Gain-of-function mutations in Aqp3a influence zebrafish pigment pattern formation through the tissue environment

The development of the pigmentation pattern in zebrafish is a tightly regulated process that depends on both the self-organizing properties of pigment cells and extrinsic cues from other tissues. Many of the known mutations that alter the pattern act cell-autonomously in pigment cells, and our knowledge about external regulators is limited. Here, we describe novel zebrafish mau mutants, which encompass several dominant missense mutations in Aquaporin 3a (Aqp3a) that lead to broken stripes and short fins. A loss-of-function aqp3a allele, generated by CRISPR-Cas9, has no phenotypic consequences, demonstrating that Aqp3a is dispensable for normal development. Strikingly, the pigment cells from dominant mau mutants are capable of forming a wild-type pattern when developing in a wild-type environment, but the surrounding tissues in the mutants influence pigment cell behaviour and interfere with the patterning process. The mutated amino acid residues in the dominant alleles line the pore surface of Aqp3a and influence pore permeability. These results demonstrate an important effect of the tissue environment on pigment cell behaviour and, thereby, on pattern formation. Summary: Dominant mutations in the water channel Aquaporin 3a cause defective pigment patterning in zebrafish, due at least in part to an effect of the mutant tissue environment on the pigment cells.


INTRODUCTION
The zebrafish adult colour pattern of alternating blue and gold stripes is the result of spatially and temporarily controlled interactions among pigment cells (chromatophores): yellow xanthophores, silvery iridophores and black melanophores. In the light stripes, tightly packed light-reflective iridophores are covered by a layer of compact xanthophores (Hirata et al., 2003(Hirata et al., , 2005. Melanophores are confined to the dark stripes, where they are overlaid by a network of blue iridophores and loose xanthophores. This pattern arises during metamorphosis (3-6 weeks post fertilization), when iridophores migrate into the skin to form the first light stripe along the horizontal myoseptum. Next, metamorphic melanophores appear in the prospective regions of the first two dark stripes. At the edges of the light stripe, the iridophores disperse and migrate dorsoventrally into the adjacent dark stripe regions. Then, upon reaching the positions of the next light stripes, they become tightly packed again. At this point, xanthophores evenly cover the skin, increasing in density in response to tightly packed iridophores and dispersing in response to melanophores (Mahalwar et al., 2014). The whole process is repeated to form the next stripes. The adult zebrafish trunk can contain up to five dark stripes interspersed with light stripes .
The formation of the pattern requires the presence of all three types of chromatophores and their mutual interactions. A number of regulators of pigment cell differentiation, survival and behaviour are known (reviewed by Singh and Nüsslein-Volhard, 2015). Many are transcription factors or signalling molecules involved in cell specification or survival and thus affect the pigment cells or their precursors directly. Others, such as ion channels, connexins and Tight junction protein 1A, are required for cell-cell communications, which are crucial for the spatial arrangement of chromatophores and the cell shape transitions of iridophores and xanthophores (Maderspacher and Nüsslein-Volhard, 2003;Watanabe et al., 2006;Irion et al., 2014;Fadeev et al., 2015;Mahalwar et al., 2016).
