Seeing Picasso: an investigation into the visual system of the triggerfish Rhinecanthus aculeatus

ABSTRACT Vision is used by animals to find food and mates, avoid predators, defend resources and navigate through complex habitats. Behavioural experiments are essential for understanding animals' perception but are often challenging and time-consuming; therefore, using species that can be trained easily for complex tasks is advantageous. Picasso triggerfish, Rhinecanthus aculeatus, have been used in many behavioural studies investigating vision and navigation. However, little is known about the molecular and anatomical basis of their visual system. We addressed this knowledge gap here and behaviourally tested achromatic and chromatic acuity. In terms of visual opsins, R. aculeatus possessed one rod opsin gene (RH1) and at least nine cone opsins: one violet-sensitive SWS2B gene, seven duplicates of the blue–green-sensitive RH2 gene (RH2A, RH2B, RH2C1-5) and one red-sensitive LWS gene. However, only five cone opsins were expressed: SWS2B expression was consistent, while RH2A, RH2C-1 and RH2C-2 expression varied depending on whether fish were sampled from the field or aquaria. Levels of LWS expression were very low. Using fluorescence in situ hybridisation, we found SWS2B was expressed exclusively in single cones, whereas RH2A and RH2Cs were expressed in opposite double cone members. Anatomical resolution estimated from ganglion cell densities was 6.8 cycles per degree (cpd), which was significantly higher than values obtained from behavioural testing for black-and-white achromatic stimuli (3.9 cpd) and chromatic stimuli (1.7–1.8 cpd). These measures were twice as high as previously reported. This detailed information on their visual system will help inform future studies with this emerging focal species.

Over the past decade, a coral reef triggerfish, Rhinecanthus aculeatus (Linnaeus, 1758) (commonly known as Picasso or Lagoon triggerfish), has been used in a number studies on visual processes, including the role of double cones (two single cones joined together) (Pignatelli et al., 2010), colour discrimination thresholds (Champ et al., 2016; system of this emerging focal species, which we hope will inform future studies of visual perception and navigation.

Study species and specimen collection
Rhinecanthus aculeatus are common reef fish found on shallow sub-tidal reef flats across the Indo-Pacific region and are generalist omnivores known to feed on a varied diet including algae, detritus, molluscs, and crustaceans (Randall et al., 1997). Individuals (n = 23) were caught using clove oil and hand nets from shallow reefs surrounding Lizard Island, Great Barrier Reef, Australia, or obtained from an aquarium supplier (Cairns Marine Pty Ltd; Cairns, Australia) between 2017 and 2020. Fish ranged in size from 8.1 to 20.6 cm (standard length, SL) and were collected under a Queensland General Fisheries Permit (183990) and Great Barrier Reef Marine Park Authority Permit (G16/38497.1). Sex and age were not determined. Experiments were approved by the University of Queensland Animal Ethics Committee (2017/AEC000077 and QBI/304/16).
For molecular and anatomical techniques, fish (n = 15) were anaesthetized with clove oil (10% clove oil; 40% ethanol; 50% seawater) and then euthanised. Eyes were enucleated, then the cornea and lens were removed. This species has a yellow corneal pigment (Siebeck and Marshall, 2001), the density of which increases across the cornea during the day (N. F. Green, unpublished observations). Depending on the analysis, the eyes were preserved in different fixative solutions. Eyes that were allocated to retinal mapping were fixed overnight in 4% paraformaldehyde (PFA) and stored at 4°C. Following this, these eyes were stored in 0.1 M phosphate-buffered saline (PBS; pH 7.4) until further analysis. For RNA sequencing, eretinas were dissected out of the eyecup and preserved in RNAlater (Ambion) at −20°C until extraction. For FISH, retinas were prepared following the protocol in Barthel and Raymond (2000) and fixed in 4% PFA at 4°C overnight, washed twice for 5 min in PBS and rinsed briefly in 70% MetOH, before being transferred to 100% methanol and stored at -20°C until use. One individual was fin clipped and the tissue was preserved on 100% ethanol for genome sequencing.
For behavioural experiments, fish (n = 8) were transported to the University of Queensland and housed in individual tanks (89 x 41 x 22 cm). Each tank had continuously flowing water supplied by a sump system and air stones to oxygenate the water.

