidae) possess an area in both the nasal and temporal part of the retina, which correlates with two main visual tasks: feeding (temporal) and predator avoidance (nasal) (Collin and Pettigrew, 1988b; Ito and Murakami, 1984). A horizontal streak is defined by an increase in cell density along the meridian. Most horizontal streaks are found in the central meridian, but sometimes they can also be located more ventrally or dorsally (Collin and Shand, 2003). The streak maintains a high spatial resolving power (SRP) throughout the horizontal section of the retina and is thought to be used to scan the horizon. It leads to an elongated sampling of the visual environment without continuous eye movements (Collin and Shand, 2003). Teleost fishes possessing a horizontal streak are commonly found in open water environments such as sandy bottoms or pelagic open ocean environments (Collin and Pettigrew, 1988b).

Variability in retinal structure and opsin gene repertoire does not only exist between species; both visual features may also change throughout the life of an individual. This is especially true for species that undergo substantial habitat changes during ontogeny such as coral reef fishes. The life of most coral reef fishes starts in the shallower zone of the open ocean as larvae (Helfman et al., 2009; Job and Bellwood, 2000), where resources may be high and the risk of predation is low (Fortier and Harris, 1989). At this stage, pelagic fish larvae feed typically on zooplankton (Boehlert, 1996) and rely on vision primarily for fundamental tasks such as predator avoidance and feeding (Leis and Carson-Ewart, 1999). After their oceanic phase, reef fish larvae typically find a suitable coral reef patch to settle on and again vision is one of the main senses used (Lecchini et al., 2005a,b). During this settlement phase, reef fish larvae undergo metamorphosis and reach the juvenile stage in which they usually already possess all basic morphological features of the adult form (Holzer et al., 2017). Following settlement on the reef, juvenile fishes are challenged with visual cues that are much more complex than in the open ocean, varying in both chromaticity and luminance. Hence, at this stage (or slightly before; Cortesi et al., 2016) the visual system of coral reef fishes is expected to undergo changes in both morphology and physiology (Helfman et al., 2009).

To date, changes in arrangement (i.e. mosaic) and distribution of the photoreceptor cells throughout ontogeny have been documented only in a few coral reef fishes (Shand, 1997). These changes are thought to enhance survival by increasing feeding success and facilitate predator avoidance in reef stages (juveniles and adults; Shand, 1997). Along with changes in morphology, ontogenetic changes in opsin gene expression have also been reported for a handful of species (Cortesi et al., 2015b, 2016). For example, in the dottyback Pseudochromis fuscus, the number and type of opsin genes that are expressed differs between larval, juvenile and adult stages. Along with the change in opsin gene expression, the visual system may also transform to a more complex colour-processing capability, such as dichromacy to trichromacy or even up to tetrachromacy. This increase in chromaticity ultimately requires behavioural testing to confirm and is likely to reflect major habitat transitions throughout development equipping fishes such as dottybacks with a more complex visual system as they grow and mature (Cortesi et al., 2015b, 2016). In comparison, while some freshwater cichlid species show a similarly dynamic change in opsin gene expression through ontogeny, other species do not change gene expression much (neoteny) or directly develop from the larval to the adult gene-expression pattern (Carleton et al., 2008; Härer et al., 2017). We currently do not know whether a progressive developmental change of the visual system, as found in the dottyback (Cortesi et al., 2016), for example, is a common feature shared among coral reef fishes, or whether some species also show different developmental modes, similar to what is found in cichlid fishes.

In this study, we investigated ontogenetic changes in retinal topography and opsin gene expression in three life stages (larval, juvenile, adult) of the spotted unicornfish, Naso brevirostris (Cuvier 1829), from the surgeonfish family (Acanthuridae) (Fig. 1). Naso brevirostris is known to shift both diet and habitat during ontogeny (Choat et al., 2002; Randall et al., 1997). Pelagic larvae feed on zooplankton before settling on the reef, where they mainly feed on algae as juveniles. As adults, N. brevirostris migrate to the reef slope, returning to a mostly zooplanktivorous diet (Choat et al., 2002, 2004; Randall et al., 1997). We therefore hypothesized that the visual system of N. brevirostris would show a ‘classic’ developmental mode, linked to either changes in habitat or diet, or both, and with a progression from larval, to juvenile and finally adult traits. Moreover, N. brevirostris develops an elongated rostral snout during maturation, and this prominent morphological change may also affect its visual requirements as it might obstruct the visual field of the fish.

Individuals of N. brevirostris were collected on the Northern Great Barrier Reef, Australia, under the Great Barrier Reef Marine Park Association (GBRMPA) permits G17/38160.1 and G16/38497.1, Queensland Fisheries permit no. 180731, or in French Polynesia. Adults [n=5; standard length (SL): range 27.7–33 cm, mean±s.d. 30.4±1.9 cm] were collected with a spear gun from No Name Reef (14°65′S, 145°65′E) on the outer Great Barrier Reef, Australia, in February 2018. Juveniles (n=8; SL, 6.5–19.5 cm, 14.5±3.7 cm) were collected using barrier nets, spear guns or clove oil and hand nets from reefs surrounding Lizard Island (14°40′S, 145°27′E) on the Great Barrier Reef between February 2016 and February 2018, and one juvenile fish was acquired through the aquarium supplier Cairns Marine (http://www.cairnsmarine.com/). Larval fish (n=5; SL: 3.1–3.3 cm, 3.1±0.1 cm) were captured using a crest net on Tetiaroa Island, French Polynesia (16°99′S, 149°58′W) in March 2018 (Besson et al., 2017; Lecchini et al., 2004). All animals were anaesthetized using an overdose of clove oil (10% clove oil, 40% ethanol, 50% seawater) and killed by decapitation under an animal ethics protocol of The Queensland Brain Institute (QBI/236/13/ARC/US AIRFORCE and QBI/304/16).

