The minimal gap-junction network among melanophores and xanthophores required for stripe pattern formation in zebrafish

Gap junctions mediate direct intercellular communication involving the transfer of molecules such as <1000 Da molecules, metabolites and ions between adjacent cells (Simpson et al., 1977). Connexin is a gap-junction protein; six connexins form a hemichannel called the connexon, and the docking of opposing hemichannels between neighboring cells generates the gap junction (Fig. S1A) (Bruzzone et al., 1996; Kumar and Gilula, 1996; Saez et al., 2003). Connexin is a four-transmembrane protein, the N- and C-terminal domains of which are located in the cytoplasm (Unger et al., 1999; Willecke et al., 2002). Whereas the N terminus functions as a voltage sensor and controls the channel opening/closing machinery (Gonzalez et al., 2007; Verselis et al., 1994), the C terminus performs multiple functions and controls the assembly of connexins, the opening/closing of gap junctions, and the degradation of gap junctions by recruiting proteins such as ZO-1, actin, tubulin and protein kinases (Giepmans et al., 2001; Smyth et al., 2012; Toyofuku et al., 1998; Wang and Peracchia, 1997). In the chordate genome, connexin genes form a large family: ∼20 connexin genes have been identified in the mammalian genome (Kosakovsky Pond et al., 2007) and ∼40 connexin genes are predicted in the teleost genome (Eastman et al., 2006).

Recent studies have enhanced our understanding of bioelectric signaling in not only neuronal cells but also in organ development and regeneration (Levin, 2007; Plotkin et al., 2016). For example, loss of connexin 43 (Cx43) gap-junction function correlates with fin shortening in zebrafish (Hoptak-Solga et al., 2007), and aberrant Cx43 hemichannel activity causes a reduction of vertebral length along the anterior-posterior axis (Misu et al., 2016). Moreover, in the Xenopus embryo, gap-junction function is crucial for left-right patterning (Levin and Mercola, 1999). Although the importance of gap junctions is well recognized, the functions and mechanisms of action of gap junctions in vivo remain largely unknown. The challenge associated with defining gap-junction functions can be attributed to the complexities of connexin assembly and gap-junction networks (Mathews and Levin, 2017; Theis et al., 2005). Distinct types of connexins can potentially form heteromeric and heterotypic gap junctions, which means that different types of connexins could form heterogeneous connexon hexamers. In general, six identical connexins form a homomeric-connexon (Fig. S1B), whereas a mixture of different types of connexins forms a heteromeric-connexon (Fig. S1C). Furthermore, the same types of two homomeric connexons form a homomeric-homotypic (Fig. S1D) or heteromeric-homotypic (Fig. S1E) gap junction, although it is difficult to know whether a heteromeric-homotypic gap junction is actually formed in the manner shown in Fig. S1E. However, different types of two connexons form a homomeric-heterotypic (Fig. S1F) or heteromeric-heterotypic (Fig. S1G) gap junction (Theis et al., 2005). In the case of connexin 41.8 (Cx41.8) and connexin 39.4 (Cx39.4), which are the focus of this study, we have previously shown that these connexins possibly form a homomeric-heterotypic and a heteromeric-homotypic/heterotypic gap junction between Xenopus oocytes (Watanabe et al., 2016). In addition, the directional control of gap junctions complicates the gap-junction network. Heterotypic gap junctions occasionally yield a rectifying property, which is caused by the anion/cation selectivity of each connexon (Suchyna et al., 1999), whereas rat Cx40 exhibits polyamine-dependent rectification properties (Musa and Veenstra, 2003). Typically, a homomeric-homotypic gap junction shows bidirectional current flow (Fig. S1D, S1H). In the case of rat Cx40, polyamines, such as spermine and spermidine, were found to bind to the N terminus of connexin and block the outward flow through gap junctions (Lin et al., 2006; Musa et al., 2004). Simply put, polyamine injected into one side of a paired cell generates unidirectional current flow through the gap junction (Fig. S1I), although this has only been detected in in vitro experiments.