Transcriptome and genome sequencing, quality filtering and de novo assembly
The retinal transcriptomes of five R. aculeatus were sequenced according to (Musilova et al., 2019). One individual was euthanised immediately after capture from the reefs off Lizard Island, Great Barrier Reef, and four individuals were euthanised after being in aquaria at the University of Queensland for between 10 and 18 months. In summary, total RNA was extracted using a RNeasy midi Kit (Qiagen), and the quality and concentration of the RNA was checked using a Eukaryotic Total RNA Nanochip on an Agilent 2100 BioAnalyzer. The transcriptome sequencing and RNAseq libraries (paired end, 150bp insert) was outsourced to Novogene (novogene.com). The transcriptome filtering and de novo assembly was performed following the protocol described in (de Busserolles et al., 2017).
To identify the visual opsin genes from the transcriptomes, further analyses were performed following the detailed protocol in de Busserolles et al. (2017). The assembled transcripts of R. aculeatus were mapped to the opsin gene sequences of the Dusky dottyback (Pseudochromis fuscus: GenBank accession No.: KP004335.1) using the medium sensitivity (30% max mismatch between transcripts) in Geneious v.2020.1.2 (www.geneious.com).
Pseudochromis fuscus was chosen because of its relatively close phylogenetic relationship with R. aculeatus, as well as having representatives from all five visual opsin gene classes.
To ascertain that all opsin genes were correctly assembled, and lowly expressed genes were picked up in the assembly, the filtered, unassembled transcriptome reads were then backmapped against the extracted opsins' sequences using medium-low sensitivity (20% max mismatch between reads). For RH2, we detected extra copies that were only partially assembled or that were misassembled. Hence, in this case we used a read mapping approach to disentangle similar gene copies as per de Busserolles et al. (2017) and Musilova et al. (2019). Filtered, unassembled reads were mapped against the P. fuscus RH2 reference using medium-low sensitivity settings (20% max mismatch between reads). We then extracted Journal of Experimental Biology • Accepted manuscript copy-specific reads by moving along the reference from single-nucleotide polymorphism (SNPs) to SNP. Paired-end information allowed us to bridge the gaps between the SNPs. If the coding region was not completely recovered, we re-mapped unassembled reads using the consensus as a template with low sensitivity settings (0-2% max. mismatch between reads). This allowed us to reconstruct the whole coding region of all expressed RH2 copies.
To make sure that we did not miss any visual opsin genes that were not expressed in our samples, we sequenced a draft genome for R. aculeatus (ID Olaf) using Illumina shortread technology. Genomic DNA was extracted from a fin clip using the Qiagen DNeasy Blood & Tissue kit according to the manufacturers protocol (qiagen.com). DNA quality control (Agilent 5400, agilent.com), library preparation, and sequencing (paired end 150 bp, 350 bp insert) was outsourced to Novogene (novogene.com). Visual opsin genes were subsequently extracted from the genomic raw reads (212'097'824 paired end fragments, 31.8 Gb) using the read-mapping approach as outlined above and detailed in (Musilova et al., 2019). The only difference was that for the genome, we used the raw-reads and mapped them against the single exons of the P. fuscus opsins to avoid long-repetitive intronic sequences. The extracted opsin genes from the genome were then combined with the ones mined from the transcriptomes to generate a R. aculeatus visual opsin gene dataset.
We used BLAST (http://blast.ncbi.nlm.nih.gov/) to confirm the opsin gene identity and by phylogenetic reconstruction. To obtain the opsin gene phylogeny, the opsin gene sequences of R. aculeatus were aligned with a reference data set (obtained from GenBank https://www.ncbi.nlm.nih.gov/genbank) using MAFFT v.7.450 (Katoh and Standley, 2013) with the L-INS-I settings. On the CIPRES platform, jModeltest v.2.1.6 was used to select the most appropriate model of sequence evolution. Following this, MrBayes v.3.2.7a (Ronquist et al., 2012) was used to infer the phylogenetic relationship between the genes. The following parameters were applied: GTR+I+γ model, with two independent MCMC searches (four chains each), 10 million generations per run, a tree sampling frequency of 1,000 generations, and a burn-in of 25%.

Opsin gene expression
For quantitative gene expression, the filtered reads were mapped to the coding regions of the identified opsin genes with high-specificity settings (98% identity, 80bp minimum read overlap) as per the protocol in de Busserolles et al. (2017). The gene-specific proportional expression was calculated as the fraction of all expressed single cone genes for SWSs, double Journal of Experimental Biology • Accepted manuscript cone genes for RH2 and LWS, and all opsin genes for RH1 (see  for a detailed discussion on opsin gene expression calculations).
For each opsin gene ( ), the read count ( ) was normalised to the length of its coding sequence ( ): The proportion of rod opsin expressed, was calculated as the proportion of RH1( ), relative to the total normalised opsin expression ( ).
The proportional expression of single cones ( ), and double cones ( ), was calculated by the proportion of the normalised opsin expression ( ), out of the total normalised expression for single ( ) or double cones ( ). Depending on what type of cone opsin that gene was expressed in, the two following formulas were used:
Previously extracted retinal mRNA was reverse transcribed using a High-Capacity RNA-to-

Prediction of visual pigment maximal absorbance
Maximal absorbances (λ max ) of expressed R. aculeatus visual pigments were predicted by comparing opsin amino acid sequences to those of percomorph fish species for which the peak spectral sensitivities of their (A1 chromophore-based) visual pigments were known from in-vitro pigment reconstitution (Oryzias latipes, RH1: 502 nm λ max , AB180742.  , 2006) and by applying tuning effects to their respective peak spectral sensitivities for substitutions at known tuning sites documented in the literature or substitutions at known tuning sites that match the polarity shift of documented substitutions (e.g., for reviews see Takahashi and Ebrey, 2003;Yokoyama, 2008;Yokoyama and Jia, 2020). For an overview of considered sites and applied effects see  (Palczewski et al., 2000).

