Photoreceptor distributions, visual pigments and the opsin repertoire of Atlantic halibut (Hippoglossus hippoglossus)

Fishes often have cone photoreceptors organized in lattice-like mosaic formations. In flatfishes, these lattices undergo dramatic changes during metamorphosis whereby a honeycomb mosaic of single cones in the larva is replaced by a square mosaic of single and double cones in the adult. The spatio-temporal dynamics of this transition are not well understood. Here, we describe the photoreceptors and mosaic formations that occur during the larva to juvenile transition of Atlantic halibut from the beginning of eye migration to its completion. To gauge the possibility of colour vision, visual pigments in juveniles were measured by microspectrophotometry and the opsin repertoire explored using bioinformatics. At the start of eye migration, the larva had a heterogeneous retina with honeycomb mosaic in the dorsonasal and ventrotemporal quadrants and a square mosaic in the ventronasal and dorsotemporal quadrants. By the end of metamorphosis, the square mosaic was present throughout the retina except in a centrodorsotemporal area where single, double and triple cones occurred randomly. Six cone visual pigments were found with maximum absorbance (λmax, in nm) in the short [S(431) and S(457)], middle [M(500), M(514) and M(527)], and long [L(550)] wavelengths, and a rod visual pigment with λmax at 491 nm. These pigments only partially matched the opsin repertoire detected by query of the Atlantic halibut genome. We conclude that the Atlantic halibut undergoes a complex re-organization of photoreceptors at metamorphosis resulting in a multi-mosaic retina adapted for a demersal life style.

Flatfishes (order Pleuronectiformes) are peculiar among teleosts in that they undergo a pronounced metamorphosis whereby a bilaterally symmetric, pelagic larva transforms into a flattened, demersal juvenile with both eyes on the same side of the head 1,2 . During this period of transformation, which can last from days to weeks depending on the species 3 , the animal experiences a progressive change in the perception of its surroundings resulting from the shifting visual fields of both eyes and increasing proximity to the ocean floor. In parallel, the diet changes from a preponderance of pelagic plankton consumed by pre-metamorphic larvae to small benthic invertebrates favoured by post-metamorphic juveniles 3 . In some species, like the common sole (Solea solea) and the Senegalese sole (Solea senegalensis), metamorphosis further entails a shift in circadian rhythm from diurnal to nocturnal 4 .
Despite these profound alterations to the anatomy and ecology of flatfishes, their visual system during metamorphosis has seldom been examined and is poorly understood. Morphological investigations into the larval retina of winter flounder, Pseudopleronectes americanus, and Atlantic halibut, Hippoglossus hippoglossus, have shown the presence of a single morphological type of cone photoreceptor prior to the start of eye migration 5,6 . This cone forms a honeycomb (or hexagonal) mosaic throughout the retina characterized by six neighbours surrounding each photoreceptor. Microspectrophotometric measurements from cone photoreceptors of premetamorphic winter flounder larvae revealed a single visual pigment with maximum wavelength of absorption (λ max ) at 519 nm 7 . Post-metamorphosis, the retina comprised two types of morphological cone types (single and doubles) forming a square mosaic whose repeating unit consisted of a single, centre cone, surrounded by four double cones (the latter made up of two cones apposed together sharing a double membrane partition) 5 . The visual pigment in the single cones peaked at 457 nm, whereas the double cones could be equal members, with a visual pigment peaking at 531 nm, or unequal, with one member possessing a visual pigment with λ max at 531 nm and the other at 547 nm 7 . Thus, the retina of winter flounder transformed its cone mosaic and visual pigment

Results
Eye morphology and migration. The youngest Atlantic halibut examined (96 days post-fertilization or 720 ATU; ATU are the Accumulated Temperature Units, i.e. the product of mean rearing water temperature (7.5 ˚C) and days from fertilization, 96) were bilaterally quasi-symmetrical with a compressed, oval shaped body and similar pigmentation on either side of the body (Fig. 1A,E). In the majority of animals examined (6 of 10), one of the eyes (usually the left eye) had recently started migrating as evidenced by its slightly closer proximity to the midline with respect to the other eye (Fig. 1E). Migration of the eye proceeded throughout the study period ( Fig. 1A-D) such that between 773 and 825 ATU (i.e., 103 and 110 days post-fertilization, respectively) the migrating eye had reached the mid-line (Fig. 1C,F) or proceeded beyond, to the other side of the head. The oldest halibut examined (878 ATU, or 117 days post-fertilization) had both eyes on the same side of the head (the ocular side, Fig. 1D,G) whereas the opposite side (the blind side) rested on the bottom and had reduced pigmentation.
All fish examined had elliptically shaped eyes and retinas (Fig. 1). The eye dimensions increased from a mean ± SD (n = 10) of 1.6 ± 0.18 mm for the long axis and 1.3 ± 0.11 mm for the short axis at 960 ATU to 2.2 ± 0.13 mm and 1.6 ± 0.15 mm, respectively, at 1170 ATU. The mean lens diameter ± SD (n = 10) increased from 0.44 ± 0.04 mm to 0.64 ± 0.02 mm during the same time period. The extracted retinas showed an embryonic fissure that projected from the medial ventral periphery to the central retina (location of the optic nerve head) pointing slightly to the temporal retina. The optic nerve head was localized in the ventral half of the retina (Fig. 1G), below the theoretical crossing of short and long retinal axes.
