Heterogeneous fates and dynamic rearrangement of regenerative epidermis-derived cells during zebrafish fin regeneration

One of the important roles of the epidermis is to protect the internal homeostasis of multicellular organisms from external stresses, including mechanical and chemical damage, pathogens and radiation (Chuong et al., 2002). In mammals, the epidermis is composed of several distinct cell layers. The horny layer at the surface contains layers of keratinized cells. The granular layer provides a hydrophobic barrier underneath the horny layer. The spinous layer is a thick layer that contains a number of differentiating keratinocytes. In the deepest part of the skin, a layer of basal stem cells constantly produces new keratinocytes to maintain epidermal tissue homeostasis (reviewed by Fuchs, 2008).

Teleost fish, including zebrafish, have a similar layered structure in their epidermis, although the outermost layer is not covered by a horny layer, but rather by a mucous layer (Rakers et al., 2013). Apart from structural similarity, the process of cell turnover in zebrafish is nearly the same as that observed in mammals, in that the basal stem cells produce all the strata of the epidermis (Lee et al., 2014). Thus, the zebrafish epidermis is thought to be a useful model for investigating the regulatory mechanisms involved in skin homeostasis and regeneration. In particular, zebrafish fins, which are devoid of dermal tissues, including the typical connective tissues and muscles, have an advantage for fluorescent live-imaging using a standard confocal microscope.

Additionally, unlike mammals, which do not completely regenerate tissues after massive injury, but instead form scars, teleost fish can perfectly reform their lost appendages through a process called epimorphic regeneration. In addition, the layered epidermal structure is also completely reformed without the formation of scars. An analysis of this scar-less skin regeneration process in zebrafish could lead to the development of treatments aimed at perfect skin regeneration.

In the process of epimorphic regeneration, epithelial cells migrate to cover the wound after amputation and form a thick epithelium, which is referred to as the wound epidermis (WE). However, the term of WE has been used without clear definition. Numerous studies have shown that regeneration is impaired by disturbing the normal formation of the WE and by surgically removing the WE (Goss, 1956; Thornton, 1957; Mescher, 1976; Tassava and Garling, 1979), suggesting that the WE is an epidermal tissue that is required for inducing blastema formation, and is essential for regeneration (reviewed by Campbell and Crews, 2008). Recent studies in zebrafish have suggested that the WE directs blastema formation through fibroblast growth factor signalling (Shibata et al., 2016), and that the alignment of osteoblast progenitors during fin regeneration is regulated by the Sonic hedgehog signalling in the WE (Armstrong et al., 2017). However, the definition of the WE is controversial, because the term WE was sometimes used as the epithelial layers formed within a few hours following injury. In this paper, we refer the epidermal cells covering the regenerate during regeneration as the ‘regenerative epidermis’ (RE).

Thus, an increasing body of data suggests that the RE plays an important role in tissue regeneration. However, no studies have investigated the origin and fate of RE cells, and whether corresponding cells are found during mammalian skin regeneration. Elucidation of the origin and fate of the RE cells during/after regeneration is an important goal in understanding the mechanism of perfect epidermal regeneration. Recently, Chen et al. (2016) have generated Skinbow transgenic zebrafish, in which they tracked superficial epidermal cells and demonstrated the dynamic behaviour of superficial epidermal cells. However, thus far, no studies have addressed the origin and fate of the RE cells, and how the cells required for the formation of the regenerated epidermis are supplied during regeneration.

Here, we reveal the origin and the fate of RE cells using fibronectin 1b (fn1b) as a regenerative epidermis marker and genetic cell fate tracking. We found that RE cells are mainly recruited from epidermal cells of the inter-ray regions, and that they contribute to the regenerated epidermis, with the exception of the early recruited cells that are immediately lost by apoptosis following wound healing. Unexpectedly, after contributing the regenerated epidermis, many of the epidermal cells derived from the RE are lost from the fin by a dynamic caudal movement of the epidermis, and the epidermal cells are replenished by a global activation of cell proliferation. We further show that de-differentiation of non-basal epidermal cells into the basal epidermal cells does not occur during regeneration. Our study has revealed heterogeneous fates of the RE cells and a dynamic rearrangement of the epidermis during and after regeneration.