Preparation of retinal wholemounts
Retinal wholemounts were prepared according to standard protocols (Coimbra et al., 2006;Stone et al., 1981;Ullmann et al., 2012). The orientation of the retina was kept by noting the position of the falciform inside the eyecup once the cornea and lens were removed.
In all our R. aculeatus specimens, the falciform process ended in the ventral margin of the retina. Each retina was bleached overnight at room temperature in a solution of 3% hydrogen peroxide in 0.1 M PBS. For photoreceptor analysis, retinas were whole mounted (photoreceptor layer up) in 100% glycerol on a microscope slide. For ganglion cell analysis, retinas were whole mounted, ganglion cell layer facing up, on a gelatinised slide, left to dry overnight in formalin vapour to improve fixation and cell differentiation (Coimbra et al., 2006;Coimbra et al., 2012), stained in 0.1% cresyl violet (Coimbra et al., 2006) and cover slipped with Entellan New (Proscitech). Possible shrinkage during staining was considered negligible and if present confined to the retina margin, since the retinal wholemount was attached to the slide during the entire staining process (Coimbra et al., 2006).

Stereological analyses and topographic map construction
Following the protocols described in (de Busserolles et al., 2014a;de Busserolles et al., 2014b), topographic distribution of single cones, double cones, total cones and ganglion cells were assessed using the optical fractionator technique (West, 1991) modified by (Coimbra et al., 2012;Coimbra et al., 2009). Briefly, cone photoreceptors and ganglion cells were randomly and systematically counted using the parameters listed in Table 1 (Table 1). To get a more accurate estimate of the retinal ganglion cell peak density, sub-sampling was performed in the highest cell density area using the same counting frame but with half the grid size from Table 1.
Single cones and double cones were easily distinguished and counted separately and simultaneously using two different markers to generate data for single cones alone, double cones alone, and the two cell types combined (total cones). Ganglion cells were arranged in a single layer within the ganglion cell layer that also comprised displaced amacrine cells and glial cells. Since amacrine cells were not easily distinguished from ganglion cells using cytological criteria alone (Collin and Collin, 1988;Hughes, 1975), especially in high density areas, they were included in the cell counts and only glial cells were excluded. While the inclusion of amacrine cells in the analysis usually does not influence the overall topography (Collin and Pettigrew, 1989), it may contribute to a slight overestimation of the peak density of ganglion cells and ultimately to a slight overestimation of spatial resolving power.
Topographic maps were constructed in R v.2.15.0 (R Foundation for Statistical Computing, 2012) with the results exported from the Stereo Investigator Software according to Garza-Gisholt et al. (2014). The Gaussian kernel smoother from the Spatstat package was used (Baddeley and Turner, 2005) and the sigma value was adjusted to the grid size.

Estimation of spatial resolving power
The upper limit of the spatial resolving power (SRP) in cycles per degree was estimated for each individual using the peak ganglion cell density (PDG in cells/mm) as described by (Collin and Pettigrew, 1989). Briefly, the angle a, subtending 1 mm on the retina can be calculated as follows: Where f, the focal length or the distance from the centre of the lens to the retina, is 2.55 (Matthiessen's ratio, (Matthiessen, 1882)) times the radius of the lens. Knowing a, the PDG and the fact that two ganglion cells are needed to distinguish a visual element from its neighbour, the SRP in cycles per degree (cpd) can be calculated as follow:

