Short-term colour vision plasticity on the reef: changes in opsin expression under varying light conditions differ between ecologically distinct fish species

Detection of visual cues is often critical for behavioural tasks such as mate choice, escape from predators or foraging (Detto, 2007; Foote et al., 2004; Miyagi et al., 2012; Rick et al., 2006; Sandkam et al., 2015; Stuart-Fox et al., 2003). Therefore, tuning of photoreceptor spectral sensitivities to specific visual tasks and/or parts of the light spectrum relevant for such behaviours may be important for maintaining optimal performance (Price, 2017). This is particularly evident in fish, which have dispersed and adapted to habitats profoundly different in their light environment, including freshwater lakes and rivers, marine coastal reefs, pelagic zones and the deep sea. Considering the light conditions in these environments, fish visual systems have adapted to the overall environmental illumination of their habitat (Cronin et al., 2014; Lythgoe, 1979). However, it may be necessary for fish to adjust their visual system in adaptation to changes in seasonal light regime (Loew and McFarland, 1990; McFarland, 1990), microhabitat differences (Marshall et al., 2003), depth (Jerlov, 1977; Loew and McFarland, 1990; McFarland, 1990) or activity period (Loew and McFarland, 1990). Spectral sensitivity tuning in fish can be facilitated by various mechanisms, including structural changes, such as optical filtering of specific wavelengths (Siebeck and Marshall, 2001), or variation of photoreceptor size, number and distribution (de Busserolles et al., 2014; Taylor et al., 2015; Wagner and Kröger, 2005); or physiological changes to the properties of the light-absorbing photopigments contained in the photoreceptors (Bowmaker, 2008).

The wavelength of maximum absorbance (λ_max) of each photoreceptor depends primarily on two components of the visual pigment: a vitamin A-derived light-absorbing chromophore [A1 or A2 (Toyama et al., 2008; Yokoyama and Yokoyama, 1996)], and the opsin, a transmembrane protein that is covalently bound to the chromophore (Hunt and Collin, 2014). Visual opsin genes in vertebrates are classified according to their photoreceptor specificity and wavelength-dependent spectral sensitivity into one rod opsin (rhodopsin, RH1) used for dim-light vision and five classes of cone opsins used for colour vision: SWS1 (short wavelength-sensitive 1, ultraviolet), SWS2 (short wavelength-sensitive 2, violet–blue), RH2B (medium wavelength-sensitive 2B, blue–green), RH2A (medium wavelength-sensitive 2A, green), and LWS (long wavelength-sensitive, yellow–red) (Cronin et al., 2014; Yokoyama, 2008).

Various genetic mechanisms affecting the sequence structure and repertoire of opsin genes (for review see Bowmaker, 2008; Carleton et al., 2016), the type of chromophore used (A1 or A2) and differences in qualitative and/or quantitative expression of opsin genes (for a review, see Carleton, 2009) render photopigments the foundation of a versatile system for adaptation to varying environmental lighting demands (Hauser and Chang, 2017). Importantly, however, besides chromophore substitution, only qualitative or quantitative differential opsin gene expression may contribute to visual system adaptation within the same species in developing or mature fish. Such differences in expression hold the potential for highly adjustable, possibly short-term, visual system adaptation to changes in the prevailing light habitat (Carleton, 2009; Marshall et al., 2015).

Qualitatively differential opsin gene expression occurs most commonly between ontogenetic transitions, i.e. from larval to adult stages, and is often accompanied by migration between different light habitats or a change in diet (Archer et al., 1995; Carleton et al., 2008; Cheng and Novales Flamarique, 2004; Cortesi et al., 2015a; Cottrill et al., 2009; Loew et al., 2002; Shand et al., 2008; Temple et al., 2008). Quantitative differences in opsin expression profiles have been shown in various freshwater and marine species, including Lake Victoria cichlids (Seehausen et al., 2008) and marine damselfish (Stieb et al., 2016) inhabiting different depths, as well as killifish inhabiting spectrally distinct streams (Fuller et al., 2004), resulting in a shift of λ_max to match specific light conditions. Furthermore, when reared under different artificial light, plasticity in opsin expression has been shown in cichlids (Hofmann et al., 2010), black bream (Shand et al., 2008), guppies (Ehlman et al., 2015; Sakai et al., 2016) and killifish (Fuller et al., 2005). Opsin expression plasticity in adult fish has been shown in killifish (Fuller and Claricoates, 2011; Fuller et al., 2010) and African cichlids (Nandamuri et al., 2017), which alter opsin expression levels within only a few days (Fuller and Claricoates, 2011) when exposed to changed habitat light. The variable nature of these findings suggests that opsin expression plasticity may be highly species specific rather than based on a general controlling mechanism mediating opsin expression based on photoreceptor quantum catch. There are two ways in which opsin genes can mediate visual pigment performance: (1) changes in opsin expression may adjust visual sensitivities to regions of the spectrum where light is abundant (Hofmann and Carleton, 2009) [as reported for killifish (Fuller et al., 2004), black bream (Shand et al., 2008), Lake Malawi cichlids (Hofmann et al., 2010), damselfish (Stieb et al., 2016)], or (2) changes in opsin expression can lead to a decline of sensitivity in regions of the spectrum where light is abundant [reported for the blue acara (Kröger et al., 1999; Wagner and Kröger, 2000)] – a compensatory mechanism that helps maintain colour constancy (Wagner and Kröger, 2005).

The opsin repertoire in coral reef fish is less well studied, but offers excellent conditions to investigate mechanisms of visual system adaptation. With regard to species and colour richness, coral reefs are one of the most spectrally diverse ecosystems on Earth – a diversity reflected in the complexity of visual communication among reef fish (reviewed in Marshall et al., 2006; Marshall et al., 2015). Considerable variation of photoreceptor spectral sensitivity (Losey et al., 2003), ocular media transmittance (Losey et al., 2003; Siebeck and Marshall, 2001) and opsin repertoire (Cortesi et al., 2015b; Hofmann et al., 2012; Phillips et al., 2016; Stieb et al., 2017) add to the challenge in understanding this system (Marshall et al., 2015). Stieb et al. (2016) demonstrated that subtle depth-dependent differences in environmental illumination correlate with different opsin expression profiles in some damselfish species whereas others showed a stable expression profile. These findings highlight the possibility that opsin gene expression in reef fish in general may be highly species specific, perhaps because of different ecological and visual demands or phylogenetic constraints. However, it is unclear whether such changes occur during developmental stages (i.e. priming during settlement) or can also occur post-settlement in mature fish. Furthermore, it remains unknown whether changes in opsin expression under pronounced differences in environmental lighting are consistent between species, and more specifically, between ecologically distinct species.

