General explanations of colour change trials
All experimental procedures were approved by the Mokpo National University Institutional Animal Care & Use Committee (approval number: MNU-IACUC-2019-008). All methods were performed in accordance with the guidelines for the use of animals in research and teaching29. We used 32 locally-purchased D. japonicus that were originally collected near Yeonggwang, South Korea (N35.26°, E126.52°) for the experiment in 2019. These frogs were used for all experiments except for the individuality experiment. We purchased an additional 20 individuals for the individuality experiment performed in 2021. Each frog was kept independently in a plastic cage under 25°C with a photoperiod of 12L:12D. We provided water and food (calcium and vitamin D-powdered mealworms and juvenile crickets) ad libitum. All colour change trials were conducted in a temperature/humidity chamber (HB-303DH, Hanbaek, South Korea) between 0900 and 1800. We maintained 70% humidity in the chamber during the trials (the temperature depended on the experiment type; see below). Fluorescent lamps were attached along the inner side of the chamber. Before conducting all experiments, we confirmed that the intensity of the light illuminated into each experimental container (see below) was similar using a photometer.
During the colour change trials, each frog was kept in a cylindrical container (experimental arena; 14.5 cm radius × 7 cm height) and remained in the chamber for 2 h. Two hours has been shown to be sufficient for D. japonicus to complete a colour change process17. We covered the inside of the arena using one of the grey-toned papers we created (see below), and a lid covered the top of the arena. The lid was transparent, with wire mesh in the centre area that allowed airflow. After spending 2 h in the chamber, we immediately moved each frog to the photographic zone and photographed them. We placed a cylindrical container (same shape as the container used for the colour change trial, with the inside covered by grey-toned papers) in the centre of the photographic zone, in which we placed each frog to photograph its dorsum. Two bulbs illuminated the photographic zone (True-Light 23W 5500K, True-light International, Frankfurter), which were placed approximately 80 cm above the container. For near-infrared photography, we placed an additional bulb above the container (Halogen work light 400W, Korea Electric, Bucheon). All photography was performed using a full-spectrum converted Nikon D7000 camera (Lifepixel, Mukilteo, WA; transmission range up to 1050 nm) equipped with a Jenoptik UV-VIS-IR 60 mm Apo Macro Lens (transmission waveband: 290–1500 nm). We attached either a Baader UV/IR cut filter for visible range (400–680 nm) or an IR-pass filter for near-infrared range (670–1050 nm) photography. We did not photograph frogs in the ultraviolet range because ultraviolet reflection is negligible in D. japonicus17. A 99% reflectance standard (WS-1-SL, Ocean Optics, Dunedin, FL) was placed next to the container and photographed together for image calibration. We saved all images in raw format. After completing all experiments, we released the frogs in the location where they were initially collected.
Colour change in response to gradual changes in external conditions
In this experiment, we used various background lightness and temperature conditions to examine the frogs’ response to gradually changing external conditions. For background lightness manipulation, we used eight different grey colours. To create these colours, we first printed white (non-printed) and black colours (R = G = B = 0) using Xerox Premium Nevertear Paper and photographed both papers alongside a 99% reflectance standard in raw format. Then, we converted the images to linearised reflectance TIFF images by removing the non-linearity of the images (using DcRaw v. 9.2730) and rescaling each colour channel value so that the 99% reflectance standard had the corresponding R, G, and B values31. After these processes, the R, G, and B values (0–255) in each image correspond to the reflectance (0–100%). Then, we measured the average reflectance (the average of the R, G, and B values) of the white (non-printed paper; 80% mean reflectance) and black-printed papers (4% mean reflectance) using ImageJ 1.53e (opensource program, National Institute of Health). Afterwards, we exhaustively searched for grey colours (equal R, G, and B values) with mean reflectance from 10 to 80% for every 10% after printing. Therefore, we created nine different grey-toned backgrounds with average reflectance values of 4, 10, 20, 30, 40, 50, 60, 70, and 80% (± 1%). We used each printed paper to cover the inside of the experimental arena and performed the colour change trials. Each frog experienced all nine background colours in a random order. We maintained a 25°C temperature during the colour change trials.
For temperature manipulation, we used the same procedure under various temperature conditions: 5, 10, 15, 20, and 25°C. Each frog experienced all five temperature conditions in a random order. We did not test below 5 nor above 25°C for ethical reasons; exposing frogs to these extreme conditions for 2 h without any food or water could cause harm. The tested temperature conditions were within the range that D. japonicus experiences in their natural habitats during the active seasons. We used non-printed papers (white) for the background colour because preliminary tests showed that colour change in response to temperature was more prominent under a white than black background.
The interactive role of background lightness and temperature
In this experiment, we performed a 2 × 2 factorial experiment using background colour and temperature to examine (i) frogs’ responses when the demand for camouflage and thermoregulation conflict and (ii) the change in near-infrared reflectance. We used white and black colours for the background treatments and 5 and 25°C for the temperature conditions. Each frog experienced all four colour-temperature combinations in a random order. When photographing each frog’s dorsum, we also photographed the frog under the near-infrared range (see above for details concerning the near-infrared photography setup).