Experimental setup
We used eight triggerfish (8.1 to 16.5 cm, SL) to measure achromatic and chromatic acuity using a pairwise discrimination behavioural task with square wave gratings (4 x 4 cm; Figure S2). Fish were allowed to acclimatise to their tanks for 2-3 weeks and fed twice a day with a mixture of blended squid, prawns, fish flakes and peas. Tanks were illuminated with 240V/50Hz LED Batten 20W lights (FL2527; Fuzion Lighting, QLD), which were hung at 50 cm above the end of the tank at which the target stimuli were placed. The side-welling irradiance of each tank was measured with an Ocean Optics USB2000 spectrometer with a 400 µm fibre and a cosine corrector ( Figure S2). Measurements were taken in the middle of the water column at 10 cm from the end of the divider pointing towards the stimuli.
Each tank was divided into three sections: a holding area and two passages separated by a 50-cm long white opaque Corflute board, down which fish could swim to reach the target stimuli ( Figure S2). During testing, a transparent Perspex board was initially placed across the tank at the end of the Corflute divider, which allowed the fish to swim back and forth in front of both stimuli several times before the transparent Perspex board was lifted and they could swim down one passage, which was considered a choice by the fish. This set the point of decision to 50 cm.
We tested black-white square wave achromatic gratings and two combinations of chromatic square wave gratings: green-yellow to mainly stimulate the medium and long cones, and pink-purple to stimulate the short and long cone. Stimuli were between 0.5 to 5 for black-white achromatic stimuli (n = 10 stimuli) and 0.5 and 3 cpd for chromatic stimuli increasing in increments of 0.5 cpd (n = 6 stimuli). A control ('solid') stimulus was made using a square wave grating of 11 cpd, which was deemed unresolvable by the triggerfish from previous data (Champ et al., 2014). Overall, the total number of fish that were tested for each colour combination was: achromatic (n = 4), green-yellow (n = 5), pink-purple (n = 5).
Square wave gratings were printed on Steinbeis TrendWhite paper using a Canon LaserJet Pro 400 printer and laminated with 80-micron gloss laminating sheets. We tried to ensure colours within a chromatic combination were as isoluminant as possible (greenyellow, and pink-purple) based on quantum catches for an average of double cone members, which are thought to process luminance information (Pignatelli and Marshall, 2010 standard was used to calibrate the spectrophotometer and a piece of black velvet covering the end of the fibre was used as the dark standard. Quantum catches of photoreceptors were quantified using photoreceptor spectral sensitivity of triggerfish , illumination and reflectance spectra of printed colours as per equation (1) in Vorobyev and Osorio (1998

Training
In the first round of testing, each fish was randomly assigned to one of three treatments: black-white (n = 2), green-yellow (n = 3), or pink-purple (n = 3) stimuli (fish were retrained to another treatment in a second round of testing). Within each of these groups, one fish was assigned the control (11 cpd) stimulus as the rewarded (S+) stimulus and the remaining individuals were assigned the striped (0.5 cpd) stimulus as the rewarded (S+) stimulus. This was done to account for any bias in the type of stimulus (control/striped) being learnt. Fish were trained using operant conditioning to approach and peck at a stimulus to receive a food reward. Initially, a small piece (< 5 mm) of squid mantle was stuck (using the natural adhesive properties of squid) on the rewarded stimulus, while the unrewarded stimulus was left without. Once fish had associated the S+ stimulus with a reward, the squid was removed from the stimulus and fish were then only presented with a reward from above on forceps once they had pecked the S+ stimulus.
Before commencing each trial, an opaque corflute board was placed in front of the transparent Perspex board to ensure the fish could not see the stimuli being placed at the end of each passage. Stimuli were placed in the left or right passage using a random number generator to ensure that fish did not develop a side bias. Once stimuli were in place, the opaque board was lifted, and the fish were given 20 seconds to view the stimulus through the transparent Perspex board. The transparent board was then lifted and the side that the fish swam down was recorded. Fish did not peck the stimulus in all cases but were given a food reward for approaching the correct stimulus. In instances when a fish swam down one passage, but then changed direction and then went down the other passage during a trial, only the original choice was recorded. If fish had not swum down one passage within one Journal of Experimental Biology • Accepted manuscript minute of the transparent board being lifted, the fish was deemed unmotivated and trials for that fish were stopped for the rest of that session and continued again in the next session.
Once fish successfully chose the S+ stimulus > 80 % of the time over 4 sessions, fish moved on to the testing phase. It took 3-4 weeks to train most fish (total of 20-29 sessions for each); however, one fish only took 1 week (9 sessions) to reach the required standard for testing.

Testing
Testing trials were conducted as per training methods, and 1-2 sessions were conducted each day with a minimum of two hours between sessions. The training stimulus (0.5 cpd) was used as reinforcement through the testing phase and was presented 20-98 times per fish, depending on the individual. Each grating between 1.0 and 3.0 cpd was presented to each fish 7-20 times for black-white achromatic stimuli and 9-30 times for chromatic stimuli. Once fish had completed testing with their first treatment stimuli (blackwhite, green-yellow, or pink-purple), they were randomly assigned to one of the other two treatments, then trained and tested again as above. However, two fish only completed one treatment due to the time taken for training and testing. In total, 2418 trials were conducted with each grating presented 110 + 14 times (mean + s.e.) (Table S4).

