Eggs and sperm from Atlantic salmon (Salmo salar) were obtained from MOWI, Tveitevågen, Norway. Fertilization took place at an approved laboratory facility at High Technology Center, University of Bergen, Norway, where alevins were raised until the stages, after hatching at 555 day degrees (dd) and just before first feeding at 720 dd. These stages correspond approximately to relative ages 310 and 390 in Gorodilov (1996). The “dd” nomenclature refers to the average temperature per day (6° with a standard deviation of 0.5) multiplied by the number of days. All experiments described followed local animal care guidelines and were given ethical approval by the Norwegian Food Safety Authority. The study complied with ARRIVE guidelines (Percie du Sert et al., 2020).

Eyes were collected from Atlantic salmon alevins at the stage just before first feeding, where total RNA was isolated by TRI reagent (Sigma, St. Louis, MO, United States), then treated with DNase I by using a TURBO DNA-free™ Kit (ThermoScientific, Waltham, MA, United States). Complementary DNA (cDNA) was reversely transcribed by using a SuperScript III kit (Invitrogen, Carlsbad, CA, United States) as recommended by the manufacturer’s instructions. Initial in silico analyzes of the Atlantic salmon genome database ICSASG_v2 (Lien et al., 2016) was conducted by using BLASTP or BLASTN via NCBI (Sayers et al., 2019) with zebrafish (Danio rerio) visual opsin sequences being used as bait for gene mining. To generate full-length visual opsins, primers were designed that anneal to published or predicted untranslated regions (UTRs) or to regions containing start/stop codons. An Advantage^®2 PCR Kit (TaKaRa, Japan) was used to perform amplification reactions by applying PCR or nested PCR approaches (see Supplementary Table 1) over 35 cycles. PCR products were extracted from agarose gels using a MinElute^®Gel Extraction Kit (Qiagen, Germany), before being cloned into StrataClone PCR Cloning vector pSC-A-amp/kan (Agilent Technologies, LA Jolla, CA, United States). Correct clones were sequenced at the University of Bergen Sequencing Facility, with nucleotide sequences being deposited to the GenBank database with Accession Numbers (ON456431-ON456441).

In silico search in salmonid genome databases [brown trout (Salmo trutta), Chinook salmon (Oncorhynchus tshawytcha), coho salmon (Oncorhynchus kisutch), rainbow trout (Oncorhynchus mykiss), sockeye salmon (Oncorhynchus nerka)] and in the northern pike (Esox lucius) was performed to identify the visual opsins of the respective species. The search was performed by BLASTP or BLASTN on NCBI (Sayers et al., 2019) or Ensembl (Cunningham et al., 2018) using the Atlantic salmon visual opsins in the search. In addition, the sequences used for Atlantic salmon rh1-2.1 and rh1-2.2 are from Lin et al. (2017). The coding sequences of annotated or putative opsin genes in the genomes were aligned to the Atlantic salmon visual opsins using Vector NTI9 software (Invitrogen, Carlsbad, CA, United States) and ClustalX2 (Larkin et al., 2007) to evaluate the coding sequence. The genomes of Atlantic salmon, brown trout, Chinook salmon, rainbow trout, coho salmon, and an updated version of northern pike were published on Ensembl (version 104), and recently updated versions of Atlantic salmon and rainbow trout have been published (version 106). The most recent sequences from NCBI and Ensembl have been used in the evaluation of open reading frame (ORF). Sequences determined to yield intact ORFs, cloned sequence for Atlantic salmon and either NCBI or Ensembl sequence for the other species, were used in the phylogenetic analysis, whereas partial sequences and/or sequences derived from pseudogenes were omitted. A codon-matched nucleotide sequence alignment of 90 ray-finned fish opsin coding regions was generated by ClustalW (Higgins et al., 1996) and manually manipulated to refine the accuracy of cross-species comparison. Specifically, the alignment incorporated opsin coding sequences of visual photopigments identified in the genomes of several salmonid species and a phylogenetically related species (i.e., E. lucius, northern pike) compared to orthologous visual opsin sequences expressed in the retina of D. rerio (zebrafish). All five visual opsin classes (i.e., lws, sws1, sws2, rh2, and rh1) were included, with zebrafish vertebrate ancient (va) opsin sequences (va1 and va2) used collectively as an outgroup given that this opsin type is a sister clade to all five visual photopigment classes. Phylogenetic analyses of 1,000 replicates were conducted in MEGA11 (Tamura et al., 2021), with evolutionary histories being inferred by using the Maximum Likelihood method and General Time Reversible model (Nei and Kumar, 2000). The percentage of trees in which the associated taxa clustered together is shown next to the branches. Initial trees for the heuristic search were obtained by applying Neighbor-Joining and BioNJ algorithms (Saitou and Nei, 1987) to a matrix of pairwise distances estimated using the Maximum Composite Likelihood (MCL) approach (Tamura and Nei, 1993). The tree was drawn to scale, with branch lengths measured in the number of substitutions per site. A total of 996 positions was present in the final dataset, with all positions with less than 95% site coverage being eliminated. That is, fewer than 5% alignment gaps, missing data, and ambiguous bases were allowed at any position.

