Developmental and comparative transcriptomic identification of iridophore contribution to white barring in clownfish

Actinopterygian fishes harbor at least eight distinct pigment cell types, leading to a fascinating diversity of colors. Among this diversity, the cellular origin of the white color appears to be linked to several pigment cell types such as iridophores or leucophores. We used the clownfish Amphiprion ocellaris, which has a color pattern consisting of white bars over a darker body, to characterize the pigment cells that underlie the white hue. We observe by electron microscopy that cells in white bars are similar to iridophores. In addition, the transcriptomic signature of clownfish white bars exhibits similarities with that of zebrafish iridophores. We further show by pharmacological treatments that these cells are necessary for the white color. Among the top differentially expressed genes in white skin, we identified several genes (fhl2a, fhl2b, saiyan, gpnmb, and apoD1a) and show that three of them are expressed in iridophores. Finally, we show by CRISPR/Cas9 mutagenesis that these genes are critical for iridophore development in zebrafish. Our analyses provide clues to the genomic underpinning of color diversity and allow identification of new iridophore genes in fish.

Among actinopterygians, coral reef fish exhibit an extraordinary set of colors and patterns, therefore providing ample material to study the cellular and genomic basis of color diversity in vertebrates. Here, we used the clownfish Amphiprion ocellaris, which exhibits an adult color pattern of distinctive, vertical white bars over a dark orange body (Salis et al., 2018). White bars exhibit a matte coloration that is qualitatively different from the iridescent light interstripes of zebrafish, and the cellular bases and genetic requirements of white barring have not been characterized. Here, we asked whether the white bars of A. ocellaris are composed of iridophores, as in zebrafish interstripes, whitish leucophores similar to those present scattered in medaka, or both types of cell. Using electron microscopy and transcriptomic analysis, we find that white cells of A. ocellaris resemble iridophores. We further show that several genes expressed specifically in white bars are required for iridophore development in zebrafish, including loci not previously identified for having such roles. Finally, using pharmacological manipulations, we show that A. ocellaris iridophores and zebrafish iridophores share a common dependence on anaplastic lymphoma kinase and leukocyte tyrosine kinase (ALK/LTK) signaling. Taken together, our results indicate that the white bars of clownfish comprise densely packed iridophores, illustrating the diverse pattern elements that can be generated by a conserved set of pigment cell classes.

| Stereomicroscope image processing
White and orange skins were imaged using Zeiss stereomicroscope (V20 discovery-Plan S objective 1.0×) using a VisiLED Brightfield Ringlight S80 for incident lighting. All the images were taken using a full homogenized lighting. However, iridescent cells of the white skin were only visible when using segmental lighting. Images were processed using a camera (Axiocam 105) with a picture resolution of

| Transmission electron microscopy
Amphiprion ocellaris 3-4 cm individuals were euthanized using 200 mg/L of MS-222 Tricaine-S. White and orange skins were dissected in a mixture of 3% paraformaldehyde/3% glutaraldehyde in 0.1 M cacodylate buffer (pH 7.2), and fixed overnight at 4°C, according to Djurdjevič, Kreft, and Sušnik Bajec (2015). After 1% OsO 4 post-fixation (1 hr at RT), individuals were dehydrated in a graded series of ethanol solutions and embedded in Epon 812 resin (Electron Microscopy Science). Ultra-thin (80 nm) sections were cut using a Leica Ultracut R microtome, contrasted with uranyl acetate and lead citrate, and examined with a Hitachi H7500 TEM.