Pattern formation cannot be attributed entirely to self-organizing properties of the pigment cells, and not all regions of the body are patterned equally. In fact, the dorsal-and ventral-most parts lack stripes, despite the presence of pigment cells. Very little is known about the role of the external milieu in determining pigment cell behaviour. Extrinsic regulators of stripe formation include the horizontal myoseptum, a general anatomical pre-pattern that directs where iridophores initially emerge in the skin to form the first light stripe . In addition, external ligands can locally activate receptors expressed in pigment cells, e.g. Csf1r in xanthophores or Ednrb1a (Ednrba) in iridophores Parichy et al., 2000b). Endocrine regulation also plays a role: thyroid hormone promotes xanthophore proliferation and represses melanophores (McMenamin et al., 2014). Finally, the communication between chromatophores via gap junctions and potassium channels may be regulated by the local tissue environment. Their conductivity depends on the polyamine spermidine (Ficker et al., 1994;Watanabe et al., 2012), and a mutation in the gene encoding spermidine synthase affects stripe width (Frohnhöfer et al., 2016).
Here, we describe zebrafish mau mutants, which have gain-offunction mutations in the gene encoding Aquaporin 3a (Aqp3a) that affect pigment patterning and fin growth. Aquaporins are tetrameric pore proteins that facilitate water diffusion through biological membranes while excluding ions. One subgroup, the aquaglyceroporins, can also transport non-polar solutes, such as glycerol, peroxide and urea. Some aquaporins can be permeable for other unconventional solutes such as ammonia and hydrogen peroxide, as well as some anions (Wu and Beitz, 2007). The zebrafish genome encodes a large repertoire of aquaporins, many of which show tissue-specific expression (Tingaud-Sequeira et al., 2010).
Point mutations in aqp3a result in a dominant phenotype with broken, undulating stripes and short fins. The most severely affected pigment cells in mau mutants are larval xanthophores, which deteriorate before metamorphosis and only partially recover during later development. We identified four alleles with altered amino acid residues that line the pore surface of Aqp3a and change pore permeability. A loss-of-function allele, generated by the CRISPR-Cas9 system, has no phenotypic consequences, indicating that Aqp3a is dispensable for normal development. In contrast to other zebrafish mutants with spotted patterns, namely leo, luc, sbr or seurat (Maderspacher and Nüsslein-Volhard, 2003;Irion et al., 2014;Fadeev et al., 2015;Eom et al., 2012), all three types of pigment cells from dominant mau mutants are capable of forming a wild-type pattern when surrounded by wild-type tissues, but wild-type pigment cells fail to do so when surrounded by mutant tissues. Our results emphasize the importance of the tissue environment in regulating pigment cell behaviour during pattern formation.