Data analysis
Data analysis for the behavioural experiment was done in RStudio v.1.3.1056 (Team, 2020). We used the R package quickpsy v.0.1.5.1 (Linares and Lopez-Moliner, 2016) to produce logistic functions for all fish for each target colour set (goodness of fit: deviance < 8.6, P > 0.47). Our response variable was whether the fish chose (1) or did not choose (0) the positive S+ stimulus. We then used the function 'threshold' to interpolate the 62% threshold, which corresponds to a correct choice frequency significantly different from random behaviour, assuming a binomial distribution of the pooled data per grating (n = 110, p < 0.01, one-tailed binomial test). Once thresholds were calculated for each fish, we tested for significant differences between treatments using a General Linear Mixed Model with the lme4 package v. 1.1-27.1 (Bates et al., 2015) and the function lmer. Fish ID was included as a random variable. We produced p-values using the lmerTest package v.3.1-3 (Kuznetsova et al., 2017). There was no difference in whether fish were trained to receive a food reward from the control stimulus (11 cpd) or the grated stimulus (t 6.4 = 0.15, p = 0.88); fish did not perform differently within a colour treatment between the first and Journal of Experimental Biology • Accepted manuscript second sessions (t 3.3 = 0.60, p = 0.59) and there was no effect of the size of the fish on the visual acuity (z = 0.67, p = 0.51); therefore these factors were excluded from the final analysis.

Opsin gene repertoire
The R. aculeatus genome contained one rod opsin gene (RH1) and at least nine cone opsins: one violet-sensitive SWS2B gene, seven duplicates of the blue-green-sensitive RH2 gene (RH2A, RH2B, and RH2C1-5), and one red-sensitive LWS gene. For RH2B and RH2C-4 and RH2C-5, only the first two exons could be annotated using the whole-genome readmapping approach. This is because the intron between exons 2 and 3 of RH2 is too long to be reconstructed using the short reads alone (for details see . A high-resolution R. aculeatus genome based on long-read technology will be needed in the future to resolve these genes and to assess if even more copies are present. Except for RH2B and three RH2C copies (RH2C3-5), all other opsins were recovered from the transcriptomes.
The phylogenetic reconstruction confirmed the class of each opsin gene identified (Figure 1).

Opsin gene expression
The retina of R. aculeatus was rod dominated, where RH1 was expressed on average 68.9% ± 3.75% (n = 8, mean ± s.e.m.). SWS2B accounted for 100% of the cone opsin expression in single cones. For the double cone opsins, there was change in expression of the RH2 and LWS genes in relation to whether fish were housed in aquaria or sampled directly from the field. Aquarium fish kept for behavioural trials showed higher RH2A expression (n = 4, 60.30 ± 1.65%) compared to individuals sampled directly from the field (n = 4, 35.04 ± 4.38%). However, RH2C-2 was expressed at higher levels for individuals collected directly from the field (45.89 ± 2.12%) compared to aquarium individuals (27.38 ± 3.81%). RH2C-1 was expressed similar between field site collected (17.34 ± 4.55%) and aquarium individuals (12.14 ± 2.17%). In addition, LWS was expressed at low levels for both aquarium and field site collected individuals (field site: 1.70 ± 1.23 %, aquarium: 0.18 ± 0.09%) (Figure 2).

Fluorescent in-situ hybridisation (FISH)
Double labelling RNA fluorescence in-situ hybridisation of expressed opsin mRNAs in R.
aculeatus wholemount retinas showed that medium wavelength opsins identified from the retinal transcriptome, RH2A and RH2Cs (RH2C probes did not discriminate between the two identified RH2C paralogs) were expressed in opposite members of double cone photoreceptors across the retina (Figure 3, A-D). The sole single cone opsin identified from the retinal transcriptome, SWS2B, was expressed exclusively in single cones (Figure 3, E+F).

Prediction of visual pigment maximal absorbance
Amino acid sequence similarity between R. aculeatus opsins and the used reference opsins ranged from 85.5% (51 variable amino acids) for RH2C-1 and RH2C-2 to 94.1% (21 variable amino acids) for RH1 (Table S3). Among variable amino acids, 33% (RH1) to 47% (LWS) (data not shown) lay outside transmembrane regions and were thus unlikely to impact spectral tuning. The numbers of variable amino acids across transmembrane regions ranged from 15 in RH1 to 32 in SWS2B, the majority of which were substitutions that did not incur a physicochemical change. One (RH1), five (SWS2B), nine (RH2A), 12 (RH2C-1) and 11 (RH2C-2) were at known tuning sites. However, many of these were undocumented substitutions that did not incur a polarity shift, resulting in consideration of one (RH1), four (SWS2B), three (RH2C-1), two (RH2C-2) and two (RH2A) sites for λmax prediction (Table   S3). R. aculeatus LWS did not differ from the reference LWS at any known tuning sites.
Predicted R. aculeatus visual pigment peak sensitivities only in part matched the cone and rod sensitivities obtained via MSP  We identified several substitutions that, although not previously investigated via site directed mutagenesis (SDM) and in vitro pigment regeneration, due to their incurred polarity shifts and the site of their occurrence, may explain the discrepancies between MSP and predicted λ max (see Table S3). In RH1, S166A incurred a polarity shift at a site close to the known RH1 tuning site A164S (Chan et al., 1992) and is also a substitution known to be involved in RH2 tuning . R. aculeatus SWS2B showed three polarity changing substitutions, C163F, S166F, and S168A in immediate proximity to the documented tuning site G164A . In both RH2Cs, V185C caused a polarity shift, whereas previously SDM in zebrafish RH2s had shown a -4 nm blue shift for substitution from T to C at this site . C98A incurred a polarity shift at a site adjacent to the powerful RH2 tuning site T97A (-8 nm;. In R. aculeatus RH2A, A151T incurred a polarity shift at a site at which SDM from N to S caused a +4 nm red shift in zebrafish RH2 .