Our previous study has shown that fn1b is highly upregulated in the RE during zebrafish fin regeneration (Yoshinari et al., 2009). A detailed expression analysis of fn1b in the adult zebrafish fin showed that mRNA expression was first detected at 0.5 day post amputation (dpa) in the epithelial cells near the amputation plane and then expanded to the entire region of the newly formed RE at 1 dpa (Fig. 1A). At 2 dpa and thereafter, fn1b mRNA expression became weaker in the outermost layer and distal region, and was confined to the basal and suprabasal cells of the inter-ray RE and the suprabasal cells of the ray RE (Fig. 1A; Fig. S1). Thus, the expression of fn1b is consistently observed in the epidermis covering the blastema, indicating that fn1b is a useful RE marker that is expressed in many RE cells.

To assess the behaviour of RE cells, we generated a BAC transgenic (Tg) line, Tg(fn1b:egfp), that recapitulates fn1b expression (Fig. 1B; Fig. S2). In the Tg, EGFP expression was low in the uncut fin (Fig. 1C), but following amputation, strong EGFP expression, which resembled that detected by in situ hybridization analysis of fn1b expression, was observed from 1 dpa and thereafter (Fig. 1D). Though the timing of EGFP expression is slightly delayed, this is reflective of the time required for the EGFP protein to mature. The intensity of EGFP expression gradually decreased after 4 dpa, although fluorescence was still slightly detectable at 6 dpa (Fig. 1E). EGFP expression in the RE was further confirmed by immunohistochemical staining of the section of the Tg fin with an antibody against E-cadherin, an epithelial cell marker (Fig. 1F).

EGFP expression was high in the inter-ray regions at 1 dpa, and further confined to the inter-ray regions at 2 dpa (Fig. 1D). Moreover, the regenerating tissue showed a striped wavy pattern of EGFP⁺ cells (Fig. 1G), suggesting a possibility that many of the EGFP⁺ cells were derived from the inter-ray regions and had migrated distally and laterally to cover the whole amputation plane. To further confirm this, we performed time-lapse observation of EGFP⁺ cells over 300 min from 24 h post amputation (hpa). The analysis supported the notion that EGFP⁺ cells indeed migrated from the inter-ray epidermis and spread over the ray region (Fig. 1H, white arrowheads). Considering the fn1b expression in the ray epidermis at 0.5 dpa, a stage before the migration of fn1b⁺ inter-ray cells (Fig. 1A), it is thought that the epidermal cells from ray and inter-ray regions together form the RE in the ray region.

Thus, the observation made in the Tg(fn1b:egfp) suggested that fn1b⁺ RE cells are mainly recruited from the inter-ray epidermis. However, it is still possible that cells dynamically change fn1b expression during tissue regeneration. To determine the actual cell fate of fn1b⁺ cells, we developed an additional Tg line that expresses CreERt2 under the regulation of the fn1b promoter (Fig. 2A). CreERt2 is a version of Cre bacterial recombinase that has a high affinity for the oestrogen analogue 4-OH tamoxifen (TAM). When double Tg(fn1b:creERt2;Olactb:loxP-dsRed2-loxP-egfp) fish (Yoshinari et al., 2012) were treated with TAM to induce Cre-loxP recombination following fin amputation, we successfully observed the appearance of EGFP⁺ cells (Fig. 2B), whereas EGFP⁺ cells were rarely seen in uncut fins (data not shown) or amputated fins without TAM treatment (Fig. 2C). More importantly, EGFP⁺ cells were observed in both the basal and superficial epidermal cells, suggesting that there is no biased Cre-loxP labelling of fn1b⁺ cells to specific epidermal cell types (Fig. 2D).