To address this, we investigated the capacity of spectral visual system adaptation in three reef fish species, Pomacentrus amboinensis Bleeker 1868 (ambon damselfish), Pomacentrus moluccensis Bleeker 1853 (yellow damselfish) and Ostorhinchus cyanosoma Bleeker 1853 (yellow-striped cardinalfish). These species belong to two of the most speciose and abundant coral reef fish families, the damselfish (Pomacentridae) and cardinalfish (Apogonidae). Importantly, these families, despite sharing several ecological traits including strong association with coral, and, generally, strong site fidelity and resilience to habitat disruption (Gardiner, 2010; Marnane, 2000), feed at different times of day: damselfish are strictly diurnal whereas cardinalfish are predominantly crepuscular or nocturnal feeders (Emery, 1973; Marnane and Bellwood, 2002). With a few exceptions, damselfish and cardinalfish go through an oceanic and pelagic larval phase, after which they settle on the reef (Leis, 1991; Victor, 1991). As adults, the three species are found in small to large aggregations in clear lagoons, and coastal or seaward reefs (Randall et al., 1990). The fact that all three species co-occur in the same shallow-water coral reef zones, but are active at different times and light levels, makes them particularly interesting to test and compare mechanisms of spectral tuning.

The damselfish visual system, including opsin gene repertoire, is well understood (Hawryshyn et al., 2003; Loew and Lythgoe, 1978; Losey et al., 2003; Marshall et al., 2006; McFarland and Loew, 1994; Siebeck and Marshall, 2001; Siebeck et al., 2008; Siebeck et al., 2010; Stieb et al., 2017). SWS opsins appear to be expressed exclusively in single cones, whereas RH2 and LWS opsins appear to be expressed exclusively in double cones (our unpublished data). However, as the cardinalfish visual system is less well known, we investigated the repertoire of expressed opsin genes in the cardinalfish O. cyanosoma using RNA sequencing. We then examined whether changes in spectral tuning via opsin gene expression plasticity are possible in adult reef fish and whether this is similar or different between nocturnal cardinalfish and diurnal damselfish. To do this, we exposed the two species of damselfish and the one species of cardinalfish to different lighting conditions in terms of colour and intensity for up to 6 months. We then used quantitative real-time polymerase chain reaction (qRT-PCR) experiments to quantify opsin expression. Finally, we were interested in understanding whether opsin expression follows dynamics that maximize signal strength, or functional maintenance and energy efficiency. Signal strength maximization would suggest photoreceptors shift opsin genes to increase photon catch under light conditions that provide relatively more photons at the relevant wavelengths than a reference light environment. In contrast, minimizing energy expenditure would suggest photoreceptors decrease expression of the respective opsin gene to reduce the cost of neural transmission. To address this question, we modelled the λ_max values of photopigments based on the identified opsin gene sequences, and used these to calculate the quantum catch of visual pigments under the experimental and natural light conditions.

Adult specimens (P. amboinensis, N=45; P. moluccensis, N=61; O. cyanosoma, N=82) were obtained between 2015 and 2017 from an aquarium supplier (Cairns Marine Pty Ltd, Cairns, Australia), and shipped to the University of Queensland as quickly as possible (e.g. on the same day or within a few days; as per information provided by the supplier). Additionally, one adult individual of O. cyanosoma used for de novo opsin gene sequencing was collected in February 2015 on the reefs surrounding Lizard Island (14°40′S, 145°27′E), Australia, using SCUBA and hand nets, and was collected under Great Barrier Reef Marine Park Permit (G12/35005.1) and Queensland General Fisheries Permit (140763).

After undergoing light treatments, fish were anaesthetized with an overdose of clove oil (10% clove oil; 40% ethanol; 50% seawater), killed by decapitation, and retinas were dissected from the eyecup and stored in RNA-later (Ambion) for subsequent molecular analysis. For opsin studies, tissues were sampled around midday (between 11:00 h and 14:00 h), and the date and time of dissection were noted. All experimental procedures were approved by The University of Queensland Animal Ethics Committee [QBI/223/10/ARC/US AIRFORCE (NF) and QBI/192/13/ARC].

At the start of the experiment, a subset of individuals for each species (P. amboinensis, N=8; P. moluccensis, N=18; and O. cyanosoma, N=18) were euthanized immediately upon arrival in our lab and used as a baseline (time point=0). Only one specimen of P. moluccensis was excluded from the baseline group as its eye was damaged and discoloured. All remaining individuals were kept under 12 h light:12 h dark lighting conditions in aquaria filled with 200 litres saltwater and subjected to altered light (colour) and control (intensity) environments. The three tested species were kept in the same treatment aquaria. All tanks were illuminated by broad-spectrum high intensity LED (Radion™, Ecotech Marine, Australia) and fluorescent black (FLH0T8BL/36, Toshiba, Japan) aquarium lights. Light environments (red, green and blue) were generated using spectral filter sheets (182 Light Red, 124 Dark Green, 172 Lagoon Blue; Lee Filters) (Fig. 1A).

Fish living in the three different colour habitats were euthanized after 1 month (time point=1, O. cyanosoma, N=23; P. moluccensis, N=9; P. amboinensis, N=22), 4 months (for O. cyanosoma, N=18) and 6 months (for P. amboinensis, N=15 and P. moluccensis, N=17). We decided to terminate the experiment for O. cyanosoma after 4 months to avoid potential health effects as previous husbandry of O. cyanosoma has occasionally proven difficult, reflecting the delicate nature of most cardinalfish species.