| RNA extraction and purification for qPCR and transcriptomic skin experiments
Amphiprion ocellaris and Amphiprion frenatus adults (3-4 cm) were euthanized prior to dissection in MS222 at 200 mg/L (three individuals per experiment). A shallow incision was made with a razor in the white body bar and orange skin (posterior to the white body bar) regions from both sides of the body fish A. ocellaris (Supporting Information Figure S2) and in the head white bar and orange body skin regions of the body fish A. frenatus. The skin was slowly pulled back with forceps while using the razor blade to make sure the skin separated cleanly from the underlying tissue and conserved Significance Coral reef fish exhibit an extraordinary set of colors and patterns, therefore providing ample material to study the cellular and genomic basis of color diversity in vertebrates.
Here, we asked whether Amphiprion ocellaris clownfish's white bars depend on iridophores, as in zebrafish interstripes, or on leucophores, as found in medaka. Using electron microscopy, transcriptomic analysis, and genetic approaches, we reveal that white cells of A. ocellaris are iridophores. Our data illustrate the diverse pattern elements that can be generated by a conserved set of pigment cell classes.
in RNAlater until RNA extraction. The RNA was extracted using TRIzol (TRIzol Reagent 15596-026 Kit; Ambion) followed by DNase treatment (DNA-free AM1906 Kit; Ambion) and then purified with 0.025 µm dialysis membranes to remove reverse transcriptase inhibitors.

| RNA-Seq library preparation and sequencing
RNA-Seq libraries were generated with TruSeq Stranded mRNA Sample Preparation Kit (Illumina) from 400 ng of total RNA using skin samples according to manufacturer's instructions. Surplus PCR primers were removed using AMPure XP beads (Beckman Coulter).
Final cDNA libraries were checked for quality and quantified using capillary electrophoresis. Libraries were loaded in the flow cell at 2 nM. Clusters were generated in the Cbot and sequenced on an Illumina HiSeq 4000 as single-end 50 base reads or paired-end 100 base reads.

| Quantification, differential expression, and gene set enrichment analyses
We used kallisto v0.43.0 to align reads originating from single-end skin samples to the reference transcriptome and to quantify transcript abundance. Estimated counts were imported in r using tximport package (Soneson, Love, & Robinson, 2015

| Phylogenetic analyses
To ensure the identity of the 10 most differentially expressed genes (DEGs) in each condition, we conducted phylogenetic analyses.
Using S. partitus sequences as query, sequences corresponding to best blastn hits were retrieved in multiple vertebrate species using Ensembl. Alignment was performed using Clustal Omega implemented in seaview v4.6.1 using default parameters. After manual curation, phylogenetic trees were built using maximum likelihood and the LG model of amino acid substitution as implemented in phyml v3.1 within Seaview. Support for the most likely tree was based on the SH-aLRT algorithm (Anisimova, Gil, Dufayard, Dessimoz, & Gascuel, 2011).

| qRT-PCR assays
qRT-PCR primers were designed to anneal on different exons (see Supporting Information Table S1), and rpl32 was used as a normalization gene. qRT-PCRs were performed in 96-well plate with KAPA SYBR Fast ABI ReadyMix Kit in 20 μl of final reaction per well following the manufacturer ratios using a StepOnePlus™ Real-Time PCR System.

| In situ hybridization
Digoxigenin RNA probes were synthesized using the Sp6/T7 Transcription Kit (Roche; see Supporting Information Table S2).
Adult and 3-month-old clownfishes were euthanized in MS222 at 200 mg/L. For each clownfish, the skin was dissected in a way to get in a same sample a part of the white body bar, the black bar, and the orange skin (posterior to the white body bar; Figure 4a). Adult skin samples were then fixed 24 hr in 4% paraformaldehyde diluted in PBS (phosphate-buffered saline). Samples were subsequently dehydrated stepwise in PBS/ethanol, and then put four times 10 min in butanol 100% and finally embedded in paraffin overnight. Embedded skin samples were sectioned transversally with a thickness of 8 µm using Leica Biosystems RM2245 Microtome. The samples were then treated as in Thisse, Thisse, Schilling, and Postlethwait (1993).