By using these double Tg, we labelled the fn1b⁺ RE cells by treatment with TAM at 0-24 hpa and tracked their cell fate between 24 and 48 hpa (Fig. 2E). As in the Tg(fn1b:egfp), Cre-labelled cells were first detected in the inter-ray epidermis and also migrated to cover the fin rays and form the RE (Fig. 2E). Although a previous study has suggested that RE cells are recruited from the proximal epidermal cells (Nakatani et al., 2008), our analysis further showed that many epidermal cells that migrate from the inter-ray region contribute to RE formation.

To further investigate the initial process of RE formation earlier than 24 hpa, we re-amputated the regenerating fins in which the RE cells were labelled by Cre recombination. As we describe in the later section (see Fig. 4), a group of Cre-labelled RE cells contribute to the regenerated epidermal cells. Consistent with the above observation at 24-48 hpa, the epidermal cells in the inter-ray region contributed to both the ray (44%) and inter-ray RE (56%), whereas those in the ray region only contributed to the ray RE (Fig. S3). These observations support the observation that the epidermal cells from ray and inter-ray regions together form the ray RE.

To dissect the processes of RE formation and the fate of its component cells, we performed pulse labelling of fn1b⁺ cells and tracked their fate. First, labelling was performed with 0.1 µM TAM over the wound healing and early RE formation stage at 9-10 hpa (Nechiporuk and Keating, 2002). Cre-labelled cells began to be detected at 1 dpa, and were clearly observed at 2 dpa in distally localized superficial regions of the RE (Fig. 3A).

When these cells were tracked at subsequent stages, they were further confined to the distal fin edge and disappeared by 5 dpa (Fig. 3A,B). Many of the Cre-labelled cells were positive for TdT-mediated dUTP nick end labelling (TUNEL) staining of apoptotic cells, indicating that these cells are discarded through apoptosis (Fig. 3C). Thus, these data suggest that the early fn1b-expressing cells found during wound healing and early RE formation stages are mainly recruited from the inter-ray epidermis to transiently cover the wound, but they are unable to form the regenerated epidermis and soon vanish from the regenerating epidermis by undergoing apoptosis.

We next tracked later-arising RE cells by inducing Cre recombination at 24-25 hpa, a stage when RE formation is already evident. After labelling, Cre-labelled cells were detected by 2 dpa, mainly in the inter-ray epidermis; in contrast to the early labelling at 9-10 hpa, EGFP⁺ cells and their progeny were observed in both the basal and superficial layers of the RE, and survived beyond 5 dpa, although the number of cells decreased as the stage proceeded (Fig. 4A). In subsequent stages, these surviving EGFP⁺ cells proliferated to form colonies within the regenerated epidermis at 14 dpa.

When Cre recombination was induced at a further later stage of fin regeneration (i.e. during 2-5 dpa), the overall fate of the RE cells was similar to that seen in Fig. 4A. The later RE cells and their progeny formed a number of colonies by 1 week post amputation (wpa); however, 80% of colonies had disappeared by 3 wpa (Fig. 4B,C). In contrast to the decrease in the number of colonies, the size of the colonies increased ∼6.5 times between 1 and 3 wpa (Fig. 4D). After 3 wpa, the number of colonies further declined; however, some progeny remained in the fin epidermis and maintained their size for more than 4 months (Fig. 4E). Taken together, the fate-tracking analysis of later-arising RE cells revealed that a population of progeny of the RE cells actually contribute to the regenerated epidermis, although many of the RE-derived colonies disappear from the regenerated epidermis by 3 wpa.