Each individual was photographed with a ruler in the frame, to enable us to extract the SL later on using Fiji v.1.0 (Schindelin et al., 2012). The eyes were enucleated and the cornea and lens removed using micro-dissection scissors. A small dorsal cut was made to keep track of the eye's orientation. The samples collected for retinal mapping were fixed in 4% paraformaldehyde in 0.1 mol l⁻¹ phosphate-buffered saline (PBS; pH 7.4) and stored at 4°C, and the eyes used for RNA sequencing were kept in RNAlater (Sigma) and stored at −20°C. For each eye, the lens diameter was measured after dissection and fixation.

Retinal wholemounts were prepared according to standard protocols (Coimbra et al., 2006, 2012; Stone and Johnston, 1981). Briefly, each eye cup was cut radially multiple times, to enable us to flatten it on a microscopy glass slide without damaging the tissue. The retina was oriented using the falciform process that extends ventrally. The sclera and choroid were gently removed and the retinas were bleached overnight in the dark at ambient temperature in a 3% hydrogen peroxide solution (in PBS). Large-sized adult retinas, which have a more developed retinal pigment epithelium, were bleached in the same solution but with a few drops of potassium hydroxide (Ullmann et al., 2012). While potassium hydroxide accelerates the bleaching process by increasing the pH of the solution, this type of bleaching is quite aggressive for the tissue. Therefore, these retinas were only bleached for 2–3 h in the dark.

For ganglion cell analyses, retinas were mounted with the ganglion cell layer facing up on a gelatinized slide and left to dry overnight at room temperature in formalin vapours (Coimbra et al., 2006, 2012). Wholemounts were then stained in 0.1% Cresyl Violet (Nissl staining) following the protocol of Coimbra et al. (2006) and then mounted with Entellan New (Merck). Shrinkage of the retina using this technique is usually deemed negligible and, if present, restricted to the borders of the retina (Coimbra et al., 2006). In this study, however, the retinas did not show equal shrinkage because of major differences in retina size between developmental stages. As such, the smaller retinas (larval stage) were more affected by shrinkage, because of their smaller surface (i.e. higher proportion of retinal borders), than those from the other stages (adult and juvenile stages). Shrinkage in these retinas was easily identified under the microscope and was taken into consideration in the data interpretation.

For photoreceptor analyses, retinas were wholemounted in 80% glycerol in PBS, on non-gelatinized slides with the inner (vitreal) surface facing downwards. Contrary to ganglion cell mounting, photoreceptor mounting shows negligible shrinkage as it takes place in an aqueous medium (Peichl et al., 2004).

The topographic distribution and the total number of ganglion cells, single cones, double cones and total cones in the three life stages of N. brevirostris were assessed using the optical fractionator technique (West et al., 1991), modified for wholemount retina use by Coimbra et al. (2009, 2012). A computer running Stereo Investigator software (v2017.01.1, 64-bit, Microbrightfield) coupled to a compound microscope (Zeiss Imager.Z2) equipped with a motorized stage (MAC 6000 System, Microbrightfield) and a digital colour camera (Microbrightfield) was used for the analysis. The contour of each retina wholemount was digitized using a ×5 objective (numerical aperture 0.16) and cells were counted randomly and systematically, using a ×63 oil immersion objective (numerical aperture 1.40) and the parameters summarized in Table S1. The total number of cells for each sample was then estimated by multiplying the sum of the neurons (ganglion cells or photoreceptors) counted by the area of the sampling fraction (Coimbra et al., 2009; West et al., 1991).

The counting frame and grid size were chosen carefully in order to achieve an acceptable Shaeffer's coefficient of error (CE), while maintaining the highest level of sampling. The CE measures the accuracy of the estimation of the total cell number and is deemed acceptable below 0.1 (Glaser and Wilson, 1998; Slomianka and West, 2005). The counting frame was adjusted between life stages to reach an average count of around 40 and 80 cells per sampling site for ganglion cells and photoreceptors, respectively, but was kept identical for individuals of the same life stage (Table S1). As fish of similar life stages can have a wide variation of SL, the grid size was adjusted for all individuals to allow sampling of around 200 sites (±10%) (de Busserolles et al., 2014a,b).