| F 0 CRISPR/Cas9 mutagenesis and iridophore counts
Wild-type strain ABb zebrafish (a derivative of inbred Ab wp ; Eom et al., 2015) were housed under standard conditions. Embryos were injected with T7-transcribed single-guide RNAs (sgRNAs) and Cas9 protein as described (Shah, Davey, Whitebirch, Miller, & Moens, 2015). sgRNA protospacers were modified to "g" at 1st and 2nd 5′ positions to facilitate in vitro transcription as necessary. To increase likelihoods of biallelic knockdown, two sgRNAs were designed to target an early exon for each candidate gene. Embryos were injected at the one cell stage except for fhl2a and fhl2b, for which embryo survival was <10% at 4 days post-fertilization (dpf); sgRNAs targeting these loci were therefore injected into a single blastomere after early cytoplasmic bridges had closed (late 16-32 cell stage; Kimmel, Ballard, Kimmel, Ullmann, & Schilling, 1995) to diminish the numbers of clonal lineages carrying mutations and enhance survivorship.
Mutagenicity of each sgRNA was confirmed by Sanger sequencing of amplicons from whole embryo genomic DNA. sgRNA targets are described in Supporting Information Table S4. F 0 embryos were raised at ~28°C for 4 days at which time larvae were anesthetized with MS222 and iridophores at the dorsal midline were counted manually under incident light. Analyses of iridophore counts were performed using jmp 14.0 statistical software (SAS Institute, Cary, NC, USA).

| RNA-Seq library preparation and sequencing
Information on sequencing methods as well as average read number and quality per sample is provided in Supporting Information Table   S6.

| White pigment cells in Amphiprion ocellaris are morphologically similar to iridophores
To determine the type of pigment cell responsible for the white color of A. ocellaris bars, we first compared white and orange skins microscopically. Both classical types of reflecting pigment cells known in vertebrates, leucophores and iridophores, can be distinguished by two main criteria: Iridophores are generally round and contain flattened reflecting platelets, and leucophores are dendritic with smaller and rounded organelles (Fujii, 1993;Schartl et al., 2015). Using light microscopy, we observed a white matte coloration overall within which scattered, rounded iridescent cells and dendritic melanophores could be observed occasionally. Given the apparent density of cells contributing to the white matte color itself, individual pigment cells contributing to it were not discernable by light microscopy (Figure 1a). We therefore used transmission electron microscopy (TEM) to understand the nature of the white coloration. In the skin, the epidermis covers the scales whose ends lie inside the dermal tissue that contain the pigment cells (Figure 1b,d;Hirata et al., 2003). Within the dermis, pigment cells were abundant in the stratum laxum, which surrounds the inner and outer faces of the bony scales (Figure 1b,d, Sire, 1988).
In skin from regions with white bars, the overwhelming majority of pigment cells had rounded bodies and contained oblong, ca.
0.5 µm platelets arranged parallel to one another in discrete stacks organized concentrically around the nucleus (Figure 1c and inset).
The morphology and organization of these platelets were consistent with the cells being iridophores, which appear iridescent when platelets are precisely organized or white when platelets are less organized (Fujii, 1993;Schartl et al., 2015). Melanophores were observed occasionally as well. Although sparsely distributed iridescent cells were evident by light microscopy (Figure 1a), we were unable to identify a distinct population by TEM that might correspond to these cells, either because their features could not be distinguished from other pigment cells ultrastructurally, or, more likely, because they occur at densities too low to have been sampled in thin sections. Given that all unmelanized pigment cells in white bars were of a rounded (rather than dendritic) morphology and contained flattened (rather than rounded) organelles similar to those of iridophores, we conclude that white coloration of A. ocellaris is likely conferred by iridophores, rather than leucophores. In orange skin, we found no platelet-containing cells and observed instead only elongated xanthophores and melanophores (Supporting Information Figure S1).

| Comparative transcriptomics of orange versus white bars
To identify gene expression signatures specific to white and orange skin and their associated pigment cells, we generated transcriptomes from skin samples of each type from each of three adult A. ocellaris (Supporting Information Figure S2), using a reference transcriptome as well as whole genome data for A. ocellaris generated for comparative analyses (A. Marcionetti and N. Salamin, in preparation).
Principal component analysis confirmed overall differences in gene expression between white and orange skin samples (axis 1, 41% of the variance; Supporting Information Figure S3). Individuals also differed markedly from one another at the transcriptomic level, which is not surprising given that individual fish were of heterogeneous, rather than inbred, genetic backgrounds (axis 2, 35% of the variance; Supporting Information Figure S3). Comparison of white and   Table S3).

