Consuming your colors: carotenoid and retinol availability affects expression of color related genes in the mimic poison frog (Ranitomeya imitator) Authors: Andrew O. Rubio, Mallory de Araujo Miles, Ashley M. Dye, Adam M. M. Stuckert, Kyle Summers Published: 2025 ABSTRACT BackgroundColor and patterns in animals are closely tied to reproduction and survival, and have long been of interest to evolutionary biologists. For instance, bright conspicuous coloration plays a vital role in aposematic organisms as these phenotypes help communicate to potential predators that they are chemically defended and therefore unpalatable. Although the field has made significant progress in identifying genes associated with the production of conspicuous coloration, our knowledge of the genes associated with carotenoid metabolism and their roles in pigmentation is incomplete. Here we attempt to fill this knowledge gap by investigating how dietary carotenoid and retinol availability affects levels of gene expression and color formation in two color morphs of the mimic poison frog (Ranitomeya imitator), the orange-banded Sauce morph and the red-headed Varadero morph which exhibit bright orange and red coloration, respectively.ResultsWe found that froglets raised under a carotenoid/retinol-enriched diet showed elevated expression levels of genes associated with the retinol pathway (e.g., dhrs3, pnpla3) whereas froglets raised under a carotenoid/retinol-deficient diet showed elevated expression levels of genes associated with the retinoic acid pathway (e.g., aldh1l1, aldh6a1, cyp7a1). In addition, we found that froglets raised on a carotenoid/retinol-enriched diet displayed brighter skin and higher ratio of bright colored to black skin patches when compared to froglets raised on a carotenoid/retinol-deficient diet in both the Sauce and Varadero morph.ConclusionsThe genetic basis of carotenoid metabolism and its role in pigmentation remain poorly understood in amphibians. Our results provide insights into the key genes associated with carotenoid/retinol metabolism and their roles in pigmentation which should be investigated further and experimentally manipulated.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12864-025-12381-4. Background The vast array of coloration seen in the animal kingdom has intrigued evolutionary biologists because such phenotypes have evolved as critical components of key ecological and evolutionary processes, including species and morph recognition, mate choice, intraspecific aggression, and predator avoidance [1–5]. Natural and sexual selection play key roles in driving coloration and color phenotypes and are frequently under selection pressure from both abiotic and biotic factors [4, 6]. Evolutionary biologists have long studied the mechanisms behind these phenotypes, given the critical role of coloration in survival and reproduction. In vertebrates, coloration is linked to chromatophores, which are specialized organelles such as melanophores, iridophores, and xanthophores [7]. Melanophores contain the pigment melanin, which is responsible for the production of black and brown coloration, and contribute to green coloration [8, 9]. The iridophores are commonly characterized by structural elements, such as guanine platelets. The size, shape, and orientation of these elements can have substantial effects on the wavelength of light reflected and hence on coloration [10–12]. For instance, thin platelets reflect short wavelengths, resulting in dark blue coloration, while thick platelets reflect longer wavelengths, leading to the appearance of orange coloration [13]. Pigments such as pteridines and carotenoids are found within xanthophores and are largely responsible for the production of yellow, orange, and red coloration [8]. Twomey et al. [13] found that differences in brightness and hue in the red and orange coloration of Ranitomeya poison frogs can be attributed to the combined effects that various pigments and structural elements (such as guanine platelets) have on spectral reflectance. In animals, pteridine pigments are actively produced de novo as they are synthesized during the production of purines [10]. The synthesis of guanine, a purine base found in DNA and RNA, is a fundamental intrinsic physiological process, and therefore, pteridines are continuously replenished. However, carotenoid pigments cannot be produced by animals and can only be acquired through their diet. Carotenoids, such as beta-carotene, lycopene, lutein, and zeaxanthin, can only be synthesized by autotrophs (e.g., plants, algae, and photosynthetic bacteria) and therefore, animals must accrue them through their diet [14, 15]. In animals, Carotenoids and retinol are crucial for many developmental and physiological processes, including developmental differentiation, immunity, reproduction, vision, and pigmentation [16, 17]. Carotenoids are transported by lipoproteins in the plasma, and distributed to various tissues as needed [18]. Carotenoids like beta-carotene can be converted to retinal, which can then be further converted to retinoic acid. Retinol (vitamin A), is crucial for numerous aspects of metabolism and immunity. Vitamin A can also be converted to retinal, and then to retinoic acid, as needed for development. The retinoic acid pathway is a key pathway for basic developmental processes, such as the formation of the hindbrain, spinal cord, skeleton, heart, eye, and pancreas [19, 20]. Another key pathway involves the conversion of retinol to retinyl ester for storage, which is crucial for many metabolic processes and is important given that excess vitamin A is toxic [21, 22]. A comprehensive overview of these metabolic pathways, including schematic diagrams of retinoid interconversion and storage, is provided in Dewett et al. [23]. Numerous studies indicate that provitamin A promotes the development and maintenance of bright coloration (e.g., red, orange) [24, 25]. For instance, variation in plumage coloration in captive male house finches reflects the type and quantity of carotenoids ingested, such that males had bright red plumage when fed a carotenoid-based diet as opposed to a carotenoid-deficient diet [26]. This research led some scientists to propose that carotenoid-based coloration could represent an example of an honest signal of genetic quality [27]. However, recently this hypothesis has been modified in light of new data on the biochemistry of carotenoid processing, suggesting that carotenoid coloration is a condition dependent trait from the moment it emerges