The zebrafish possesses yellow and black stripes on its body (Fig. S2A); yellow stripes consist of xanthophores, which have yellow pigments, such as pteridin derivatives (Odenthal et al., 1996), and black stripes consist of melanophores, which have melanin granules in the cells (Milos et al., 1983). In zebrafish, a third type of pigment cell, iridophores, exists, which have a glossy appearance. In the stripe region of the fish body, iridophores spread over and underneath melanophores, whereas in the inter-stripe region, iridophores are located under xanthophores. In fish fins, only a few iridophores exist (Frohnhöfer et al., 2013; Hirata et al., 2003, 2005). All three types of pigment cell are involved in the stripe pattern formation on the trunk of zebrafish (Fadeev et al., 2015, 2016, 2018; Frohnhöfer et al., 2013; Krauss et al., 2013; Lang et al., 2009; Patterson and Parichy, 2013), although iridophores are not involved in the stripe patterning of the fin region (Singh et al., 2015; Watanabe and Kondo, 2015a). The zebrafish stripe is recognized as a highly favorable model for investigating pattern formation in not only experimental studies (Eom et al., 2015; Mahalwar et al., 2014; Singh et al., 2014; Yamanaka and Kondo, 2014) but also in theoretical studies (Nakamasu et al., 2009; Volkening and Sandstede, 2018). A mathematical model, the reaction-diffusion (R-D) model, which was originally presented by Alan Turing, effectively explains pattern formation (Turing, 1952). In this model, two diffusible substances interact with each other to form a pattern. More recently, Meinhardt and Gierer reported that the R-D model for pattern formation requires only a network that includes short-range positive feedback and long-range negative feedback (Meinhardt and Gierer, 1974, 2000), which generates patterns in a cell-autonomous manner. This model effectively reconstructs many biological patterns, e.g. patterns observed in vertebrate skin or on seashells and arrangements of feathers on the bodies of chicks (Kondo and Asai, 1995; Meinhardt, 1995; Prum and Williamson, 2002). Accordingly, among pigment cells in zebrafish, several interactions that generate stripes have been identified (Frohnhöfer et al., 2013; Maderspacher and Nusslein-Volhard, 2003; Mahalwar et al., 2016; Patterson and Parichy, 2013; Takahashi and Kondo, 2008; Yamaguchi et al., 2007; Yamanaka and Kondo, 2014). Repulsive interactions between xanthophores and melanophores, and interactions that mediate activation signaling from xanthophores to melanophores, are crucial for the generation of an equally spaced pattern of stripes by xanthophores and melanophores (Nakamasu et al., 2009; Watanabe and Kondo, 2015b; Yamanaka and Kondo, 2014). Regarding the repulsive interactions, melanophores occur in the yellow stripe region of xanthophores, which are usually eliminated by the repellant movement of melanophores to the melanophore region (Sawada et al., 2018; Takahashi and Kondo, 2008), or are occasionally removed by xanthophores (Nakamasu et al., 2009). This repulsion was effectively reconstructed in vitro where the repellant movement of melanophores from xanthophores was observed in culture (Inaba et al., 2012; Yamanaka and Kondo, 2014). At the metamorphosis stage in zebrafish, melanoblasts in the yellow stripe region move to the melanophore region in a macrophage-dependent manner (Eom and Parichy, 2017). The requirement of the activation signal from xanthophore to melanophore was observed in vivo. Removal of xanthophore from a region on the trunk of a fish by laser ablation caused the reduction of melanophores surrounding the ablated xanthophore space (Nakamasu et al., 2009). Zebrafish mutants that harbor a temperature-sensitive mutation in csf1ra support this feature; the elimination of xanthophores by heat treatment of mutant fish causes the reduction of the number of melanophores on the body of these fish (Parichy et al., 2000; Parichy and Turner, 2003). The involvement of Notch-Delta signaling in the activation signal is expected (Hamada et al., 2014).

We and others have sought to understand the molecular mechanisms underlying this stripe patterning (Eom et al., 2012; Fadeev et al., 2015; Inoue et al., 2014; Iwashita et al., 2006). Recent studies have revealed that gap-junction proteins are crucial molecules for the cell-cell interaction among pigment cells (Irion et al., 2014; Mahalwar et al., 2016; Watanabe et al., 2006, 2016; Watanabe and Kondo, 2012). RT-PCR analysis has previously shown that cx39.4 and cx41.8 (gja5b) are expressed in both xanthophores and melanophores (Watanabe et al., 2016), and whereas mutations in Cx39.4, a teleost-specific connexin, cause a labyrinth or irregular stripe formation (Fig. S2B) (Irion et al., 2014; Watanabe et al., 2016), a Cx41.8-null mutation generates a spotted pattern (Fig. S2C) (Watanabe et al., 2006) instead of stripes (Fig. S2A). Notably, double knockout of cx39.4 and cx41.8 (WKO) causes a loss of the typical skin pattern (Fig. S2D) (Irion et al., 2014; Watanabe et al., 2016). Transplantation experiments clarified that the gap junctions among melanophores and xanthophores are crucial for gap junction-dependent stripe patterning and that iridophores are not involved (Irion et al., 2014; Maderspacher and Nusslein-Volhard, 2003). Furthermore, Cx39.4 and Cx41.8 harbor a predicted polyamine-binding motif in their N-terminal domains that might function in skin pattern formation (Watanabe et al., 2012). Supporting this notion, ectopic overexpression of a spermine/spermidine metabolic enzyme [spermidine/spermine-N(1)-acetyltransferase; encoded by ssat] in melanophores perturbed the stripe pattern (Watanabe et al., 2012). Intriguingly, the stripe pattern was also disrupted by the loss of spermidine synthase (encoded by idefix) but not spermine synthase, which indicates that spermidine, but not spermine, contributes to stripe-pattern formation in zebrafish (Frohnhöfer et al., 2016).

Here, to elucidate the function of gap junctions in the mechanism underlying stripe pattern formation, we reconstructed the gap-junction network among pigment cells and, ultimately, identified the minimal requirement of connexin expression in pigment cells for stripe formation. Moreover, we performed electrophysiological analyses and detected spermidine-dependent rectification in Cx39.4-containing gap junctions.