Photoreceptor types and mosaics. The morphological trends were the same for the migrating and nonmigrating eye. At 720 ATU, the retina had two main types of cone photoreceptors: single and double cones ( Fig. 2A,C,E,G). The single cones were round in cross section and formed a honeycomb mosaic in the ventrotemporal and dorsonasal regions of the retina (Fig. 2C,E). The honeycomb mosaic consisted of six neighbouring cones flanking a single cone ( Fig. 2C; white contours) thereby forming a hexagon if the centres of neighbouring cones were to be connected. The inner segments of some single cones stained darker (Fig. 2C,E; double black arrowheads) than the rest of the cone population, and an accessory outer segment was prominent at the base of the outer segments ( Fig. 2E; white arrowheads).
In the ventronasal and dorsotemporal quadrants, double cones were present as part of square mosaics of varying regularity ( Fig. 2A,G). The basic square mosaic unit consisted of four double cones surrounding a single (centre) cone ( Fig. 2A; white contours). There was, however, substantial deviation from a lattice formation ( Fig. 2A; white rectangle). Single (corner) cones, i.e., those facing the partitioning membranes of neighbouring double cones, were absent or extremely rare. Some double cone members in the ventronasal retina stained darker than their counterparts ( Fig. 2A; red arrow) and, in the dorsotemporal retina, accessory outer segments were prominent and appeared to make contact between cone types ( Fig. 2G; white arrowheads).
At 773 ATU, double cones were present in the centroventrotemporal retina forming a square mosaic region of varying disposition (Fig. 2D). The honeycomb mosaic was restricted to the dorsonasal retina (Fig. 2F). In regions with double cones, an arrangement consisting of a single cone flanked by three double cones was common (Fig. 2B,D,H; white circles), as were irregular formations of single and double cones (Fig. 2B, white rectangle). The left eye has migrated half way between the blind side (bottom part of the fish following metamorphosis) and ocular side (top part of the fish following metamorphosis). (G) Head of fixed specimen at 878 ATU showing completed eye migration and the extracted right eye without lens. The optic nerve (on) exits the retina below the crossing of theoretical short and long ellipse axes that characterize this elliptically-shaped eye and retina. The optic nerve head coincides with the central end of the embryonic fissure (ef) which extends to the ventral periphery. The panel illustrates the direction of sectioning when full eyes were cut, revealing tangential sections of central retina and more oblique sections toward the periphery, and coronal cuts when a head was sectioned, revealing radial views of the retina throughout the lens region and tangential views at the nasal and temporal ends. The schematic shows the division of a right eye retina into quarters (VT, ventrotemporal; VN, ventronasal; DT, dorsotemporal; DN, dorsonasal). Other abbreviations: V, ventral; D, dorsal; N, nasal; T temporal. Scale bar in (A) equals 1 cm and holds for (B-D) and the lateral views in (E); the other scale bars represent 1 mm and are associated with the figures closest to them. www.nature.com/scientificreports/ , cones in the periphery were more tightly packed and the mosaic appeared more regular (Fig. 4B). Cone distributions differed the most from a lattice formation in the region approaching the ventronasal retina, where double cone members with darker stain (red arrow) were present (Fig. 4D).
As per the trends in the dorsonasal retina, the ventral retina at 878 ATU consisted of square mosaics with variable jitter in the lattice (Fig. 5A,C,D; white rectangles and circle). Double cone members with darker stain (red arrows) than their counterparts were more frequent in the ventral retina than in other areas of the retina, as were single corner cones (double white arrowheads) (Fig. 5A,B,D). At this developmental stage, rod photoreceptors were present; their outer segments were larger and displaced further into the retinal pigment epithelium with respect to those of cones (white closed contours, Fig. 5E,F).
To confirm that rod photoreceptors were associated with the appearance of double cones and square mosaic formation, in-situ hybridization experiments were performed with two previously established riboprobes, one tagging rh2 opsin transcript and the other rod (rh1) opsin transcript 9 . At all locations, whether characterized by the honeycomb mosaic or by the presence of double cones, the retina of Atlantic halibut at 720 ATU labelled prominently with the rh2 riboprobe along the inner segment myoids of cones (Fig. 6A,B; green arrows). In contrast, the rh1 riboprobe failed to label cryosections characterized by a honeycomb mosaic (Fig. 6C,D). At 878 ATU, both the rh2 and rh1 riboprobes labelled, respectively, the myoid region of double cone members (Fig. 6E) and those of rods (Fig. 6F). Parallel experiments with corresponding sense probes resulted in no labelling (Supplementary Fig. S1).