Retinal topography and anatomical spatial resolving power
The topographic distribution of ganglion cells and cone photoreceptors was analysed in four individuals (two individuals per cell type). Since intraspecific variability in topography pattern was very low for both ganglion cells and cone photoreceptors, only one representative map is presented in Figure 4 and individual maps are provided in the supplementary material ( Figure S1).
Ganglion cell topographic distribution revealed the presence of a well-developed horizontal streak with a peak cell density located in the temporal part of the retina (Figure 4) that ranged from 37,161 to 40,125 cells/mm 2 in the two individuals analysed (Table 2).
Based on these peak cell densities, estimated spatial resolving power for R. aculeatus ranged from 6.4 to 7.2 cycles per degree ( Table 2).

Journal of Experimental Biology • Accepted manuscript
Cone photoreceptors in R. aculeatus were arranged in a regular retinal mosaic where one single cone was surrounded by four double cones, resulting in a double cone to single cone ratio of 2:1. This mosaic pattern was consistent across the entire retina, resulting in similar topographic distributions for each cone type. As a result, we only present and describe the distribution pattern for the total cone population (Figure 4). Maps for each cone type are provided in the supplementary material ( Figure S1). The total cone topography varied slightly from the ganglion cell pattern by having a less developed horizontal streak and a more pronounced area temporalis (Figure 4). However, the peak cell density of total cones which ranged from 64,167 to 70,278 cells/mm 2 ( Table 2)  suggested a summation ratio of total cones to ganglion cells in the peak density area of around 2:1.

Discussion
Previous microspectrophotometry (MSP) measurements of R. aculeatus photoreceptors suggested three spectrally distinct cone types: one blue sensitive single cone (λ max = 413 nm), one blue-green sensitive double cone member (λ max = 480 nm) and a second, green sensitive double cone member (λ max = 528 nm), as well as a single spectral type of rod photoreceptor (λ max = 498 nm) . Our transcriptome analysis largelybut not fullyreflects this by revealing the expression of a single rod opsin (RH1) and for the cone opsins, a short-wavelength-sensitive opsin (SWS2B), yet four medium-long-wavelengthsensitive cone opsins (RH2C-1, RH2C-2, RH2A and LWS). Except for one field caught individual, LWS expression was at levels barely high enough to reconstruct this gene's coding sequence, suggesting that measured double cone λ max were primarily (if not entirely) due to

Journal of Experimental Biology • Accepted manuscript
the three identified RH2 opsins. Amino acid-based λ max predictions clarified this picture, indicating that the RH2A visual pigment with a λ max of 526 nm was likely to account fully (i.e., not co-expressed) for the green sensitive double cone member, and this was further supported by our FISH analysis. Predictions also indicated that the two RH2C paralogs were very close in terms of their λ max , and as both showed high expression, either one or both may account for the blue-green sensitive cone member. Unfortunately, due to high mRNA sequence similarity (97.3% identity) our FISH analysis did not allow a distinction between the two expressed RH2C paralogs. Co-expression of either of the two RH2C paralogs with RH2A, in theory, could have explained the discrepancy between predicted RH2C λ max and measured λ max of the blue-green sensitive MWS double cone, a pattern seen for example in some freshwater cichlids (Dalton et al., 2014;Torres-Dowdall et al., 2017). However, our FISH analysis did not show evidence of RH2A/RH2C co-expression. Instead, it showed these opsins expressed in opposite double cone members, a pattern also observed in other reef fish such as some anemonefishes (Stieb et al., 2019). Taken together, the transcriptome data, visual pigment λ max predictions, and FISH analysis corroborate the existing, MSP-based picture of the Picasso triggerfish photoreceptor spectral sensitivity repertoire.
However, although the number of spectrally distinct photoreceptors might be limited, differences in RH2 expression between individuals that were sampled in the field and the ones that were kept in aquaria shows that opsin gene expression in adult R. aculeatus is plastic over time. As for some freshwater fishes (e.g., cichlids; Nandamuri et al., 2017) and reef fishes (e.g., cardinalfishes and damselfishes; Luehrmann et al., 2018) the change in opsin gene expression in R. aculeatus was likely caused by differences in the light environment between the reef and the aquarium. The higher expression of RH2A in aquarium held fishes suggests an overall quantitative shift to longer wavelength sensitivity i.e., more longwavelength-sensitive cones when kept under longer-shifted green/red-dominated LED lights.
While it remains to be investigated whether a change in opsin gene expression truly causes a change in visual discrimination, the findings do highlight the importance of assessing an animal's visual systemfrom molecule to behaviourin its specific light environment.