As noted in the preceding section, Cx39.4 and Cx41.8 are involved in stripe pattern formation, and mutations in these proteins generate labyrinth and spot patterns instead of the typical stripe pattern (Fig. S2A-C). However, whether other connexins are involved in the stripe patterning remains unclear, and expressions of connexins in pigment cells of trunk have not been investigated. To address these issues, we examined connexin expression in melanophores and xanthophores both from the skin and fin. After homogenizing the trunk and fins in adult fish, 100 melanophores and 100 xanthophores were manually collected separately. RNA was extracted from the isolated pigment cells, and cDNA libraries were generated. We then performed RT-PCR and detected that cx39.4 and cx41.8 were expressed in melanophores and xanthophores from both the trunk (Fig. 1A) and fin (Fig. 1B) using gene-specific primer sets (Table S1). We did not detect expression of the other connexin genes in melanophores or xanthophores from either the trunk or fin (Fig. S3A,B). Based on these results, we concluded that only cx39.4 and cx41.8 are detectably expressed in melanophores and xanthophores.

To understand connexin/gap-junction functions in pigment cells, we generated transgenic lines expressing Cx39.4 and/or Cx41.8 in melanophores and/or xanthophores. To induce cell type-specific gene expression in melanophores and xanthophores, we used the promoters of mitfa (microphthalmia-associated transcription factor a) and aox5 (aldehyde oxidase 5), respectively (Fig. 2A) (Lister et al., 1999; Parichy et al., 2000). The expression of mitfa is occasionally detected in xanthophores (Fig. S2I - S2I″) (Saunders et al., 2019) because xanthophores and melanophores are derived from common progenitor cells, and mitfa expression continues during melanoblast differentiation into melanophores (Dooley et al., 2013; Lister et al., 1999; Usui et al., 2018). We used the IRES-H2BRFP fluorescence-reporter cassette to monitor gene expression driven by the mitfa or aox5 promoter [Fig. S2E-H″, S5A-A″ (white arrowheads); RFP (red fluorescent protein) signal marks nuclei in melanophores (Fig. S2E″, F″) and xanthophores (Figs S2G″,H″, S5A″); yellow arrowhead indicates xanthophore autofluorescence (Fig. S2F″)]. By using this system, we collected zebrafish lines that did not show any detectable RFP signal in undesired cells (Figs S2E-H, S5A).

In Fig. 2B-R, we present the phenotypes of wild-type, mutant and transgenic zebrafish lines generated in this study. On the left side of each panel, black and yellow boxes represent melanophores and xanthophores, and green and red small boxes indicate connexons formed by Cx39.4 and Cx41.8, respectively; thus, a pair of two green small boxes indicates homotypic Cx39.4 gap junctions, whereas a pair of green and red boxes indicates heterotypic gap junctions formed by Cx39.4 connexons and Cx41.8 connexons. In this model, the possible formation of heteromeric gap junctions was ignored. In wild-type zebrafish (Fig. 2B), both Cx39.4 and Cx41.8 are expressed in both melanophores and xanthophores, which supports the potential existence of homotypic gap junctions formed by Cx39.4 and Cx41.8, as well as heterotypic gap junctions formed by Cx39.4 and Cx41.8. These heterotypic gap junctions can be of two types: Cx39.4(M)-Cx41.8(X) or Cx39.4(X)-Cx41.8(M) (M or X indicates melanophore or xanthophore, respectively). In Fig. 2C,D, we show the connexin mutants cx39.4^−/− (luchs) and cx41.8^−/− (leopard) (Irion et al., 2014; Watanabe et al., 2006, 2016). In these mutants, Cx39.4 gap junctions or Cx41.8 gap junctions exist among the pigment cells. Fig. 2E shows the double mutant of Cx39.4 and Cx41.8 (WKO or cx39.4^−/−, cx41.8^−/−), which lacks gap junctions among melanophores and xanthophores. To evaluate the effect of the connexins expressed in pigment cells, we quantified and compared the number of melanophores, the stripe width, the spot size and the xanthophore density (Table 1, Fig. 3).

Fig. S4A-I shows representative examples of the variation of fish phenotypes analyzed in Fig. 3. The left column of each panel in Fig. S4 shows maximal/strong phenotypes, and the right column shows minimal/weak phenotypes of melanophores number/width in stripe/spot of each fish line. For example, Fig. S4A shows the variation of stripe width of the wild-type zebrafish, which is reflected as the width of the box chart in Fig. 3D. To simplify the counting of melanophores, melanosomes in melanophores were aggregated with epinephrine, so that the black stripe or spot on the fish trunk appeared bright (Fig. S4A-I). The phenotypes of the mean in width/spot size among each fish line are shown in Fig. 2. In addition, the instability of the fin pattern was particularly noticeable. Fig. S4J-L depict the differences of fin patterns between siblings of wild-type (Fig. S4J) and mutant fish (Fig. S4K,L). Even the wild-type zebrafish occasionally shows a broken stripe pattern (Fig. S4J, right), and two cx39.4^−/− mutant fish showed different patterns, both in the trunk and fin. This might be because various phenotypes from the same transgenic line were obtained and the recoveries in fin stripe appear incomplete (Fig. 2). Another possibility is that the recovery of the fin pattern was less than that of the trunk, which might be caused by the promoters, making it difficult to compensate for the endogenous promoter activity with mitfa- and aox5-promoters over a long period of time (>1 month).