mau mutants have missense mutations in aqp3a
In several screens for zebrafish mutants with pigment pattern defects we identified four mau mutant alleles (tVE1, tXGU1, tVBU1, tWGU2), which exhibited similar phenotypes with undulating, broken stripes and short fins ( Fig. 1A-E,K-M). The mutations are dominant, with heterozygous fish showing a similar but weaker phenotype (Fig. 1F,H). The mutations were mapped to a 3.7 Mb region on chromosome 5 between markers z9568 and z13641. Because this region shows a very low frequency of sequence polymorphisms among our mapping strains, we performed wholegenome sequencing of fish homozygous for the allele tVE1. The only candidate for a phenotype-causing mutation in the entire region was a missense mutation in exon 5 of the aqp3a gene. Sequencing of aqp3a in the other three mau mutant alleles also identified missense mutations (Table 1, Fig. 2A).
The mutations affect conserved amino acid residues of Aqp3a: R220, which is mutated in the strongest allele tVE1, invariably follows the NPA motif that forms the water selectivity filter (Sui et al., 2001); V201, which is mutated in tXGU1, and R225, which is mutated in tVBU1 and tWGU2, are conserved among aquaglyceroporins ( Fig. 2A) (Tingaud-Sequeira et al., 2010).
The pore permeability of Aqp3a is affected in mau mutants We modelled the zebrafish Aqp3a protein structure based on the known structure of the E.coli glycerol transporter GlpF, which is homologous to animal aqua-glyceroporins (Fu et al., 2000;Tajkhorshid et al., 2002). The model predicts that all of the affected amino acids line the pore surface (Fig. 2B) and are thus likely to affect pore permeability. To assess this, we tested the permeability for water, glycerol and peroxide, which are known cargos of Aqp3 (Ma et al., 1994;Echevarria et al., 1994;Miller et al., 2010;Tingaud-Sequeira et al., 2010), of wild-type and mutant zebrafish Aqp3a in Xenopus oocytes. Permeability for water and glycerol was significantly reduced in all mutants, whereas peroxide permeability increased by 80% in tVE1, but was reduced in the other alleles (Fig. 2C,D). Taken together, these results suggest that the structure and functionality of the Aqp3a channel are altered in the dominant mau mutants. To test if the dominance observed in mau mutants is due to haploinsufficiency or caused by a gain-of-function, we used the CRISPR-Cas9 system (Hwang et al., 2013) to target aqp3a exon 4, which is upstream of the identified point mutations. A 4 bp deletion induced in mau embryos homozygous for the tVE1 allele leads to complete reversion of the phenotype, giving rise to heterozygous and homozygous fish of wild-type appearance (Fig. 1G,I). This deletion is predicted to be a null allele, as it causes a frameshift followed by a premature stop codon resulting in a truncated 158 amino acid polypeptide that lacks the last three transmembrane domains. The mutation leads to the downregulation of the corresponding transcript (Fig. 2E,F), possibly owing to nonsense-mediated mRNA decay. A comparable truncation, caused by exon 5 skipping, has been The loss-of-function mutant aqp3a −/− , generated in the background of aqp3a tVE1/tVE1 , has a wild-type phenotype in homozygous and heterozygous fish. (J) The aqp3a tVE1/− phenotype is similar to that of aqp3a tVE1/+ . (K) Ratio of caudal fin length to standard body length (SL) in mau mutants. For each allele, seven to nine fish were measured. Mean±s.d. **P<0.01 for mutant versus wild type (one-way ANOVA followed by Tukey-Kramer multiple comparisons test). (L) Number of fin ray segments in the third ray of the caudal fin in wild-type and aqp3a tVE1/tVE1 fish. n=8. Mean±s.d. ***P<0.0001 (unpaired t-test). (M) Length of fin ray segments in the third ray of the caudal fin in wild-type and aqp3a tVE1/tVE1 fish. n=8. Mean±s.d. n.s., not significant (unpaired t-test).
reported in humans where it leads to a complete loss of AQP3 in red blood cells (Roudier et al., 2002).
To test whether the point mutations in Aqp3a might affect protein folding, we fused wild-type and mutant proteins to Strep-tag and super-folded GFP (sfGFP). When expressed in HeLa cells (Fig. S1A,B) or in the fin fold of zebrafish embryos (Fig. 3), wild-type and tVE1 fusion proteins localized to the plasma membrane; however, the remaining three mutant alleles showed either partial (tXGU1) or strong (VBU1 and WGU2) retention in the endoplasmic reticulum (ER). Interestingly, the amount of the mutant protein delivered to the plasma membrane, bypassing ER retention, appeared to correlate with the strength of the mau phenotype.
Together with the observations that heterozygous mau mutants have a weaker phenotype than homozygotes, and that the phenotype of aqp3a tVE1/− fish is similar to that of aqp3a tVE1/+ (Fig. 1F,J), these data suggest that the aqp3a mutations in mau mutants are neomorphic and act dominantly in a dose-dependent manner.