in a population and remains condition dependent even in the absence of natural or sexual pressure [28]. Within the past decade, a notable amount of effort has been put into uncovering the molecular mechanisms producing the array of colors seen in the animal kingdom [29, 30], however, many of these studies have been focused on melanin-based coloration [31, 32]. In particular, numerous studies have attempted to pinpoint the molecular mechanisms responsible for aposematic coloration in poison frogs in the family Dendrobatidae [13, 24, 33–38], which is associated with signalling unpalatability to predators. These studies have identified key candidate genes associated with the production of aposematic coloration, paving the way for future studies to test hypotheses of the evolution of color pattern across aposematic vertebrates. In addition, several studies identified color related genes that have evolved under positive selection [35, 36, 39]. Although we have made considerable progress in identifying genes associated with color development in poison frogs, very little is known about the genes responsible for carotenoid metabolism and their roles in pigmentation. In this study, we investigated the effects of dietary carotenoid/retinol availability on levels of gene expression in the liver and dorsal skin from two captive bred color morphs of the mimic poison frog (Ranitomeya imitator), the Sauce (orange-banded) and Varadero morph (red-striped heads with reticulated blue legs) morphs (Fig. 1). The rationale for investigating carotenoid/retinol availability is that R. imitator tadpoles raised in the field are likely to have access to both carotenoids (from algae in phytotelmata) and to retinol (from trophic eggs provided by the mother in this egg-feeding species). To accomplish this we raised froglets under one of two different diets, a carotenoid/retinol-enriched diet and a carotenoid/retinol-deficient diet. We quantified coloration by measuring the effects of dietary carotenoids/retinol on the color (spectral reflectance) and the production of bright conspicuous coloration in relation to black on the dorsum of representative individuals.Fig. 1Ratio of color to black pigmentation of Sauce and Varadero Ranitomeya imitator froglets raised on either a carotenoid/retinol-enriched or carotenoid/retinol-deficient diet. Violin plots display the full distribution and density of color to black ratios within each treatment group, while boxplot summarizes the median and interquartile range. Sample sizes were: Sauce: carotenoid/retinol-enriched n = 7, carotenoid/retinol-deficient n = 7; Varadero: carotenoid/retinol-enriched n = 11, carotenoid/retinol-deficient n = 11) Ratio of color to black pigmentation of Sauce and Varadero Ranitomeya imitator froglets raised on either a carotenoid/retinol-enriched or carotenoid/retinol-deficient diet. Violin plots display the full distribution and density of color to black ratios within each treatment group, while boxplot summarizes the median and interquartile range. Sample sizes were: Sauce: carotenoid/retinol-enriched n = 7, carotenoid/retinol-deficient n = 7; Varadero: carotenoid/retinol-enriched n = 11, carotenoid/retinol-deficient n = 11) We predicted that genes responsible for carotenoid/retinol metabolism would be upregulated in the liver and skin tissue of the Sauce and Varadero morphs when raised under a carotenoid/retinol-enriched diet as opposed to those raised under a carotenoid/retinol-deficient diet. The sites where carotenoid and retinol processing occurs appear to vary among taxa, and at least in birds can occur in the liver or in the integument and feathers, with substantial variation between species [40, 41]. Processing (especially packaging) also takes place in the intestines. For example, carotenoids may be bound and transported by specific transporter proteins in the intestines (e.g. scarb1), and carried to other tissues by those transporters [42, 43]. These genes may also be expressed in other tissues, such as the liver [43]. We chose to focus on gene expression in the liver and skin, as these tissues have been implicated as major areas of carotenoid and retinoid metabolism in previous studies, including in poison frogs [24, 40]. In addition, we predicted the froglets raised under a carotenoid/retinol-deficient diet would have duller dorsal aposematic color and less area covered by dorsal aposematic color compared to froglets raised under a carotenoid/retinol-enriched diet. Methods Sample collection We established morph-specific breeding pairs for Ranitomeya imitator frogs, using individuals of either the Sauce or Varadero morphs, which we obtained from Understory Enterprises, LLC. Each enclosure included two water-filled PVC tubes (~ 25 mL) for tadpole deposition. Tanks were monitored regularly (three times weekly), and any newly discovered tadpoles were transferred into individual containers holding 100 mL of clean water. Full water changes were performed twice a week and tadpoles were raised on either a carotenoid/retinol-enriched or carotenoid/retinol-deficient diet. The carotenoid-enriched diet consists of Omega One Marine Flakes fish food and the carotenoid-deficient diet consists of nutritional yeast (Bragg, Santa Barbara, CA). We note that Omega One Marine Flakes contains both carotenoids such as beta-carotene and lutein (from spirulina algae) and preformed vitamin A (retinol) from fish. The source form of vitamin A (beta-carotene or retinol) that tadpoles receive in the wild is not known with certainty, since tadpoles may have access to both carotenoids and retinol. Ranitomeya imitator parents feed their offspring trophic eggs [44], which could contain either or both nutrients, and tadpoles may also have access to beta-carotene from algae in phytotelmata. Hence, we provided supplemented food that contains both nutrients (beta-carotene and retinol). The nutritional yeast we used contains inactive dry yeast and a mix of B-complex vitamins but does not contain carotenoids. It was used in a different study to investigate the effect of a carotenoid-deficient diet on coloration in copepods [45]. Once tadpoles reached developmental Gosner stages 43 or 44 [46], they were euthanized using topical 20% benzocaine followed by cerebral pithing to ensure death. At this point in development, individuals had fully formed limbs and exhibited the distinct coloration and patterning associated with their parental morph. Following euthanasia, dorsal skin regions displaying distinct pigmentation were carefully dissected. In