We counted the melanophores present within the area demarcated by the red solid line and indicated by the green arrow in Fig. 3A (Fig. 3B,C). Stripe width was measured around the center region of the 1V stripe (Fig. 3A,D) (Hamada et al., 2014). Cluster size of melanophore spots was determined by counting the number of melanophores in each spot (Fig. 3E). Xanthophore densities were measured from four randomly selected areas (∼0.5-1.4 mm² square) in the X0 yellow stripe, which is demarcated by the red solid line and indicated by the blue arrow (Fig. 3A). The means were calculated in each fish line (Fig. 3F,G). The two regions indicated by blue and green arrows are separated by a purple dotted line, which indicates the myoseptum. Serial numbers were assigned to the transgenic fish as Tg-1–Tg-14 (Fig. 2, Figs S2, S5).

In Fig. 2F-I, Tg-1–Tg-4 represent the phenotypes of transgenic zebrafish in WKO background in which Cx39.4 or Cx41.8 expression was induced using pigment cell-specific promoters (as indicated): Cx39.4 was expressed in either melanophores (Fig. 2F, Fig. S4B) or xanthophores (Fig. 2H, Fig. S4C), and Cx41.8 was expressed in either melanophores (Fig. 2G) or xanthophores (Fig. 2I, Fig. S4D). Cx39.4 expression in melanophores (Tg-1) restored the WKO phenotype to the skin pattern present in cx41.8^−/− (Fig. 2D-F). Not only was the melanophore density on the fish trunk restored but so was the size of each spot, as in cx41.8^−/− (Table 1, Fig. 3B,E). This result indicates that Cx39.4 functions not between melanophores and xanthophores but between melanophores; however, the possibility remains that Cx39.4 also forms gap junctions with unidentified connexons on other cells or functions solely as a hemichannel. By contrast, single Cx41.8 expression in melanophores (Tg-2) was insufficient for pattern formation (Fig. 2G). In this fish, we observed no change in phenotype from that of WKO. In the case of single expression of Cx39.4 or Cx41.8 in xanthophores (Tg-3, Tg-4), the mutant phenotype was not restored, although melanophore numbers (Fig. 2H,I, Fig. S4C,D) and xanthophore densities (Fig. 3G) were increased (Table 1). Given this, we hypothesized that gap junctions in xanthophores might function as adhesion molecules in order to make the xanthophore area compact, which would allow melanoblasts in the free space to readily differentiate without interference from xanthophores. To confirm this, we expressed Cx43 in xanthophores (Fig. S5A-A″), which revealed that xanthophore density was increased, although the recovery of xanthophore density was small (Fig. S5B). Intriguingly, Cx43 expression did not cause recovery of the number of melanophores (Fig. S5A).

By means of mating among the transgenic lines Tg-1–Tg-4 (Fig. 2F-I) or by mating them with the mutant lines cx39.4^−/− or cx41.8^−/−, we generated double and quadruple transgenic lines in a WKO background (Tg-5–Tg-9; Fig. 2J-N) and single-knockout background (Tg-10–Tg-13; Fig. 2O-R). As expected, quadruple-transgenic zebrafish showed the stripe pattern (Tg-5; Fig. 2J). This result supports the view that Cx39.4 and Cx41.8 expression in melanophores and xanthophores was sufficient, and that the expression of these connexins in other cells was not necessary for stripe formation in the WKO-background fish. Intriguingly, in Tg-5, the black stripe was thinner than that in wild type, but the underlying reason remains unclear (Figs 2J and 3D).

Next, we examined the effects of connexin pairs expressed between melanophores and xanthophores. When Cx39.4 was expressed in both melanophores and xanthophores in the WKO background (Tg-6), the spot pattern was generated as in cx41.8^−/− (Fig. 2D,K). This agrees with the result obtained with Tg-1. Conversely, when Cx41.8 was expressed in both melanophores and xanthophores (Tg-7), the WKO phenotype was effectively restored to the cx39.4^−/− phenotype (Fig. 2L), which indicates that mitfa and aox5 promoters compensated for the cx41.8 promoter in melanophores and xanthophores (Watanabe and Kondo, 2012). Tg-8, in which Cx39.4 was expressed in melanophores and Cx41.8 was expressed in xanthophores, showed the stripe pattern (Fig. 2M). This key finding forms the basis for the conclusion of this study that, for stripe formation, the minimal requirement of connexin expression in pigment cells is Cx39.4 in melanophores and Cx41.8 in xanthophores. However, the black stripe in Tg-8 was thinner than that in wild type (Fig. 3D), and the stripe in the fin was incomplete. The Tg-9 phenotype appeared intermediate between the Tg-7 and Tg-8 phenotypes, featuring partially broken and narrow stripes (Fig. 2N), although this phenotype was not particularly informative because, as noted above, Cx41.8 expression in xanthophores is necessary for generating stripes. The Tg-10 phenotype strongly supported the conclusion here regarding the requirement of Cx41.8(X)-Cx39.4(M) (Fig. 2O), and in Tg-10, the stripe pattern was clearer than that in Tg-8, which suggests that Cx41.8 expressed in melanophores performs a supportive function in stripe formation. Tg-11 fish showed wavy and partially broken stripes as in the cx39.4^−/− mutant (Fig. 2P), and the xanthophore density in Tg-11 was almost the same as that in cx39.4^−/− (Fig. 3F). This agrees with the result indicating that Cx39.4 expression in xanthophores is not required. The phenotype of Tg-12 was the same as that reported previously (Figs 2Q and 3C) (Watanabe and Kondo, 2012), although individual fish showing broken stripes to large spot patterns were also obtained. Tg-13 satisfied the connexin expression required for stripe formation, but the stripes were partially broken and narrower than in wild type (Fig. 2R). Cx41.8 might be necessary for generating narrow/normal interstripes. When Cx39.4 was co-expressed with Cx41.8 in xanthophores, the stripe pattern was unstable, and the spots or stripes were smaller or narrower than those in transgenic fish expressing only Cx41.8 in xanthophores (Tg-10 versus Tg-5, Tg-8 versus Tg-13).