Cone densities and measures of visual acuity. A summary topographical schematic showing mean
cone densities in the Atlantic halibut retina at the two extreme developmental time periods examined (720 and 878 ATU) revealed profound retinal transformations ( Fig. 7). At 720 ATU, the greatest cone density within the central retina was found in the ventrotemporal quadrant and the lowest in the dorsotemporal quadrant (Fig. 7A). Correspondingly, the greatest theoretical spatial resolving power was found in the ventrotemporal quadrant and the lowest in the dorsotemporal quadrant (Fig. 7B). This translated into a mean increase of 14 mm in the distance at which the ventrotemporal retina would resolve a 0.5 mm target with respect to the dorsotemporal retina (Fig. 7B). Based on total density of cones (where each cell was counted individually), the greatest resolving power was found in an area devoid of double cones and square mosaics.
At 878 ATU, the region of the central retina with the greatest cone density had shifted to the dorsotemporal quadrant (Fig. 7C). This region, with theoretical resolving power similar to that of the peripheral retina, also had the greatest density of triple cones (Fig. 7D). Despite an overall reduction in cone density in the central retina during development from 720 to 878 ATU (Fig. 7A,C), theoretical resolving power generally improved (Fig. 7B,D) as a result of increased lens size.
Visual pigments. Absorbance measurements from the outer segment of individual photoreceptors from four juvenile Atlantic halibut revealed six cone visual pigments and one rod visual pigment (Fig. 8). The cone visual pigments consisted of two short wavelength (S) types with maximum wavelength of absorbance (λ max ) ± SD at 431 ± 7 nm (n = 8) (Fig. 8A) and 457 ± 6 nm (n = 7) (Fig. 8B), three middle wavelength (M) types with λ max ± SD at 500 ± 5 nm (n = 23) (Fig. 8C), 514 ± 8 (n = 35) (Fig. 8D), and 527 ± 6 (n = 19) (Fig. 8E), and one long wavelength (L) type peaking at 550 ± 11 nm (n = 4) (Fig. 8G). The rod visual pigment had λ max ± SD at 491 ± 7 nm (n = 25) (Fig. 8F). The S visual pigments were located in the single cones whereas the M and L visual pigments were mostly confined to the double cones, one per double cone member. The S(431) absorbance curve showed  www.nature.com/scientificreports/ www.nature.com/scientificreports/ www.nature.com/scientificreports/  To assess whether cones in non-lattice mosaics had some level of organization, their spatial statistics were compared to those of twenty random distributions of corresponding mean density and constrained by soma size (the latter was the mean diameter of 2.2 µm for the single cone, mean short axis distance of 3.0 µm for the double cone, and mean median of 4.1 µm for the triple cone; n = 30; Figs. 9P-S, 10K-N, 11F-I). None of the comparisons to the random distributions were significantly different, or the means were greater for the random distributions. Thus, all cone types (single, double, and triple) in mixed cone areas with high triple cone density (i.e., where the proportion of triple cones was statistically similar to that of double cones; Figs. 10F, 11A) were distributed randomly.
Autocorrelation and density recovery profile analyses. The autocorrelograms of morphological cone types that were part of square or honeycomb mosaics revealed a variety of lattice-like arrangements characterized by periodic foci with variable jitter (Figs. 12, 13). The density recovery profile, derived from the aurocorrelogram, showed a region surrounding the origin (the effective radius) where like-type cones were absent (Figs. 12B,C,J,K, 13B,C). This exclusion zone for single cones in the honeycomb mosaic was slightly greater than the diameter of a single cone (Fig. 12K). For double cones of the square mosaic, it approximated the size of the long axis of a double cone ellipsoid cross section (Fig. 13C). These results reflect the tight packing of these cone types in their respective lattices. The greatest distance between single cones (compared to double cones) in the square mosaic resulted in a larger effective radius for this cone type (Fig. 12C). www.nature.com/scientificreports/  Table 1 for corresponding fit parameters). www.nature.com/scientificreports/

Genomic identification and phylogenetic analyses of Atlantic halibut opsins. Queries of the
Atlantic halibut genome predicted eight opsin genes corresponding to five opsin classes: SWS1, SWS2, RH1, RH2, and LWS (Supplementary Table S1). Two genes were predicted for rh2 and sws2, whereas one gene was predicted for each of the other opsin classes. Coding sequences corresponding to rh2, sws2, and lws were located on chromosome 5; sequences for sws1 and rh1 were located on chromosomes 23 and 7, respectively (Supplementary Fig. S2A-D).
Compared to the annotated genomes of turbot (Scopthalmus maximus) and zebrafish (Danio rerio), the positions of the predicted opsins and their flanking genes were in similar orientations ( Supplementary Fig. S2A-D). Predicted RH2 opsins were located in tandem and were flanked by slc66a22.2 and synaptoporin (synpr) (Supplementary Fig. 2A). Initially, the annotated genome predicted three rh2 coding sequences; however, one of these sequences (accession: XP_034442263.1) was incorrectly annotated as it had an unusual six exon structure and overlapped the regions of two other rh2 sequences (Supplementary Table S1, Supplementary Fig. S3A). We hypothesized that this incorrect sequence corresponded to a rh2 pseudogene, with only the first three exons being correctly annotated ( Supplementary Fig. S3B), and did not consider it in further analyses.
The sws2-lws gene cluster on chromosome 5 was flanked by hcfla and gnl3l, consistent with similar genomic regions in turbot and zebrafish ( Supplementary Fig. S2B). Within this cluster, two sws2 genes were separated by a sws2 pseudogene.