Currently, it is known that λ_max values of vertebrate photopigments are influenced by ∼40 known amino acid tuning sites (Davies et al., 2012; Musilova et al., 2019). Using conventional bovine rod opsin (Accession Number NP001014890) numbering, spectral peak of absorbance values were reliably predicted for Sws1, Rh1, and Lws photopigments using manual interrogation as outlined in Musilova et al. (2019). Particular attention was made at site 86, which is important for UV-sensitivity (UVS) in Sws1 photopigments (Cowing et al., 2002). For Rh1 photopigments, sites 83, 90, 96, 102, 113, 118, 122, 124, 132, 164, 183, 194, 195, 207, 208, 211, 214, 253, 261, 265, 269, 289, 292, 295, 299, 300, and 317 were analyzed as being spectrally important (Sakmar et al., 1991; Chan et al., 1992; Hope et al., 1997; Yokoyama et al., 1999, 2007, 2008; Yokoyama, 2000, 2008; Hunt et al., 2001; Janz and Farrens, 2001; Davies et al., 2007, 2012; Kuraku et al., 2008). Similarly, spectral tuning sites 164, 181, 261, 269, and 292 were investigated to calculate Lws photopigment λ_max values (Yokoyama, 2000; Davies et al., 2012). To determine the λ_max values of four Rh2 and an Sws2 visual photopigment, molecular dynamics simulations-based modeling approaches were applied (Patel et al., 2018, 2020). In this approach, molecular dynamics simulations were carried out using 3D homology models of vertebrate Rh2 and Sws2-type visual photopigments for which λ_max values have been experimentally measured. Molecular simulations were then used to develop simple statistical models using parameters describing conformation and fluctuation of the 11-cis retinal chromophore and attached lysine residue, which predicted λ_max values with high accuracy. Once determined, representative dark spectra for all visual photopigments were generated using a standard A₁-based rhodopsin template (Govardovskii et al., 2000).

RNA sequencing reads of the developmental series (whole embryos and alevins) were obtained by downloading from SRA on NCBI (Sayers et al., 2019) (BioProject PRJNA72713, BioSample SAMN02864156-SAMN02864171) associated with the publication of the Atlantic salmon genome (Lien et al., 2016). The developmental stage, day degrees (dd), was identified by calculating the number of days associated with the BioSamples and the incubation temperature (9.4°C) (von Schalburg et al., 2014). The samples were aligned to the published Atlantic salmon genome (GCF_000233375.1) using Bowtie2 (Langmead and Salzberg, 2012), read counts were generated using Samtools (Li et al., 2009) and the reads where normalized to the sample with the lowest number of reads. A heatmap presented by the logfold2 change of individual visual opsin genes was made by pheatmap (Kolde, 2019).