In previous studies, we predicted that gap junctions mediate signal transfer from xanthophores to melanophores, and that this represents a cue provided by the gap junction network for pattern formation (Watanabe et al., 2012). Here, we sought to perform electrophysiological analysis of Cx39.4 gap junctions by using the patch clamp technique to determine whether Cx39.4 expressed in melanophores possesses a rectification property and exhibits the ability to control the direction of signal transfer through gap junctions. First, we designed a plasmid construct that expresses EGFP-tagged Cx39.4 and thus allows gap junction plaque visualization in vitro and in vivo. We cloned the gene encoding the fusion protein composed of Cx39.4 (full-length ORF) and EGFP in pIRES2-DsRed vector and transfected the plasmid into HeLa and Neuro2A (N2a) cells. In parallel, we introduced the gene encoding EGFP-tagged Cx39.4 into the cx39.4^−/− mutant by using the same method described in Fig. 2A. We found that EGFP-tagged Cx39.4 did not function in vivo or in vitro: EGFP-tagged Cx39.4 neither formed gap junction plaques between cultured cells nor restored the cx39.4^−/− mutant phenotype of zebrafish (data not shown). We struggled to find the construct of functional EGFP-tagged Cx39.4. In humans, plaque formation of GFP-tagged C-terminal truncated Cx37 has been reported (Kumari et al., 2000). Referring to this, we added an EGFP tag to the 256th position on Cx39.4, generating Cx39.4del_256-341EGFP. In this case, we found that fewer than 10% of N2a cell pairs formed a gap junction plaque between N2a cells (data not shown). Next, expecting the improvement of plaque formation rate, we added the C-terminal fragment of Cx39.4 at the C terminus of the EGFP sequence in Cx39.4del_256-341EGFP because putative ZO-1 binding residues are included in the C-terminal region of Cx39.4, and the importance of the ZO-1 binding domain for Cx39.4 function has been predicted (Fadeev et al., 2015). Consequently, we obtained a construct encoding Cx39.4exc_256-321EGFP, in which a 66-residue stretch, from amino acids 256 to 321, at the C-terminal domain of Cx39.4 was exchanged with the EGFP fragment (Cx39.4exc_256-321EGFP) (Fig. 4A). We confirmed that Cx39.4exc_256-321EGFP was functional in vivo: Cx39.4exc_256-321EGFP expression, which was driven by a mitfa promoter, restored the cx39.4^−/− mutant phenotype (Fig. 4B). Furthermore, the EGFP signal was observed in cells and localized at cell membranes, and gap junction plaques were detected between melanophores (arrowhead in Fig. 4C; compared with negative control in Fig. 4D). This gene fragment was also cloned into the pIRES2-DsRed vector for transfection of cultured cells, and our results revealed successful formation of gap junction plaques between N2a cells (Fig. 4E,E′); however, gap junction plaques could not be detected between transfected HeLa cells.

Zebrafish connexins rarely assemble on the mammalian cell membrane, which precludes precise analysis of their gap junction properties; therefore, the oocyte-clamp technique is typically used for studying zebrafish gap junctions (Hoptak-Solga et al., 2007; Klaassen et al., 2016; Misu et al., 2016; Watanabe et al., 2016). The oocyte-clamp technique offers the advantage that a large current value is obtained when compared with that in patch-clamp experiments; however, analyzing polyamine sensitivity of gap junctions using this method is challenging. The large volume of an oocyte (estimated as 1 μl compared with the 1 pl volume of a HeLa cell) (Ferrell and Machleder, 1998; Fujioka et al., 2006) prevents the analysis of polyamine sensitivity to gap junctions. Thus, no previous study has successfully examined the polyamine sensitivity of zebrafish gap junctions. Here, we were able to reconstruct gap junction plaques and visualize Cx39.4 gap junctions between N2a cells, and we therefore used this system to perform patch-clamp experiments. At 24-72 h post-transfection, we selected, for analyses, pairs of N2a cells exhibiting the EGFP signal as an indicator of gap junction plaques. We clamped both cells initially at −40 mV and applied a series of transjunctional voltages (Vi, −140 to +60 mV in 20 mV increments) to one of the cells and recorded the transjunctional current (Ij) in the other cell (Fig. 4F,G; upper set of lines in Fig. 4G). Fig. 4G presents examples for each experimental condition. As a negative control, we used cell pairs that were not transfected with the plasmid, and we detected no currents here (Fig. 4G, second set of lines). In Fig. 4G, the third line represents transjunctional current traces of Cx39.4 gap junctions. As previously shown in voltage-clamp experiments performed using Xenopus oocytes, slow deactivation of gap junctions was observed (Watanabe et al., 2016).