Chromatophores are affected indirectly in mau mutants
To test whether the patterning defects in mau mutants are caused by pigment cells, we transplanted mau mutant blastomeres into embryos of nacre (mitfa), pfeffer (csf1ra) and transparent (mpv17) mutants, which lack melanophores, xanthophores and iridophores, respectively (Lister et al., 1999;Odenthal et al., 1996;Krauss et al., 2013) (Fig. 4A-C). In all three cases, mutant pigment cells contributed to the wild-type pattern in the resulting chimeric animals ( Fig. 4D-F), indicating that the mau phenotype is not based on mutant pigment cells alone. By contrast, when wild-type blastomeres carrying Tg(TDL358:GFP)-labelled iridophores were transplanted into mau mutant embryos, they failed to form proper stripes (Fig. 4G). This suggests that although mau pigment cells are normal and capable of forming the wild-type pattern, the mutant tissue environment prevents them from doing so.

Development of the mau phenotype
The early larval pigment pattern of mau mutants appears similar to wild type (data not shown). To understand the changes in pigment cell development and behaviour that create the mau pattern, we followed the development of homozygous tVE1 fish carrying markers for the different pigment cells. The first pigment cells that will later contribute to the adult pattern are the larval xanthophores (Mahalwar et al., 2014). At 2 days post fertilization (dpf ) they cover the entire flank of mau fish, similar to wild type. In wild type the larval xanthophores persist and begin to proliferate at the onset of metamorphosis (Mahalwar et al., 2014); however, in mau, the xanthophores lose contact with each other (visible at 10 dpf ), begin to deteriorate, and by 15 dpf mostly disappear from the flank. Later on, they partially recover and persist to adulthood (Fig. 5A,B).
During early metamorphosis, iridophores are similar in number and appearance in mau and wild-type fish [ Fig. 5C, standard length (SL) 5.9 mm]. Later, the light stripe boundaries in mau are often less well defined compared with wild type (Fig. 5C, SL 6.9 mm). The proliferation rates of iridophores are also comparable between wild type and mau (Fig. 5D). Clear differences become apparent by the time the second dorsal light stripe is formed. Melanophores are present in normal numbers ( Fig. 5F) in mau, and their appearance and density are normal; however, they are more dispersed (Fig. 5C, SL 9.3, 10.6 and 12.7-13 mm; Fig. 5E). Dense iridophores do not adopt a contiguous arrangement in the region of the second light stripe, even though patches of dense iridophores are present in this area ( Fig. 5C, SL 12.7-13 mm).
Although all three types of chromatophores initially appear normal, the sharpening of dark and light stripes is affected. Transient loss and recovery of larval xanthophores is the most striking difference in chromatophore development between mau and wild type.

Effect of the tissue environment on the behaviour of pigment cells
Aqp3a is expressed in most tissues in zebrafish, including skin and muscle (Hamdi et al., 2009;Tingaud-Sequeira et al., 2010). To address which tissues affect the behaviour of the chromatophores in mau mutants, we expressed the tVE1 variant (i.e. Aqp3aR220Q) fused to sfGFP under the ubiquitous ef1a promoter (Fig. 6A) and analyzed the phenotypes arising from mosaic overexpression of the protein in random tissues.
We observed pattern irregularities, the appearance of ectopic melanophores in the light stripe regions, and melanophore and iridophore death when large epidermal clones coincided with large muscle clones (Fig. 6B, top). Neither large muscle clones (Fig. 6B, bottom) nor epidermal clones alone were sufficient to affect the pigment pattern. Additionally, the expression of Aqp3aR220Q in xanthophores did not affect stripe formation (Fig. 6B, bottom), substantiating our transplantation experiment results. No effect on pigment patterning was observed upon overexpression of the wildtype variant of Aqp3a (Fig. 6C). We also created a transgenic line expressing sfGFP-Aqp3a wild type (WT) or R220Q under the epidermis-specific keratin promoters krtt1c19e [basal layer of epidermis, which gives rise to all the strata of adult skin (Lee et al., 2014)] and krt4 [also known as krt8, enveloping layer and periderm-specific (Gong et al., 2002)]. Neither of these transgenes had any effect on the trunk patterning, although Tg(krtt1c19e: sfGFP-Aqp3aR220Q) fish showed slight defects in fin pigmentation (e.g. the stripes do not reach the margin of the caudal fin). Tg(krt4: Aqp3aR220Q) fish occasionally exhibited spotted fins resembling the pigmentation of heterozygous tVE1 fish and weaker mau mutant alleles (Fig. 6D,E). Tg(krt4:sfGFP-Aqp3aR220Q) fish have short fins, similar to aqp3a tVE1/+ (Fig. 6F). Although the krt4 promoter is known to drive expression in the periderm and simple epithelia (Imboden et al., 1997;Gong et al., 2002), we observed weak fluorescence in fin ray segments of post-metamorphic and adult fish (Fig. 6G,H), and this expression is likely to be responsible for the short fin phenotype.