Skin samples include pigment cells and a variety of other cell
types. To gain further insight into the transcriptomic signature of A. ocellaris white skin, we compared our results to an available set of 346 genes specifically enriched in zebrafish iridophores (Higdon, Mitra, & Johnson, 2013). Of these 346 loci, 237 were detectable in A. ocellaris skin samples (Supporting Information Table S5). Using this dataset, we carried out two types of analysis: (a) we first performed a gene set enrichment analysis using all A. ocellaris genes detectably expressed in both sample types. The expression levels of all genes were correlated to one or the other skin color. Each gene in the "iridophore gene set" (i.e., the clownfish orthologs of 237 zebrafish iridophore markers) was then ranked among all expressed genes ("Ranked gene list" in Supporting Information Figure S4a). The results show that the expression of the genes in this gene set is overall correlated to the white skin phenotypic information. Hence, a negative enrichment score is observed. The zebrafish iridophore gene set is thus enriched in white skin (p-value <0.01, 1,000 permutations); (b) second, we compared the zebrafish iridophore gene set to the 86 A. ocellaris genes expressed at significantly higher levels in white skin than in orange skin (Supporting Information Table S3). This comparison revealed 12 genes in common between A. ocellaris white skin and zebrafish iridophores, but only one gene between A. ocellaris orange skin and zebrafish iridophores (Supporting Information Figure S4b), a difference significantly greater than expected by chance (chi-square test, p < 1e-16).
Included among these 12 genes were four genes inferred previously to function in iridophore development, or likely to do so, including genes encoding products involved in the synthesis of purines, crystals present in iridophore reflecting platelets conferring their iridescence or white coloration (Fujii, 1993). For example, pnp4a encodes an enzyme required for purine synthesis mutated in the medaka iridophore mutant guanineless (Kimura et al., 2017) and is used as a marker of iridophores in zebrafish (Eom et al., 2015;Patterson & Parichy, 2013), whereas prtfdc1 encodes an enzyme whose human ortholog is involved in purine metabolism (Welin et al., 2010). tfec encodes a basic helix-loop-helix transcription factor that marks the embryonic iridophore lineage in zebrafish (Lister, Lane, Nguyen, & Lunney, 2011) and fhl2a is a teleost-specific duplicate of fhl2b, encoding a four-and-half LIM domain transcription factor that has been recruited evolutionarily for iridophore-containing ornamental "eggspots" on the fins of cichlids (Santos et al., 2014).
Only one gene in the zebrafish iridophore gene set was found among the 83 DEGs in orange skin (Supporting Information Figure   S4b): fmn2, encoding a formin protein, with no known pigmentation function (Higdon et al., 2013). We conclude that white bar pigment cells in A. ocellaris share similarities with iridophores at a gene expression level. Moreover, the genes differentially expressed in the orange skin provide a clear signature of xanthophores and could therefore be a useful resource to study this cell type in clownfish (not shown).

| Iridophores are important for white color formation
We then used a pharmacological approach to test whether cells related to iridophores are responsible for white coloration in clownfish. In zebrafish, iridophores require leukocyte tyrosine kinase (Ltk) signaling (Lopes et al., 2008) and iridophores are ablated when fish are treated with TAE684 (TAE; Colanesi et al., 2012;Fadeev et al., 2016;Rodrigues et al., 2012), an inhibitor of Ltk and anaplastic lymphoma kinase (Alk). Transcriptomic data from white skin of A. ocellaris revealed expression of both alk and ltk, with alk being over-expressed in white skin compared to orange skin (Supporting Information Table S3)