Three cell types can be found in the ganglion cell layer: ganglion cells, displaced amacrine cells and glial cells. These can usually be distinguished based on cytological criteria (Collin, 1988; Collin and Pettigrew, 1988c; Hughes, 1975), with ganglion cells having an irregular shape, an extensive nucleus and a larger size compared with the smaller, rounder amacrine cells, which have a darker stained appearance, and glial cells, which have an elongated shape relative to the other two cell types (Fig. 2A). However, as amacrine cells were often difficult to distinguish from ganglion cells in N. brevirostris, especially in high-density areas, amacrine cells were included in all counts and only glial cells were excluded. The inclusion of amacrine cells in the analysis should not interfere with the overall topography, as the distribution of amacrine cells has been shown to match the ganglion cell distribution in other animals (Bailes et al., 2006; Coimbra et al., 2006; Collin, 2008; Collin and Pettigrew, 1988c; Wong and Hughes, 1987), and the density of displaced amacrine cells in N. brevirostris was relatively low. However, the inclusion of amacrine cells in the ganglion cell counts will contribute to a slight overestimation of ganglion cell density and ultimately to a slight overestimation of SRP. For ganglion cell analysis, sub-sampling was performed in the regions of highest cell density to allow a more accurate estimation of the peak ganglion cell density. The same counting frame parameters as for the whole retina were used for sub-sampling, but the grid size was reduced by half.

Photoreceptor cells, in contrast, could be distinguished unambiguously as single and double cones (Fig. 2B). The two cone types were counted separately and simultaneously using two different markers to acquire data for single cones alone, double cones alone and total cones (single and double cones).

Topographic maps were created using the statistical program R v.3.4.1 (R Foundation for Statistical Computing, 2012) with the results exported from Stereo Investigator and the R script provided by Garza-Gisholt et al. (2014). As for previous retinal topography studies on teleost fishes (Dalton et al., 2016; de Busserolles et al., 2014a,b), the Gaussian Kernel Smoother from the Spatstat package (Baddeley and Turner, 2005) was chosen and the sigma value was adjusted to the distance between points (i.e. grid size) for each map (Fig. 3).

The retinas from different life stages of N. brevirostris (adult, n=3; juvenile, n=6; larvae, n=3) were dissected out of the eye cup, total RNA was extracted, and their retinal transcriptomes were sequenced according to Musilova et al. (2019). Briefly, total RNA of larval and smaller juvenile retinas was extracted using the RNeasy Mini Kit (Qiagen), and for larger juvenile and adult retinas the RNeasy Midi Kit (Qiagen), according to the manufacturer's instructions, which included a DNase treatment. Total RNA concentration and quality were determined using a Eukaryotic Total RNA NanoChip on an Agilent 2100 BioAnalyzer (Agilent Technologies). Juvenile transcriptomes were sequenced in-house at the Queensland Brain Institute's sequencing facility. For these samples, strand-specific libraries were barcoded and pooled at equimolar ratios and sequenced at PE125 on a HiSeq 2000 using Illumina's SBS chemistry version 4. Library preparation (strand-specific, 300 bp insert) and transcriptome sequencing (RNAseq HiSeq PE150) for larval and adult individuals was outsourced to Novogene (https://en.novogene.com/).

Retinal transcriptomes were filtered, and de novo assembled following the protocol described in de Busserolles et al. (2017). Briefly, raw-reads of transcriptomes were uploaded to the Genomics Virtual Laboratory (GVL 4.0.0) (Afgan et al., 2015) on the Galaxy Australia server (https://usegalaxy.org.au/), filtered by quality using Trimmomatic (Galaxy version 0.36.4) (Bolger et al., 2014) and then de novo assembled using Trinity (Galaxy version 2.4.0.0) (Haas et al., 2013).

Following the protocol in de Busserolles et al. (2017), the N. brevirostris transcriptomes were mined for their visual opsin genes. Briefly, using the opsin coding sequences from the dusky dottyback, Pseudochromis fuscus (Cortesi et al., 2016), we searched for the N. brevirostris opsin genes by mapping the de novo assembled transcripts to the P. fuscus reference genes using Geneious v.11.1.3 (www.geneious.com). Pseudochromis fuscus was chosen because it is relatively closely related to N. brevirostris, and because it possesses orthologues from all of the ancestral vertebrate opsin genes (Cortesi et al., 2016).

Assemblies based on short-read libraries tend to overlook genes expressed at low levels and similar gene copies, and/or short-reads may be misassembled (chimeric sequences); for that reason, a second approach was used to confirm the visual opsin genes of N. brevirostris. A manual extraction of the gene copies was performed by mapping raw-reads against the P. fuscus references and then moving from single nucleotide polymorphism (SNP) to SNP along the gene, taking advantage of paired-end information to bridge the gaps between SNPs. The extracted reads were then de novo assembled and their consensus was used as a template against which unassembled reads were re-mapped to elongate the region of interest; this approach eventually led to a reconstruction of the whole coding region (for details on this approach, see de Busserolles et al., 2017; Musilova et al., 2019).

Opsin gene identity was then confirmed using BLAST (http://blast.ncbi.nlm.nih.gov/) and by phylogenetic reconstruction to a reference dataset obtained from GenBank (www.ncbi.nlm.nih.gov/genbank/) and Ensembl (www.ensembl.org/) (as per de Busserolles et al., 2017; Fig. 4; Fig. S4). The opsin gene phylogeny was obtained by first aligning all opsin genes, i.e. the reference dataset and N. brevirostris genes, using the L-INS-I settings as part of the Geneious MAFFT plugin v.1.3.7 (Katoh and Standley, 2013). jModeltest v.2.1.10 (Darriba et al., 2012) was subsequently used to select the most appropriate model of sequence evolution based on the Akaike information criterion. MrBayes v.3.2.6 (Ronquist et al., 2012) as part of the CIPRES platform (Miller et al., 2010) was then used to infer the phylogenetic relationship between opsin genes using the following parameter settings: GTR+I+G model; two independent MCMC searches with four chains each; 10 million generations per run; 1000 generations sample frequency; and 25% burn-in.