After results of colour treatments were known, more individuals of O. cyanosoma (N=23) and P. moluccensis (N=17) were placed in three additional light treatments to test the effects of light intensity. P. amboinensis could not be obtained because of a coral bleaching event on the Great Barrier Reef in 2016. One experimental group was exposed to unfiltered light (Fig. 1A). Two additional groups were exposed to light attenuated by neutral density filters (298, 0.15 ND; 210, 0.6 ND; Lee Filters) reducing light intensity by 60–80% (filter 298) or 20–30% (filter 210) while not completely blocking any part of the spectrum (Fig. 1A). Intensity treatments were run for 1 month (see Table S1 for a summary of specimens used), as this had been shown to be long enough to induce expression changes.

For P. amboinensis and P. moluccensis, opsin genes and their classification have been determined previously (Hofmann et al., 2012; Stieb et al., 2016). As no such data was available for O. cyanosoma, and to verify the opsin genes previously identified in the damselfish, we initially sequenced retinal transcriptomes of three specimens and validated opsin gene classification using phylogenetic methods. In order to quantify opsin gene expression, we performed quantitative real-time polymerase chain reaction (qRT-PCR). All retinas were homogenized using a TissueLyser LT (Qiagen, The Netherlands) and total RNA was extracted with the RNeasy Mini Kit (Qiagen) following the manufacturer's protocol. An optional DNase digestion step was performed to eliminate traces of genomic DNA. Retinal RNA was reverse transcribed using the High Capacity RNA-to-cDNA kit (Applied Biosystems).

RNA was quality checked with an Agilent 2100 BioAnalyzer 6000 NanoChip (Agilent Technologies). RNA-seq libraries were made using the TruSeq RNA Sample Preparation Kit v.2 (Illumina, San Diego, USA) and the retina-specific transcriptomes were sequenced as 125 bp paired reads on the Illumina platform (HiSeq2000 v4) by the sequencing facility within the Queensland Brain Institute at the University of Queensland, Australia. Samples were multiplexed at 12 samples per lane obtaining 20–30 million sequenced reads per sample.

Data were processed using the online Bioinformatics platform Galaxy (Research Computing Centre, The University of Queensland, Australia) (Afgan et al., 2015). Reads were quality checked using FastQC, and high copy sequences, such as primers and library indices were removed using Trimmomatic. Furthermore, regions with quality scores below 20 were removed by cropping at the start and end of a read, as well as by using a sliding window quality crop. Finally, reads with lengths less than 80 bp were dropped from the analysis. Trinity was used for de novo assembly of transcripts, with a group pair distance of 250 bp, and minimum inchworm kmer coverage of 2.

Further bioinformatics analyses were performed using Geneious software (version 9.0.4). For P. amboinensis and P. moluccensis, assembled transcripts were mapped to species-specific known and publicly available opsin genes (P. amboinensis: SWS1, GenBank accession no. HQ286506; SWS2B, HQ286516; RH2B, HQ286526; RH2A, HQ286536; LWS, HQ286546; RH1, HQ286556; P. moluccensis: SWS1, KU745428; SWS2B, KU745427; RH2B, KU745429; RH2A, KU745430; LWS, KU745432; RH1, KU745431). To identify SWS2A opsin genes of O. cyanosoma, the assembled transcripts were also mapped to reference opsin gene sequences from the dusky dottyback (Pseudochromis fuscus: KP004335.1). For each species and all opsin genes identified, we further followed the methods described in de Busserolles et al. (2017) to manually check for gene duplications. Briefly, after identification of candidate gene coding sequences, unassembled reads were mapped to the opsin gene repertoire of the species using medium sensitivity settings (70% identity threshold). Deviating reads were then extracted by working from single polynucleotide polymorphism (SNP) to SNP by exploiting paired-end matching to cover gaps, and their consensus sequence was used as species-specific reference for repeated high-specificity (100% identity) mapping of unassembled reads until maximum obtainable sequence length was reached.

To confirm the assignment of the newly identified opsin genes of O. cyanosoma to the known opsin classes, we aligned their amino acid sequences with the opsin genes of the zebrafish (Danio rerio), the Japanese ricefish (Oryzias latipes), the bluefin killifish (Lucania goodie), the Lake Malawi cichlid (Metriaclima zebra) and the Nile tilapia (Oreochromis niloticus) (see Fig. 2 for GenBank accession numbers). We then estimated maximum likelihood phylogenies for each gene based on the amino acid sequences using RAxML 8.2.10 (Stamatakis, 2014) on the web-based platform CIPRES (Miller et al., 2010), followed by a rapid bootstrap analysis with 1000 replicates. The highest scoring tree was selected as the best tree.

We quantified relative opsin gene expression using qRT-PCR [SYBR Green master (Rox) dye (Roche)] on a StepOnePlus Real-Time PCR System (Applied Biosystems, USA). Following previously described methods (Carleton and Kocher, 2001; Stieb et al., 2016, 2017), relative opsin gene expression was calculated from the efficiency and critical cycle number (C_t), as relative single (SWS opsin genes) and double (RH2 and LWS genes) cone opsin expression as a fraction of the total of single, or double cone opsin genes expressed and relative rod opsin gene expression as the fraction of all opsin genes expressed. The distinction of single and double cone opsin was chosen as in situ hybridization assays on several damselfish and other reef fish species suggest this expression is specific to particular cell types (unpublished data), and because it prevents the masking of expression changes in the less abundant single cone opsins by highly abundant double cone opsins. All percentage values are reported as median and interquartile range.

For qPCR reactions targeting opsin genes of P. amboinensis and P. moluccensis, we used publicly available and already validated primers (Table S2) from Stieb et al. (2016). For O. cyanosoma, we followed the methods as per Stieb et al. (2016, 2017) and designed unique primers (Table S2) for each opsin gene with either the forward or the reverse primer spanning an exon–exon boundary (except for the intronless RH1) to ensure only cDNA would be amplified with a product length of 60–100 bp. Primer efficiencies (Table S2) were tested using a five orders of magnitude dilution series of a species-specific opsin pool. The opsin pool contained equal ratios of fragments of each opsin gene (molarity of fragments was measured using an Agilent 2100 BioAnalyzer High Sensitivity DNA Chip, Agilent Technologies) that were amplified from cDNA using pool primers (see Table S2) specifically designed for each opsin gene identified in the O. cyanosoma transcriptome. Products were separated via gel electrophoresis, then cut out from the gel and purified using the QIAquick PCR Purification Kit (Qiagen). The RH1 amplicon was additionally Sanger sequenced for sequence verification. All experiments were carried out with three technical replicates, and samples originating from the different light and control experiments were randomly assigned to each qPCR plate.

