Skip to main content
Dryad

Ultraviolet vision in anemonefish improves color discrimination

Cite this dataset

Mitchell, Laurie et al. (2024). Ultraviolet vision in anemonefish improves color discrimination [Dataset]. Dryad. https://doi.org/10.5061/dryad.wwpzgmsp9

Abstract

In many animals, ultraviolet (UV) vision guides navigation, foraging, and communication, but few studies have addressed the contribution of UV vision to color discrimination, or behaviorally assessed UV discrimination thresholds. Here, we tested UV-color vision in an anemonefish (Amphiprion ocellaris) using a novel five-channel (RGB-V-UV) LED display designed to test UV perception. We first determined that the maximal sensitivity of the A. ocellaris UV cone was at ~386 nm using microspectrophotometry. Three additional cone spectral sensitivities had maxima at ~497, 515, and ~535 nm, which together informed the modelling of the fish’s color vision. Anemonefish behavioral discrimination thresholds for nine sets of colors were determined from their ability to distinguish a colored target pixel from grey distractor pixels of varying intensity. We found that A. ocellaris used all four cones to process color information and is therefore tetrachromatic, and fish were better at discriminating colors (i.e., color discrimination thresholds were lower, or more acute) when targets had UV chromatic contrast elicited by greater stimulation of the UV cone relative to other cone types. These findings imply that a UV component of color signals and cues improves their detectability, which likely increases the salience of anemonefish body patterns used in communication and the silhouette of zooplankton prey.

README: Anemonefish have finer color discrimination in the ultraviolet

S2 Data.csv - Tables containing the absolute cone quantum catches (Q) calculated for both target colours and grey distractors (in separate tabs). 'u' = ultraviolet cone, 'm1' = medium-wavelength-sensitive cone 1, 'm2' = medium-wavelength-sensitive cone 2, and 'l' = long-wavelength-sensitive cone. Relative LED intensities are also provided for each colour and grey.

S3 Data.csv - Individual pre-bleach MSP measurements of photoreceptor spectral absorbance, and lens spectral transmission measurements in A. ocellaris. Photoreceptor light absorbance is reported across the wavelength range of 300 - 700nm. The title of each column/scan is given as the peak wavelength absorbance value as determined for the individual scan, where a number contained in brackets indicates whether >1 measurement of the same peak wavelength sensitivity was obtained. Best fit (A1/A2 based) visual pigment absorbance curve templates are also provided as determined by regression analyses using mean photoreceptor spectral absorbance values. Note, all are A1 based except for the U cone which best fit an A2 based template. R2 values from fitting the visual pigment templates can be found in the Supplementary Material.

blue Ishihara task.mkv - video clip showing an anemonefish successfully complete the trained task of pecking a target (blue in this example) against a background of grey distractors. The fish then moves out-of-frame to receive a food reward on the other side of the tank that becomes divided. After refreshing the display and randomizing the target position the next trial commences.

R scripts and associated data

Visual modelling.R - R file containing script to run visual modelling of anemonefish colour vision, including the receptor noise limited model which returns chromatic distances for target colours relative to an average grey distractor. Code is also included to generate plots presented in Figures 1 and 2.

Psychometric curve fitting.R - An R file containing script to fit and plot (in Figure 2A) individual psychometric response curves and thresholds for colour discrimination in anemonefish. Additional lines of code perform fitting of secondary/post-experiment psychometric curves (and thresholds) and plots both the preliminary and secondary curves for retested fish, as seen in Supplementary Figure 5.

Mixed effects modelling and plotting.R - R file containing script to run linear mixed effects model and generalised linear mixed-effects model analyses reported in the manuscript. Code is also included to run plots presented in Figure 3 and Supplementary Figure 7.

choice_data.csv - Behavioural dataset for colour discrimination experiment.

  • 'JND_TETRASPACE' = delta S value or colour distance of the target away from the average grey distractor.
  • 'SUCCESS' = binary score for whether a fish was successful '1' or unsuccessful '0' in making a correct choice during the trial.
  • 'INCORRECT' = the number of incorrect choices made by a fish during a trial, that ranged from 0 to 2 (an unsuccessful trial).
  • 'LATENCY' = the time (in seconds) from passing through the tank divider to pecking the stimulus.

all_mean_thresholds.csv - Dataset containing the mean colour discrimination thresholds calculated per colour set ('COLOUR_LINE') and receptor noise scenario ('NOISE'). Single noise values indicate a global noise value, while two values seperated by an underscore indicate different single cone and double cone noise values, respectively. 'M2/M3' refers to switching the input of the M2 cone with the M3 cone. 0.95 CI are given by their 'lower' and 'upper' bounds.

all_targets_normalised.csv - Normalised spectral radiance of each target colour per colour set.