Next, we examined the polyamine-dependent rectification property of Cx39.4 gap junctions. The contribution of polyamines to zebrafish skin pattern formation has been investigated. As mentioned in the introduction, ectopic overexpression of ssat in melanophores perturbs the stripe pattern (Fig. 4H) (Watanabe et al., 2012), and ssat expression in xanthophores does not affect skin patterning (Fig. 4I). In mammals, two ssat genes, ssat1 and ssat2, exist in the genome. SSAT1 is related to polyamine metabolism, whereas SSAT2 is not (Coleman et al., 2004; Vogel et al., 2006). In zebrafish, three ssat1 homologs, sat1a.1, sat1a.2 and sat1b, exist (Lien et al., 2013). We performed RT-PCR to investigate gene expression of three ssat1 genes in pigment cells and found that only sat1a.2 is expressed in melanophores and xanthophores (Fig. S6). This result indicates that Sat1a.2 may function in homeostasis of melanophores and xanthophores, and the disturbance of its function in melanophores caused an irregular stripe patterning. Moreover, loss of spermidine synthase (idefix), but not of spermine synthase, was shown to disrupt the stripes, which indicated that spermidine but not spermine is involved in stripe pattern formation in zebrafish (Frohnhöfer et al., 2016). Thus, we analyzed the spermidine sensitivity of Cx39.4 by adding 15 or 30 mM spermidine into the pipette solution and recorded transjunctional currents by using the same procedure as described in the preceding paragraph. Notably, our results demonstrated asymmetric gap junction gating (Fig. 4G, bottom two sets of lines; Fig. 4J, normalized I-V curve). Thus, we conclude that Cx39.4 exhibits dose-dependent spermidine sensitivity.

To elucidate the function of gap junctions in pattern formation in zebrafish skin, we reconstructed the gap junction network among pigment cells. By controlling the expression of the connexins Cx39.4 and Cx41.8 specifically in melanophores and xanthophores (Fig. 2), we determined that the minimal requirement of connexin expression for stripe pattern formation is Cx39.4 in melanophores and Cx41.8 in xanthophores. Furthermore, we successfully generated an EGFP-tagged Cx39.4 construct that formed gap junction plaques in vitro and restored the mutant phenotype in vivo. Last, by using this EGFP-tagged Cx39.4 construct and the patch-clamp technique, we showed that Cx39.4 gap junctions exhibit a spermidine-dependent rectification property. However, spermidine sensitivity of Cx41.8 gap junctions has not been examined because Cx41.8 does not form gap junction plaques between cultured cells. Instead, previously we used rat Cx40, a mammalian orthologue of Cx41.8, to examine the necessity of the polyamine-binding motif at the N terminus of connexin (Watanabe et al., 2012). Rat Cx40 is reported to be blocked by 5-15 mM spermidine (Musa and Veenstra, 2003) and an amino acid substitution at the polyamine-binding motif inhibits the rectification property of the rat Cx40 gap junction (Lin et al., 2006; Musa et al., 2004). We showed the polyamine-binding motif on rat Cx40 is required for the stripe pattern formation of zebrafish in the cross-species transgenic experiments (Watanabe et al., 2012), which supports that connexins expressed in melanophores require polyamine-binding properties for the stripe pattern formation. According to our previous mathematical and experimental models for stripe pattern formation, the signal that is required for melanophore survival is transferred from xanthophores to melanophores. Here, we show the possibility that this signal could be transferred from xanthophores to melanophores through a heterotypic gap junction, which was made by a docking Cx41.8-connexon in xanthophores with Cx39.4-connexon in melanophore and has spermidine-dependent rectification properties. The phenotypes from the transgenic experiment in which Cx41.8 was expressed in melanophores may have a function in stabilizing the stripe.

In Fig. 5, we present a model for the gap junction network among pigment cells. In this model, Kir7.1 is also described because it is sensitive to spermidine and crucial for skin pattern formation (Inaba et al., 2012; Iwashita et al., 2006). We found here that 30 mM spermidine blocked outward flow through the Cx39.4 gap junctions. However, the concentration at which spermidine blocks zebrafish Kir7.1 has not been determined. In terms of the effect of spermidine on Kir-family potassium channels, the effective spermidine concentration measured for Kir2.1 was 0.1-50 μM, which blocked the outward flow of potassium ions through the channel (Liu et al., 2012). Previous examination of spermidine concentrations and distribution in cells revealed that the spermidine concentration in the atrium and ventricle in total was ∼100 μM, and that the free spermidine concentration in cells was ∼10 μM (Miyamoto et al., 1993; Watanabe et al., 1991). The spermidine concentration in melanophores is currently unknown, as is the number of spermidine molecules that colocalize with gap junctions at the cell membrane and block gap junction function; however, Cx39.4 and Cx41.8 gap junctions might be blocked partially and over time, to different extents, because the spermidine sensitivities of these two types of gap junctions were found to be lower than that of Kir potassium channels and the spermidine concentration is expected to be ∼100 μM in melanophores. This insufficient blockage of gap junctions might enable signal molecules to spread among melanophores through Cx39.4 gap junctions. In the model presented in Fig. 5, melanophores are shown to extend long projections. Previously, we observed that long filopodia from melanophores extended to xanthophores, and we hypothesized that melanophores use these filopodia to make direct contact with xanthophores at the adult stage in zebrafish (Hamada et al., 2014). Because gap junctions are formed between cells and mediate direct cell-cell interaction, Cx41.8(X)-Cx39.4(M) gap junctions could exist at the tips of melanophores (Fig. 5).