To test whether opsin gene expression changed as a consequence of exposure to different light and control treatments, we used the beta regression method based on the R package BETAREG (Cribari-Neto and Zeileis, 2009). This regression allows handling of non-transformed data to model percentages and proportions. The beta distribution can be of variable shape, and is, therefore, suitable for the analysis of relative opsin gene expression as our dependent variable. To identify relationships between relative opsin expression and different time points, each time point was used as a categorical factor. To test for differences between different light and control treatments, categorical factors were set for each treatment. Repeated hypothesis testing was accounted for by using a Bonferroni correction considering six tested hypotheses for comparisons between treatments and baseline, and five tested hypotheses for comparisons between treatments and no filter treatment (one per treatment group). P-values were thus calculated according to P=α/m, with m=6 or m=5, respectively. Analyses were performed in R (https://www.r-project.org/) using the interface RSTUDIO (version 0.98.1062).

The spectral sensitivities of the different cone and rod opsin classes identified in O. cyanosoma, P. amboinensis and P. moluccensis were modelled using the translated amino acid sequences. Opsin amino acid sequences were aligned with bovine rhodopsin (GenBank accession no.: NP_001014890.1). This allowed us to infer the loci of transmembrane regions and the identification of known retinal chromophore binding pocket sites, as well as previously discovered tuning sites, based on the protein structure (summarized in Hunt et al., 2001; Yokoyama, 2008).

Furthermore, amino acid comparisons of the identified opsin genes to those of other fish species with known opsin λ_max, was used to infer tuning effects where possible. Our opsin gene sequences were aligned with those of the cichlids Metriaclima zebra and Oreochromis niloticus, the killifish (Lucania goodei), the Japanese ricefish (Oryzias latipes), as well as the dusky dottyback (Pseudochromis fuscus). The opsin genes found in these species have been studied extensively, including in vitro opsin protein expression studies to assess pure protein spectral absorbance and microspectrophotometry (MSP; Carleton, 2009; Cortesi et al., 2015a). Damselfish photoreceptor absorbance had previously been measured using MSP (summary in Stieb et al., 2016).

We focused on variable amino acid residues that occurred either in areas corresponding to the retinal binding pocket and were substitutions that resulted in a change in polarity, or at known tuning sites (Dungan et al., 2016; Takahashi and Ebrey, 2003; Yokoyama, 2008). Final calculations were based on pure opsin spectral absorbance of the photopigments of Oreochromis niloticus (Spady et al., 2006), Lucania goodei (Yokoyama et al., 2007) or Oryzias latipes (Matsumoto et al., 2006; RH1). Tuning effects of the amino acid residues present at the sites in question were inferred directly at sites of known tuning effects. Tuning effects of other variable amino acid residues between the species were inferred based on the λ_max differences of the known sequences and the variable sites present (see Table S5 for a summary of bovine rhodopsin aligned site effects considered). Errors in inferred spectral sensitivities will have small effects on quantum catch calculations.

It is notable that we did not test chromophore composition at any stage in our experiment as it is unlikely to occur in our test species. Evidence suggests a limitation to the A1 chromophore in marine fish, on which we based our λ_max calculations. Shifts from one chromophore to another are thought to occur only in fish that undergo extreme habitat changes that bring about a drastic change in environmental conditions, such as eels or salmon (Beatty, 1966, 1975, 1984; Wood et al., 1992). Our test species, however, are confined to a marine environment throughout their adult life history.

The light spectrum in each treatment tank, and, for comparison, on a shallow reef (2 m) off Lizard Island around midday in March 2015, were determined by measuring 45 deg downwelling radiance using a UV/VIS 100 μm optic fibre (Ocean Optics) connected to a USB2000 spectrophotometer (Ocean Optics), and the software Spectrasuite (Ocean Optics). For absolute radiance measurements, the fibre and spectrophotometer were calibrated using a Xenon Arc calibration light source (Ocean Optics).

The λ_max values, as determined by amino acid-based modelling (SWS2B, SWS2A, RH2B, RH2A, LWS) and MSP (SWS1), were used to generate photoreceptor absorbance curves and to calculate the quantum catch of hypothetical photopigments, one for each identified cone opsin class and under each lighting condition. As an additive sites model for site effects in SWS1 genes is not well supported, we conclude that SWS1 λ_max gained from MSP measurements undertaken in P. amboinensis (Losey et al., 2003) are more robust than any estimate. As SWS1 is only expressed in damselfish and SWS1 sequences of both test species are identical in tuning sites, we therefore used those SWS1 λ_max values for both P. amboinensis and P. moluccensis. We are also aware that increasing evidence suggests opsin gene co-expression in cones (Dalton et al., 2014, 2015, 2017; Torres-Dowdall et al., 2017); however, as this information is unavailable for our study species, we believe that for the overall conclusion of our study, our estimated quantum catches will provide a good enough first estimate.

To test for correlation between relative opsin expression and opsin specific photopigment quantum catch, we used Kendall's τb correlation using the Kendall package in R (2005). Quantum catch data was ln transformed for visualization purposes only. Kendall's τb allows the correlation of non-normally distributed data by assigning and correlating the ranked data.

RNA sequencing and de novo transcript assembly reconstructed complete coding sequences of five different opsin gene classes in O. cyanosoma. Their identity was confirmed by amino acid-based maximum-likelihood phylogenetic inference that grouped the newly discovered genes with those of previously well-studied fish species (Fig. 2): four cone opsins (SWS2A, RH2B, RH2A, LWS) and one rod opsin (RH1). Two sister copies of the SWS2A gene (SWS2Aα, SWS2Aβ), and one sister gene each of the RH2B (RH2B2) and RH2A (RH2A2) genes were identified. However, as both RH2 duplicates were expressed at low levels and could not be assembled to full coding sequence length, these two genes were omitted from further analyses.

RNA sequencing and de novo assembly of the retinal transcriptome of P. moluccensis and P. amboinensis confirmed previous reports of six opsin genes (encoding five cone opsins: SWS1, SWS2B, RH2B, RH2A, LWS; and one rod opsin: RH1) found in damselfish (Hofmann et al., 2012; Stieb et al., 2016), and furthermore allowed complete reconstruction of their coding sequence (Fig. 2).