DISCUSSION
We describe novel mutations in aqp3a in zebrafish that result in an aberrant pigmentation pattern as well as short fins. We show that pigment cells are not directly affected, but that the tissue environment is changed by the mutations. Aqp3a is known to facilitate the transmembrane transport of water and small non-polar solutes such as glycerol, peroxide and urea (Tingaud-Sequeira et al., 2010). The functionality of the pore is determined by six transmembrane domains and two semitransmembrane loops, which work as a water selectivity filter and exclude ions from the pore (Sui et al., 2001). All four dominant mau mutant alleles have amino acid exchanges in the pore surface of the protein. The strongest allele (tVE1) replaces an invariant arginine residue of the water selectivity filter with glutamine, while the three weaker alleles affect other conserved residues. Each mutation alters the pore permeability for water and glycerol, which are known cargos for Aqp3 in mammals (Ma et al., 1994;Echevarria et al., 1994;Miller et al., 2010). However, the decrease in water and glycerol transport cannot explain the dominant mau phenotype, as the complete loss-of-function mutation reverts the mau phenotype to wild type. Our preliminary data suggest that there are alterations in calcium homeostasis in the skin of mau mutants (A.E., U.I. and C.N.-V., unpublished). In the three weaker alleles we find ER retention of the mutant proteins, implying that protein folding might be affected in addition to permeability, resulting in reduced amounts of protein reaching the membrane. This might lead to the attenuation of the phenotype compared with tVE1.
Mutations in the human gene leading to truncations of AQP3 also have no phenotypic consequences (Roudier et al., 2002). An Aqp3 null mutation in mice results in a weak phenotype with excess urine production and mild defects in skin thickness and hydration (Ma et al., 2000;Hara-Chikuma and Verkman, 2008). The fact that the aqp3a loss-of-function mutation that we generated by CRISPR-Cas9 shows no phenotype might also be due to redundancy between aqp3a and aqp3b, the two paralogues present in the zebrafish genome (Tingaud-Sequeira et al., 2010).
All three types of chromatophores from mau mutants can contribute to a normal pigment pattern in a wild-type environment. The development of pigment cells in mau shows that all three cell types initially appear normal, but later fail to form a proper striped pattern: the boundaries between the light and dark stripes are not well defined and the stripes break up into spots. Interactions between chromatophores are required for their proliferation, survival and normal behaviour, and are essential for pigment pattern formation (Frohnhöfer et al., 2013;Singh and Nüsslein-Volhard, 2015;Parichy and Spiewak, 2015;Walderich et al., 2016;Mahalwar et al., 2016). In mau mutants the most severely affected cells are the larval xanthophores, which lose contacts and mostly disappear from the flanks of the fish shortly before metamorphosis. Larval xanthophores give rise to metamorphic xanthophores, which in mau are present in lower numbers and with an altered morphology. As xanthophores are   3. GFP-Aqp3a localization in zebrafish fin fold cells. mRNA of GFP-Aq3a was microinjected at the 1-cell stage and the protein localization was imaged at 1dpf. Wild-type and Aqp3aR220Q (tVE1) proteins reach the plasma membrane, whereas Aqp3aR225L (tVBU1) and Aqp3aR225H (tWGU2) mutants are retained intracellularly. Aqp3aV201G (tXGU1) shows partial retention. Nuclei are stained with Hoechst 33342. Scale bars: 25 µm.  Development (2017Development ( ) 144, 2059Development ( -2069Development ( doi:10.1242 required for shaping the dark stripes (Parichy and Turner, 2003;Frohnhöfer et al., 2013;Iwashita et al., 2006), the abnormal distribution of melanophores that we find in mau could be the result of impaired xanthophore-melanophore communication. In pfeffer (csf1ra) mutants, which completely lack xanthophores, the melanophores are scattered outside of the dark stripe region and their number is reduced (Parichy et al., 2000a;Parichy and Turner, 2003). However, unlike pfeffer mutants, mau mutants have the normal number of melanophores, suggesting that there are enough xanthophores to support the survival of melanophores. It is conceivable that the abnormal positioning of melanophores during stripe formation in mau mutants prevents iridophores from aggregating and forming a regular second light stripe, ultimately yielding a broken striped pattern. Consistent with this, pfeffer and mau double mutants are similar to pfeffer mutants in terms of pigment pattern (data not shown). Thus far, no mutants have been described that resemble mau in terms of larval xanthophore development. However, temporal inactivation of Csf1ra during the early stages of xanthophore development, using a temperature-sensitive csf1ra allele, and then reactivation during metamorphosis leads to a scattering of melanophores across the flank. Interestingly, fish carrying a temperature-sensitive allele and a null allele for csf1ra show stripe breakages (Parichy and Turner, 2003) resembling the pattern that we observe in heterozygous tVE1 fish. Nonetheless, the mau phenotype is stronger than phenotypes caused by temporary loss of xanthophores and of Csf1ra signalling (Parichy and Turner, 2003), implying that the cellular behaviour of the other two pigment cell types (melanophores and iridophores) is also affected in the mau mutant environment.