The fate-tracking studies described above showed that many of the RE progeny are lost from the regenerated tissue, with the exception of a fraction of them. We next questioned how these different fates arise. To do this, we conducted single-cell fate tracking of RE cells by inducing sparse Cre labelling. It has been reported that fulvestrant (ICI182780; hereafter ICI), an antagonist of the oestrogen receptor, induces CreERt2 recombination in cultured cells (Metzger et al., 1995; Feil et al., 1997). By applying 2.5 µM ICI from 2 to 3 dpa, we successfully labelled a small number of RE cells (Fig. 5A). Well-isolated EGFP⁺ colonies, each of which is considered to be a clone derived from a single RE cell, were randomly selected from both the proximal and distal regions at 5 dpa, and their fates were tracked every day by comparing their positions and morphologies. Furthermore, cell types and numbers composing the respective clones were determined every other day by analysing z-stack confocal images (Fig. S4; Table S1).

The results of the single clone tracking experiment were similar to those in Fig. 4B. Many of EGFP⁺ clones (27 out of 41, 65.9%) disappeared by 15 dpa (Fig. 5B). At 5 dpa, one of them (2.4%) contained only surface cells, whereas the rest were composed of either basal (68.3%) or suprabasal (29.3%) cells. As regeneration proceeds, the clones containing only the suprabasal cells became the clones containing both suprabasal and surface cells by 7 dpa, and finally became the clones containing only the surface cells at 15 dpa (Fig. 5C). Many of the basal cell clones at 5 dpa also produced suprabasal and surface cells during regeneration, but they retained basal cells at 15 dpa, resulting in clones containing all three epidermal cell types (Fig. 5C). These observations suggest that epidermal cellular turnover begins immediately after formation of the regenerated epidermis, as it does in normal epidermis, and that the presence of basal cells is essential for maintaining long-lasting epidermal clones.

With respect to cell proliferation, some clones without basal cells (circles filled in blue, Fig. 5B) showed increases in cell numbers during regeneration (numbered 6, 10, 23 and 24 in Fig. 5B), but they stopped to proliferate or disappeared by 15 dpa. Although almost all of the clones that were retained at 15 dpa (Fig. 5B, marked with the thick blue outline) were basal cell-containing clones (Fig. 5B, circles filled in red), different proliferation patterns were observed between them; e.g. there were clones with almost no proliferation (numbers 18, 32 and 40), clones with continuous proliferation (numbers 12, 25, 26, 34, 36, 38, 39 and 41), and clones that initially proliferated, but later decreased their size (number 27, 33 and 35). Significantly, it appeared that these clones with different proliferative profiles were distributed at random, suggesting that the RE-derived epidermal clones stochastically adopt heterogeneous proliferative fates.

Importantly, fate tracking of RE-derived epidermal clones showed that many of the clones, even clones with basal cells and/or an active cell proliferation profile, also disappeared by 15 dpa (Fig. 5B). By plotting the initial position within the fins of respective clones that had survived or disappeared at 15 dpa, we found that clones surviving beyond 15 dpa (circles with a blue edge, Fig. 5B) were mostly situated within the proximal half of the regenerates in 5 dpa fins (Fig. 5D), and that clones in the distal region were moved caudally and expelled from the regenerated fin, revealing that many of the epidermal cells derived from the later RE transiently contribute to the regenerated epidermis and are lost from fin as a result of dynamic epidermal movement to the distal direction. It is thought that cells reaching the fin edge disappear either by apoptosis or by simple detachment from the fin edge; however, we were unable to verify these possibilities owing to the apoptosis that always occurred at the fin edge.

In homeostatic epidermal cell turnover, suprabasal cells, committed progenitor cells or differentiated keratinocytes do not produce basal stem cells in mammalian skin. However, studies have reported that during epimorphic regeneration of amphibian limbs and fish fins, de-differentiation of mature muscle cells or osteoblasts to form proliferative progenitor states can occur (Sandoval-Guzmán et al., 2014; Knopf et al., 2011; Geurtzen et al., 2014). Thus, it is possible that de-differentiation of suprabasal cells or surface epidermal cells, and the resulting supply of basal cells, could enable perfect skin regeneration.