Prediction of visual pigment maximal absorbance
We do note that the reliability of predicted visual pigment λ max based on amino acid sequences is inherently limited despite tuning sites and their effects having been extensively studied in many taxa, and in fish arguably more so than in any other vertebrate group (for reviews see Takahashi and Ebrey, 2003;Yokoyama, 2000;Yokoyama and Jia, 2020).
Omission of yet unknown tuning sites, effects (if any) of unknown substitutions at known tuning sites, and over-or underestimation of tuning effects due to unknown synergistic effects (including effects that affect visual processing in non-spectral ways, e.g. via consider predictions more robust for more conserved genes (e.g., RH1) or genes in which the extent of λ max variability across many different taxa could be nearly fully described by variations at only a few sites, as is the case for many LWS pigments ('Five-sites rule'; Yokoyama and Radlwimmer, 1998). In contrast, λ max of SWS pigments (SWS1, SWS2As, SWS2B) and RH2 pigments are notoriously difficult to predict due to a plethora of variable sites that have to some extent been implicated with spectral tuning, however complicated by apparent yet not fully understood, synergistic effects (Shi et al., 2001;Yokoyama and Jia, 2020;. In future, additional in vitro expression and pigment regeneration assays, particularly in non-model organism species, as well as further investigation of possible tuning sites via SDM, could improve the reliability of spectral sensitivity predictions.

Retinal topography and anatomical acuity
In this study, ganglion cell topography showed a complete and well-developed horizontal streak and no additional high-density areas outside the streak. The total cone topography had a less developed horizontal streak and a more pronounced area temporalis.
The peak ganglion cell density was found in the temporal retina with an average of 38,643 cells/mm 2 , resulting in an acuity estimate of 6.8 cycles per degree. These estimates fit within

Journal of Experimental Biology • Accepted manuscript
the typical range of shallow water reef fishes, which have an anatomical estimate of visual acuity between 4-27 cpd (Collin and Pettigrew, 1989).
The well-defined streak pattern of R. aculeatus found in this study is very similar to the pattern found for another triggerfish, Balistoides conspicillum (Collin and Pettigrew, 1989). According to the 'terrain theory' (Hughes, 1977), the topography of cells across the retina represents the symmetry of the habitat. Triggerfishes live in open flat areas where they scour the seafloor for potential prey items and have an uninterrupted view of the sand-water horizon; therefore, a horizontal streak may help them scan their environment for potential predators while searching for prey.
For cone densities and distribution, our results were similar to a previous study on R.
aculeatus (Champ et al., 2014); however, they substantially differed for the ganglion cell analysis, especially in density and acuity estimates. Ganglion cell topography in Champ et al. (2014) revealed the presence of a partial horizontal streak that extends from the nasal to the central retina, as well as several other high-density areas outside the streak. A peak cell density of 12,450 cells/mm 2 was found in the temporal retina, resulting in an acuity estimate of 3.4 cycles per degree. These differences in topography pattern, ganglion cell density and acuity estimate between the two studies could be due to several factors such as differences in individual size, the methods of analysis used and/or issues encountered during sample preparation.
In teleost fishes, ganglion cell densities and topography patterns usually show very little intraspecific variability (e.g. de Busserolles et al., 2014b), except between individuals of different life stages and therefore sizes (Shand, 1997;Stieb et al., 2019;Tettamanti et al., 2019). In terms of cell densities, while larger individuals with larger eyes usually have a higher total number of ganglion cells in their retina, the number of ganglion cells per retinal area is usually smaller (Stieb et al., 2019;Tettamanti et al., 2019). According to this, the lower cell densities found in the previous study compared to ours could have been explained if they had used larger individuals. However, Champ et al. (2014) used similar sized individuals or slightly smaller ones compared to our study (7-12 cm vs 10-17 cm SL), indicating that size is not a likely explanation for the differences in densities observed between the two studies.
In both studies, displaced amacrine cells were included into the ganglion cell counts.
Contrary to the present study, Champ et al. (2014) subsequently applied a multiplication factor of 0.76 to all their ganglion cell counts to account for the inclusion of displaced amacrine cells, which could explain some of the discrepancies in densities between the two Journal of Experimental Biology • Accepted manuscript studies. However, even if we take this correction into consideration in our study, our peak ganglion cell density is still at least twice what was reported in the previous study. We chose not to apply a correction factor in this study as displaced amacrine cells in coral reef fishes are mainly present in peripheral and non-specialised areas of the retina, and their inclusion in ganglion cell retinal topographic analyses usually have very little impact on the general topography pattern or the peak cell density estimation (Collin and Pettigrew, 1988).
Consequently, even though the inclusion of amacrine cells in our study might slightly overestimate the peak density of ganglion cells and therefore acuity, we believe it provides a more accurate estimate than applying a general correction factor.
The most likely explanation for the differences in ganglion cell densities and topography pattern observed between the two studies is sample processing. During the preparation of our retinal wholemounts we found that R. aculeatus had very thick vitreous that proves challenging to remove, especially in the centre of the retina along the retinal meridian. During several attempts, which were not included in this study, leftover vitreous resulted in very weak ganglion cell staining making counting unreliable or even unfeasible in certain areas. Similar challenges might have been encountered in Champ et al. (2014) resulting in an under-sampling of the ganglion cell population along the streak. This could explain why an incomplete streak and patchier topography pattern was found in the previous study, as well as why relatively similar ganglion cell densities (i.e. ~ 5,000 cells/mm 2 ) were found in the dorsal and ventral part of the retina in the two studies, but much lower densities were found along the streak in the previous study.