RESEARCH ARTICLE
The failure of wild-type pigment cells to organize into stripes when transplanted into mau mutants suggests that they receive the wrong cues from non-pigment cells or the extracellular environment. As Aqp3a is a pore protein and the mutant variants act in a dose-dependent manner, it seems likely that they affect the pigment cells and fin growth locally. Skin pigment cells in zebrafish are located in the hypodermis. They are overlaid by layers of epidermal keratinocytes and loosely packed dermal fibroblasts (Hirata et al., 2003(Hirata et al., , 2005. Beneath them are thin layers of slow and fast muscle bundles (van Raamsdonk et al., 1982). We found that expression of the tVE1 variant (Aqp3aR220Q) in large areas of both the epidermis and the underlying muscles leads to some stripe disruption and the appearance of ectopic melanophores within the light stripe region, resembling the mau phenotype. Smaller patches of cells do not yield such a phenotype, nor do clones that are restricted to muscle or epidermal cells only, suggesting that a minimal number of cells in the adjacent tissues is needed to affect the behaviour of the chromatophores.
Pigmentation in zebrafish presents an attractive system with which to study the cellular basis of pattern formation processes in vivo. The cell-autonomous interactions between the three types of chromatophores, which are necessary to produce the stripes, occur in a tissue-specific context. Our study emphasizes the role that the cellular environment plays in pigment cell behaviour and pattern formation, and the role that it might play in the immense diversification of pigment patterns in fish and other ectothermic vertebrates.

Genetic mapping
Heterozygous offspring of a cross between tWGU2 and WIK were used for meiotic mapping as described (Nüsslein-Volhard and Dahm, 2002). The mutation was mapped to the region between microsatellite markers z9568 (36.7 cM) and z13641 (41.4 cM) on chromosome 5.
For next-generation sequencing, a genomic DNA library was produced from a single homozygous tVE1 mau mutant and WIK sibling fish using the NEBNext DNA Library Kit (NEB) and sequenced on a HiSeq 2000 Sequencing System (Illumina). The sequences were assembled with CLC Genomics Workbench (Qiagen), using Zv9 (GCA_000002035.2) as a template.