Opsin expression values (for a summary see Table S3) are presented in relative percentage, were normalized within cone types (single cones, double cones), and are presented as median percentage and interquartile range. At time 0 (baseline), the yellow-striped cardinalfish expressed SWS2Aα, SWS2Aβ, RH2B, RH2A, LWS and RH1. The total retinal opsin repertoire was dominated by RH1 opsin, making up 91.2% (3.9) of all expressed retinal opsin (Fig. 3A). In single cones, O. cyanosoma almost exclusively expressed SWS2Aβ (90.6%, 7.5). In double cones, RH2A was expressed strongest (82.5%, 8.6), RH2B at lower levels (17.1%, 8.5) and LWS expression was very low (0.4%, 0.9). Variability in expression was greatest among RH2 (i.e. RH2B: 7.1–43.6%) and small among SWS2A genes.

At time 0 (baseline), both damselfish species expressed SWS1, SWS2B, RH2B, RH2A, LWS and RH1. Rod opsin made up only 63.2% (11.0) (P. moluccensis) and 58.7% (6.5) (P. amboinensis) of total opsin expressed (Fig. 3B,C). Among single cone opsins, both species expressed predominantly SWS1 (P. amboinensis: 86.6%, 9.6; P. moluccensis: 85.0%, 12.5). It is notable that levels of SWS genes showed large variability, particularly in P. amboinensis. In double cones, both species expressed comparable amounts of RH2B and RH2A (P. amboinensis: RH2B=45.0%, 4.0; RH2A=54.5%, 3.9; P. moluccensis: RH2B=43.4%, 3.1; RH2A=52.2%, 2.7), but differed in LWS expression (P. amboinensis: LWS=0.4%, 0.4; P. moluccensis: LWS=3.6%, 2.1, P<0.00017).

Under altered illumination we observed major shifts in opsin expression in the three investigated reef fish species. We observed effects following altered light spectrum, light intensity and treatment duration, as well as differences in effects between opsin gene classes and fish family. A summary of the beta-regression statistics for the tested comparisons of baseline and treatment expression levels is provided in Table S4. The ratio of single cone opsins in damselfish, namely SWS1 and SWS2B, showed the largest changes (Figs 4–7). In colour treatments low in short wavelength radiation, rapid shifts from SWS1 (370 nm) to SWS2B (408 nm) (see Table 1) expression in damselfish were observed (Figs 4–7, panels C–F). Generally, these shifts occurred rapidly after only 1 month, and were even stronger after 6 months. P. moluccensis showed a shift in SWS1 to SWS2B expression ratio similar in nature to that observed in P. amboinensis, but this was not statistically significant, perhaps due to the low number of replicates. Rapid shifts towards SWS2B were also observed in the intensity treatments, but the extent of the shift was less when more UV radiation was available (Fig. 7D–F). In O. cyanosoma, we also observed rapid shifts among single cone opsins (Figs 4–7). Here, the ratio of SWS2Aα (448 nm) to SWS2Aβ (468 nm) (see Table 1) shifted to the longer-tuned photopigment after 1 month of red treatment (Fig. 6A). Under the two brightest intensity treatments (Fig. 7A,B; no filter and 0.15 ND), the ratio of SWS2Aα to SWS2Aβ expression shifted to the shorter-tuned photopigment. Under blue, green and 0.6 ND treatment, single cone opsin expression remained unchanged.

Expression levels among double cone opsins were generally more rigid, showing, where present, only delayed changes in opsin expression under the different treatment conditions. There were also differences in affected genes between damselfish and cardinalfish. In O. cyanosoma, the ratio of RH2B to RH2A opsin expression (476 nm vs 518 nm, see Table 1) shifted towards RH2A in the colour treatments (Figs 4–6, panels A,B); however, this effect only showed in the 6 month treatment groups. The no filter and 0.6 ND group showed a shift from RH2A to RH2B after 1 month; in the 0.15 ND group double cone opsin expression was unchanged (Fig. 7A,B). LWS (544 nm, see Table 1) expression was unaffected by changed light conditions. In damselfish, double cone opsin expression remained comparably stable following changes in lighting conditions. Observed changes that were statistically significant were generally small, or reverted back to baseline levels – and are thus likely to be mostly attributable to natural variability. However, LWS (554 nm, see Table 1) expression in P. moluccensis saw a small increase after 6 months of blue light, 1 and 6 months of green light, 6 months of red light, and 1 month of unfiltered aquarium light. RH2 opsin gene expression in both species was largely unaffected by colour treatments or reverted back to pre-exposure levels (i.e. Fig. 5E,F), except for reduced RH2A (518 nm, see Table 1) expression in P. moluccensis in the no filter and the 0.15 ND groups.

An overview of the considered amino acid positions in each investigated gene, as well as the known or estimated substitution effects, are given in Table S5. On the whole, there were few variable tuning sites compared to other species with known λ_max, suggesting our predictions are reasonable approximations. In addition, because of the tuning site similarities for both Pomacentrus species, their gene predictions are identical for all opsins.

A summary of our calculated λ_max values and any reported MSP measurements from the same or related species are given in Table 1. We calculated the resulting photopigments in damselfish to be maximally sensitive to 370 nm (SWS1), 408 nm (SWS2B), 480 nm (RH2B), 518 nm (RH2A) and 554 nm (LWS). These calculations fit well with values previously obtained using MSP. In O. cyanosoma, we calculated the photopigments to be maximally sensitive to 448 nm (SWS2Aα), 468 nm (SWS2Aβ), 476 nm (RH2B), 518 nm (RH2A) and 544 nm (LWS). To our knowledge, MSP measurements in Apogonidae have been reported only on Pristiapogon kallotperus (Losey et al., 2003); thus we list these for reference.