the Sauce morph, orange and black skin regions were separated, while in the Varadero morph, red and black regions were separated. Because the skin of froglets yields relatively low RNA quantities and limited tissue volume, skin samples from individuals produced by the same breeding pair were pooled to obtain sufficient material for RNA extraction. In addition, we dissected ≤ 25% of liver from each froglet and pooled tissue from individuals produced by the same breeding pair. For both the Sauce and Varadero morphs, we used four tadpoles from each of three breeding pairs for the carotenoid/retinol-enriched and carotenoid/retinol-deficient treatments (total n = 48 tadpoles; Table S1). Samples were stabilized in RNA later (Qiagen) at 4° C for 24 h before being transferred to −20° C. We extracted RNA using a hybrid protocol that included Trizol reagent (Ambion), DNase and RNAsin treatment for DNA removal and RNA stabilization, followed by purification with an RNeasy spin column (Quagen). We used an Agilent 2100 Bioanalyzer at the Genomic Core Facility at East Carolina University (NC, USA) to assess RNA quality. RNA samples were submitted to Novogene for cDNA library preparation and sequencing. Libraries were paired-end sequences with 150 bp read lengths. In total, we sequenced 36 libraries, 18 per treatment (6 libraries for black skin tissue, 6 libraries for color skin tissue, and 6 libraries for liver tissue). Differential gene expression We followed the RNA-seq data processing pipeline described in Rubio et al. [37], which included adaptor trimming with Trimmomatic version 0.36 [47], sequence alignment to the Ranitomeya imitator genome (R_imi_1.0 (GCA_958301615.1)) [38] using STAR version 2.7 [48], and gene quantification with htseq-count [49]. Gene identities were obtained from the annotation of the Ranitomeya imitator R_imi_1.0 genome, GCA_958301615.1. Because this annotation is generated using homology-based methods, ortholog and paralog assignments are putative, and functional interpretations should be made cautiously. Transcript accession numbers for all genes included in the RNA-seq analysis are provided in Table S2. We conducted gene differential expression analyses in R version 4.4.2 “Pile of Leaves” (R Core Team 2024) by using the package DESeq2 [50]. In R, we merged all counts for each gene from htseq-count, twelve gene expression libraries for liver tissue and twenty-four gene expression libraries for skin tissue (Table S3). To avoid potential biases from genes with low expression, we removed any genes that had a total experiment-wide sum expression level of ≤ 50. We applied Wald’s test to identify differentially expressed genes, using a Benjamini and Hochberg correction for multiple testing with a significance threshold of 0.05 [51]. We conducted eight different comparisons; two comparisons of liver tissue (sauce: carotenoid/retinol-enriched vs. carotenoid/retinol-deficient; varadero: carotenoid/retinol-enriched vs. carotenoid/retinol-deficient), two comparisons of all skin tissue (sauce: carotenoid/retinol-enriched vs. carotenoid/retinol-deficient; varadero: carotenoid/retinol-enriched vs. carotenoid/retinol-deficient), two comparisons of color (orange and red) skin patches (sauce: carotenoid/retinol-enriched vs. carotenoid/retinol-deficient; varadero: carotenoid/retinol-enriched vs. carotenoid/retinol-deficient), and two comparisons between colored skin patches shared across morphs (carotenoid/retinol-enriched: Sauce orange skin patches vs. red skin patches; carotenoid/retinol-deficient: Sauce orange skin patches vs. Varadero red skin patches) (Table S4). We included all twelve liver gene expression libraries when conducting analyses related to liver tissue and all twenty-four gene expression data when conducting analyses related to skin tissue. For each model, we accounted for overall differences between morphs by including morph as a fixed variable. Because each froglet contributed both black and colored skin tissue, we controlled for the paired nature of the data by accounting for individual identity nested within morph. This ensures that color differences are elevated within individuals while controlling for baseline variation among morphs. Because sex cannot be determined at the Gosner stages at which samples were euthanized, we were unable to evaluate potential sex-related effects. We then used a curated list of genes associated with coloration across vertebrates [52, 53] to identify differentially expressed genes associated with color pattern development. In addition, we searched through all genes that were found to be differentially expressed in each comparison to ensure that genes related to color pattern development were not missing from the curated list and, more importantly, to highlight genes that are close paralogs and related gene families of well known genes. This is important as evidence suggests that paralogous genes often retain overlapping or similar biological functions [54–56]. Skin ratio of bright conspicuous color to black We quantified color from all froglets in this study. Before we sacrificed froglets, we used a Canon camera to photograph the dorsal surface of froglets raised on the two treatment diets for each morph. We used the program ImageJ version 1.53t [57] to analyze the ratio of colored skin patches to black skin patches on the dorsum of 34 froglets (Varadero n = 22 and Sauce n = 14). First, we standardized body measurements to a snout-vent length (SVL) of 17.7 mm to control for variation in pixel size, pixel counts, and image quality. We used the polygon selection and measurement functions in ImageJ to measure colored skin dorsum area and total dorsum area. We then calculated the black skin dorsum area by subtracting the colored skin dorsum area from the total dorsum area. To calculate the ratio of colored skin patches to black skin patches on the dorsum we divided the color dorsum area by the black dorsum area. Prior to statistical analysis, we assessed the normality of our data using a Shapiro-Wilk test [58]. For each morph, we then used a two-sample t-test to determine whether there were significant differences in the ratio of colored skin patches to black skin patches on the dorsum between froglets raised under the carotenoid/retinol-enriched and carotenoid/retinol-deficient diets. In addition, to test for size-effects of diet treatment, we used a two-sample t-test to determine whether there were differences in total dorsum area for each morph. Spectral reflectance We used an Ocean Optics USB4000 