Opsin gene mining and phylogenetic reconstruction revealed, amongst a number of other visual opsin genes, two N. brevirostris RH2 paralogues, of which one clustered within the RH2A clade of other percomorph fishes. However, the phylogenetic placement of the second paralogue could not be fully resolved using this approach alone (Fig. 4). Therefore, in order to resolve a more detailed relationship between the two N. brevirostris RH2 paralogues, we took advantage of the phylogenetic signal within the single exons of the displaced paralogue (as per Cortesi et al., 2015b). The five N. brevirostris RH2-2(B) exons were obtained by annotating the coding regions of the gene with a P. fuscus RH2 orthologue. The single exons were separated from one another and inserted as ‘single genes’ in the alignment in a reduced (RH2 genes only) reference dataset, along with the N. brevirostris RH2A gene. The RH2-specific phylogeny was then reconstructed using MrBayes on the CIPRES platform using the same parameters as before (Fig. 5A).

To support our phylogenetic analyses, the amino acid sequences of the RH2 paralogues were aligned to the amino acid sequence of bovine rhodopsin (GenBank accession no. NM001014890) using MAFFT in Geneious with parameters: L-INS-I, BLOSUM62, gap open penalty 1.53, offset value 0.123. Referring to bovine rhodopsin as the reference, this allowed us to compare site 122 between RH2s, which has been found to be key to distinguish between longer-shifted (RH2A in percomorphs) and shorter-shifted (RH2B in percomorphs) RH2 copies (Chinen et al., 2005; Fig. 5C).

Topographic maps of ganglion cells (including amacrine cells) for the three life stages of N. brevirostris were constructed from Nissl-stained retinal wholemounts. Little variation in topographic distribution of ganglion cells was observed within the same ontogenetic stage. Therefore, the topographic map of only one individual per life stage is presented here (Fig. 3A), and the results for the remaining individuals are shown in Fig. S1.

Differences in retinal topography were mainly found between the larval stage and the two later stages (Fig. 3A). In general, the larval retina showed less specialization compared with juvenile and adult retinas. In the larval retina, the onset of a horizontal streak was observed with the highest cell density found in the central meridian of the retina (1.5 times increase compared with areas with the lowest cell densities). Within this weak streak, three areas of high cell density were found: in the nasal, central and temporal parts of the retina. However, these areas of high cell density should be taken with caution because of the limitations of the Nissl-staining protocol for very small retinas. Larval retinas were challenging to prepare and analyse because of their small size and thus the higher amount of shrinkage present after staining. After several attempts with different larvae, only one larval retina was deemed acceptable for analysis. Even for this individual, the areas of high cell density in the nasal and temporal part of the retina are questionable, as they are very close to the retinal borders and therefore could be the result of shrinkage. A prominent horizontal streak along with a centralized area centralis (the area centralis had a 2.5–3 times increase in cell density compared with areas with the lowest cell densities) was present in the juvenile and adult individuals. Similar to the larvae, the streak in juveniles and adults was located on the central meridian of the retina, extending to the nasal and temporal margins. Although slightly different patterns were found for each life stage, they all showed a higher ganglion cell density in the central area close to the optic nerve, accompanied by a horizontal streak (Fig. 3A).

The total number of ganglion cells increased with the size of fish and ranged from 208,975 cells for the larval individual, to ∼1,600,000 cells for juveniles and ∼2,100,000 cells for adults (Table 1). Conversely, the mean cell density decreased with the size of the fish, ranging from 19,439 cells mm⁻² in the larval individual, to ∼8,500 cells mm⁻² in juveniles and ∼5000 cells mm⁻² in adults. Peak cell density also decreased through development, from 30,400 cells mm⁻² in the larval individual, to ∼23,000 cells mm⁻² in juveniles and ∼20,500 cells mm⁻² in adults.

Based on the peak ganglion cell density, the SRP of N. brevirostris ranged from 2.98 cycles deg⁻¹ in the larval individual, to ∼8.0 cycles deg⁻¹ in juveniles and a maximum of 11.0 cycles deg⁻¹ in adults (Table 1). Overall, SRP or visual acuity in N. brevirostris increased with the size of the fish with very little variation found within ontogenetic stages (Fig. S2).

The density and topographic distribution of cone photoreceptors (double and single cones) was assessed in the three life stages of N. brevirostris. Double and single cones were arranged in a square mosaic, with a single cone at the centre of four double cones (Fig. 2B). This pattern was consistent throughout the entire retina, thus providing a ratio of double cones to single cones of 2:1. As a consequence of this regular arrangement, the topographic distribution of single cones, double cones and total cones was identical. Moreover, similar to the ganglion cell topography, little variation in topographic distribution of cone photoreceptors was observed within the same ontogenetic stage. Therefore, the total cone topographic map of only one individual per life stage is presented here (Fig. 3B). The remaining maps (i.e. for single and double cones separately, and maps of all individuals) are provided in Fig. S3 and in Dryad (https://doi.org/10.5061/dryad.k6djh9w37).