Preparation of digoxigenin DIG-labeled riboprobes for the visual opsins of Atlantic salmon were prepared following the manufacturer’s instructions (Roche Diagnostics, Germany). In the synthesis of riboprobes, specific PCR products were used as template for the reaction (Thisse and Thisse, 2008), where synthesized probes were precipitated by LiCl and ethanol together with tRNA (Roche Diagnostics, Germany). Nucleotide sequence alignments showed that sequence identity between visual opsin targets was between 50 and 65%, thus ensuring no cross-hybridization between opsin classes. For the four rh2 genes, the identity in the coding region between the four genes is ∼87% and cross-hybridization was likely to occur. Therefore, the overall Rh2 class expression was analyzed using all four rh2 as templates in the probe synthesis reaction to generate a probe mix of all genes. For the four lws genes, the probe targets the coding region with an overall sequence identity of ∼87%. As such, both lws2 and lws3 were used as template, however, the probe hybridized to all lws genes as the individual probes have a sequence similarity above 90% to paralogues not used during the generation of the probe by PCR. Alevins (555 and 720 dd, hatched Atlantic salmon larvae living on their yolk sac) were euthanized with an overdose of tricaine (MS-222, Sigma, United States) before fixation in 4% paraformaldehyde-buffered PBS and further processed as described in Eilertsen et al. (2014). Cryosectioning of consecutive frontal sections (10 μm) was performed on a Leica CM 3050S cryostat (Leica Biosystems, Germany) and before storage at −20°C, the tissues were air dried for 1 h at room temperature and then incubated for 30 min at 65°C. RNA in situ hybridization was carried out as described by Sandbakken et al. (2012). Images were taken with a Leica DFC 320 digital camera (Leica Microsystems, Germany) attached to a Leica DM 6000B microscope (Leica Microsystems, Germany). Adobe Photoshop CC (San Jose, CA, United States) was used to adjust image brightness and contrast, as well as for displaying schematic overviews of visual opsin expression patterns at different developmental stages.

In silico gene mining of the Atlantic salmon genome and reviewing the literature allowed for the identification of rh1-1, rh1-2.1, rh1-2.2, sws1-1, sws1-2, sws2, rh2-1, rh2-2, rh2-3, rh2-4, lws1, lws2, lws3, and lws4 visual opsin genes (Table 1, Supplementary Table 2, and Supplementary Figures 1–7). However, rh1-2.1, rh1-2.2, and sws1-2 were evaluated to be pseudogenes by Lin et al. (2017) and this present study (Supplementary Figures 2, 6, 7). An evaluation of sws1-2 showed a 6 base pair (bp) deletion and a premature stop codon, indicating that Sws1-2, if translated, would be truncated. The results of this study show that an additional lws gene exists in the salmon genome, here named lws4, when compared to the findings of Lin et al. (2017), as indicated in Musilova et al. (2019). Using molecular cloning, the sequence of 11 full-length visual opsins were identified to be expressed in Atlantic salmon.