Individual consistency experiment
To test whether the colour change capacity of individual frogs persists over time (i.e., whether inter-individual differences in colour change capacity reflect individuality), we purchased an additional 20 frogs in February 2021; we used new frogs because those used for the above experiments had been released in 2020. We housed the frogs under 25°C and 12L:12D conditions.
In this experiment, we primarily focused on their capacity to change colour in response to background lightness because frogs showed higher plasticity in response to changes in background lightness compared to temperature. The colour change capacity was defined as the difference in dorsal lightness when against black and white backgrounds. We tested each frog’s colour change capacity seven times on different days: the day we started the experiment (day 0), one day later, and 1, 2, 4, 9, and 17 weeks later. On each experimental day, we exposed each frog to two different background conditions (black and white) and photographed them after two hours. We maintained a 25°C temperature during the trials. Colour capacity was calculated as the dorsal lightness of the frogs against the white background subtracted by the dorsal lightness against the black background (see Image analysis section for details about the measurement of dorsal lightness).
Image analysis
We first converted all raw images into linearised TIFF images using Dcraw 9.27. Then, we generated 8-bit linear reflectance images using the 99% reflectance standard as a reference by scaling the R, G, and B values of each colour channel. After this process, each pixel’s R, G, and B values represented the reflectance ranges from 0 to 255, corresponding to a reflectance of 0–100%. Using the linear reflectance images, we selected three random dorsal areas (but only outside the dark-patterned area) on the frog’s dorsum and measured the median R, G, and B values. Then, we calculated the median dorsal reflectance (hereafter referred to as the dorsal lightness) as the average of the measured R, G, and B values divided by 2.55. We also calculated the hue and chroma according to the methodology of Smith et al.24. However, hue changes were negligible, and lightness change largely predicted chroma change17; thus, we focused our analysis on lightness. Our primary interest was to compare the extent of dorsal colour change in frogs, not how predators or conspecifics perceive frogs’ dorsal colour; therefore, we did not assume any specific receiver visual systems32.
To measure dorsal reflectance in the near-infrared region, we first converted the raw near-infrared images into 8-bit linear reflectance TIFF images using the procedure described above. Then, we used only R channel pixel values for the analysis (after dividing them by 2.55), which showed the most sensitive responses in the near-infrared range.
Data analysis
To compare dorsal lightness across various grey-toned backgrounds with different reflectances, we used linear mixed models (LMMs) with dorsal lightness as the response variable and background reflectance as a discrete predictor. Frog ID was used as a random factor in all mixed model analyses. Similarly, to compare dorsal lightness across the various temperature conditions, we used LMMs with dorsal lightness as the response variable and temperature condition as a discrete predictor. For temperature data, there was a noticeable linear and gradual change in frog dorsal reflectance as temperature increased; therefore, we additionally fitted another LMM with dorsal lightness as the response variable and temperature condition as a continuous predictor. For post-hoc comparisons of the LMMs, we used the Tukey’s multiple pairwise comparisons implemented in the “glht” function of the “multcomp” package33. We controlled for false discovery rates whenever multiple comparisons were conducted34.
We analysed visible-range colour and near-infrared reflectance separately for the interactive role of colour and temperature experiment. For the visible-range analysis, we fitted LMMs with visible-range dorsal lightness as the response variable, and the two treatments (background colour and temperature) as well as their interaction as predictors. For the near-infrared range analysis, we fitted additional LMMs with the same predictors, but with near-infrared range lightness as the response variable. In addition, because near-infrared lightness correlated highly with visible-range lightness26,35, we also analysed whether any “special” adaptations existed in near-infrared range lightness after accounting for their high correlation with visible-range lightness. To do this, we fitted another LMM using near-infrared lightness as the response variable, colour and temperature treatment and their interaction as predictors, and visible-range lightness as a covariate. When there were multiple predictors, we used the information theoretic approach for model inference36. We first generated all candidate models and ranked each based on the Akaike Information Criterion corrected for small sample sizes (AICc). Then, we selected models with ΔAICc < 4 and conditionally averaged (i.e., averaged over the models where the parameter appeared) the parameters of all selected models and estimated P-values based on the Wald z statistic.
Using the same data, we also tested whether frogs with higher colour capacity in response to background lightness also showed higher colour capacity in response to temperature. We calculated each frog’s colour change capacity in response to both background (the lightness difference between when they were against black and white backgrounds) and temperature (the difference between when they were under 5 and 25°C conditions). Then, we performed a Pearson’s correlation test on the two variables.
To analyse individual consistency in colour change capacity, we estimated intraclass correlation coefficients (ICCs)37. ICCs are a type of correlation, but operate on data structured as groups. Here, we tested whether each frog’s colour change capacity was similar across different trials (ICC for absolute agreement). P-values and confidence intervals were estimated using a F-test. Additionally, we fitted LMMs using each frog’s colour change capacity as a response, trial as a discrete predictor, and frog ID as a random factor; we performed this additional analysis to (1) extract the amount of variance explained by the random factor (individual frog ID) and (2) test whether there was a directional trend in frogs’ colour change capacity across trials. We present our results with a gradual notion of evidence38.