An overview of light environment spectra used (Fig. 1A), visual pigment absorbance curves generated from λ_max calculations for O. cyanosoma (Fig. 1B), P. amboinensis and P. moluccensis (Fig. 1C), as well as calculated quantum catches (Fig. 1D–G) and correlation plots with opsin gene expression (Fig. 1H) are shown. A summary of all calculated quantum catches is provided in Table 2. Calculated quantum catches in treatments were of similar magnitudes to those calculated for a natural reef setting, except for the red, ND 0.15 and ND 0.6 treatments in which quantum catch was up to 20-fold lower. Overall, under reef illumination, quantum catch was dominant for double cones, slightly less for single cones in O. cyanosoma, and markedly less in P. moluccensis and P. amboinensis.

Under blue treatment illumination, quantum catch was short-wavelength shifted. As a result, single cone opsins collected relatively more photons than double cone opsins in O. cyanosoma. However, in P. moluccensis, single cone opsin quantum catch was very low for SWS1 pigment, but remained comparably high for SWS2B pigment. In both species among double cone opsins, RH2B collected the most photons. Under green illumination, quantum catch in both species was long-wavelength shifted, such that double cone quantum catch dominated single cone quantum catch. In these conditions, RH2A was best tuned to the available light spectrum in both species. The strong light-attenuating properties of the red filter led to conditions such that in both species quantum catch for all photopigments except the LWS-based one was up to 20-times lower than under a reef spectrum. Owing to UV filter properties of the colour filters used, neither of the three colour treatments strongly stimulated the SWS1-based pigment.

All three light intensity treatment spectra resulted in similar quantum catch ratios between the opsin-based pigments. Differences in total caught quanta matched the expectation based on the light intensity. In O. cyanosoma, quantum catch of the five pigments was almost equal, suggesting a relative increase in single cone opsin expression compared with double cone opsin, and among single cone opsin, quantum catch was similar for SWS2Aα and SWS2Aβ. Total quantum catch, however, was between two times greater (no filter) and five times (0.6 ND) lower than under a reef spectrum. In P. moluccensis, relative quantum catch between the five opsin-based pigments was similar under the different conditions, albeit at different total numbers of absorbed quanta.

Test results for whether shifts in opsin expression are proportionate to the quantum catch resulting from the different light treatments using Kendall τ correlation coefficients are given in Table 3. Owing to the small extent of changes in RH2 opsin genes in P. moluccensis and lack of changes in LWS expression in O. cyanosoma, we emphasize the relationship of single cone opsin, RH2 opsins in O. cyanosoma and LWS in P. moluccensis. Overall, in P. moluccensis and O. cyanosoma, single cone opsin expression strongly correlated with quantum catch (Fig. 1Hi–iii). SWS1 expression correlated positively with quantum catch under colour and intensity treatments (Fig. 1Hi,ii), whereas SWS2Aα expression correlated positively with quantum catch under colour treatments only. In O. cyanosoma, neither of the RH2 opsin genes showed correlation with quantum catch across all treatments (intensity+colour). However, analysis of intensity and colour separately revealed that under colour treatments, decreased RH2B expression coincided with decreased RH2B quantum catch, and increased RH2A expression coincided with increased RH2A quantum catch. In intensity treatments, on the other hand, quantum catch did not correlate with opsin expression, thus masking these effects in the combined analysis. In P. moluccensis, LWS opsin gene expression correlated with quantum catch in colour treatments but not in intensity treatments (Fig. 1Hi,ii,iv).

In this study, we investigated the possibility that phenotypic plasticity in opsin expression affects photoreceptor spectral sensitivity in adult coral reef fish by comparing species with a diurnal versus nocturnal feeding activity. We first identified the previously unknown opsin complement in cardinalfish (Fig. 2). We then exposed damselfish and cardinalfish to altered light conditions in the lab (Fig. 1A) and measured the levels of expressed opsin genes in the retinal tissue. We found that in the cardinalfish retina, the total opsin pool is dominated by rhodopsin (RH1 ∼90%) and thus different from damselfish in which rhodopsin only makes up ∼60% of total opsin (Fig. 3). This is consistent with their different ecologies, as nocturnal cardinalfish expressed more of the low light-sensitive rhodopsin compared with diurnal damselfish. Furthermore, we found that the damselfish cone opsin repertoire is tuned to shorter wavelengths than the cardinalfish opsin repertoire, by means of UV-sensitive SWS1 and SWS2B opsins (Fig. 1B,C). Our experiments further demonstrate that opsin gene expression can vary within several months or weeks, but that this plasticity differs between damselfish and cardinalfish (Figs 4–7). Our data further indicate that environmental light can induce opsin expression changes via variation of intensity and spectrum; however, this is true only in cone opsin genes as rhodopsin levels remain unaffected. We found that single cone opsin gene expression demonstrated a more rapid change under most conditions compared with double cone opsin gene expression. Furthermore, correlation analysis indicates that quantitative opsin expression does correlate with quantum catch; however, there are differences between specific opsin gene classes (Fig. 1D,H; Tables 2 and 3).

RNA sequencing confirmed the damselfish opsin repertoire previously described (Hofmann et al., 2012; Stieb et al., 2016, 2017). Five cone opsins (SWS1, SWS2B, RH2B, RH2A, LWS), as well as one rod opsin (RH1), are expressed in the damselfish retina. The retinal opsin gene repertoire we identified in the yellow striped cardinalfish, O. cyanosoma, comprises five cone opsins (SWS2Aα, SWS2Aβ, RH2B, RH2A, LWS) and one rod opsin (RH1).

Whereas our calculation of damselfish visual sensitivities match well with values previously obtained using MSP, those calculated for the cardinalfish only match in part (Table 1). For cardinalfish, there is little data on photoreceptor spectral sensitivity (Cronin et al., 2014; Losey et al., 2003). In the iridescent cardinalfish, only one type of single cone and two spectrally distinct double cone types were identified using MSP (Losey et al., 2003). The deviations of sensitivity from our calculations may be explained by opsin co-expression, which is not accounted for in our modelling, or if genes expressed at very low levels are not functional (LWS, 1% of double cone opsin). However, MSP measurements of single cones in the dusky dottyback (Cortesi et al., 2015a) concur mostly with our SWS2A calculations.