Using in silico analyzes, visual opsins were identified in six additional salmonids as listed above, as well as in the northern pike, a member of the closest related diploid sister-group to salmonids (Rondeau et al., 2014; Table 1 and Supplementary Table 2). In brown trout, rh1-1, rh1-2.1, rh1-2.2, sws1-1, sws1-2, sws2, rh2-1, rh2-2, rh2-3, rh2-4, lws1, lws2, lws3, and lws4 were identified. Closer inspection showed that rh1.2.1 and rh1.2.2 are pseudogenes as it possesses several indels within the ORF, which is consistent with the same gene being a pseudogene in Atlantic salmon (Supplementary Figure 6). In addition, sws1-2 had the same 6 bp deletion as present in Atlantic salmon sws1-2, as well as additional indels in the ORF (Supplementary Figure 7). As such, these genes were not included in the phylogenetic analyzes. The following genes were identified in Chinook salmon: rh1-1, rh1-2.1, sws1-1, sws1-2, sws2, rh2-1, rh2-2, rh2-3, rh2orf-4, lws1, lws2, lws3, and lws4; however, both lws3 (partial sequence) and rh1-2.1 contained several indels (Supplementary Figure 6) and were not included in the phylogenetic analyzes. In coho salmon, rh1-1, rh1-2.1, sws1-1, sws1-2, sws2, rh2-1.1, rh2-1.2, rh2-2, rh2-3, rh2-4, lws1, lws2, lws3, and lws4 were identified, revealing an additional rh2 when compared to Atlantic salmon. Closer inspection showed that the ORF of rh1-2.1 had several indels and is, therefore, a pseudogene as in Atlantic salmon (Supplementary Figure 6). The lws3 gene was not included in the phylogenetic analyzes as it lacks sequence at the 5′ end and contains several indels. In rainbow trout, rh1-1, rh1-2.2, sws1-1, sws1-2, sws2, rh2-1.1, rh2-1.2, rh2-2, rh2-3, rh2-4, lws1, lws2, lws3, lws4, and lws4.2 genes were found. As in Atlantic salmon, rainbow trout rh1-2.2 contained several indels and has become pseudogenized (Supplementary Figure 6). In this study, an extra rh2-1.1 gene was identified compared to Lin et al. (2017) but the lws3 gene was evaluated to be a pseudogene, as in Lin et al. (2017), with several indels and a truncated 3′ end. At Ensembl version 106, an extra lws4 was identified at chromosome 15 (named lws4.2). It has a 33 bp deletion in the third transmembrane region and were consider as not functional. In the sockeye salmon genome, the following visual opsin genes were identified: rh1-1, sws1-1, sws1-2, sws2, rh2-1.1, rh2-1.2, rh2-2, rh2-3, rh2-4, lws1, lws2, and lws4. As in coho salmon and rainbow trout, an additional rh2 gene was also identified. The sockeye salmon genome is currently only hosted by Ensembl Rapid Release, and the search for opsins was done at NCBI, and a search for rh1-2.1 and rh1-2.2 genes contained within NCBI databases failed to yield any hits. The lws4 gene was shown to possess an indel in the ORF and was not included in the phylogenetic analyzes. As revealed in Lin et al. (2017), the northern pike genome contains rh1-1, sws1-1, rh2-1, rh2-2, rh2-3, rh2-4, lws1, lws2, lws3, and lws4, and in the updated pike genome at Ensembl database (version 104, Eluc_v4 assembly) a fourth rh2-4 is included.

A maximum likelihood tree based on a codon-matched alignment of nucleotide sequences was generated to show the putative evolutionary relationship between differing salmonid visual opsin gene copy numbers and related visual opsin genes of the northern pike. Specifically, Figure 1 shows the presence of several lws, sws1, sws2, and rh2 cone opsin genes, with all species having one functional rh1 gene. All salmonids have an additional sws1 gene compared to northern pike (Table 2); however, the second sws1-2 sequence was only included in the phylogenetic analyzes for Chinook salmon, coho salmon, rainbow trout, and sockeye salmon since these genes are intact and likely to be functional photopigments. As shown in Lin et al. (2017), northern pike lack a sws2 gene in the genome, while salmonids all possess one intact sws2 gene. The lws1 and lws2 genes were present for all salmonids and northern pike, while the lws3 gene was only included for Atlantic salmon, brown trout and northern pike since it was not identified, present as a partial sequence or assigned as a pseudogene in other salmonids. The lws4 gene was not identified in sockeye salmon but were present for the other species. The maximum likelihood tree of the rh2 opsin class revealed that salmonids possess 3 to 5 rh2 genes: coho salmon, rainbow trout, and sockeye salmon all have five rh2 genes, with phylogenetic analyzes showing that there has been an extra duplication of the rh2-1 gene in these species. The rh2 opsin genes in salmonids are tandemly duplicated in a gene array where each gene is located in close proximity to each other on the same chromosome for a particular species (Supplementary Table 2). Phylogenetically, rh2-1/rh2-2 and rh2-3/rh2-4 appear to branch together.