spectrometer with an LS-1 tungsten-halogen light source and the associated Ocean Optics software (OceanView Version 1.6.7) to collect spectral reflectance data. We used a 45° angle tip on the probe to standardize distance and angle to frog skin. To account for lamp drift, we calibrated the spectrometer by using a WS-1-SL white standard prior to collecting data as well as between each tadpole. We collected spectral reflectance measurements from six brightly colored patches of skin (orange in the Sauce frogs and red in the Varadero frogs) on the dorsum of each frog prior to euthanasia (Figure S1). We used the R package pavo [59] to clean and analyze collected spectral reflectance data. For each data point, we fixed negative values by setting them to zero using the function “fixneg”. We chose four color variables: mean brightness (B2), hue (H1), yellow (550–625 nm) (S1Y), and red chroma (605 nm-lambda max) (S1R) [59]. For each color morph, we used a Welch’s t-test to investigate if there were significant differences in the values of the color variable between froglets raised under carotenoid-enriched and carotenoid-deficient diets. Results Skin ratio of bright conspicuous color to black We assessed normality in our data using a Shapiro-Wilk test [58] and found no significant difference between the distribution of our data and a normal distribution for ratios of color to black (Sauce carotenoid/retinol-enriched (n = 7): W = 0.944, p-value = 0.683; Sauce carotenoid/retinol-deficient (n = 7): W = 0.830, p-value = 0.080; Varadero carotenoid/retinol-enriched (n = 11): W = 0.953, p-value = 0.685; Varadero carotenoid/retinol-deficient (n = 11): W = 0.983, p-value = 0.982) or total dorsum area (Sauce carotenoid/retinol-enriched (n = 7): W = 0.909, p-value = 0.433; Sauce carotenoid/retinol-deficient (n = 7): W = 0.978, p-value = 0.943; Varadero carotenoid/retinol-enriched (n = 11): W = 0.969, p-value = 0.876; Varadero carotenoid/retinol-deficient (n = 11): W = 0.912, p-value = 0.258), and so proceeded to employ parametric tests. For froglets of the Sauce morph, we found a significant difference in the ratio of colored skin patches to black skin patches on the dorsum between froglets raised on a carotenoid/retinol-enriched and carotenoid/retinol-deficient diet (t = 2.470, df = 10.774, p-value = 0.031) (Fig. 1). For froglets of the Varadero morph, we also found a significant difference in the ratio of colored skin patches to black skin patches on the dorsum between froglets raised on a carotenoid-enriched and carotenoid-deficient diet (t = 2.842, df = 18.165, p-value = 0.010) (Fig. 1). We found no significant difference between the total dorsum area between diet treatments in either morph (Sauce: (t = 0.508, df = 7.110, p-value = 0.626); Varadero: (t = −0.694, df = 13.294, p-value = 0.499)). Quantifying coloration When comparing color variable values from Sauce froglets raised on either a carotenoid/retinol-enriched or carotenoid/retinol-deficient diet, we found a significant difference in mean brightness (t = 5.010, p-value = 1.499e-06) and yellow chroma (t = 2.660, df = 93.037, p-value = 0.009); however, we did not find a significant difference in hue (t = −1.4743, p-value = 0.1441) and red chroma (t = 1.482, df = 85.741, p-value = 0.141) (Fig. 2a). When comparing color variable values from Varadero froglets raised on either a carotenoid/retinol-enriched or carotenoid/retinol-deficient diet, we found a significant difference in mean brightness (t = 5.743, df = 136.660, p-value = 5.753e-08); however, we did not find a significant difference in hue (t = 0.379, df = 141.570, p-value = 0.7047), yellow chroma (t = 0.62574, df = 167.5, p-value = 0.5323), or red chroma (t = 0.229, df = 169, p-value = 0.819) (Fig. 2b).Fig. 2Spectral reflectance of Ranitomeya imitator froglets raised on either a carotenoid/retinol-enriched or carotenoid/retinol-deficient diet. Mean reflectance spectra including standard deviation of (a) the Sauce morph and (b) the Varadero morph Spectral reflectance of Ranitomeya imitator froglets raised on either a carotenoid/retinol-enriched or carotenoid/retinol-deficient diet. Mean reflectance spectra including standard deviation of (a) the Sauce morph and (b) the Varadero morph Overall gene expression differences We generated an average of 22.04 million (± 1.9 SD) paired end sequences per library. Of these, an average of 83.24% (± 1.64% SD) of the reads mapped to the R. imitator genome. Gene expression counts data are all available in the Zenodo repository for this project. For liver tissue, we found a total of 21,869 genes to be expressed, 15,041 of which had a total experiment-wide expression level of ≥ 50. For skin tissue, we found a total of 23,084 genes to be expressed, 17,570 of which had a total experiment-wide expression level of ≥ 50. Differentially expressed genes in the liver We identified 106 differentially expressed genes between the liver tissue of Sauce froglets that were raised on a carotenoid/retinol-enriched diet and on a carotenoid/retinol-deficient diet (Table S5). Of these, eighteen genes have been putatively linked to pigment production in poison frogs or other taxa (Table S6). We highlight nine genes with well-known functions, four of which were up-regulated in the liver of sauce froglets that were fed a carotenoid/retinol-enriched diet (tmem135, phgdh, lrat, akr1d1) and five of which were up-regulated in the liver tissue of sauce froglets that were raised on a carotenoid/retinol-deficient diet (cyp7a1, aldh1l1, xdh, cyp27a1, aldh6a1) (Fig. 3a).Fig. 3Heatmap summarizing differentially expressed genes between the liver of froglets raised on a carotenoid/retinol-enriched and carotenoid/retinol-deficient diet. Each column represents an individual library. Orange and red colors indicate higher relative expression of a gene, whereas grey indicates lower relative expression. Liver tissues were obtained by dissecting ≤ 25% of the liver from each froglet, and liver tissue from froglets originating from the same breeding pair was pooled prior to RNA extraction. For each morph, three breeding pairs (four froglets per pair) were used for both treatments. Panels show results for the (a) Sauce morph and (b) Varadero morph. These are a subset of genes that were found to be differentially expressed. The full list of genes and their log2FoldChange values can be found in Table S5 Heatmap summarizing differentially expressed genes between the liver of froglets raised on a carotenoid/retinol-enriched and carotenoid/retinol-deficient diet. Each column