The topographic distribution of cone photoreceptors varied between stages, with a continuous increase in specialization from the larval to the adult stage (Fig. 3B). Larvae had a weak horizontal streak in the central meridian with two large areas of high cell density in the nasal part and the temporal part of the retina. One of the two analysed larval individuals also showed a dorsal increase in cell density (Fig. S3f). However, this apparent increase in cell density was probably caused by an artefact from not properly flattening the dorsal part of the retina during mounting and should therefore be disregarded. Compared with the larvae, juveniles had a more pronounced horizontal streak in the central meridian with a peak density of cells in the temporal part of the retina. Moreover, a weak vertical streak could be seen in the temporal part of the retina, extending from the dorso-temporal area to the ventro-temporal area. In adults, the horizontal streak in the central meridian was still present but did not extend as far into the nasal part as in the juveniles. Moreover, the vertical streak was more prominent compared with that found in juveniles, resulting in a very large area of high cell density in the temporal region (Fig. 3B). The continuous nature of the transition between juvenile and adult specializations is highlighted by the topography of individuals of intermediate sizes (Fig. S3). For example, the horizontal streak was less pronounced in the nasal part of a larger (SL 19.5 cm; Fig. S3c) compared with a smaller juvenile (SL 13.8 cm; Fig. S3d). Conversely, the vertical streak in a smaller adult (SL 27.7 cm; Fig. S3b) was still developing compared with that found in a larger adult (SL 29.7 cm; Fig. S3a).

Similar to the ganglion cells (Table 1), the total number of photoreceptors increased with the size of the fish, ranging from ∼650,000 cells in larvae, to ∼4,300,000 cells in juveniles and ∼5,700,000 cells in adults (Table 2). A large difference in the total number of photoreceptors was found between the two juvenile individuals. This difference is probably due to the size difference between these individuals. Photoreceptor peak cell density decreased with the size of the fish, ranging from ∼69,000 cells mm⁻² in larvae, to ∼51,000 cells mm⁻² in juveniles and ∼34,000 cells mm⁻² in adults (Table 2).

The total number of cone photoreceptors was greater than the total number of ganglion cells, indicating a high summation ratio between the two cell types. For one individual (juvenile ID1), the distribution of both ganglion cells and photoreceptors was analysed, which allowed us to estimate the summation ratio between photoreceptors and ganglion cells in low- and high-density areas. For this individual, the summation ratio was found to be as low as 2.3 in the central part and as high as 5.4 in the ventro-temporal part of the retina.

Naso brevirostris were found to mainly express four opsin genes in their retinas. Independent of ontogeny, these were the blue–violet opsin SWS2B, the green opsins RH2B and RH2A, and the rod opsin RH1. The red opsin LWS was also found to be expressed, albeit at very low levels in all stages (0.1–6.5% of total double cone opsin expression; Fig. 6A; Table S2). The phylogenetic reconstruction based on the full coding regions of the genes confirmed the positioning of all genes within their respective opsin class (Fig. 4). However, for RH2B in particular, the resolution between RH2-specific clades was poor (Fig. 4; Fig. S4). This was resolved using the exon-based approach, which showed the placement of some of the N. brevirostris RH2B exons within a greater percomorph RH2B clade (Fig. 5A). We also found evidence for substantial gene conversion affecting this gene, with the placement of exons 1 and 2 close to, or within, the RH2A clade (Fig. 5B). Amino acid comparison between the two RH2 paralogues corroborated our phylogenetic placement of genes and revealed the conserved sites E122 and Q122 for the longer- (RH2A) and shorter-shifted (RH2B) copies, respectively (Fig. 5C).

Quantitative opsin gene expression revealed that SWS2B was the only single cone gene and thus was expressed at 100% in all developmental stages (Table S2). Of the double cone opsins, there was a change in expression for the RH2 genes with ontogeny. During the larval stage (n=3), RH2B (mean±s.e.m., 36.2±4.8%) was less highly expressed compared with RH2A (63.6±4.8%). The opposite pattern was found in the juvenile (n=6) and adult stages (n=3), where RH2B was the most highly expressed of all double cone opsins genes (juvenile: 56±1.3%, adult: 56.1±1.9%). RH2A in the juvenile (41.2±1.4%) and adult (38.1±1.6%) stages was less highly expressed. Despite LWS being expressed at low levels in all stages, there was a noticeable increase in expression with development (larva: 0.2±0.0%, juvenile: 2.8±0.7%, adult 5.8±0.4%; Fig. 6A). Rod opsin (RH1) expression was substantially higher compared with cone opsin expression in all stages (82–86% for all stages; Fig. 6B).

The visual systems of fishes often change through development when transitioning from one habitat to another. These changes are usually associated with a shift in light environment, e.g. when moving from the open ocean to a coral reef, but possibly also with changes in diet and predation pressure (Sale, 2013). Our objective was to assess the visual system development in the spotted unicornfish, N. brevirostris. Naso brevirostris experiences multiple changes in habitat, diet and morphology throughout ontogeny (from larval to adult stages; Fig. 1) making it a prime candidate to study visual system changes on the reef.