In terms of function, Cx39.4 and Cx41.8 were found to play divergent roles in pattern formation in the two types of pigment cells. When Cx39.4 or Cx41.8 was expressed only in melanophores, markedly distinct phenotypes appeared (Fig. 2F,G), although why Tg-1 and Tg-2 fish showed such phenotypes is unclear. Previous comparison of the electrophysiological properties of Cx39.4 and Cx41.8 gap junctions (Watanabe et al., 2016) revealed that Cx41.8 gap junctions exhibit higher sensitivity to transjunctional voltage and more rapid time-dependent inactivation relative to Cx39.4 gap junctions. Moreover, as noted above, the efficiency with which spermidine blocks these two types of gap junctions might differ, and this property would result in divergent effects being produced on the cell-cell communication between melanophores, even in the presence of spermidine. Conversely, when Cx39.4 or Cx41.8 was expressed only in xanthophores, the number of melanophores was increased, although the increase rate differed (Table 1, Fig. 2H,I). Regarding this matter, we investigated the possibility that gap junctions in xanthophores might function as adhesion molecules. Ectopic expression of Cx43 in xanthophore partially recovered the density of xanthophore; however, the number of melanophores was not restored (Fig. S5). Further investigation is required to define how Cx41.8 in xanthophores contributes to the development of melanophores.

In this study, several conditions that might be important for gap junction formation were ignored to simplify the study model. For example, the possibility exists that Cx41.8 or Cx39.4 forms heterotypic gap junctions with other connexin-connexon channels in other cells. Tjp1a mutant zebrafish (schachbrett), in which a mutation is present in the protein ZO-1a, showed a spotted pattern similar to that in the cx41.8^−/− mutant (Fadeev et al., 2015). The PDZ domain in ZO-1 bound to the C terminus of Cx41.8 and Cx39.4, which raised the possibility that Cx41.8 or another connexin in iridophores contributes to the pattern formation (Fadeev et al., 2015). However, our results here indirectly contradict the notion that Cx39.4 or Cx41.8 expression in iridophores is not required for pattern formation (Irion et al., 2014; Maderspacher and Nusslein-Volhard, 2003; Mahalwar et al., 2016). Although large-scale mutant screening of zebrafish has not revealed the involvement of other connexins in pattern formation, further investigation might yield clues to resolving the issue of whether a third connexin in surrounding cells is involved in skin pattern formation.

All experiments were approved by the Animal Experiments Committee and Gene Modification Experiments Safety Committee of Osaka University (permit numbers 04294 and FBS-14-002-1). Zebrafish (Danio rerio) were maintained under standard conditions at 28.5°C and a 14/10 h light/dark cycle.

Pigment cells were collected as previously described (Yamanaka and Kondo, 2014). Briefly, three tail fins were cut off from anesthetized adult zebrafish and washed with phosphate-buffered saline (PBS); the fin clips were then treated with a trypsin solution [2.5 mg/ml trypsin (TRL; Worthington), 1.2 mg/ml bovine serum albumin (BSA) (Sigma) and 1 mM EDTA in PBS] for 1 h at 28°C. After washing five times with PBS, the fin clips were treated with collagenase solution [1 mg/ml collagenase I (Worthington), 0.1 mg/ml DNase I (Worthington), 0.1 mg/ml soybean trypsin inhibitor (Worthington) and 1.2 mg/ml BSA in PBS] for 1 h at 28°C. To isolate the trunk pigment cells, fish skin was peeled off from anesthetized zebrafish, using scissors and forceps, and washed with PBS. The collected skin was dissected into ∼3 mm squares with a knife and then treated with the trypsin solution for 20 min at 28°C. After washing five times with PBS, the dissected skin fragments were treated with the collagenase solution for 20 min at 28°C. Next, melanophores and xanthophores from the fin or trunk, respectively, were manually collected individually from the collagenase-treatment solution by using glass capillaries. In each experiment, 100 melanophores or 100 xanthophores were collected from both the trunk and fin. Other organs, including the brain, eye, intestine, ovary and testis, were also collected using forceps. mRNAs were purified from the cells and organs by using an RNeasy purification kit (Qiagen) and used for cDNA library synthesis. RT-PCR was performed using connexin-specific or ssat-specific primers (Table S1). actb1 (encoding β-actin) was the positive control for RT-PCR, and aox5 was the positive control of the xanthophore-specific marker genes. dct was used as a melanophore-specific marker gene instead of mitfa because mitfa is occasionally expressed both in melanophores and xanthophores (Dooley et al., 2013; Lister et al., 1999; Usui et al., 2018). cDNAs synthesized from RNA isolated from brain, eye, testis, ovary and skin tissues were used for positive controls to check gene-specific primer sets.

Transgenic fish were generated as reported previously. Briefly, the Tol2 transposon-based transgenesis system was used, and 1.5 kb of mitfa promoter and 1.7 kb of aox5 promoter were used to induce connexins (Lister et al., 1999; Parichy et al., 2000). Connexin-coding fragments were PCR amplified (Watanabe et al., 2016) and cloned into the pTol2 plasmid (Kawakami et al., 1998, 2000).