represents an individual library. Orange and red colors indicate higher relative expression of a gene, whereas grey indicates lower relative expression. Liver tissues were obtained by dissecting ≤ 25% of the liver from each froglet, and liver tissue from froglets originating from the same breeding pair was pooled prior to RNA extraction. For each morph, three breeding pairs (four froglets per pair) were used for both treatments. Panels show results for the (a) Sauce morph and (b) Varadero morph. These are a subset of genes that were found to be differentially expressed. The full list of genes and their log2FoldChange values can be found in Table S5 We identified 191 differentially expressed genes between the liver tissue of Varadero froglets raised on a carotenoid/retinol-enriched diet and a carotenoid/retinol-deficient diet (Table S5). Of these, Twenty-seven genes have been putatively linked to pigment production in poison frogs or in other taxa (Table S6). We highlight nine genes with well-known functions, two of which were up-regulated in the liver of Varadero froglets that were fed a carotenoid/retinol-enriched diet (lrat, creb3l1) and of which were up-regulated in the liver tissue of Varadero froglets that were fed a carotenoid-deficient diet (pnpla3, gramd1c, tyr, xdh, aldh1l1, aldh1a3, sult1a1, cyp7a1) (Fig. 3b). Differentially expressed genes in the skin We identified 262 differentially expressed genes in the skin (black and orange skin patches) between Sauce froglets raised on a carotenoid/retinol-enriched diet and on a carotenoid/retinol-deficient diet (Table S7). Of these, twenty genes have been putatively linked to pigment production in poison frogs or in other taxa (Table S8). We highlight five genes with well-known functions, three of which were up-regulated in the skin of Sauce froglets that were fed a carotenoid/retinol-enriched diet (phgdh, creb3l1, dhrs3) and two of which were up-regulated in the skin of Sauce froglets that were fed a carotenoid/retinol-deficient diet (sult1a1, aldh1a3) (Fig. 4a). In addition, we found thirteen differentially expressed genes in orange skin between Sauce froglets raised on a carotenoid/retinol-enriched diet and on a carotenoid/retinol-deficient diet (Table S9). Of these, two genes have been putatively linked to pigment production in poison frogs or in other taxa. We highlight two genes with well-known functions, one of which was up-regulated in the skin of Sauce froglets that were fed a carotenoid/retinol-enriched diet (tf-a) and one of which was up-regulated in the skin of Sauce froglets that were fed a carotenoid/retinol-deficient diet (bcl6) (Table S10).Fig. 4Heatmap summarizing differentially expressed genes between the skin (black and colored skin patches) of tadpoles raised on a carotenoid/retinol-enriched diet and carotenoid/retinol-deficient diet. Each column represents an individual library. Orange and red colors indicate higher relative expression of a gene, whereas grey indicates lower relative expression. For each morph, three breeding pairs (four froglets per pair) were used for both treatments. Panels show results for the (a) Sauce morph and (b) Varadero morph. These are a subset of genes that were found to be differentially expressed. The full list of genes and their log2FoldChange values can be found in Table S7 Heatmap summarizing differentially expressed genes between the skin (black and colored skin patches) of tadpoles raised on a carotenoid/retinol-enriched diet and carotenoid/retinol-deficient diet. Each column represents an individual library. Orange and red colors indicate higher relative expression of a gene, whereas grey indicates lower relative expression. For each morph, three breeding pairs (four froglets per pair) were used for both treatments. Panels show results for the (a) Sauce morph and (b) Varadero morph. These are a subset of genes that were found to be differentially expressed. The full list of genes and their log2FoldChange values can be found in Table S7 We identified 86 differentially expressed genes in the skin (black and red skin patches) between Varadero froglets raised on a carotenoid/retinol-enriched diet and a carotenoid/retinol-deficient diet (Table S7). Of these, nine genes have been putatively linked to pigment production in poison frogs or in other taxa (Table S8). We highlight two genes with well-known functions, which were up-regulated in the skin of Varadero froglets that were fed a carotenoid/retinol-deficient diet (sult1a1, ttc32) (Fig. 4b). In addition, we found forty-nine differentially expressed genes in the red skin between Varadero froglets raised on a carotenoid/retinol-enriched diet and on a carotenoid/retinol-deficient diet (Table S9). Of these, eight genes have been putatively linked to pigment production in poison frogs or in other taxa (Table S10). We highlight five genes with well-known functions, two of which were up-regulated in the skin of Varadero froglets that were fed a carotenoid/retinol-enriched diet (creb3l1, tf-a) and three of which were up-regulated in the skin of Varadero froglets that were fed a carotenoid/retinol-deficient diet (mc1r, impdh1, akr1b1). Differentially expressed genes between colored skin patches shared across morphs We identified 116 differentially expressed genes between the orange skin patches of Sauce froglets raised on a carotenoid/retinol-enriched diet and the red skin patches of Varadero froglets raised on a carotenoid/retinol-enriched diet (table S11). Of these, 25 genes have been putatively linked to pigment production in poison frogs or in other taxa (Table S12). We highlight thirteen genes with well-known functions, four of which were up-regulated in the orange skin of Sauce froglets that were fed a carotenoid/retinol-enriched diet (tyrp1, pmel, kcnj13, mlana) and nine of which were up-regulated in the red skin of Varadero froglets that were fed a carotenoid/retinol-enriched diet (gmps, akr1b1, dgat2, impdh1, mc1r, pax3-a, gch1, pax7, xdh). We identified 178 differentially expressed genes between the orange skin patches of Sauce froglets raised on a carotenoid/retinol-deficient diet and the red skin patches of Varadero froglets raised on a carotenoid/retinol-deficient diet (table S11). Of these, 34 genes have been putatively linked to pigment production in poison frogs or in other taxa (Table S12). We highlight fifteen genes with well-known functions, four of which were up-regulated in the orange skin of Sauce froglets that were fed a carotenoid/retinol-deficient diet (tyrp1, pmel, kcnj13, mlana) and