For Lws, Sws1, and Rh1 photopigments, it is possible to accurately determine spectral peak of absorbance (λ_max) values using manual interrogation (e.g., Davies et al., 2012; Musilova et al., 2019). Specifically, analyses of the ∼40 known amino acid tuning sites showed that Sws1 was predicted to have a λ_max value at 360 nm due to the presence of Phe86, (Cowing et al., 2002), whereas the spectral peak of Atlantic salmon Rh1 was determined to be 509 nm (Figure 2 and Table 2), assuming the use of a vitamin A₁-based retinal chromophore in both cases. Similar inspection of the four Lws photopigments in Atlantic salmon predicted a 14 nm span in their respective λ_max values based on the presence of spectral tuning sites 164, 181, 261, 269, and 292 (Yokoyama, 2000; Davies et al., 2012). Specifically, these spectral peaks ranged from 546 nm (for Lws3) to 560 nm (for Lws1), whereas the λ_max value was predicted to be 553 nm for both Lws2 and Lws4 photopigments (Figure 2).

As λ_max values for Sws2 and Rh2 photopigment classes are difficult to accurately predict from amino acid sequences alone, molecular dynamics simulations-based modeling approaches were applied (Patel et al., 2018, 2020). A two-term statistical model, 475.628 + (−8.720 ×Torsion 15) + (34.925 ×RMSF_(LYS+RET)), predicted the λ_max values of four Rh2 cone visual photopigments expressed in Atlantic salmon (Figure 3). In this model, Torsion 15 was a median value of the dihedral angle formed by C7–C6–C5–C18 atoms of 11-cis retinal (Figure 3C) and RMSF_(LYS+RET) was a value of area under the root mean square fluctuation curve of the chromophore bound to lysine residue in the chromophore binding site (Figure 3D; Patel et al., 2018). Conversely, the λ_max value of Atlantic salmon Sws2 cone visual photopigment was predicted by a three-term model: 2677.5348 − (−17.052 ×Angle 3) + (5.1634 ×Torsion 3) + (2.3642 ×Torsion 12), where Angle 3 (C3–C7–C8), Torsion 3 (C15–C14–C13–C20), and Torsion 12 (C19–C9–C8–C7) were the median values obtained from molecular dynamics simulations (Figures 3E–G; Patel et al., 2020). To do this, molecular dynamics simulations using each visual photopigment homology model were performed as previously described (Patel et al., 2018, 2020). Specifically, 100 ns long molecular dynamics simulation for all five visual photopigment systems were performed using a GROMACS simulation package (Van Der Spoel et al., 2005). RMSF and median values of the angles were then calculated from each 100 ns simulation and these values were applied to the statistical models to predict all λ_max values. Modeling results showed that Sws2 had a λ_max value at 420 nm, whereas the predicted λ_max values of Atlantic salmon Rh2 photopigments span a range of almost 40 nm, from 472 to 511 nm, where Rh2-1/Rh2-2 have λ_max values around 475 nm and Rh2-3/Rh2-4 around 510 nm (Figure 2 and Table 2).

The relative expression profiles of visual opsins in whole embryos and alevins are shown in Figure 4, where the developmental series ranges from the eye pigmentation stage to first feeding. In addition, Table 3 shows the mean relative level of expression by normalized counts at each developmental stage, where reads were normalized to the sample with the lowest number of reads (Supplementary Table 3 shows the individual normalized counts of the developmental series). The results reveal that the expression of visual opsins is substantial at 800 dd, corresponding to the developmental stage around first feeding. However, already at 410 dd, before hatching, opsin expression levels increased within the rh2 and rh1-1 classes. At 800 dd, rod opsin was at a maximal expression level (Table 3). Among the four rh2 opsin genes, rh2-3 and rh2-4 were expressed at higher levels at all stages when compared to rh2-1 and rh2-2. While lws2 had the highest expression level within the lws opsin class, both lws1 and lws3 exhibited little or no expression at embryo and alevin stages. Overall, changes in expression levels during development (Figure 4) revealed that the largest shift in transcript levels occurred between 410 and 800 dd.