The primer set used to generate cx43 clones from zebrafish genomic DNA was: cx43_SalI_F, AAAGTCGACGCCACCATGGGTGACTGGAGTGCGTT; and cx43_NotI_R, AAAGCGGCCGCTAGACGTCCAGGTCAG. Amplified fragments were digested with SalI and NotI, and then ligated into pTol2-aox5 promoter plasmid. To generate pTol2-cx39.4exc_256-321EGFP plasmid, three fragments, zfcx39.4N (N-terminal domain), EGFP cassette and zfcx39.4CT (C-terminal domain) sequences were amplified, respectively, using three primer sets: zfcx39.4N (zfcx394EcoF, AAATTTGAATTCGCCACCATGTCCAGAGCTGACTGGGG; and zfcx39.4N255EGFPR01, TCCTCGCCCTTGCTCACCATCACCTCTGCCTGATATTTCTTCTCT); EGFP (39.4N255EGFPF01, AGAAATATCAGGCAGAGGTGATGGTGAGCAAGGGCGAGGAGCTGT; and 394_GFPN321R02, TTTAGGAAGATTGTTTTTGTTCTTGTACAGCTCGTCCATGC); and zfcx39.4CT (39.4EGFPN321F02, GCATGGACGAGCTGTACAAGAACAAAAACAATCTTCCTAAA; and zfcx394NotR01, AAATTTGCGGCCGCTCAAACATAATGTCTCGGTT). The amplified fragments were mixed and re-amplified using the primer set zfcx394EcoF and zfcx394NotR01 to generate a full-length cx39.4exc_256-321EGFP fragment. The 1.6 kb fragment obtained was then digested with EcoRI and NotI, and cloned with a pTol2 plasmid vector.

Total RNA was extracted from fish brain by using the RNeasy kit, and a brain cDNA library was generated through reverse transcription performed using Super Script III (Invitrogen). The following primers were used to generate a sat1b clone from the brain cDNA library: sat1b_SalI_F, AAAGTCGACGCCACCATGGCCAATTTTAATTTGCG; and sat1b_NotI_R, AAAGCGGCCGCTCACTCTTCAGCAGACATTTTC. Amplified fragments were digested with SalI and NotI, and then individually ligated into a pTol2 plasmid. Transgenic zebrafish lines were generated as described previously (Kawakami et al., 1998, 2000).

To count melanophores, epinephrine-treated zebrafish were photographed under an MZ16FA stereoscopic microscope (Leica). Epinephrine (10 mM) was used to aggregate melanosomes in melanophores, which facilitates counting. Melanophore numbers and xanthophore densities were calculated using the measurement and particle analyzer features of ImageJ software. Magnified and high-resolution bright-field images of xanthophores were acquired using a BZ-X710 inverted microscope (Keyence).

The mouse neuroblastoma cell line Neuro2A (N2a, JCRB Cell Bank) was used for electrophysiological experiments. N2a cells were maintained in Dulbecco's modified essential medium supplemented with nonessential amino acids, antibiotics and 10% fetal bovine serum (Hyclone). The cx39.4exc_256-321EGFP fragment (in which a part of Cx39.4 C-terminal domain, from the 256th to 321st amino acid, was exchanged with an EGFP fragment) was subcloned into a pIRES2-DsRed vector (Clontech) using the following primer set: cx39.4_EcoRI_F, CTTCGAATTCGCCACCATGTCCAGAGCTGACTGGG; cx39.4_BamHI_R, TTTGGATCCTCAAACATAATGTCTCGGT. N2a cells were transfected with pIRES2-cx39.4exc_256-321EGFP plasmid by using the FuGENE transfection reagent (Promega) according to the manufacturer's instructions, and Cx39.4exc_256-321EGFP expression in N2a cells was detected using fluorescence microscopy at 24 h post-transfection. The N2a cells were washed three times with PBS and then placed on the stage of an inverted phase-contrast microscope. The bath buffer consisted of 142 mM NaCl, 1.3 mM KCl, 0.8 m MgSO₄, 0.9 mM NaH₂PO₄, 1.8 mM CaCl₂, 4.0 mM CsCl, 2.0 mM TEACl, 5.5 mM dextrose and 10 mM HEPES (pH 7.5, adjusted using 1 N NaOH). The pipette solution contained 140 mM KCl, 4.9 mM CsCl, 2.0 mM TEACl 3.0 mM CaCl₂, 5.0 mM K₄BAPTA, 1.0 mM MgCl₂ and 25 mM HEPES (pH 7.5, adjusted using 1 N KOH). MgATP was added to a final concentration of 3.0 mM before analyses. Junctional currents were recorded using the double whole-cell recording technique by using an EPC 10 USB Double (HEKA Elektronik). Patch electrodes featured a tip resistance of 4-6 MΩ. All experiments were performed at room temperature (25°C). To investigate the polyamine sensitivity of Cx39.4, spermidine (15 or 30 mM) was added into one side of the pipette.

Results are presented as mean±s.d. of the number of independent experiments indicated in each figure legend. P-values were calculated using Student's t-test and a 95% confidence level was considered significant. Statistical analysis was performed using Origin (OriginLab).

We thank H. Soma for his technical assistance in the electrophysiological experiments.