eleven of which were up-regulated in the red skin of Varadero froglets that were fed a carotenoid/retinol-deficient diet (gmps, dgat2, akr1b1, gpr143, dgat2, impdh1, mc1r, pax3-a, gch1, pax7, xdh). Discussion We found that froglets raised on a carotenoid/retinol-enriched diet displayed brighter skin when compared to froglets raised on a carotenoid/retinol-deficient diet in both the Sauce and Varadero morph. In addition, we found that froglets that were fed a carotenoid/retinol-enriched diet displayed a higher ratio of bright colored to black skin patches when compared to froglets that were raised on a carotenoid-deficient/retinol diet in both the Sauce and Varadero morph. Numerous studies indicate that carotenoids promote the development and maintenance of bright coloration (e.g., red, orange) (e.g [24, 25]). Recently, a number of studies have identified specific genes that appear to regulate various aspects of yellow, orange and red skin and feather coloration, especially in fish and birds (see below for details). However, none of these genes showed a clear association of differential expression with color variation in our study. Genes previously found to be associated with carotenoid coloration include fatty acid translocase (cd36), tetratricopeptide repeat domain 39b gene (ttc39b), cytochrome p450 2J19 (cyp2j19), diacylglycerol O-acyltransferase 2 (dgat2), perilipin 6 (plin6), gram domain containing 1B gene (gramd1b), patatin-like phospholipase domain containing 2 (pnpla2), among others [25, 60–62]. We did not find differential expression of any of these candidate genes in our study, although we did find differential expression of closely related paralogs or members of the same gene family, as described below. Because functional information on the differentially expressed genes in our study, especially in the context of carotenoid metabolism, is limited, we drew evidence from better-studied paralogs and related gene families. This approach is supported by studies showing that paralogous genes often retain overlapping or similar biological functions following duplication [54–56]. We also found that the carotenoid/retinol-deficient diet treatment subjects showed elevated levels of expression of genes associated with the retinoic acid pathway, whereas we found upregulation of genes associated with the retinol pathway in froglets raised under the carotenoid/retinol-enriched diet. Lipid metabolism and carotenoid metabolism The candidate genes most likely to be involved in carotenoid processing and transport based on previous studies are the gram domain containing 1 C gene (gramd1c) and the patatin-like phospholipase domain containing 3 (pnpla3) genes. Recent research identified the protein product of the gram domain containing 1B gene (gramd1b) as a key transporter of carotenoids in mammals [63]. Gramd1c, like gramd1b, contains a StART-like (VASt) lipid-transfer domain and participates in non-vesicular cholesterol transport, suggesting a conserved role in intracellular lipid trafficking [64]. Given their established roles in lipid metabolism and cholesterol transport, gramd1 paralogs, such as gramd1c, may also contribute to intracellular carotenoid trafficking, a possibility that merits further investigation. We identified over two logfold upregulation of the gramd1c gene in the livers of the Varadero morph tadpoles raised on a carotenoid/retinol-deficient diet. The protein product of this gene is also intimately involved in lipid metabolism and cholesterol transport [64], and may also be associated with carotenoid transport [65]. Recent research has also identified the patatin-like phospholipase domain-containing 2 (pnpla2) gene as a key catalyst for the hydrolysis of triglyceride esters into diacylglycerol and a fatty acid, and the hydrolysis of retinyl esters [25]. This gene was found to play an important role in carotenoid-mediated coloration in cichlid fish [25]. The pnpla3 gene has also been shown to play key roles in the metabolic processing of retinol, and specifically the hydrolysis of retinyl palmitate to retinol and palmitic acid [66, 67]. This gene is also associated with the processing of lipid droplets [66], which can contain either retinol or carotenoids [68, 69]. It is possible that the pnpla3 gene plays a role in carotenoid metabolism and coloration in R. imitator, as pnpla2 appears to do in cichlids. We found the pnpla3 gene to be up-regulated in the liver of Varadero morph tadpoles raised on a carotenoid/retinol-deficient diet. We note that both gramd1c and pnpla3 genes were upregulated under the carotenoid/retinol-deficient diet, which was opposite to our expectations for genes affecting carotenoid metabolism and transport. However, it is possible that when carotenoids are scarce, these genes may be upregulated to more effectively process and transport whatever carotenoids are available. In our experiments, the carotenoid/retinol-deficient diet treatment had (at most) low levels of carotenoids. Although the inactivated yeast diet we used lacks carotenoids, we cannot exclude the possibility that some could be obtained in small amounts from algae growing in the pools. However, algal growth, if it occurred, was at low levels (i.e. algal growth was not visible in the pools). In addition to these two candidate genes, other candidates may also have been involved in carotenoid metabolism, as described below. We found that akr1d1 (aldo-keto reductase family 1 member D1) showed elevated expression in the livers of Sauce morph tadpoles that were fed a carotenoid/retinol-enriched diet. This gene is a member of the aldo/keto reductase superfamily, with over 40 members. These enzymes catalyze the reduction of various aldehydes. Ketoreductases have been implicated in the conversion of yellow to red carotenoids (ketocarotenoids) in various species, including two species of Ranitomeya. However recent research [70] indicates that this kind of red pigment is not found in R. imitator. Curiously, a closely relatedgene(akr1b1) showed the oppositeexpression pattern (higher in the carotenoid-deficient-diet treatment). The reason for this difference is not clear, and requires further research. Previous studies have found that genes associated with processing of lipids or cholesterol (or both) are also involved in the metabolism of carotenoids or retinol (or both). The transmembrane protein 135 (tmem135) was differentially upregulated in the livers of the Sauce morphs raised on a carotenoid/retinol-enriched diet. The exact function of this gene is unknown, but