The retinal expression patterns of visual opsins in Atlantic salmon after hatching and just before first feeding (555 and 720 dd) were revealed by RNA in situ hybridization using opsin class-specific riboprobes (Figure 5 and Supplementary Figure 8). Strong expression of rod opsin (rh1) was detected in the central retina after hatching, with expression that spread dorsally and ventrally as development progressed (Figures 5A1–A3). Both sws1 (Figures 5B1–B3) and sws2 (Figures 5C1–C3) were also expressed centrally after hatching, with opsin-positive cones spreading dorsally and ventrally in a similar manner to that of rh1; however, the number of sws2-positive cones were sparse when compared to those expressing sws1, especially after hatching. The expression of the four rh2 genes after hatching was more widespread than the other visual opsin genes, even though rh2 expression was detected in the central retina (Figures 5D1–D3). Before first feeding, rh2 opsins were also expressed dorsally and ventrally (Figures 5D1–D3). For the Lws class, opsin expression was initially located centrally, spreading dorsally, and ventrally as development progressed (Figures 5E1–E3), as observed with other visual opsin profiles. For both rh2 and lws opsin classes, expression patterns before first feeding were located proximal to the retinal pigment epithelium (Figures 5D3, E3).

The salmonid-specific fourth whole genome duplication, Ss4R, initially provided the common ancestor of all salmonids with a doubling of the complete genome sequence. However, subsequent Ss4R duplicate gene loss was predominated by pseudogenization (Berthelot et al., 2014; Lien et al., 2016) since the genomes of both Atlantic salmon and rainbow trout do not appear to contain the expected doubling of opsin genes compared to teleosts that did not undergo a 4R duplication event (e.g., northern pike) (Lin et al., 2017). In recent years, several genome assemblies of salmonids and a revised genome assembly of northern pike, Atlantic salmon, and rainbow trout have been deposited to Ensembl (Cunningham et al., 2018), which forms the basis here for analyzing salmonid genomes for the presence/absence of visual opsin genes. Indeed, this study identified a number of intact visual opsins, as well as several pseudogenes. As indicated in Lin et al. (2017) and Musilova et al. (2019), there are at least two genes in the Rh1 class for salmonids, but rh1-2 paralogues have become pseudogenes. The current study showed that there is only one sws2 gene in all salmonids analyzed and northern pike lack any orthologue of the sws2 gene class (Lin et al., 2017). By comparison, multiple rh1, two sws1 and two sws2 paralogue genes were identified in Cyprinus carpio (common carp) and Sinocyclocheilus spp., i.e., fishes where the common ancestor also underwent a 4R event (Lin et al., 2017). Here, two sws1 genes were found in all salmonids analyzed; however, in Atlantic salmon and brown trout the second copy of sws1 seems to have undergone pseudogenization due to presence of indels (e.g., a 6 bp deletion) that would cause in frame protein truncations. The analyzes of the rh2 opsin genes indicate that salmonids have 3–5 rh2 genes; synteny analyses (Lin et al., 2017), and a close chromosomal location indicates that these rh2 genes are tandem duplicated genes and not ohnologues. In northern pike four rh2 genes are similarly located in an array on the same chromosome. Interestingly, four rh2 genes were also identified in C. carpio and Sinocyclocheilus spp. but they do not localize to the same chromosome (Lin et al., 2017), suggesting that these genes either do not derive from a tandem duplication or that sufficient genomic rearrangements have occurred to reposition these rh2 genes to different chromosomes. Phylogenetic analyzes revealed that rh2-1/rh2-2 and rh2-3/rh2-4 branch together, and as described in zebrafish (Chinen et al., 2003) sequence comparison showed a closer relation of rh2-3 and rh2-4 than rh2-1 and rh2-2, indicating that tandem duplication of rh2-3/rh2-4 occurred more recently. Coho salmon, rainbow trout and sockeye salmon have an additional rh2-1 gene, with high sequence similarity, suggesting an even more recent duplication event in these species. Comparative analyzes of the lws opsin class showed that there are 3–4 genes in salmonids, with all species seeming to possess intact functional lws1 and lws2 genes that are located close to each other on the same chromosome and with the sws2 gene in proximity as shown in Lin et al. (2017). Conversely, several salmonid lws3 and lws4 genes are not identified at all, present as partial sequences, or has been assigned as a pseudogene. Notable, the updated genomes of Atlantic salmon and rainbow trout (Ensembl version 106) show that lws3 and lws4 are located on different chromosomes, while for the other salmonids under investigation lws4 is not linked currently to a particular chromosome. By comparison, northern pike has four lws genes located on the same chromosome, where lws1 and lws2 and lws3 and lws4 form distinct clades. In contrast to salmonids, there are two lws genes in C. carpio and Sinocyclocheilus spp. located on different chromosomes (Lin et al., 2017).