Cell culture, transfection and immunostaining
HeLa cells (ATCC CCL-2) were cultured in DMEM (Sigma-Aldrich) containing 10% fetal calf serum (FCS), 2 mM glutamine, 100 U/ml penicillin and 100 µg/ml streptomycin. Transfection of Strep-tagged Aqp3a (KY379154, KY379155, KY379156, KY379157, KY379158) was achieved using Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol. Twenty-four hours after transfection, the cells were fixed in 4% paraformaldehyde, washed with 0.1% Triton X-100 in PBS (PBST), blocked in 10% normal goat serum (NGS) in PBST and incubated with anti-strep II antibody (1:200; IBA, 2-1507-001). For subsequent Fig. 5. Development of the mau phenotype in aqp3a tVE1/tVE1 . (A) Xanthophores labelled by Tg(fms:GAL4; UAS:mCherry) in wild-type and in aqp3a tVE1/tVE1 (mau) larvae. By 15 dpf, the numbers and the area covered by xanthophores are reduced in the mutants, but xanthophores later recover. (B) Numbers of xanthophores per segment in the mid-trunk of premetamorphic fish (n=8-10 fish); data points represent mean±s.e.m. ***P<0.0001, unpaired t-test. (C) Representative example of recurrent imaging of individual fish carrying Tg(TDL358:GFP) (green, labelling iridophores) and Tg(sox10:mRFP) (magenta, labelling all the pigment cells) in wild type and in mau. Standard length (SL) is indicated. White arrows point to contracted xanthophores [labelled in mau, Tg(sox10:mRFP)], whereas the wild-type xanthophores are spread and cover the entire flank of the fish. Yellow arrows indicate the position of the second light stripe. In mau, iridophores are present in the area but fail to form a continuous light stripe. (D) Increase in iridophore numbers per segment over the course of 4 days. The quantifications were obtained from recurrent imaging in the mid-trunk segments of the fish during the initial formation of the light stripe (SL 6-6.5 mm), and upon completion of the first light stripe (SL 6.5-7.5 mm). (E) mau melanophores are scattered and not confined to the dark stripe region. (F) Numbers of flank melanophores per segment in the mid-trunk (two segments above the anus) imaged in metamorphic wild-type and mau fish. Each data point represents the number of melanophores in one segment. Scale bars: 50 µm in A; 250 µm in C.
We expressed sfGFP-Aqp3a variants transiently in zebrafish embryos to assess protein localization. sfGFP was fused in-frame to the coding sequence of Aqp3a and cloned into pCS2+ (the resulting constructs were KU645301, KU645302, KU645303, KU645304, KU645305). The linearized plasmids were transcribed in vitro using the SP6 mMessage Machine Kit (Ambion), and 100 ng/µl of mRNA was injected into 1-cell stage embryos.

Transplantations
Chimeric fish were produced as described (Nüsslein-Volhard and Dahm, 2002) by transplanting 20-40 blastomeres from donor to host embryos at the 1000-cell blastula stage.

Image acquisition and processing
Fish were anaesthetized for imaging with 0.004% MS-222 (Sigma). Embryos at 1-5 dpf were embedded in 0.5% low-melting agarose. Images of embryos and metamorphic fish were acquired on a Zeiss LSM 780 NLO confocal microscope and processed with Fiji (Schindelin et al., 2012) and ZEN 2012 (Zeiss). Confocal z-stacks are shown as maximum intensity projections. Follow-up of pigment cell development was by time-lapse imaging of metamorphic fish, imaging the mid-body area once a day. Adult fish were anaesthetized as above and photographed with a Canon EOS 5D MarkII camera and a Macro 100 objective.