example spectra.csv - Nominal spectral radiance used in plotting examples of LED emission in Figure 1D.

fish_thresholds.csv - Compiled individual discrimination thresholds for running the linear mixed effects model separated by column indicating the sign direction of UV contrast (negative = 'UV_neg', 'neutral', and positive = 'UV').

illum_ptfe.csv - Spectral irradiance of background ptfe screen of the LED display measured as sidewelling illumination.

ocellaris_specsens1.csv - Four cone spectral sensitivities (U = ultraviolet cone, S = M1 cone, M = M2 cone, l = L cone).

ocellaris_specsens2.csv - Three cone spectral sensitivities (S = M1 cone, M = M2 cone, l = L cone).

ocellaris_specsens3.csv - Three cone spectral sensitivities (S = U cone, M = M2 cone, l = L cone).

ocellaris_specsens4.csv - Three cone spectral sensitivities (S = U cone, M = M1 cone, l = L cone).

ocellaris_specsens5.csv - Three cone spectral sensitivities (S = U cone, M = M1 cone, l = M2 cone).

retest_data.csv - Behavioural dataset, as per 'choice_data.csv' but with the addition of secondary or post-experiment trials used to analyse the effect of experience and test order.

scatterdot_RGB_target.csv - RGB values that supplies a colour key for the customised plotting of colour loci in tetrahedral space (in Figures 1 and 2).

supporting data file 1.xlsx - Contains seven tabs including:

  • individual cone absolute quantum catches (Q) for target stimuli ('non-UV targets' and 'UV targets') and 'Grey distractors' ('relative LED intensities' are given for each colour from the short/UV- to long/red-wavelength channel separated by '_');
  • results for Linear Mixed Effects models 'LMM#1' showing colour discrimination thresholds compared between colour sets with adjusted p-values for multiple comparisons (β = coefficient estimate, SE = standard error, Z = Z-value, p = adjusted p-value) and LMM#2 testing for whether the latency (log-transformed time/seconds) of target detection differed significantly between colour sets and with increasing contrast (ΔS);
  • results for a Generalised Linear Mixed-Effects model 'GLMM output' which tested how each colour set influenced the response of fish to the test (β = coefficient estimate, SE = standard error, Z = Z-value, p = adjusted p-value, secondary p = adjusted p-value from the post-experiment/secondary testing);
  • and 'retest threshold comparison' showing a summary of experimental and post-experimental colour discrimination thresholds for fish which were retested to assess the effect of colour set test order or experience ('thre' = discrimination threshold/ΔS, 'prob' = threshold probability, 'threinf' = lower CI bound, 'thresup' = upper CI bound).

Methods

Lens transmission of A. ocellaris

For the measurement of lens transmission in A. ocellaris, the lenses (n = 3 fish) were isolated from the hemisected eyecup and rinsed in PBS to remove any blood and vitreous. Spectral transmission (300–800 nm) was measured by mounting the lens on a drilled (1.0 mm diameter hole) metal plate between two fibers (50, 100 µm diameters) connected to an Ocean Optics USB4000 spectrometer and a pulsed PX2 xenon light source (Ocean Optics, USA). Light spectra were normalized to the peak transmission value at 700 nm, and lens transmission values were taken at the wavelength at which 50% of the maximal transmittance (T50) was attained. No pigmented ocular media was observed.

Photoreceptor spectral sensitivities of A. ocellaris

The spectral absorbance of A. ocellaris photoreceptors was measured using single-beam wavelength scanning microspectrophotometry (MSP). In summary, small pieces (~1 mm2) of tissue were excised from the eyes of two-hour dark-adapted fish, then immersed in a drop of 6% sucrose (1X) PBS solution and viewed on a cover slide (sealed with a coverslip) under a dissection microscope fitted with an infra-red (IR) image converter. A dark scan was first taken to control for inherent dark noise of the machine and a baseline scan measured light transmission in a vacant space free of retinal tissue. Pre-bleach absorbance measurements were then taken by aligning the outer segment of a photoreceptor with the path of an IR measuring beam that scanned light transmittance over a wavelength range of 300–800 nm. Post-bleach scans were then taken after exposing the photoreceptor to bright white light for 60 seconds and then compared to pre-bleach scans to confirm the presence of a labile visual pigment. Confirmed visual pigment spectral absorbance data was then analyzed using least squares regression that fitted absorbance data between 30% and 70% of the normalized maximum absorbance at wavelengths that fell on the long-wavelength limb. The wavelength at 50% absorbance was then used to estimate the maximum absorbance (λmax) value of the visual pigment by fitting bovine rhodopsin as a visual pigment template. This absorbance curve fitting was performed in a custom (Microsoft Excel) spreadsheet, where the quality of fit of absorbance spectra between A1- and A2-based visual pigment templates was also visually compared. Individual scans were binned on their grouping of similar (≤10 nm difference) λmax values, and then averaged and reanalyzed across fish to create mean absorbance spectra.