Considering the opsin repertoire of O. cyanosoma – high SWS2Aβ and RH2A, low RH2B and SWS2Aα, and very low LWS – it is possible that their colour vision system operates using a dichromatic opponency mechanism under photopic light conditions. However, without further insight into how and where in the retina these opsins are expressed, this remains unclear. Dichromacy has been suggested to be highly efficient in environments that offer two main colours, such as algae and corals, or in a restricted light environment (Chiao et al., 2000; Lythgoe, 1979; Marshall et al., 2003). Cardinalfish are nocturnal, spending the night primarily feeding on benthic or planktonic invertebrates (Barnett et al., 2006; Marnane and Bellwood, 2002). With a rod opsin to cone opsin ratio of ∼9:1, the cardinalfish visual system shows typical adaptations to life in dimly lit environments (de Busserolles and Marshall, 2017; Hunt et al., 2001; Wikler and Rakic, 1990). This increased rhodopsin (RH1) ratio suggests that colour vision in cardinalfish may be restricted in its functionality, and adapted to subserve mostly specific colour tasks during the day, such as social interaction (Kuwamura, 1985).

The expression patterns we found in damselfish, high SWS1, RH2B and RH2A, and low LWS and SWS2B, confirmed those previously reported in wild specimens (Stieb et al., 2016, 2017), and are used to discriminate between colours (Siebeck et al., 2008). The difference between opsin repertoires in both families with the greatest functional impact is probably the presence of transcripts for UV-sensitive SWS1 and violet-sensitive SWS2B opsin in damselfish in place of blue-sensitive SWS2Aα and SWS2Aβ in cardinalfish, shifting spectral sensitivity in damselfish into ultraviolet wavelengths. UV-reflecting body patterns are common in both families (Marshall, 2000; Stieb et al., 2017), supporting hypotheses that small reef fish may benefit from a covert short-range communication channel invisible for larger, UV-blind predatory fish (Siebeck et al., 2010). All damselfish investigated to date express an SWS1 opsin gene (Stieb et al., 2017), forming visual pigments sensitive to UV light (Hawryshyn et al., 2003; Loew and Lythgoe, 1978; Losey et al., 2003; Marshall et al., 2006; McFarland and Loew, 1994) and possess ocular media transmitting UV light (Siebeck and Marshall, 2001). P. amboinensis has been shown to use these features to identify both conspecifics and heterospecifics based on UV-reflecting facial markings (Siebeck et al., 2010).

Our results provide evidence for the presence of phenotypic opsin expression plasticity in coral reef fish from two families. In diurnal damselfish, UV-sensitive (SWS1) and violet-sensitive (SWS2B) single cone opsins appeared most susceptible to changes in light conditions. Similarly, but to a lesser extent, blue-sensitive single cone opsins (SWS2Aα, SWS2Aβ) in nocturnal cardinalfish responded to changes in light conditions, suggesting a high degree of plasticity and adaptability in these opsins. Double cone opsins were less plastic, showing large shifts only in RH2 genes in the nocturnal cardinalfish. It is important to highlight that expression changes were also observed in fish exposed to the unfiltered (no filter) aquarium light treatment when compared with our baseline. In P. moluccensis, these changes were not significantly different to those observed under the other treatments, thus clouding the distinction between effects of light environment and potential other confounding factors for this species, e.g. stress, season or dietary changes (Table S4). This is further complicated since the no filter spectrum can hardly be considered a negative control as it also differs from the conditions present on the reef. Additionally, holding times may have affected opsin gene expression. For example, time spent in the supplier's facility reportedly did not exceed a few days; however, as they were provided by an external supplier, we cannot be certain that this was consistent for all individuals.

Changes in opsin expression could serve as a rapid mechanism to shift photoreceptor spectral sensitivity in order to adapt to altered optical conditions of the environment, thus maintaining optimal visual perception of vital cues in their immediate surrounding. Altered conditions of the light environment can be associated with seasonal change, a phenomenon particularly common in freshwater systems, but also known to occur in marine habitats, particularly in coastal regions and tropical coral reefs due to increased terrestrial organic matter run-off as a result of wet-season rainfall (Lythgoe, 1979; McClanahan, 1988; Munz and McFarland, 1977). Such effects are commonly associated with algal blooms, or increased amounts of other particulate matter in the water column, and a result of prolonged daylight duration, increased water temperature, increased rainfall and increased land run-off, variable solar radiation intensity or a combination of these factors (Lythgoe, 1979; Munz and McFarland, 1977; McFarland and Munz, 1975). Increased particulate organic matter, or increased phytoplankton, generally leads to a long-wavelength shift of the available light spectrum due to light absorption, suggesting that the visual system response observed in this study under altered lighting conditions could be found in nature. In fact, there are several accounts reporting seasonal periodicity in fish visual system characteristics, such as altered spectral sensitivity in the three-spined stickleback (Gasterosteus aculeatus) (Cronly-Dillon et al., 1968), altered opsin gene expression in the damselfish (Pomacentrus nagasakiensis) (Stieb et al., 2016) and the Japanese ricefish (Oryzias latipes) (Shimmura et al., 2017). Whether the expression changes observed in P. nagasakiensis by Stieb et al. (2016) were in fact due seasonal change remained unclear. Thus, further investigation in this direction is needed to clarify the effects of season on reef fish opsin expression.

Differences in ambient light at different depths is also important to consider. Both damselfish species and O. cyanosoma reportedly occur at up to 40 m depth (Allen et al., 2003; Randall et al., 1990). Stieb et al. (2016) showed that damselfish collected from different habitat depths did differ in opsin expression as a result of adaptation to changed lighting conditions. However, it remained unclear whether the observed changes resulted from plasticity during juvenile settlement, or whether the animals retained expression plasticity in adulthood. As our data suggests, the tested damselfish may have developed those expression changes after they reached maturity. Nevertheless, damselfish and cardinalfish generally display high site fidelity (Gardiner and Jones, 2005; Marnane, 2000; Petersen, 1995), making change of habitat depth or geographical relocation unlikely candidates to demand visual system adaptation in adults of these species.