Retinal topography is an effective method to identify visual specializations and recognize the area of the visual field of greatest importance to a species as it visually samples its environment (Collin, 2008; Collin and Pettigrew, 1988a; Hughes, 1977). In marine fishes, visual specializations have been found to correlate with the structure and symmetry of the environment they live in and/or with their feeding behaviour (Caves et al., 2017; Collin and Pettigrew, 1988a,b; Ito and Murakami, 1984; Shand, 1997). In this study, we show that the N. brevirostris eye possesses a horizontal streak in all life stages (Fig. 3A). This type of specialization has previously been found in species living in open environments where an uninterrupted view of the horizon, defined by the sand–water or air–water interface, is present (Collin and Pettigrew, 1988b). As N. brevirostris spends much of its life (larval and adult stages) searching for prey in the water column, having a pronounced horizontal streak is likely to increase feeding and predator surveillance capabilities by allowing it to scan the horizon without using excessive eye movements (Collin and Shand, 2003). Moreover, at the larval stage, this type of specialization may also enable fish to scan the environment when searching for a reef habitat to settle on. In contrast, a horizontal streak does not seem to match the visual needs at the juvenile stage during which N. brevirostris lives in close association with the reef, i.e. in a more enclosed 3D environment. At this life stage, we would have expected to find one (or multiple) area centralis and no horizontal streak – a common feature in fish that live close to or within the reef matrix (Collin and Pettigrew, 1988a,c). Compared with the lifespan of these fishes (up to 20 years), the juvenile stage is relatively short (∼3 years; Choat and Axe, 1996), which may explain the maintenance of the horizontal streak throughout development.

The peak ganglion cell density in the juvenile and adult stages is in the central part of the retina (i.e. at the centre of the streak; Fig. 3A). This is quite unusual, as the area of high cell density in coral reef fishes is normally found in the temporal zone, which receives information from the nasal visual field, and thus is usually correlated with feeding and predator avoidance in front of the fish (Collin and Pettigrew, 1988a, 1989; Fritsches et al., 2003; Fritsch et al., 2017; Shand et al., 2000). The type of specialization found in N. brevirostris could potentially be associated with its unusual visual behaviour, as these fish examine objects side-on (V.T., unpublished observations). A possible explanation for this peculiar behaviour is that because of its protruded snout, which grows through development, the frontal image might be partially blocked and stereoscopic vision may be impaired or even impossible (Purcell and Bellwood, 1993). Although the visual field of N. brevirostris was not investigated in their study, Brandl and Bellwood (2013) suggest that the protruding snout found in many Naso species indeed prevents an overlap of their horizontal field of view. Similar to the monocular vision found in hammerhead sharks (Lisney and Collin, 2008; McComb et al., 2009), increasing visual acuity in the central part of the retina would thus maximize a sideward oriented visual field. Together with a visual streak, a high cell density in the central part of the retina is likely to enable N. brevirostris to accurately navigate both within the complexity of the reef and in open water.

The photoreceptor topography showed a similar trend to the ganglion cell topography, with bigger changes occurring between the larval and juvenile stage and only minor adjustments thereafter (Fig. 3B). In addition to a weak horizontal streak, larval fish had two peaks of high cell density in the nasal and temporal zones, which comply with two of their main ecological needs: feeding (temporal; looking forward) and predator avoidance (nasal; looking backwards) (Boehlert, 1996; Collin and Pettigrew, 1988a; Fortier and Harris, 1989). These high-density regions were not matched by the ganglion cell topography and, as such, are likely to provide areas of higher sensitivity (i.e. areas of high photoreceptor to ganglion cell ratio; Walls, 1942). Moreover, although larval fishes rely mainly on olfactory cues to zoom in on a suitable habitat for settlement (Lecchini et al., 2005b), the area of high cell density in the temporal zone in particular might also assist when searching for such habitats over longer distances (Mouritsen et al., 2013).

In addition to the horizontal streak becoming more prominent in juvenile and adult stages, N. brevirostris showed a temporal vertical specialization appearing at the juvenile stage and being more pronounced in adults (Fig. 3B). Such a double streak specialization, with a vertical and a horizontal component, is the first to be found in coral reef fishes. Naso brevirostris adults live on the coral reef slope/wall, and move up and down the wall (from 2 to 122 m) while foraging and searching for mates (Mundy, 2005). As such, in line with the terrain hypothesis (Hughes, 1977), the evolution of this vertical specialization is likely to be a result of the vertical component in their visual environment.

A difference in the topography of ganglion cells and photoreceptors means that the summation ratio between the cell types, i.e. the sensitivity and spatial resolution of the retina, also differs depending on the visual field in question. For example, high photoreceptor densities and comparable low ganglion cell densities in the ventro-temporal and dorso-temporal parts of the vertical streak confer higher sensitivity to these two areas (Walls, 1942). Theoretically, this enables juvenile and adult N. brevirostris to detect even small differences in luminance, which may help to reveal well-camouflaged predators against the reef wall. A high density of both photoreceptor and ganglion cells in the centre of the retina, in contrast, confers a low summation ratio which leads to an increase in visual acuity (Walls, 1942). This area of high acuity may help fish to identify conspecifics and also to distinguish between food items (Cronin et al., 2014).