Behavioural measurements of achromatic and chromatic acuity
For achromatic gratings, we found that Rhinecanthus aculeatus had a behavioural acuity of 3.9 cpd, which was significantly lower than anatomical measurements of 6.8 cpd.
Matthiessen's ratio (2.55), which is an average value from a measure range of 2.4-2.82, is used to estimate focal length instead of using the true focal length of the fish and therefore may cause discrepancies in estimations. Second, visual acuity estimates based on ganglion cell counts only represent a theoretical upper limit of visual acuity. This is because not all ganglion cells contribute to visual acuity. For example, in primates, spatial resolving power is set by the midget ganglion cells, which only account for 70-80% of the total ganglion cell Journal of Experimental Biology • Accepted manuscript population (Wassle, 2004). Furthermore, the function of different types of retinal ganglion cells varies both temporally and spatially across the retina in zebrafish (Zhou et al., 2020).
Behavioural studies of visual acuity are likely to represent a more accurate estimate of an animal's functional visual abilities than its anatomical visual acuity. Furthermore, experiments that utilise ecologically relevant paradigms to measure acuity should give closer estimates between behavioural and anatomical measured thresholds. For example, Temple et al. (2013) found that the maximum acuity of archerfish was similar to anatomical measurements when fish spat at prey using the area centralis in the ventro-temporal region of the retina. In contrast, optomotor/optokinetic tests provide some measure of average retinal acuity but will not capture the acuity in specialised regions of the retina, such as in the area centralis or fovea. Behavioural measurements of acuity may also vary depending on the stimuli used. Triggerfish did not perform as well in acuity tests when trained to circular stimuli as opposed to grating stimuli of horizontal and vertical stripes (Champ et al., 2013).
Furthermore, different acuities were measured when honeybees were presented with radial (sectored) compared to linear (square-wave) gratings (Srinivasan and Lehrer, 1988), which could be explained by orientation specific feature detecting receptive fields (Marr & Hildreth 1980, Marr 2010. We found that achromatic acuity (3.94 cpd) of triggerfish was significantly higher than chromatic acuity for both the green-yellow (1.71 cpd) or pink-purple (1.89 cpd) stimuli.
This supports previous findings in birds, mammals, and bees (Giurfa et al., 1997;Lind and Kelber, 2011;Mullen, 1985); however, in humans and birds, differences in acuity were found between different colour channels. In budgerigars, visual acuity was lower for blue-green than for red-green gratings, which may be explained by lower numbers of SWS cones in the retina compared to MWS and LWS cones (Hart et al., 2000). Similarly in humans, acuity was lower for blue-yellow contrasts than for red-green (Mullen, 1985). However, the black-white achromatic stimulus had higher overall luminance compared to the chromatic stimuli based on double cone quantum catch; this may have also improved acuity for achromatic stimuli.
In conclusion, we hope that data from this study will inform future studies with this species investigating a range of topics including, but not limited to, teleost perception, navigation, and cognition. We believe that Picasso triggerfish will continue to play a key role in contemporary research in these areas due to their versatility as a focal study organism. We highlight the need for further information on the specific chromatic opponent mechanisms in animals, which is currently poorly understood (Baden and Osorio, 2019).  Table S6.   Fig. S2. i) Spectral reflectance of laminated stimuli used in each treatment labelled by color: pink, purple, green and yellow; irradiance of sidewelling lights measured in the testing tanks shown as dashed line; ii) Spectral sensitivities of the three cones in triggerfish, Rhinecanthus aculeatus, with a single cone sensitivity of λmax = 413 nm, and double cone sensitivities of λmax = 480 nm and λ = 528max nm multiplied by yellow cornea ; iii) Experimental tank setup; iv) An example of the achromatic (black and white) squarewave gratings: a) 1 cycle per degree; b) 5 cycles per degree.    -Table S4. Summary of stimuli presented to each fish and in which order, whether the fish was trained to receive a food reward from the Distractor (11 cpd) or the test gratings (0.5-6 cpd), the total number of trials conducted by each fish for each colour (total number of trials = 2438) and calculated discrimination thresholds at 62% correct choice. NA indicates not tested due to time taken to complete treatment 1.