Spectral properties of visual opsins in several salmonids have already been analyzed by microspectrophotometry (MSP) of photoreceptors (Cheng et al., 2006), however the results do not take into account the diversity within the Rh2 and Lws classes. Here, manual interrogation of the Sws1, Rh1 and Lws photopigments (Musilova et al., 2019) and molecular dynamics simulations-based modeling approaches for the Sws2 and four Rh2 visual photopigment (Patel et al., 2018, 2020) were applied to estimate the spectral properties of all photopigments within the five opsin classes of Atlantic salmon. For Sws1 and Rh1, the manual interrogation gave λ_max of 360 and 509 nm, respectively, and were in close accordance with the MSP measurements at 361 and 515 nm (Cheng et al., 2006). The molecular dynamics simulations-based modeling of Sws2 gave a λ_max of 420 nm while MSP measurements gave a λ_max 435 nm (Cheng et al., 2006). Notably, the spectral sensitivity of a photopigment can be altered by a vitamin A₁ to A₂ shift of the chromophore, where the A₂ give higher λ_max than A₁ (Bridges, 1972; Hárosi, 1994) and altered ratio of A₁/A₂ has been shown to vary with temperature and daylength in coho salmon (Temple et al., 2006). However, the MSP results indicated that the retinas were primarily based on vitamin A₁ (Cheng et al., 2006) and in the molecular dynamics simulations-based modeling approaches for the Sws2 and four Rh2 the λ_max used a standard A₁-based rhodopsin template, indicating that the 15 nm difference in the λ_max for Sws2 is not due to different chromophores. When comparing MSP to molecular dynamics approaches, it is possible that modeling predictions are more divergent when estimating λ_max values for opsin protein sequences that are less phylogenetically related to the template sequence (e.g., comparing Sws2 to an Rh1 template vs. Rh2 compared to Rh1 that share a higher sequence identity and are more similar in λ_max values). The absorbance maximum determined for Rh2 and Lws were 518 and 578 nm, respectively (Cheng et al., 2006) while the diversity within the Rh2 and Lws classes determined by molecular dynamics simulations-based modeling spanned from 472 to 511 and 546 to 560 nm, respectively. These results indicate that the MSP measurements were done on photoreceptor cells with Rh2-3 or Rh2-4 photopigments for the Rh2 class and Lws1 for the Lws class. Further, the 40 nm range within the Rh2 class of photopigments are in accordance with zebrafish where the four Rh2 photopigments span from 467 to 505 nm (Chinen et al., 2003; Patel et al., 2018), but the range between the Rh2-1/Rh2-2 and Rh2-3/Rh2-4 branches is greater in Atlantic salmon. As proposed in agnathans, alterations in gene copies and subsequent mutations play an important role in evolution of color vision (Davies et al., 2009). Mutagenesis studies have shown that amino acid position 122 in teleost Rh2 opsins is essential in determining green-shifted λ_max (>495 nm) when occupied by Glu (E), and blue-shifted λ_max (<495 nm) when occupied by Gln (Q). This E122Q substitution alone accounts for ∼15 nm of determined spectral shift (Chinen et al., 2005; Patel et al., 2018). Supplementary Figure 4 shows that Atlantic salmon Rh2-1 and Rh2-2 have Q122 whereas Rh2-3 and Rh2-4 have E122. Which likely explains the blue vs. green spectral shift among salmonid Rh2 photopigments. The absorbance maximum of visual opsins in the anadromous Atlantic salmon range from 360 to 569 nm with representatives in all five opsin classes, providing specialized vision for living both in rivers and in the open ocean. In comparison, the benthopelagic Atlantic cod (Gadus morhua) lack representatives in the Sws1 and Lws classes, but has a greater repertoire within the Sws2 and Rh2 classes (Valen et al., 2014).