According to the offset hypothesis, a dichromatic visual system is best tuned to an environment if one of its sensitivities is matched to the overall environmental backlight and the other is offset from this background in order to allow contrast detection (Lythgoe, 1979). This principle may explain the differences observed between single and double cone opsin gene expression, and therefore possibly, the differences observed between damselfish and cardinalfish. In the above scenario, the overall environmental backlight is likely to be the least-variable parameter. Overall, environmental backlight on coral reefs peaks around 500 nm (Marshall et al., 2003; Matz et al., 2006), a wavelength to which reef fish double cones are well matched (Losey et al., 2003). Wavelengths at either end of the spectrum, by contrast, are attenuated rapidly with increasing depth or distance (Lythgoe, 1979) while 500 nm light intensity remains comparably constant. Under these circumstances, constructing a visual system with set double cone but comparably plastic single cone spectral sensitivities, would supply the most feasible adaptive system.

Rhodopsin expression remained largely unaffected by changes in light in all tested species, regardless of shifts in spectral distribution or light intensity. For a monochromatic visual system, any change in light condition effectively only changes perceived intensity because different spectral channels for comparison are not available (Gegenfurtner and Kiper, 2003). Our study species and other investigated shallow water reef fish express only one RH1 opsin, and to date, more than one rhodopsin has been found among fish only in several deep-sea species (de Busserolles et al., 2015; Partridge et al., 1992; Pointer et al., 2007), carp (Lim et al., 1997) and the zebrafish (Morrow et al., 2017). Hence, if changed light conditions had had an effect, we would have expected to see this in all treatment groups differing in light intensity. In fact, from other systems, rhodopsin expression is known to fluctuate following a circadian rhythm (Korenbrot and Fernald, 1989); however, it seems that this is not affected in the timeframes we investigated here.

To investigate whether changes in the spectral composition of the light environment favour upregulation or downregulation of opsin gene expression by means of visual pigment performance, we calculated quantum catches of each photopigment modelled from the peak spectral absorbance calculated using its amino acid sequence. Our results suggest that both strategies are employed and that it depends on the nature of the environmental change, as well as on the affected pigment, which strategy will drive opsin expression change. For instance, SWS1 opsin gene expression in Pomacentrids correlated positively with increased quantum catch of the SWS1 photopigment. Consequently, SWS1 expression in comparison to SWS2B expression dropped in a proportionate manner upon exposure to gradually decreasing levels of UV radiation. This agrees with reports of long-wavelength shifted spectral sensitivities in black bream reared in short wavelength reduced conditions (Shand et al., 2008), but is seemingly in contrast to a reported reduction of blue-sensitive single cones in blue acara, when reared under monochromatic blue light (Kröger et al., 1999, 2003; Wagner and Kröger, 2000; reviewed in Wagner and Kröger, 2005). The non-compensatory principle apparently controlling SWS gene expression in damselfish may greatly influence visual system capabilities in an ecological context such as their ability to use UV signals for conspecific and heterospecific identification (Siebeck et al., 2010). As known habitat depth extends to 40 m (Randall et al., 1990), a significant difference in available UV illumination from high in surface waters to almost nothing at depth on the reef is a real consideration here (Cronin et al., 2014).

In O. cyanosoma, RH2B expression positively correlated with quantum catch in colour treatments, suggesting the adaptive response is driven by RH2B expression. LWS expression in P. moluccensis, in comparison, correlated negatively with quantum catch, further illustrating that the effect of changed light conditions is dependent on the nature of the change and the available opsin repertoire.

Differential opsin gene expression could be facilitated by various mechanisms. The number of photoreceptor cells or outer segment size may change and thus provide more or less room for the opsin. In cichlids reared under chromatically deprived light conditions, both mechanisms were observed (Kröger et al., 1999; Wagner and Kröger, 2000, 2005). Increasing evidence suggests that gene co-expression can also shift spectral sensitivity via differential opsin gene expression. Opsin co-expression in individual photoreceptors is known to occur in fish (Dalton et al., 2014, 2015, 2017; Takechi and Kawamura, 2005; Torres-Dowdall et al., 2017) and allows rapid adjustments of the opsin complement without making structural changes to the retinal anatomy. By yielding photoreceptors with peak spectral sensitivities intermediate between those of pure opsin-based photoreceptors, co-expression may furthermore be an effective means to achieve optimal backlight matching.

Lastly, species differences in total amounts of retinal cone opsin and the relative proportion of photoreceptors subserving colour and luminance vision, respectively, may facilitate different adaptive responses in opsin gene expression changes. If one assumes two visual systems with similar proportion of photoreceptors tasked with colour and luminance vision, but with different total cone opsin volumes, the one with less cone opsin available (i.e. cardinalfish) should show greater sensitivity to expression changes, since smaller increases or decreases of gene expression have a larger effect on relative differential gene expression. As a consequence, even if the net change in gene expression in both systems were similar, relative expression changes would be greater in the system with less cone opsin overall. If the colour task in this species were background matching, changes in lighting conditions would have a greater relative effect. Such effects could be affected by differential topographic distribution of cone opsins, a phenomenon known at the molecular level, among others, from work on cichlids and zebrafish (Dalton et al., 2015; Takechi and Kawamura, 2005). On a morphological level differential distribution of different photoreceptor classes is common in many animals, including many reef fish (Collin and Shand, 2003). Such differences are thought to relate directly to visual system demands and may be related to tasks specific for certain cone classes (e.g. background matching; Temple, 2011). Such a spatial differentiation of tasks may then result in differential distribution of opsins across the retina (as in the human fovea, which contains fewer blue-sensitive cones; Curcio et al., 1991; Roorda and Williams, 1999) and as a consequence, make these different retinal areas respond differently to altered light conditions.

Increasing evidence suggests that gene expression plasticity seems to be crucial for sensory system adaptation to environmental conditions. Our study shows that retinal opsin expression in adult reef fish is plastic, and that it can be modulated by spectral and intensity changes in environmental light. Such expression adjustments may allow rapid adaptation to changing light conditions in the wild because of changed habitat depth or seasonal variability. Our results largely concur with previous reports on opsin gene expression plasticity in fish; however, the effects observed here differ between species and opsin gene class, suggesting that interactions of this nature need to be assessed at the species level.

We would like to thank Fabio Cortesi for providing advice on RNA-seq analyses, Cairns Marine for sourcing and supplying fish, and the staff at Lizard Island Research Station for support during field work. Furthermore, we thank Janette Edson, QBI Genomics Facility, for preparing the libraries and running the RNA-seq, and the Mowry lab (QBI) for kindly providing the StepOnePlus Real-Time PCR System.