To confirm that the predicted proteins were all visual opsins, key residues from these were compared to corresponding ones from visual opsins in zebrafish, and non-visual opsins in Atlantic halibut and zebrafish (Supplementary Table S4).
Except for SWS2A, the λ max of reconstituted barfin flounder opsins 13 corresponded well with some of our microspectrophotometry records of associated visual pigments and were therefore used as references for inferring the λ max of Atlantic halibut opsins when bound to retinal as the chromophore. Therefore, the λ max values of these opsins were predicted to be around 367 nm (SWS1), 416 nm (SWS2B), 494 nm (RH1), 506 nm (RH2B), and 490 nm (RH2C) (Supplementary Table S10). For SWS2A the corresponding opsin from Japanese flounder, Paralichthys olivaceus 14 , was used. The A292S substitution for SWS2A suggested a − 8 nm shift 15 and a predicted λ max near 458 nm. Based on the A164S substitution 15 , LWS was predicted to have a λ max in the range 554-558 nm. Table 1. Best-fit Simplex derived parameters and corresponding least sum of squares (SS) for the spectral functions depicted in Fig. 8H. The k 431 and k other coupling constants are the inputs of the visual pigments to the absorbance function [k 431 corresponds to the S(431) visual pigment and k other to the other pigment in a combination]. The λ max of the visual pigments used in the fitting process are indicated in the first column. The SS statistic is derived from the sum of differences between mean absorbance data and the corresponding values predicted by the Simplex algorithm. www.nature.com/scientificreports/ www.nature.com/scientificreports/ www.nature.com/scientificreports/   www.nature.com/scientificreports/

Discussion
Spatio-temporal dynamics of photoreceptor distributions. The retina of Atlantic halibut undergoing eye migration experienced major changes in photoreceptor types and distributions. The single cone type organized in honeycomb mosaic reported at the end of the yolk sac period (40 days post-hatching) 6 started to reorganize to incorporate double cones 6 as part of square mosaics (Fig. 2) toward the beginning of eye migration. The youngest halibut examined in this study, whose eye migration had barely started, had two regions of the retina, the centroventronasal and centrodorsotemporal, that contained double cones and square mosaics. At later developmental stages, the square mosaic began to appear in the centroventrotemporal retina and, by the end of eye migration, was present throughout the retina. This progression, which swept from ventronasal and dorsotemporal regions to ventrotemporal and, lastly, to the dorsonasal retina contrasts with a previous report stating that the start of the square mosaic was most advanced at the periphery near the ora serrata (at 70 days post hatching) and reached the centre of the retina by 100 days post-hatching 6 .
The seeming controversy between our study and previous work 6 can be resolved by realizing that there are two independent mechanisms that create double cones and the square mosaic in the retinas of flatfishes 8 . One, common to all fishes, operates at the peripheral growth zone (ora serrata) through differentiation of progenitor cells into photoreceptors that become part of established (adult) mosaics, like the square mosaic 16 . The other appears to occur only among fishes with indirect development (i.e., species that undergo a pronounced metamorphosis where the larva and adult have very different form 17 ). Here, there seems to be a re-arrangement of existing cones in central parts of the retina (present prior to the onset of metamorphosis 8 ) that leads to square mosaic lattices with substantially more jitter than those forming directly at the periphery, as per our study. In other words, previous authors 6 did not disambiguate between these two mechanisms leading to potential confusion over the progression of square mosaic formation in the non-peripheral retina prior to the start of metamorphosis.
The progression of double cone and square mosaic appearance reported here agrees with similar observations in winter flounder where two patches of retina, located in the ventronasal and dorsotemporal regions, were the first to express new opsin transcripts simultaneously (i.e., were present in all samples examined) from the start of metamorphosis 8 . As in our study, these authors observed that, at the start of eye migration, the ventronasal and dorsotemporal quadrants had double cones as part of emerging square mosaics whereas the dorsonasal and ventrotemporal quadrants had single cones in honeycomb formation 8 .
Another major discrepancy between our study and previous work 6 pertains to the appearance of triple cones in the central retina as metamorphosis progressed. Upon completion of eye migration, these cones occupied a large area of central retina as part of random distributions along with double and single cones, whose distributions were also random. Triple cones have never been reported during metamorphosis of flatfishes [5][6][7][8] but are present in the centrodorsal retina of several fish species during metamorphosis [18][19][20] . Some fishes, which are demersal or live in low light environments as adults, retain a high density of triple cones in a restricted area of centrodorsotemporal retina 21 often located next to a fovea 22 . Other, nocturnal fishes, like some species of holocentrids, also have triple cones. They are primarily located in regions of high cone density, along a central meridian, but constitute less than 0.5% of the total cone population 23 .
Based on the appearance of triple cones around the time when double cones and square mosaics start forming in the retinas of some marine fishes, it has been hypothesized that triple cones may be part of a transitional mosaic between the larval honeycomb mosaic and the adult square mosaic 18 . Two observations suggest otherwise, however. First, in Atlantic halibut, triple cones were rare in the ventral retina and scarce in the nasal compared to the temporal retina. If triple cones were part of a necessary transitional mosaic leading to the square mosaic, then the ventral and nasal regions of the central retina would not have developed such a mosaic or, at least, not fully. Second, triple cones are retained in the centrodorsotemporal retina of adults of other fish species 21,22 suggesting a functional role within the random mosaics that they are a part of.