| fhl2a, fhl2b, saiyan, gpnmb, and apoD1a are differentially expressed in white skin and iridophores
To validate transcriptomic results, we assessed by qRT-PCR the expression of five genes identified as most differentially expressed between white and orange skin on independent skin samples in A.
ocellaris (Figure 3). For each locus, the directionality of the difference in expression was concordant between RNA-Seq and qRT-PCR ( Figure 3b). The only exception was clec19a for which the expression difference was non-significant. Moreover, to test the generality of these results we also examined expression in a second Amphiprion F I G U R E 5 fhl2a, fhl2b, saiyan, and gpnmb are essential for development of iridophores in zebrafish. (a) Quantification of dorsal iridophores in zebrafish 4 dpf larvae. Each dot corresponds to one individual. Asterisks indicate significant differences in iridophores complements (non-parametric comparisons with control using Steel method; **p < 0.01, ***p < 0.001). Although sizes differed between larvae that were controls and those that carried somatically induced mutations, these differences in length did not themselves account for variation in iridophore numbers as indicated by preliminary analysis of covariance that identified significant overall differences among groups (F 4,88 = 4.13, p < 0.005) even after controlling for potential overall effects of standard length (F 1,88 = 0.05, p = 0.8), as well as potential genotype × standard length interaction (F 4,88 = 1.18, p = 0.3; using inverse-transformed counts to normalize residuals to meet model assumptions). (b) Typical phenotypes of uninjected controls and individuals mosaic for somatically induced mutations, with insets to illustrate dorsal iridescent iridophores used in analyses. In wt individuals, numerous iridophores (located with the arrowhead) were present along dorsal midlines, whereas injected individuals had fewer iridophores even after controlling for differences in body length [Colour figure can be viewed at wileyonlinelibrary.com] species. In A. frenatus, whose adults contain a unique white bar positioned on the head (Salis et al., 2018) on a dark reddish background, we found that white and orange skin genes were also differentially expressed as in A. ocellaris, with the exception of ttc39b (Figure 3c).
Among the 86 genes more highly expressed in white skin than orange skin, fhl2a and fhl2b-already known to be expressed in iridophores (Santos et al., 2014)-and gpnmb had higher expression in zebrafish iridophores than melanophores (Higdon et al., 2013).
Apolipoprotein D (apoD1a) is expressed specifically in cichlids eggspots but its function in pigmentation, if any, has not been characterized (Gu & Xia, 2017). Finally, the gene with the second highest change in white skin was an ortholog to zebrafish gene ENSDARG00000055172 (Ensembl)/795494 (NCBI; Figure 3a) with no associated function in Ensembl, NCBI, or ZFIN databases. This gene is currently annotated as si:ch211-256m1:8, and we refer to it as saiyan in reference to the manga comics characters that can become shiny.
White bars appear late during post-embryonic development, starting at stage 4 and being well visible at stage 6 (N. Roux, P. Salis, and V. Laudet, in preparation; Salis et al., 2018). To test for an association between expression of these five genes and bar ontogeny, we used a transcriptomic dataset for skins across multiple post-embryonic stages. As shown in Supporting Information Figure S5, all five genes had minima of expression prior to overt bar development at stage 4, followed by increases through the climax of bar development at stage 6 and drops in expression by early juvenile stage 7.
To assess the spatial localization of these transcripts and confirm the presence of all five genes in iridophores, we performed in situ hybridization on A. ocellaris white and orange adult skin. fhl2a, fhl2b, and apoD1a were expressed in the stratum laxum of the dermis, mainly composed of iridophores as observed in Figure 1b, suggesting that those genes are effectively expressed in iridophores ( Figure 4b,d,f, arrowheads). None of those three genes were expressed in orange skin sections (Figure 4c,e,g), and no signal was observed with the sense probes in the white sections (Supporting Information Figure S6). In contrast, we could not detect a clear difference in expression by in situ hybridization between white and orange skins for saiyan and gpnmb (not shown). To confirm the association of fhl2a, fhl2b, and apoD1a with iridophores, we looked at their expression in TAE-treated fishes in which no iridophores were observed ( Figure 2a). Interestingly, those genes are not anymore expressed in the white skin of fishes treated for 20 days in TAE 0.6 µM (Supporting Information Figure S7).
Taken together, these data show that at least the three genes fhl2a, fhl2b, and apoD1a are specifically expressed in iridophores.