To examine this possibility, we generated an additional transgenic line that express CreERt2 under control of the keratin4 (ker4) promoter, which is known to be activated in the non-basal layer epidermis (Wehner et al., 2014; Fig. S5), and tracked the fate of suprabasal and/or surface cells in the epidermis. At 2 days after treatment of the double Tg(krt4:creERt2;Olactb:loxP-dsRed2-loxP-egfp) with TAM for 24 h (Fig. 6A), EGFP⁺ cells appeared in the epidermis of uncut fins (Fig. 6B), mainly in the suprabasal and surface cell layers (Fig. 6C), indicating that the suprabasal and surface cells had been successfully labelled. In contrast, TAM-independent Cre-recombination was rarely observed in uncut fins or amputated fins without TAM treatment (Fig. 6D).

At 2 days after TAM treatment, fin amputation was performed, and the fates of EGFP⁺ cells were analysed by identifying the epidermal cell types from the z-stack images of confocal microscopy. The results revealed that most of the EGFP⁺ cells were observed in the non-basal layers like the suprabasal and the surface layers (Fig. 6E). As a small proportion of the EGFP⁺ cells (2.7%) were detected among basal cells in uncut fins following TAM treatment (Fig. 6F), the numbers of EGFP⁺ cells in the basal and non-basal layers were quantified and compared with those before and after fin amputation. The fraction of basal cells within the EGFP⁺ cells was 1.7% at 3 dpa, which was unchanged from that before amputation, indicating that suprabasal and surface cells of the epidermis do not extensively de-differentiate into basal cells during fin regeneration.

The finding that many of the regenerated epidermal clones are lost within a few weeks of regeneration raised a further question about how the lost epidermal cells are replenished. It is thought that new epidermal cells are supplied by an increase in basal cells, either locally in the adjacent proximal region or globally in the entire fin.

To explore these possibilities, we performed EdU labelling of proliferating cells in 10 dpa fins, a stage when the RE-derived epidermal cells are rapidly lost (Fig. 5B), and compared the number of EdU⁺ cells in different fin regions with those in uncut fins (Fig. 7A). In the uncut side of fins, only a few EdU⁺ nuclei were uniformly detected in different fin regions (Fig. 7B), although non-basal cells have a higher proliferative index than basal cells (Fig. 7C). In the amputated side of fins, an apparent increase in the number of EdU⁺ nuclei was observed both in basal and non-basal cells of the epidermis (Fig. 7D-F). Importantly, EdU⁺ nuclei were equally distributed both in the distal region, which is adjacent to the area from where the RE-derived epidermal cells are lost, and the proximal fin regions distant from the RE-derived epidermis (Fig. 7D-G), indicating that new epidermal cells are replenished by a global activation of epidermal cell proliferation.

More importantly, when the 3D distribution of EdU⁺ cells was analysed from the captured confocal z-stack images, an increase in EdU⁺ cells in the amputated side was detected among both the basal cells and non-basal cells of the epidermis (Fig. 7G). Therefore, these data suggest that both basal and non-basal cell proliferation replenish the lost epidermal cells in the distal fin area during and after regeneration, and that the increased proliferation of the basal cells in the entire fin region is crucial for providing long-lasting epidermal clones.

It has been suggested that the epidermis at the wounded site plays an essential role in regeneration (Goss, 1956; Thornton, 1957; Mescher, 1976; Tassava and Garling, 1979). However, the fate of the RE during and after regeneration has remained unclear. In this study, we performed Cre-loxP-mediated cell fate tracking, using fn1b as a marker induced in the RE, and succeeded in revealing the fates of the RE cells. Our study revealed heterogeneous cell fates of epidermal cells derived from the major RE cells expressing fn1b and further demonstrated the existence of a dynamic and global rearrangement of the epidermis during and after regeneration.