it is strongly associated with lipid metabolism, and has been localized to lipid droplets, as well as mitochondria and peroxisomes [71–75]. As mentioned above, carotenoid-derived pigments are likely to be transported via lipid droplets. Expression of the ttc39b (tetratricopeptide repeat domain 39b) gene was found to have a key effect on orange coloration in parrot finches, and on skin color in cichlid fish [25, 61, 76]. This gene is involved in various aspects of cholesterol metabolism in humans. In our analyses, we found a member of the same broader TPR gene family, ttc32 (tetratricopeptide repeat domain 32), to be upregulated in the skin of Varadero morphs that were fed a carotenoid/retinol-deficient diet. While ttc32 gene shares the conserved TPR protein-interaction motif found in ttc39b, it has not been implicated in coloration or lipid-related processes in any taxa, and its role in these pathways remains unknown. Although key genes related to carotenoid lipid storage were up-regulated in the tissue of froglets that were fed a carotenoid enriched diet, we did not find β-carotene oxygenase enzymes (e.g., bco2 and bcmo1), which play vital roles in the breakdown of carotenoids, to be differentially expressed between froglets raised under these diet treatments. However a β-carotene oxygenase enzyme (bco1) was found to be differentially expressed between larval developmental stages of R. imitator, R. fantastica, and R. variabilis [38]. This is not surprising, as it is likely that gene expression levels typically differ over the course of development. It is plausible that the genetic mechanisms controlling carotenoid-based coloration differ between species in the genus Ranitomeya. Retinol and retinoic acid metabolism We found that the carotenoid/retinol-deficient diet treatment subjects showed elevated levels of expression of genes associated with the retinoic acid pathway, whereas we found upregulation of genes associated with the retinol pathway in froglets raised under the carotenoid/retinol-enriched diet. For example, froglets raised on a carotenoid/retinol-deficient diet showed elevated levels of aldehyde dehydrogenase (aldh) genes in liver and skin tissue. Members of this family of genes are known to be involved in the conversion of retinaldehyde to retinoic acid, and hence are critical in regulating many key processes during development. In mammals, aldh1a1, aldh1a2 and aldh1a3 are all involved in the production of retinoic acid [16]. Mutations in these enzymes can cause severe developmental abnormalities. In our study, we found higher expression levels of aldh1l1 and aldh6a1 in the livers of subjects in the carotenoid/retinol-deficient diet treatment, for both morphs. Although these genes have not previously been implicated in retinoic acid synthesis, their up-regulation under retinoid limitation indicates that these enzymes may be crucial for retinoic acid production in amphibians. One plausible explanation is a compensatory response, where developing tadpoles increase the expression of genes involved in these critical developmental processes when carotenoids are scarce. For instance, Zhai et al. [77] found that vitamin A deficiency in rats causes a 7-fold increase in raldh1 expression, demonstrating that reduced retinoid availability can induce up-regulation of retinoic acid producing enzymes. Another class of enzymes that are involved in processing retinoic acid during development are the cytochrome P450 (cyp) superfamily of genes. Members of this family of genes are well known for producing enzymes in the liver involved in detoxification. It is worth noting that cyp genes can also affect other biological mechanisms such as responding to stress [78, 79]. We found significantly higher expression (over 8 logfold) of cyp7a1 in the livers of both morphs under the carotenoid/retinol-deficient diet, and of cyp27a1 in the livers of the Sauce morph in the carotenoid/retinol-deficient diet. Other members of the cyp family were found to be upregulated in the skins of the Sauce morphs (some on the carotenoid/retinol-enriched diet, some on the carotenoid/retinol-deficient diet). Several cyp genes are involved in processing retinoic acid. For example, in mice, cyp26a1 is crucial for the metabolic breakdown of retinoic acid when it is present in excess [80]. It is possible that, in R. imitator, cyp27a1 or cyp7a1 (or both) play a similar role in regulating retinoic acid levels. Upregulation of these genes under the carotenoid/retinol-deficient diet may be associated with increased expression of genes regulating the retinoic acid pathway. Retinol is often converted to retinyl ester for storage when it is in excess. The lrat gene (Lecithin-retinol acyltransferase) is well known to play vital roles in converting all-trans-retinol (vitamin A) into all-trans-retinyl esters and has been identified in numerous studies as the primary enzyme in the esterification of retinol [81–84]. Results from Batten et al., [84] suggest that in mice, lrat is possibly the only enzyme in control of retinyl ester formation in the eye and liver. O’Byrne et al., [85] found that there was a significant reduction of retinoid absorption and storage in the liver of mice that lacked lrat. Another study found that mice were susceptible to vitamin A deficiency when lrat was knocked out [86]. Specifically, lrat plays key roles in the regulation and formation of lipid droplets [87]. We found significantly higher expression (over 6 logfold) of the lrat gene (Lecithin-retinol acyltransferase) in the livers of both morphs that were raised on a carotenoid/retinol-enriched diet. We found two genes related to retinol metabolism that were differentially expressed between froglets that were raised on either a carotenoid/retinol-enriched or carotenoid/retinol-deficient diet. The dhrs3 (retinoic acid-inducible dehydrogenase reductase 3) gene plays a vital role in retinoid metabolism by reducing all-trans-retinaldehyde to all-trans-retinol for lipid storage and ultimately restraining the biosynthesis of all-trans-retinoic acid [88–90]. This gene was upregulated in skin tissue of tadpoles of the Sauce morph that were raised on a carotenoid/retinol-enriched diet. When the dhrs3 gene is knocked out in mice, all-trans-retinoic acid levels are increased by 40% and retinol and retinyl ester levels decreased by 60% [91]. In contrast, the over expression of the dhrs3a gene results in the reduction of all-trans-retinoic acid signaling in zebrafish embryos [92]. It is possible that some genes were upregulated