Color visual capacity can be determined by several mechanisms, including changes in the type of chromophore used and/or the opsin expressed, the latter of which is influenced by opsin gene duplication or loss, as well as by changes in expression levels as a function of development or change of habitat (Carleton et al., 2020). For example, the gain and loss of UV-sensitive cones in salmonids are well studied in relation to their sea-river migration paths (Allison et al., 2003, 2006; Cheng et al., 2006). Previously, a two-step developmental profile of the duplex retina in Atlantic cod was demonstrated, where development of scotopic vision was concurrent with life-stage transition and diminishing expression of some of the rh2 genes (Valen et al., 2016). In Atlantic salmon, with a more direct developmental process, the results of this study are consistent with previous findings showing that the salmon retina is duplex from early developmental stages (Cheng et al., 2007) and that rod opsin is the dominantly expressed visual opsin class post-hatching. As described in Chinook salmon (Cheng et al., 2007), the results presented here revealed that early opsin expression in the retina is located centrally, but expands dorsally and ventrally as eye development progresses. For example, sws1 expression is topographically widely distributed before first feeding, which is in strong contrast to the restricted sws1 expression profile close to the ciliary marginal zone at the smolt stage of rainbow trout (Allison et al., 2003). The number and distribution of sws1 expressing cones in the whole retina at early stages indicates that UV light is likely to be very important for vision in rivers and lakes during early development. Unlike the marine larvae of Atlantic halibut (Hippoglossus hippoglossus) where sws1 expression is restricted to the ventral part of the retina where it detects downwelling UV light, the distribution of sws1 cones in the whole eye of Atlantic salmon suggests that UV light is detected from all directions (Helvik et al., 2001).

Due to high sequence identity between opsin genes within Rh2 and Lws classes, respectively, RNA in situ hybridization experiments were designed not to distinguish between different paralogues genes, but for opsin genes from within a class. However, RNA sequencing analyzes were able to resolve specific expression profiles for opsin genes within a class, where opsins were shown to be differentially expressed during development, with important functional implications. For example, rh2-3 and rh2-4, resulting in photopigments that maximally absorb wavelengths above 500 nm, were more highly expressed at first feeding than rh2-1 and rh2-2, with photopigments that were predicted to have λ_max 471 and 475 nm. Similarly, the expression levels of lws1 and lws3 (with photopigments that maximally absorb at λ_max 546 and 560 nm, respectively) are minimal compared to the expression of lws2 and lws4 (both photopigment types predicted to have a λ_max value at 553 nm). These patterns of differential expression might be a result of distinct expression of opsins within a class during life history transitions, as observed for sws1 (Allison et al., 2003). As such, further studies that analyze the expression profiles of opsins within both Rh2 and Lws classes through smoltification would be of great scientific and commercial interest. In conclusion, this study describes the opsin repertoire of several species of salmonids, accompanied by more in depth analyzes of spectral tuning and expression profiles of Atlantic salmon photopigment genes.