Different square mosaic formation processes in the central and peripheral retina.
Our results support two different mechanisms of square mosaic formation in the retina of Atlantic halibut during metamorphosis. At 720 ATU, the centrodorsonasal and centroventrotemporal areas had only singles cones and these were arranged in honeycomb formation. Because a square mosaic was present in these regions of the retina by the end of metamorphosis, and no new cones are known to differentiate in the main retina of flatfishes 8 , existing single cones must have coalesced to form the square mosaic (a conclusion previously reached for the winter flounder by other authors 5,8 ). Electron microscopy evidence suggesting such cone fusion has been provided for the black bream, Acanthopagrus butcheri, and the West Australian dhufish, Glaucosoma hebraicum, two marine fishes that also undergo metamorphosis 18,19 .
Near the periphery, the regularity of the square lattice contrasted with the jitter present in multiple square mosaic regions of the central retina. At the growth zone, the square mosaic contained corner cones (Supplementary Fig. S4), and these disappeared with distance from the periphery, as has been shown for the Atlantic salmon, Salmo salar 16 . These observations suggest a direct path of square mosaic formation, as established for other fishes.
Topographical changes in visual acuity match foraging ecology. In the non-peripheral retina, spatial resolving power, our theoretical measure of visual acuity, was greatest in the ventrotemporal retina of the youngest Atlantic halibut examined. At the end of metamorphosis, the area of highest resolving power had shifted to the dorsotemporal retina. The values computed were nearly identical to those published for the plaice, Pleuronectes platessa, another flatfish, where spatial resolving power (independently derived from measures of radially-oriented central retina) was ~ 0.5º at the start of metamorphosis and ~ 0.3º by the end 24 .
The shift in area of greatest resolving power conforms with the foraging needs of the fish as it transforms from a bilaterally symmetric larva to a flattened, demersal juvenile. Pelagic fishes, including larvae, often have www.nature.com/scientificreports/ the highest cone densities in the ventrotemporal retina and strike at prey that contrast against the water background located in front of them or slightly above [25][26][27] . Demersal fishes, on the other hand, often strike at prey that contrasts against the bottom of the water body 28 . As such, the observed shift in area of greatest resolving power would be expected to maximize prey capture, the strongest determinant of fish growth and survival.
A potential role for the triple cone in achromatic target detection. That the highest visual acuity in the postmetamorphic fish coincided with the area of densest triple cone presence suggests that the random mosaic associated with triple cones is not an unintentional by-product of development but serves a specific function. The adults of other demersal fishes have a similar triple cone area in the centrodorsotemporal retina, in the vicinity of where the highest cone densities are found 21,22 . Spatial analyses of flatfish mosaics show that greater cone packing can be achieved through a combination of different cone morphologies (including the triple cone) compared to that from the square mosaic 21 . It has been argued, however, that a regular mosaic should provide better chromatic visual acuity due to the regular distribution of spectral inputs across the visual field 21,29,30 . For a fish that lives in habitats with low light, visual acuity based on chromatic contrast may not be as important as improving photon catch and, potentially, achromatic detection of targets.
In the random mosaic region, the ratio of the sum of triple and double cones to single cones is the same (~ 2) as that of double to single cones in nearby square mosaic regions. As such, the total ratio of longer to shorter wavelength input should be similar regardless of the mosaic, the major difference being the regularity of chromatic input. Based on these observations and the greater cone packing achievable with the random mosaic 21 , we propose that triple cones serve an achromatic area of high visual acuity. Because of the ventral position of the optic nerve head, the area of random mosaic appears to view the forward horizon. It may therefore represent a specialized region for the early detection of large groundfish predators, like many species of elasmobranchs, that cruise the ocean floor in search of demersal prey. Such predators lack significant colour and would be readily detected achromatically by improving photon catch. Interestingly, photoreceptors in this area also have among the longest outer segments, further improving photon catch.