| fhl2a, fhl2b, saiyan, and gpnmb are required for iridophore development in zebrafish
To test whether fhl2a, fhl2b, apoD1a, gpnmb, and saiyan are essential for iridophore development, we assayed their function in zebrafish. We predicted that if these genes are required for iridophore lineage, then loss of function mutations should result in iridophore deficiencies at early stages. Accordingly, we generated fish mosaic (F 0 ) for somatically induced CRISPR/Cas9 mutations and compared the numbers of dorsal iridophores that developed with those of wild-type sibling controls. Given that fhl2a and fhl2b are paralogous loci, we tested their effects simultaneously.
At 4 days post-fertilization, fish mosaic for mutations in apoda.1 (apoD1a clownfish ortholog) did not have iridophore defects but those harboring mutations in fhl2a/fhl2b, gpnmb, and saiyan exhibited significantly fewer iridophores than uninjected siblings, even after controlling for variation in standard length ( Figure 5).
Thus, most of those DEG genes in A. ocellaris white skin were required for early iridophore development in zebrafish, consistent with A. ocellaris pigment cells of white skin representing a specific class of iridophore required for generating the striking white bars of clownfish.

| D ISCUSS I ON
Our findings suggest that clownfish have iridophores enriched in white bars and that these cells are critically important in producing the white color, though we cannot formally exclude contributions of other mechanisms or cell types (e.g., dermal cells of the stratum laxum).
First, microscopic analyses showed that cells with a typical iridophore morphology (round shape cells with flattened reflecting platelets) were abundant in white bars (Figure 1; Fujii, 1993;Schartl et al., 2015). The iridophores in clownfish white bars were similar to zebrafish S-type iridophores that contain densely stacked purine platelets with a platelet size of ca. 0.5-2 µm, as compared to L-iridophores that have larger purine platelets (ca. 10 µm) organized parallel to the dermis (Hirata et al., 2003). This latter type was not observed. The relationship between the iridescent cells that we see in light microscopy and the iridophores that we see in electron microscopy-likely responsible for the white color-is still unclear. The radial arrangement of iridophore platelets observed in TEM is a likely contributor to the matte appearance (Fujii, 1993).
However, we have not seen in our sections iridophores with a parallel or at least non-radial organization that would contribute to Tyndall scattering required for iridescence. Therefore, future work should clarify the existence of possible subtypes among clownfish iridophores.
Second, using transcriptomic analysis, we independently confirmed that cells with transcriptional profiles similar to iridophores contribute to white but not orange skin ( Figure S4). We compared genes specific to zebrafish iridophores (Higdon et al., 2013) with genes differentially expressed in clownfish white skin and showed that overlap between datasets was significantly greater than would be expected by chance, showing that white skin in clownfish had a transcriptomic signature of iridophores. In particular, 12 genes were common to both datasets and could thus constitute conserved iridophore markers within actinopterygians. Although significant, this overlap was relatively small compared to the size of respective clownfish (86) and zebrafish sets of genes (237). This presumably reflects differences in sample origins: Zebrafish RNA-Seq data were obtained from purified iridophores, whereas clownfish data were derived from whole skin in which many other cell types are present (Higdon et al., 2013). Thus, clownfish data represented a more diverse set of cells compared to the zebrafish dataset, though biological species differences in iridophore transcriptomes may also explain some of the differences between both datasets. These and other possibilities will be resolvable with transcriptomic profiling of purified populations of clownfish iridophores.
Third, we retrieved previously identified iridophore markers in the list of the most DEGs in white skin (Figure 3). Among the 12 genes found in common between the clownfish white bar and zebrafish iridophores, 4 (pnp4a, prtfdc1, tfec, and fhl2a) were already demonstrated or inferred to be important for the development or function of iridophores in teleost fishes (Kimura et al., 2017;Lister et al., 2011;McMenamin et al., 2014;Petratou et al., 2018;Santos et al., 2014;Welin et al., 2010).