To summarize, the retinal topography in N. brevirostris may be adapted to the habitat in both the larval and the adult stage. Juveniles live in a more enclosed, 3D coral reef environment compared with the other two life stages. Therefore, we would have expected the juvenile visual system to reflect its habitat by having a less-developed streak and one or more area centralis instead. The unexpected topography found in the juvenile stage may be explained by this stage only lasting a fraction of the lifespan of N. brevirostris (Choat and Axe, 1996). The relatively short period of time spent in a habitat rich in shelter and food enables the fish to grow big enough to avoid most predators (Barnes and Hughes, 1999; Lasiak, 1986). During this time, juvenile N. brevirostris mostly feed on benthic algae, which do not require a highly specialized visual system in terms of retinal topography (Caves et al., 2017; Collin and Pettigrew, 1988a,b; Randall et al., 1997). As such, instead of changing the visual system multiple times, it is probably more energy efficient to maintain (or slightly adjust) a visual system that is optimally adapted for both the larval and adult stages.

The visual acuity of N. brevirostris was found to increase through development (Fig. S2; Table 1). This seems to be a common feature in coral reef fishes, as a higher acuity often correlates with an increase in eye size during growth. The benefit of having a higher visual acuity is that, as fish grow and expand their home ranges, it increases the distance at which visual objects such as predators, conspecifics and food can be detected (Caves et al., 2017; Shand, 1997). Accordingly, as in other coral reef fish larvae (Shand, 1997), the acuity of N. brevirostris larvae was relatively poor (2.98 cycles deg⁻¹). The overabundance of zooplankton in their habitat means that larval coral reef fishes can wander instead of using a lock-and-pursuit feeding behaviour, i.e. they do not need to spot their food from a distance, but rather bump into it while floating in the plankton (Evans and Fernald, 1990; Fortier and Harris, 1989). Once settled on the reef, the visual acuity of N. brevirostris starts to increase in line with their growth (Fig. S2). Adult N. brevirostris were found to have a similar visual acuity (∼11 cycles deg⁻¹) to that of other reef fishes of comparable size, such as the clown triggerfish, Balistoides conspicillus, a species that also inhabits the reef slope and shows a pronounced horizontal streak (Collin and Pettigrew, 1989).

Phylogenetic reconstruction showed that the N. brevirostris visual opsins belong to the opsin gene clades usually found within percomorph fishes (Fig. 4). However, within the RH2 genes, an exon-based phylogeny revealed that the N. brevirostris RH2B gene is likely to have undergone substantial gene conversion (Fig. 5A). As such, it seems that parts of its first and second exon have been acquired from the RH2A paralogue, explaining its phylogenetic uncertainty when using whole coding region-based reconstructions (Fig. 4; Fig. S4). This is not that surprising, as RH2 duplicates in teleosts are commonly found in tandem (e.g. Musilova et al., 2019) and, as is the case for other teleost opsin genes (Cortesi et al., 2015b; Hofmann and Carleton, 2009), frequently experience gene conversion (Cortesi et al., 2015b; Escobar-Camacho et al., 2016; Hofmann et al., 2012). This phenomenon is thought to be one of the main mechanisms for concerted evolution in small gene families, which often originate from tandem duplications (Li, 1997; Ohta, 1983) and could help to preserve gene function by repairing null mutations (Innan, 2009) or by resurrecting previously pseudogenized gene copies (Cortesi et al., 2015b). As the RH2 opsin genes are highly expressed in N. brevirostris, they seem rather important for their visual ecology, and it is therefore likely that gene conversion played a major evolutionary role in maintaining their function.

Interestingly, the key spectral tuning site 122, despite being within the area of gene conversion, showed the RH2B-specific 122Q (Fig. 5B,C). As the remainder of the exon is almost identical to RH2A, it is likely that this site was re-gained post-gene conversion, suggesting that strong selection is maintaining two spectrally diverse RH2 copies in this species. As such, based on opsin gene expression, N. brevirostris could be behaviourally trichromatic (i.e. has three spectral sensitivities) for all three developmental stages, with the violet opsin SWS2B and the blue–green opsin RH2B and RH2A genes being expressed in sufficient quantity to enable this level of chromatic analysis. Supporting these findings, microspectrophotometry (MSP) in adults of two closely related Naso species (Naso literatus and Naso unicornis; Sorenson et al., 2013) found three cone photoreceptors with spectral sensitivities of ∼420 nm λ_max for single cones, and ∼490 nm λ_max and ∼515 nm λ_max for the accessory and principal members of double cones, respectively (Losey et al., 2003). A short-shifted visual system with high sensitivity in the violet to green range might benefit feeding on zooplankton and gelatinous prey during the larval and adult stages of N. brevirostris (Marshall et al., 2003). However, it seems at odds with the mainly algivorous diet of the juvenile stage, where a red-shifted visual system would be of advantage (Cortesi et al., 2018 preprint; Stieb et al., 2017).