Visual pigment co-expression: a potential shift toward longer wavelength absorbance. Absorbance measurements from postmetamorphic retinas indicated the presence of at least two opsins, with associated visual pigments peaking at 431 nm and 457 nm, among the single cone population. The λ max and bandwidth at half maximum of the rod visual pigment (mean ± SD = 4170 ± 234 cm -1 , n = 20) suggested retinal (the aldehyde of vitamin A 1 ) as the only chromophore in the retina 31 , as has been measured by High Performance Liquid Chromatography in starry flounder, Platichthys stellatus 32,33 . Co-expression of the S(431) pigment with another visual pigment [potentially the M(500) or M(514)] in a given single cone was common and may illustrate a progressive switch toward the longer wavelength-associated opsin as the juvenile moves to greater depths. A progression toward longer wavelength-absorbing visual pigments occurs in multiple cone types of the juvenile starry flounder 33 . Visual opsin repertoire of Atlantic halibut. The predicted proteins from the Atlantic halibut genome provided evidence for six visual opsin genes. Phylogenetic and tuning site analyses revealed that these opsins correspond to SWS1, SWS2A, SWS2B, RH1, RH2B, RH2C, and LWS. This expands the previous opsin repertoire known for Atlantic halibut which included SWS1, SWS2, RH1, RH2, and LWS 11 . The updated opsin repertoire is consistent with those of other flatfish species, with six visual opsins reported for the common sole and the Senegalese sole 21 , eight for starry flounder 32 and marbled sole, Pseudopleuronectes yokohamae 34 , and nine for turbot, Scophthalmus maximus 35 . In particular, expansion of the rh2 repertoire in Atlantic halibut follows findings in other flatfishes where rh2b and rh2c genes are common and rh2a is often absent or present as a pseudogene 13,14,21,34,35 . Tandem duplications of sws2 genes are also commonly found in flatfishes 13,14,34 and other teleosts 36 . Both the number of predicted opsins and the surmised λ max of their visual pigments when bound to retinal only partially matched microspectrophotometric records. The predicted SWS2A opsin matched the S(457) visual pigment, RH1 corresponded to the rod(491) visual pigment, and the LWS opsin matched the L(550) visual pigment. The S and M pigments found by microspectrophotometry were not a good match for the SWS2B or RH2 opsins. Previous work comparing the λ max of M visual pigments in the starry flounder 32 and the common sole 21 to those of barfin flounder 13 or marbled sole 34 (the latter two sets measured from opsin-retinal reconstitutions or predicted from opsin tuning sites, respectively) showed that the in-vivo measured pigments were significantly long wavelength shifted with respect to in-vitro determinations or predictions. With some opsins, this was the case despite having identical amino acids at all tuning sites between species 13,21 . These results suggest that alternative tuning sites or interactions between multiple tuning sites play important roles in determining the λ max of S and M visual pigments in flatfishes.

Conclusion
Retinal transformation at metamorphosis in Atlantic halibut is a complex process that prepares the animal for the visual ecology demands of a demersal life style. Photoreceptors organization changes from a honeycomb mosaic consisting of single cones to two novel formations: the square mosaic, which occupies most of the postmetamorphic retina, and a random mosaic consisting of single, double and triple cones located in a restricted region of the dorsotemporal retina. In parallel, a square mosaic of greater regularity forms directly from progenitor cells in the peripheral growth zone, and both mosaics coalesce during metamorphosis. The central region of greatest visual acuity shifts from the ventrotemporal to the dorsotemporal retina in accordance with the change in foraging behaviour that accompanies the transition from pelagic to benthic life style. The retina has the potential for colour vision containing a minimum of six cone visual pigments and one rod visual pigment. www.nature.com/scientificreports/ Such composition resembles that of other flatfishes that have been studied which show abundant expression of multiple S and M visual pigments, but diminished expression of UV and L visual pigments with age.

Materials and methods
Fish husbandry and collections. Atlantic halibut were reared from the fertilized egg in 100 L recirculating saltwater indoor tanks at a temperature of 7.5 ± 1.6 °C and maintained under artificial illumination (λ range: 350-750 nm, irradiance: 1.2 × 10 15 photons m -2 s -1 ) provided by tungsten-halogen tubes (Scotian Halibut Ltd, Nova Scotia, Canada). Peak hatching occurred at 14 days post-fertilization [~ 90 Accumulated Temperature Units (ATUs), calculated as the product of days from fertilization and mean water temperature]. Hatched larvae had a prominent yolk sac which was fully absorbed around 50 days post-hatching (~ 320 ATUs). Pre-metamorphic larvae were fed a combination of freeze-dried Gemma Diamond 0.5 and 0.8 mm pellets (Skretting), added every 30 min throughout the day at a density of 2 pellets per fish, and Artemia nauplii, added twice a day at a density of 10-15 metanauplii ml −1 day −1 . Fish nearing complete metamorphosis, or having completed it, were fed a mixture of Gemma Diamond 0.8, 1.0 and 1.2 mm every hour at a density of 3 pellets per fish and Artemia nauplii twice a day at a density of 5 to 10 metanauplii ml −1 day −1 . Clay was added to the tanks every day to recreate the turbid water conditions found in nature.
Collections of specimens took place at 82, 89, 96 and 103 days post-hatching, corresponding to 720, 773, 825 and 878 ATU. At each stage, fish were collected in the light-adapted state for in-situ hybridization analyses and, separately, for morphological observations. At 720 ATU, eye migration had just started in the majority of fish examined and, by 878 ATU, metamorphosis was complete (Fig. 1). In addition to these collections at the aquaculture facility, post-metamorphic juvenile fish were flown live to the University of Victoria aquatic facility (British Columbia, Canada) for measurements of visual pigment absorbance by microspectrophotometry. The study is reported in accordance with ARRIVE guidelines.