Fourth, our pharmacological experiments showed that treatment with TAE 684, an inhibitor of Ltk and Alk tyrosine kinase receptors expressed by zebrafish iridophores (Colanesi et al., 2012;Fadeev et al., 2016Fadeev et al., , 2018Lopes et al., 2008;Rodrigues et al., 2012), reduced the number of presumptive iridophores at an ultrastructural level and concomitantly reduced the white hue of the bars at the organismal level ( Figure 2). Both alk and ltk were expressed in cells of white skin, alk being more highly expressed in white skin than orange skin (Supporting Information Table S3).
Therefore Ltk, Alk, or both are required for iridophores. Here, we extended previous analyses by showing that TAE 684 treatment affects iridophores and the white hue of juvenile fish even after bars have formed (Salis et al., 2018). Interestingly, early blockade of bar formation prevented subsequent normal development of posterior bars, but not the most anterior ones (Figure 2b). Moreover, anterior bars that had developed during early TAE treatment were able to regenerate a nearly normal white appearance after removal of TAE, whereas posterior bars that failed to develop initially under these conditions also failed to develop subsequently.
Anterior-posterior differences in TAE sensitivity suggest intrinsic differences between bars, and perhaps distinct Ltk-or Alk-mediated processes, worthy of further investigation.
Fifth, we found that zebrafish with somatic mutations in orthologs of three to four genes most differentially expressed in clownfish white skin had significantly fewer iridophores compared to controls ( Figure 5). Owing to the mosaicism of these F 0 fish, it is likely that not all cells carried mutant alleles and so these analyses likely underestimate the magnitude of phenotypic effects attributable to loss of function mutations. Among the genes tested (fhl2a and 2b being tested together because of their sequence relatedness), only one (apoda.1) failed to generate a clear phenotype.
Lastly, our data are not consistent with clownfish white skin pigment cells being leucophores, as opposed to iridophores. Leucophores are dendritic cells with round organelles that lack an overt organization (Schartl et al., 2015), distinct from the round cells with flat, highly organized organelles that we observed in clownfish. To date, only one gene, slc2a15b, has been documented to have expression specific to leucophores . This gene was not expressed detectably in clownfish skin samples, whereas its teleost-specific paralog, slc2a15a, was expressed at comparable levels between white and orange skin. Eventual identification of additional leucophorespecific markers will enable future studies of these questions.
The comparison of white and orange colored skin provided a set of genes differentially expressed in each condition, some of which might constitute new iridophore and xanthophore markers. Among the 10 most DEGs in clownfish white bar, four were previously associated with confirmed or putative iridophore differentiation processes: fhl2a and fhl2b, tfec, and gpnmb. As discussed above, fhl2 genes were associated with iridophores found in cichlid egg-spots (Santos et al., 2014) and were expressed in zebrafish iridophores (Higdon et al., 2013). Gpnmb was also expressed in iridophores of zebrafish (Higdon et al., 2013) and cichlids (Santos et al., 2014).
Among the 10 most DEGs in white skin, we identified two other genes (gja5 and trpm1a) associated with pigmentation although not necessarily in iridophores. Mutations in gja5 result in the zebrafish leopard mutant phenotype, characterized by a spotted pattern (Irion et al., 2014;Watanabe et al., 2006). In addition, we identified several genes that, to our knowledge, have not been previously assigned roles in pigmentation. Three of those (ada2, prtfdc1, and slc2a9) are involved in purine metabolism (Pei et al., 2016;Welin et al., 2010), consistent with a function in platelet formation.
Ranked second within the most DEGs in white skin was saiyan also identified in the comparison between both zebrafish and iridophore datasets. Its role in iridophore development was functionally confirmed in zebrafish, as mosaic CRISPR/Cas9 F 0 fish mosaic for saiyan mutations harbored ~40% fewer iridophores compared to wild type ( Figure 5).

| CON CLUS ION
We have shown that the white color of an iconic fish species, the clownfish A. ocellaris, is composed of iridophores, and we identified several new genes of likely importance to the development or maintenance of these cells. This analysis opens the door to functional analyses of the extraordinarily diverse pigmentation patterns of coral reef fishes and will allow for integrated studies relating developmental genetics and evolution of pattern formation to the fascinating behaviors and ecologies of these fishes. to D.M.P.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.