Color selection and stimuli design

To estimate anemonefish photoreceptor excitation for target and distractor colors, receptor quantum catches ‘q’, were first calculated for each stimulus, ‘S’ (i.e., target and distractor radiance spectra in µM/cm-2/s-1/nm) viewed under well-lit conditions and integrated over 300 to 700nm given by:

𝑞𝑖=𝑘𝑖 ∫𝑅𝑖(𝜆)𝑆(𝜆) 𝑑𝜆, (1)

where k is a scaling coefficient for receptor adaption to the background ambient light, Sb:

𝑘𝑖= 1/∫𝑅𝑖(𝜆)𝑆𝑏(𝜆) 𝑑𝜆. (2)

Ri(λ) was the normalized spectral absorbance of a given receptor type ‘i’ (i = U, M1, M2, L) multiplied by lens transmittance, and ‘λ’ denoted wavelength (nm). Sb(λ) was the spectral radiance of the PTFE display screen (between the pixels) with all LEDs turned off and measured from 5.0 cm in the experimental tank. This approach allowed for modelling spectral emission (from LEDs) rather than more commonly calculated for reflectance. Integration was performed across the visible spectrum (i.e., 300–700 nm for A. ocellaris). Relative cone quantum catches were used to plot color loci in a tetrahedral color space.

The contrast (Δqi) for each receptor channel was calculated by,

Δ𝑞𝑖=ln 𝑞𝑖𝑡𝑎𝑟𝑔𝑒𝑡 / 𝑞𝑖𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑑𝑖𝑠𝑡𝑟𝑎𝑐𝑡𝑜𝑟 (3).

In the absence of direct noise measurements for A. ocellaris cones, we estimated cone 516 noise levels (ei) by,

𝑒𝑖=√𝜎 / 𝜂𝑖 (4),

where ‘σ’, the numerator of the Weber fraction, and ‘η’ is the ratio of the given cone type. Based on the regular mosaic of one single cone surrounded by four double cones in the A. ocellaris retina, we used a relative cone abundance ratio of 1 : 2 : 1 : 1 (U : M1 : M2 : L) for a tetrachromatic visual system and 1 : 2 : 2 for a trichromatic visual system. 

ΔS in tetrachromatic visual space was calculated by:

Δ𝑆= (𝑒1𝑒2)2(Δ𝑞4−Δ𝑞3)2+(𝑒1𝑒3)2(Δ𝑞4−Δ𝑞2)2+(𝑒1𝑒4)2(Δ𝑞3−Δ𝑞2)2+(𝑒2𝑒3)2(Δ𝑞4−Δ𝑞1)2+(𝑒2𝑒4)2(Δ𝑞3−Δ𝑞1)2+(𝑒3𝑒4)2(Δ𝑞2−Δ𝑞1)2 / (𝑒1𝑒2𝑒3)2+(𝑒1𝑒2𝑒4)2+(𝑒1𝑒3𝑒4)2+(𝑒2𝑒3𝑒4)2      (4),

and in trichromatic visual space was calculated by:

(Δ𝑆)2=𝑒21 (Δ𝑞3− Δ𝑞2)2+ 𝑒22 (Δ𝑞3− Δ𝑞1)2+ 𝑒2 3 (Δ𝑞1− Δ𝑞2)2 / (𝑒1 𝑒2)2+ (𝑒1 𝑒3)2+ (𝑒2 𝑒3)2    (5).

Grey distractor spectra (N=13) were chosen to be <1 ΔS of the achromatic point of A. ocellaris and ranged between 0.3 ΔS to 0.8 ΔS of each other. To control for the potential use of achromatic (intensity) cues when discriminating targets, we selected 6 to 10 distractor greys (from the 13) per stimulus based on all four-cone quantum catches to encompass the highest and lowest target intensities.

Alternative models calculated ΔS values using more-conservative receptor σ-values ranging from 0.05 to 0.15, to assess their fit with A. ocellaris behavioral thresholds. Lower single cone noise (σ = 0.04–0.11) than double cones (σ = 0.14) was also modelled in case of different inherent noise levels. Threshold predictions were also compared between models of trichromat and tetrachromat vision in A. ocellaris, in case this could reveal any information on the contribution of double cones to color vision. The closest model fit was determined based on which had the smallest mean difference summed across all color lines from 1 ΔS.

Training and experiment

During both training and the experiment, the LED display was presented in a section of the aquarium separated by a sliding, opaque door. This door was closed to keep fish from viewing the display while the stimulus was updated between trials, and only upon trial commencement was the door raised to allow fish to view and interact with the display. For both training and testing, a morning (09:00–11:00) and afternoon (14:00–16:00) session were run, in which fish completed between 10 to 12 trials per day.