Histology. Fish were euthanized and the entire bodies fixed in primary fixative (2.5% glutaraldehyde, 1% paraformaldehyde in 0.08 M PBS, pH = 7.4) at 4ºC for a minimum of 72 h. The retina from the migrating and non-migrating eye were individually extracted, rinsed in 0.08 M PBS, and cut into four quadrants (ventrotemporal, VT; ventronasal, VN; dorsonasal, DN; and dorsotemporal, DT). The quadrants were post-fixed in secondary fixative (1% osmium tetroxide in 0.08 M PBS) for 1 h at 4ºC. Following a brief rinse in distilled water, the tissue was dehydrated through a series of solutions of increasing ethanol concentration, infiltrated with mixtures of propylene oxide and EPON resin, and embedded in 100% EPON resin. Retinal blocks were cut tangentially, in 1-2 µm steps, and stained with Richardson's solution (1:1 mixture of 1% Azure II in dH 2 O and 1% Methylene blue in 1% NaB 4 O 7 ) to reveal the cone mosaic 23,31 . Digital images of sections were acquired with an E-600 Nikon microscope equipped with a DXM-100 digital camera. Photographs were obtained using a Plan/Apo 60X/1.40 objective and further amplified 1.5 × with an optical tourette on the microscope (total magnification 90x). Observations of mosaic types in different sectors were carried out on 10 retinas (5 each from migrating and non-migrating eyes). From these, four retinas with the largest tangential mosaic area (0.0106 mm -2 ) spanning the region of widest cone ellipsoid cross sections to the bases of the outer segments were selected for density determinations and spatial analyses.
In-situ hybridization. These experiments were undertaken on 10 retinas to verify morphological observations of rod photoreceptor appearance using riboprobes. Whole bodies of euthanized fish were fixed (4% paraformaldehyde in 0.08 M PBS, pH 7.4) at 4 °C for a minimum of 72 h. For a given fish, the head was processed as a whole or the eyes were extracted from it and processed individually. Following several rinses in 0.08 M PBS, the tissue was cryoprotected in sucrose solution (25% sucrose, 20% optimal cutting temperature [OCT] medium in 0.08 M PBS) and embedded in 100% OCT medium (Cedar Lane Laboratories). The resulting blocks were cryosectioned in 7-10 µm steps in either tangential (for individual eyes) or coronal (for heads) planes to reveal the cone mosaic or photoreceptors along their lengths in the medial retina, respectively. Serial sections were deposited cyclically on poly-L-lysine coated slides and hybridized with species-specific riboprobes against the rh1 and rh2 opsin mRNAs of Atlantic halibut 9,11 .
In situ hybridizations were performed according to procedures detailed in the literature 37,38 . Briefly, this involved rehydrating the sections, permeabilizing them in 10 µg/ml proteinase K for 5 min, exposure to 0.1 M triethanolamine containing 0.25% acetic anhydride, dehydration, and overnight hybridization at 56 °C with 4 µg of cRNA probe in 50% formamide, dextran sulfate and goat serum. Following immunodetection with anti-Digoxigenin F ab fragments conjugated to alkaline phosphatase (1:3000) for 2 h at room temperature, the riboprobe was visualized with NBT/ BCIP substrate 38 . Cross hybridization of a riboprobe with the non-corresponding opsin did not take place as evidenced by the different level of label of photoreceptor myoids in the light-adapted retina (corresponding to cones and rods; Fig. 6).
Sections were viewed with the E-600 Nikon microscope equipped for DIC imaging. Photographs were obtained with the DXM-100 digital camera at 90 × magnification using the 60x/1.40 objective and microscope tourette.
Morphometric analyses. Digital  www.nature.com/scientificreports/ logical cone type in the field of view. The coordinates were then imported into a customized Matlab program that computed the Delauney tessellation of the field, from which the nearest neighbour distance of each individual cell was determined, as was their Voronoi domain area (i.e., the area surrounding each cell that encloses the territory closer to that cell than to any of the neighbours) 39 . Each analysis excluded "border cells", i.e., those with uncertain nearest neighbour distances or Voronoi domain areas. In each case, the regularity index was defined as the mean divided by the standard deviation.
To assess whether the cone distributions analyzed had higher order (lattice-like) periodicity, a spatial autocorrelation analysis was carried out that examined the positioning of each cell with respect to all other cells across the mosaic. The Density Recovery Profile was derived from each autocorrelogram, providing a plot of the mean density of cells as a function of distance from each cell. The autocorrelogram permits the detection of higher order patterning and the Density Recovery Profile provides a measure of the exclusion zone surrounding each cell where other homotypic cells are less likely to be found than at further distances 39 .
To assess whether cell distributions with low regularity indices were different from random distributions, their spatial statistics were compared to those of simulated random fields of cells matched for density and constrained by soma size. Non-random retinal mosaics of cells show greater regularity indices and exclusion zones to those from corresponding random simulations resulting in effective radii that are larger than if determined by soma size alone 40 .
Statistical analyses to assess differences in mosaic regularity and cone density between retinal sectors were based on analysis of variance (ANOVA) with post-hoc grouping tests (Student-Neuman-Keuls, Tukey HSD) evaluated at α = 0.05 level of significance.