to compensate for the lack of carotenoids. For example, we found higher expression of the xanthodine hydroxylase (xdh) gene in the livers of both morphs raised on a carotenoid-deficient diet. This gene is well known for its role in pteridine synthesis, and this pigment is a major contributor to red coloration in several Ranitomeya species, including R. imitator [70]. However, xdh also plays vital roles in the biosynthesis of retinoic acid, mainly due to its ability to oxidize xanthine to uric acid [93, 94]. The xdh enzyme can be converted to xanthine oxidase (XO) through oxidation and together they can form the xanthine oxidoreductase (XOR) enzyme [95]. Studies suggest that xanthine oxidase, which is the O2-dependent form of xanthine oxidoreductase (XOR), has the ability to oxidize all-trans-retinol, vitamin A, to all-trans-retinaldehyde [96, 97]. This suggests that upregulation of xdh may represent a physiological response aimed at maintaining adequate levels of retinoic acid in the absence of dietary carotenoids. Such mechanisms could help mitigate the developmental and physiological consequences associated with carotenoid scarcity. We also found that the sulfotransferase family 1 A member 1 (sult1a1) was highly upregulated (above 2 logfold) under the yeast diet in the skin of both morphs and was also upregulated in the livers of the Varadero morph frogs. This gene family codes for enzymes that catalyze sulfate conjugation of many different substrates (hormones, neurotransmitters, xenobiotics), and are involved in androgen metabolism. Sulfotransferase family genes have been found to be associated with variation in coloration between red and yellow morphs of frillneck lizards (McLean et al. 2019). Specifically, they are more highly expressed in association with yellow coloration, which is thought to be associated with higher androgen levels. Whether a similar relationship holds in R. imitator deserves further investigation. It is possible that upregulation of this gene provides an alternative pathway to enhancing yellow coloration in the absence of carotenoids. Diet-dependent gene expression Our results indicate that dietary carotenoid/retinol availability has pronounced effects on the expression of genes involved in carotenoid and retinol metabolism, yet the molecular responses are largely morph-specific. Very few differentially expressed genes were shared between the Sauce and Varadero morphs within the same dietary treatment, suggesting that each morph may recruit distinct molecular pathways to process carotenoids and retinol. Despite these differences, both morphs exhibited similar phenotypic outcomes, increased brightness and higher color to black ratios under carotenoid-enriched diets, indicating that multiple, functionally convergent pathways can lead to the same pigmentation phenotype. Importantly, we identified a small set of genes whose expression changed in both morphs in response to diet, making them strong candidates for mediating the shared phenotypic effects of carotenoid availability. Under carotenoid/retinol-enriched diets, both Sauce and Varadero froglets upregulated the lrat gene, a key enzyme responsible for esterifying retinol for storage. This shared response suggests that enhanced retinol storage may be a core physiological mechanism underlying the increased brightness observed under carotenoid-rich conditions. Conversely, froglets raised on carotenoid/retinol-deficient diets in both morphs consistently upregulated genes associated with retinoic acid production, xdh, aldh1l1, cyp7a1, and sult1a1. These genes likely support the maintenance of essential retinoic acid signaling when dietary carotenoids are limited. Such compensatory upregulation provides a mechanistic explanation for the reduced brightness and lower color to back ratios observed under carotenoid scarcity. Together, these findings show that although morphs differ in their transcriptional responses to diet, both converge on a small set of shared diet-responsive genes that are strong candidates for linking carotenoid availability to pigment outcomes. The consistent upregulation the lrat under carotenoid/retinol-enriched diets, and of xdh, aldh1l1, cyp7a1, and sult1a1 under carotenoid/retinol-deficient diets, highlights a core set of genes likely central to carotenoid and retinol metabolism in Ranitomeya imitator. Conclusion The genetic basis of carotenoid metabolism and its role in pigmentation remain poorly understood in amphibians. Here, we address this gap by experimentally manipulating diet to investigate the regulation of key genes involved in carotenoid and retinol metabolism, drawing on homologs characterized in other vertebrates. We did this by raising R. imitator tadpoles of the the Sauce and Varadero morph morphs on varying levels of dietary carotenoid/retinol and found that froglets raised under a carotenoid/retinol-enriched diet showed elevated levels of expression of genes associated with the retinol pathway whereas froglets raised under a carotenoid/retinol-deficient diet showed elevated levels of expression of genes associated with the retinoic acid pathway. Although our experiment demonstrated clear gene expression differences between dietary treatments, our phenotypic data also revealed substantial within-group variation in color to black ratios, with considerable overlap between diets, particularly in the Varadero morph. Because tissues were pooled, we were unable to correlate individual variation in coloration with individual gene expression, but this pattern highlights the need for future work to directly link expression levels to phenotypic outcomes. Such analyses would help clarify whether the observed variation reflects differences in carotenoid processing, deposition, or other regulatory mechanisms. Finally, we note that gene identities were derived from homology-based annotation of the Ranitomeya imitator genome. As such, ortholog and paralog assignments are putative, and functional interpretations should be made cautiously. Future studies should aim their efforts at using knockout techniques, such as CRISPR-Cas9, to manipulate the genes highlighted in this study. For instance, the gramd1c gene may be associated with carotenoid transport and should be investigated further. These efforts will allow us to conclusively identify which genes are playing vital roles in producing and maintaining carotenoid based coloration. Supplementary Information Supplementary Material 1. Supplementary Material 1. Supplementary Material 2. Supplementary Material 2.