Rather than the anticipated long-wavelength shift in juveniles, the most notable change in opsin gene expression was found in the RH2 copies between the larval and later N. brevirostris stages (Fig. 6A). A change in expression levels of RH2B and RH2A has also been found in coral reef damselfishes (Pomacentridae) (Stieb et al., 2016). On shallow, clear coral reefs, a broad spectrum of light is available (Marshall et al., 2003). However, with increasing depth, the long and short ends of the spectrum are cut off as a result of absorption and scattering through interfering particles, resulting in a blue mid-wavelength saturated light environment (Smith and Baker, 1981). Consequently, in an attempt to maximize photon catch, some damselfish species have been found to increase the expression of the blue-sensitive RH2B gene and simultaneously decrease the expression of the green-sensitive RH2A gene with increasing depth (Stieb et al., 2016). In N. brevirostris, the shift in expression of RH2 genes occurs between the larval and juvenile stages where depth differences do not seem that relevant. In lieu of depth, individuals migrate from a pelagic blue-shifted open water environment to the more green-shifted light environment of the coral reef (Marshall et al., 2003). This could in theory explain the high RH2A expression in larval fish at the settlement stage; however, it does not explain the increase in RH2B expression post-settlement (Fig. 6A). An increasing number of fishes are being found to change their opsin gene expression to tune photoreceptors to the prevailing photic environment (e.g. Fuller et al., 2004; Härer et al., 2017; Hofmann et al., 2010; Luehrmann et al., 2018; Nandamuri et al., 2017; Shand et al., 2008; Spady et al., 2016; Stieb et al., 2016). At the opposite end of the spectrum, opsin gene expression might be pre-programmed either by phylogeny or on a species by species basis, as exemplified by only some damselfishes changing expression with depth (Stieb et al., 2016). It is therefore possible that opsin gene expression in N. brevirostris is under phylogenetic control and that changes in photic environment contribute very little to opsin gene expression in this case. Also, it is important to note that we currently do not know whether and how quantitative differences in opsin gene expression translate to the behaviour of N. brevirostris. Hence, until these results become available, it remains speculative as to what impact the changes observed here might have on the visual ecology of these fishes.

Naso brevirostris was not found to express the UV-sensitive SWS1 gene at any developmental stage. SWS1 expression is often found in larval fish and more generally in fishes feeding on zooplankton, with UV-vision thought to increase the detectability of this food source (Browman et al., 1994; Sabbah et al., 2010). As N. brevirostris feeds on zooplankton at both larval and adult stages (Choat et al., 2002, 2004), the lack of SWS1 expression seems striking. However, it does support ocular media measurements, which revealed UV-blocking lenses in both larval and adult N. brevirostris (Siebeck and Marshall, 2007). UV-blocking lenses seem common in many bigger coral reef fishes, which is thought to enhance sighting distance by reducing chromatic aberration and scatter, as well as protecting the eye from the damage caused by these high-intensity wavelengths (Siebeck and Marshall, 2001). Instead, the expression of the violet-sensitive SWS2B gene, as its spectral absorption curve reaches into the near-UV (based on MSP in closely related species; Losey et al., 2003), may be sufficient to increase the discrimination of zooplankton from the water background while foraging.

We furthermore found low expression of the LWS gene (<6%) at all developmental stages. This suggests that LWS expression is either restricted to certain areas of the retina or interspersed at low frequency across the retina, or some photoreceptors might co-express LWS with an RH2 gene (e.g. Cortesi et al., 2016; Dalton et al., 2014; Torres-Dowdall et al., 2017). MSP in related Naso species did not show any long-wavelength-sensitive photoreceptors, nor did it show any evidence for opsin co-expression (i.e. red-shifted unusually broad absorbance peaks; Losey et al., 2003). As this technique only samples a subset of the photoreceptors across the retina, it might be that the photoreceptors containing this pigment were missed because of their low number or that LWS was simply not expressed in these fishes.

It is possible that the LWS expression found here is just a by-product of the way opsin gene expression is controlled and that it does not serve any ecological function. Nevertheless, LWS expression did increase steadily with development (Fig. 6A). Hence, an alternative explanation might be that LWS is co-expressed with an RH2 gene, which has been shown to increase achromatic discrimination in cichlids (Dalton et al., 2014). Moreover, LWS expression has recently been shown to be correlated with algal feeding in damselfishes (Stieb et al., 2017), and blennies (Cortesi et al., 2018 preprint). As N. brevirostris juveniles, and to a certain extent also adults (Choat et al., 2002), feed on algal turf, an increase in LWS expression in later stages may improve feeding efficiency through the increased contrast of algae against the reef background (Cortesi et al., 2015a; Marshall et al., 2003; Stieb et al., 2017). In situ hybridization studies (e.g. Dalton et al., 2014, 2016; Stieb et al., 2019; Torres-Dowdall et al., 2017) coupled with behavioural colour vision experiments (e.g. Cheney et al., 2019) will be needed in the future to assess the distribution and function of the various opsin genes and ultimately the colour vision system of N. brevirostris.

Using a multidisciplinary approach, we analysed the ontogeny of the visual system of N. brevirostris. Changes in retinal topography and opsin gene expression were mainly found between the larval and juvenile stages, with few minor modifications thereafter. The juvenile stage in particular did not seem to match its ecology very well, which may be explained by the short duration of this stage compared with the lifespan of these fishes. This is different from the steady development that was found in other reef fishes (Cortesi et al., 2016; Shand et al., 2008; Job and Julia, 2001) and highlights the need for a comprehensive analysis of visual ontogeny across the reef fish community.

We would like to thank Sara M. Stieb, Vivien Rothenberger, Simon Dunn and the staff of Tetiaroa Society (Moana LeRohellec) and of CRIOBE (Camille Gache) for assistance with specimen collection. We furthermore thank the staff at the Lizard Island Research Station for logistical support, and Janette Edson from the Queensland Brain Institute (QBI) sequencing facility for library preparation and RNA sequencing. We also acknowledge QBI Advanced Microscopy Facility for the use of the Stereo Investigator (software v.2017.01.1), generously supported by the Australian Government through the ARC LIEF grant LE100100074.