Visual acuity. Theoretical determinations of spatial resolving power (in degrees of arc) were calculated as the inverse of the Nyquist frequency (f N ) (in cycles per degree) using the equation: where D is the photoreceptor cell density (in mm -2 ) and PND is the posterior nodal distance (in mm) 22 . PND was approximated as the product of lens radius and 2.55 (the mean Matthiessen's ratio for teleost eyes 41 ). The photoreceptor cell density was computed from the sum of all cone types measured in a 0.0106 mm -2 area, with each double cone and triple cone contributing two and three cells, respectively. Based on the resolving power, the distance at which a given region of the retina would detect a 0.5 mm size food item, such as the typical food pellet encountered in aquaculture conditions or prey copepod consumed in nature, was computed.
Microspectrophotometry. This analysis was performed on four post-metamorphic juveniles with mean weight ± SD and total length ± SD of 3.6 ± 0.8 g and 7.5 ± 0.5 cm, respectively. Individual fish were dark adapted for 8 h, following which the fish was euthanized, and the retina extracted under infrared illumination. Small pieces of retina were teased apart and prepared for viewing with the dichroic microspectrophotometer (DMSP) as per previous studies [42][43][44] . The DMSP is a computer-controlled, wavelength-scanning, single-beam photometer that simultaneously records average and polarized transmitted light fluxes through microscopic samples 42,44 . The DMSP was equipped with ultrafluar (Zeiss) objectives: 32/0.4 for the condenser and 100/1.20 for the objective. With the aid of reference measurements recorded through cell-free areas, individual photoreceptor outer segments were illuminated sideways with a measuring beam of rectangular cross section of ca. 2 × 0.6 μm. Absolute absorbance spectra were computed in 2 nm increments from the obtained transmittances (each spectrum consisted of an average of 8 scans). The solid spectra (fits) were derived from experimental data by Fourier filtering 43 . Analysis of spectra consisting of multiple visual pigments. Because the S(431) absorbance spectrum showed a "hump" in the range 450-550 nm, suggesting the presence of multiple visual pigments, the Simplex algorithm 45 was used to fit the data with a combination of visual pigment absorbance curves using the equation 46 : where R is the response (overall absorbance) function, A i p is the absorbance of pigment i at light of wavelength λ, and p is an exponent resulting from the mathematical requirement that the absorbance function to be fitted be differentiable at the origin 46 . The variables k i are coupling constants resulting from the best fit of the model to the data. The visual pigment absorbance values used in the model were generated with an eighth-order polynomial template 47 based on the λ max of visual pigments measured by microspectrophotometry. The use of templates ensures β-band representation of individual visual pigments, which is not always present in microspectrophotometry records. Table S11) were used as query sequences to search predicted proteins from the reference genome of Atlantic halibut 48 (NCBI Hippoglossus hippoglossus Annotation Release 100; GCF_009819705.1) using BLASTp (v.2.12.0). Predicted proteins with e-values < 1e-5 and percent identities ≥ 50% were retained. Duplicated results were filtered by accession number using R (v.4.1.0) 49 and RStudio (v.1.4.1717) 50 . Locations of the corresponding coding sequences and their immediate flanking genes were compared to those in the reference genomes of turbot (NCBI Scophthalmus maximus Annotation Release 100; GCF_013347765.1) 51 and zebrafish (NCBI Danio rerio, GRCz11, NCBI Annotation Release 106; GCF_000002035.6) and used to further narrow down the predicted opsins using the NCBI Genome www.nature.com/scientificreports/ Browser and Gene Graphics 52 . Coding sequences that were eliminated from further analysis can be found in Supplementary Table S12 with the corresponding justification.

Genomic analyses. Zebrafish visual opsins (Supplementary
To confirm the assigned opsin classes of the predicted proteins, opsin sequences reported previously 11 were used to query the subset of predicted Atlantic halibut proteins using BLASTp, and sequences with an e-value < 1e-5 and a percent identity ≥ 70% retained. To confirm that the predicted proteins were visual opsins, a multiple sequence alignment was performed that incorporated visual zebrafish opsins and non-visual zebrafish and Atlantic halibut opsins (Supplementary Table S4). Visual opsins were distinguished from non-visual opsins by their identity at key amino acid sites 53 . All multiple alignments were performed and edited using ClustalX (v.2.0) 54 and Qiagen CLC Sequence Viewer (v.8.0) (www. qiage nbioi nform atics. com). Amino acid residue numbers were standardized to bovine rhodopsin (Accession: NP_001014890.1) 55 .
The predicted Atlantic halibut opsins were assigned to opsin subclasses using phylogenetic analyses and tuning sites from key amino acid residues for each of the visual opsins 12,35,56,57 . Phylogenetic trees were constructed with MEGAX 58 using maximum likelihood and boostrap analyses (n = 100) and visualized using R and ggtree 49,50,59 . Sequences used in these analyses (Supplementary Tables S11, S13) were obtained from GenBank (https:// www-ncbi-nlm-nih-gov. proxy. lib. sfu. ca/ genba nk) and SoleaDB 60 , with zebrafish VA opsins used as the outgroup.
Animal experimentation approval. All animal use was approved by the Animal Care committees of Simon Fraser University (protocol # 1126B-10) and the University of Victoria (protocol # 2017-005), which abide by regulations set by the Canadian Council for Animal Care.

Data availability
Data are available in the manuscript and supplementary materials. Anything further is available from the authors upon reasonable request.