Pigment cell counts
Pigment cell counts were performed using maximum intensity projection images obtained from confocal z-stacks using the Fiji Cell Counter plugin. We counted pigment cells from two or three mid-body segments. For counts of melanophores, fish were treated with epinephrine, and only melanophores on the flank of the fish that contribute to the stripe pattern were taken into consideration. The xanthophores were counted in Tg(fms:GAL4),Tg(UAS: E1b:nfsB-mCherry) double-transgenic fish from 9 to 21 dpf. The iridophores were counted in Tg(TDL358:GFP) (Levesque et al., 2013) and Tg(sox10: mRFP) fish at SL 6-6.5 mm and 6.5-7.5 mm. The fish were imaged daily from 23 to 26 dpf. The number of iridophores per segment was normalized to the number of iridophores in the same segment at 23 dpf.
At 24 h, the oocytes were manually defolliculated, and maintained in MBS at 18°C. To determine the osmotic water permeability (P f ), the oocytes were transferred at 48 h to 10-fold diluted MBS (20 mOsm). Oocyte swelling was tracked by video microscopy using serial images at 2 s intervals for 20 s. P f values were calculated taking into account the timecourse changes in relative oocyte volume [d(V/V 0 )/dt], the partial molar volume of water (V W =18 cm 3 /mol), and the oocyte surface area (S) using the formula V 0 [d(V/V 0 )/dt]/[SV W (Osm in −Osm out )]. The surface area of the oocyte was considered to be nine times the apparent area because of membrane folding (Zampighi et al., 1995).
Glycerol permeability (P gly ) was also determined volumetrically in isotonic MBS, where NaCl was replaced by 160 mM solute. Oocyte swelling was measured by video microscopy, using serial images at 5 s intervals for 1 min. P gly was calculated from oocyte swelling using the formula [d(V/V 0 )dt]/(S/V 0 ) (Verkman and Ives, 1986). The uptake of H 2 O 2 was determined using the ROS-sensitive, cell-permeable fluorescent dye 5-(and-6)-chloromethyl-2′,7′-dichlorodihydrofluorescein diacetate, acetyl ester (CM-H 2 DCFDA; C6827, Life Technologies), following a method described by Chauvigné et al. (2015). Oocytes were incubated with isotonic MBS plus 0.5% DMSO and 200 µM CM-H 2 DCFDA for 1 h at 18°C. After washing in MBS, the oocytes were exposed to 100 µM H 2 O 2 in MBS for 30 min at 18°C. The CM-H 2 DCFDA fluorescence of each well (oocyte plus MBS) was measured at excitation and emission wavelengths of 495 nm and 525 nm, respectively, using a multiwell plate reader (InfiniteM200, Tecan). We corrected the background to remove nonspecific signals from MBS and from oocytes not exposed to the dye.
To study the relative expression of Aqp3a mutants with respect to wild type, constructs carrying the human influenza hemagglutinin (HA) tag at the N-terminus (nucleotide sequence: 5′-TACCCATACGATGTTCCAGATT-ACGCT-3′ or 5′-TATCCATATGATGTTCCAGATTATGCT-3′) were PCR amplified and 15 ng of the corresponding cRNAs was injected into Xenopus oocytes. For western blotting, total membranes were isolated from groups of ten oocytes as previously described (Kamsteeg and Deen, 2001). Protein samples equivalent to the total membrane extract of two oocytes per lane were denatured at 95°C for 10 min in Laemmli buffer, electrophoresed on a 12% polyacrylamide SDS gel, and then blotted onto nitrocellulose membranes (Bio-Rad). Membranes were blocked for 1 h with TBST (20 mM Tris, 140 mM NaCl, 0.1% Tween, pH 8) containing 5% nonfat dried milk (Sigma-Aldrich), and then incubated with 1:1000 HA polyclonal rabbit antibody (PA1-985, Life Technologies) in TBST with 5% nonfat dried milk at 4°C overnight. Horseradish peroxidase-coupled goat antirabbit IgG antibody (Santa Cruz Biotechnology, sc-2004) was used at 1:5000. Reactive protein bands were detected using enhanced chemiluminescence western HRP substrate (ECL, Millipore).

Homology modelling
HHpred (Söding et al., 2005) was used to identify homologues of known structure. HHpred searches were conducted in default settings against the protein data bank (PDB) (Berman et al., 2000), as available on 1st February 2016, and clustered at 70% pairwise sequence identity. Searches of the D. rerio Aqp3a sequence retrieved glycerol uptake facilitator protein from E. coli (PDB identifier 1LDF) and aqua-glyceroporin from P. falciparum (PDB identifier 3C02) as the closest homologues. These structures served as templates for generating homology models with Modeller (Sali and Blundell, 1993).

Statistical analysis
Statistical analysis was performed using InStat (GraphPad Software).