Fish were initially enticed to peck the LED display by presenting a pseudo-randomly chosen high-contrast pixel (blue, green, red, or UV) with a small piece of prawn meat smeared on it. Over a week, we gradually reduced the size of the smeared food and transitioned towards a food reward (Formula One Ocean Nutrition pellets) delivered by forceps when fish pecked the single target pixel. Once anemonefish readily approached and pecked at the display without enticement, we introduced the grey distractor pixels alongside the target pixel. Fish were only rewarded when they correctly chose/pecked the target color within 60 seconds. They were deemed to have reached the training criteria for the discrimination task after maintaining a correct choice probability of 0.75 over five consecutive sessions. 11 anemonefish met this criterion (mean number of training trials ± sd = 8.0 ± 4) and underwent experimental testing.

For testing, like training, fish were only rewarded for pecking the target pixel. Trials were terminated if fish made more than one incorrect choice or exceeded 60 seconds, upon which fish were returned to behind the divider (starting position) without reward. Note, because of the numerosity of pixels (n=38) per stimulus and the potential for distractions, each fish was permitted to make up to one incorrect choice per trial. For each trial, we recorded whether fish made a correct or incorrect choice, time (seconds) after fish entered through the door till target detection (i.e., latency), tested color set, and target ΔS.

Each color set was tested using five or six individual anemonefish that completed a minimum of eight trials per target color per assigned set (mean ± sd = 10 ± 1.0). Fish were divided into two groups assigned different color sets, including: 1) Fish IDs 19, 20, 33, 34, and 36 which were assessed in order of testing with green, UV, purple, and UV-red, and 2) Fish IDs 21, 22, 24, 31, 32, and 35 which were assessed in order of testing with blue, UV-blue, violet-green, red, and orange.

Between each trial the target pixel contrast was pseudo-randomly assigned from a list of LED intensity values for each color set. Throughout the experiment, we included control trials (n=10) to ensure that no other cues were created by the controller or code when choosing the target pixel, this determined the random chance of fish making a correct choice by displaying a target pixel of zero contrast (i.e., grey). In none of the control trials did fish correctly peck the control target. 

To verify that differences in discrimination thresholds were not influenced by the order in which each of the color sets was tested, we reassessed each of the nine sets at the end of the experiment using two anemonefish from each group. Behavioral thresholds and psychometric functions from this secondary assessment were then compared with the primary assessment. Although we found evidence that experience effects had influenced the shape or incline of the psychometric function for some color sets (e.g., UV, blue, green, and UV-blue), there was none indicating that experience had contributed to differences in color discrimination thresholds that remained unchanged in the reassessment. The direction and size of differences among color discrimination thresholds did not vary systematically over the course of the study.

Usage notes

All statistical analyses and color modelling were conducted using the statistical program R (v. 4.0.2). Color distances were calculated using the RNL model and plotted in a tetrahedral space using the package ‘PAVO 2’. Discrimination thresholds were determined by the point at which fish had a 0.5 probability of making a correct choice approximately at the inflection or steepest point of a sigmoid curve fitted to the behavioral data using the package ‘quickpsy’.

The effect of color sets on anemonefish discrimination thresholds was assessed using a linear mixed-effects model (LMM) run using function ‘lmer’ in the package ‘lme4’. Individual threshold ΔS value was treated as the response variable, color sets as the fixed factor, and fish ID was the random effect. A post-hoc, pair-wise analysis controlled for multiple comparisons of threshold ΔS values across all possible combinations of color sets using Bonferroni adjustment (p.adjust, R base package ‘stats’). Another set of LMMs tested the effect of color on the duration (latency) of trials, where trial latency (in seconds) was the response variable, color set, target ΔS and the first-order interaction between color set and target ΔS were fixed effects, and fish ID was a random effect. Nine models were run (one per color set), where the intercept was assigned to a specific color to enable pairwise comparisons with the others.

To test whether the color set influenced the relationship between ΔS and the proportion of correct choices (i.e., did color influence how fish responded to the test and resulting shape of the psychometric curve), a generalised linear mixed effects model (GLMM) was run using R package 'lme4'. Variables included anemonefish choice (0 = incorrect, 1 = correct) used as a binomial response variable, ΔS, color set and the first-order interaction between the two variables treated as fixed factors, and Fish ID entered as a random effect. Model p-values were corrected for multiple comparisons via Bonferroni adjustment (‘p.adjust’, base R package ‘Stats’).

Funding

Australian Research Council, Award: DP18012363

Australian Research Council, Award: FT190100313

Australian Research Council, Award: DE200100620