In this study, we successfully used fn1b as a marker of the RE to label many of the RE cells. In amphibian limb regeneration, the RE has been defined by its thick and layered morphology and the expression of characteristic genes such as MMPs, FGFs and collagens (Campbell and Crews, 2008). In zebrafish fin regeneration, previous studies have shown that genes such as igfr1 (Chablais and Jaźwińska, 2010), laminin beta 1a (Chen et al., 2015), fgf20a (Shibata et al., 2016) and sonic hedgehog (Armstrong et al., 2017) are expressed in the RE; however, the expression of these genes is restricted to the basal layer of the RE. As fn1b is expressed in all strata of the RE (Fig. 1A,D), fn1b is a better marker for labelling all types of the RE cells. As we have previously reported that junb is also expressed in the RE from an early stage of regeneration (Yoshinari et al., 2009), junb could also be an alternative gene for labelling RE cells.

Our live-imaging experiment using the fn1b Tg revealed that RE cells are recruited mainly from the inter-ray region. A previous study by Poleo et al. (2001) used DiI to trace the short-term fate of the RE and suggested that epidermal cells far proximal to the amputation site migrate to form the RE. Recently, Chen et al. (2016) have traced the fates of the superficial epidermal cells using Skinbow zebrafish and reported that superficial epidermal cells in the inter-ray region have a higher cellular mobility than those of ray superficial cells during wound healing. This observation is consistent with ours, in that many of the RE cells originate from the superficial inter-ray epidermis.

Using cell-fate tracking, we also demonstrated that RE cells recruited early in the wound healing process disappear through apoptosis by 5 dpa. The occurrence of such early RE cell apoptosis is consistent with an observation reported by Gauron et al. (2013), in which they detected apoptotic cells among epidermal cells during regeneration and wound healing. Because the early (i.e. before 12 hpa) migrating epidermal cells display little or no cell proliferation (Nechiporuk and Keating, 2002; Jaźwińska et al., 2007), it is thought that the epidermal cells around the wound are recruited to the wounded site without cell proliferation, in order to efficiently make a transient epidermal seal over the wound. It is possible that these short-lived cells could also play an additional role in regeneration, because Gauron et al. (2013) have reported that inhibition of apoptosis in epidermal cells impaired blastema formation.

Our study has revealed for the first time that the RE cells actually contribute to the regenerated epidermis. However, unexpectedly, we also found that the RE-derived cells in the distal region are expelled by movement of the epidermal sheet in the caudal direction, and, in turn, following loss of these epidermal cells, they are replenished by a global activation of cell proliferation in nearly the entire area of the fin. These results raise further questions about why the distal RE-derived epidermal cells are discarded and how a cell proliferation signal reaches proximal fin regions far away from the wound; these questions will be the subjects of future studies.

Furthermore, it is also intriguing to note that the proliferation of the respective RE-derived epidermal clones occurs at random, independently of their positions (Fig. 5B). It has also been recently suggested that the basal layer contains heterogeneous stem cell populations in mice, which show different patterns of proliferation and differentiation during homeostasis and regeneration (Sada et al., 2016). It is possible that the basal cells are heterogeneous in advance and that their proliferative abilities could be related to their ageing status and/or numbers of cell division they have undergone. Alternatively, it is also conceivable that proliferative and differentiation capacity is stochastically determined within homogeneous progenitor cells (Clayton et al., 2007).

More significantly, cell tracking of non-basal cells suggested that de-differentiation of non-basal cells into basal cells does not occur during regeneration (Fig. 6). In some cases, such as regeneration of muscle during newt limb regeneration (Sandoval-Guzmán et al., 2014) or of osteoblasts during zebrafish fin regeneration (Knopf et al., 2011; Geurtzen et al., 2014), proliferating cells are generated by cell de-differentiation during regeneration. However, we did not detect an apparent and extensive de-differentiation of the non-basal epidermal cells into basal stem cells during fin regeneration, suggesting the importance of self-renewal of basal stem cells for epidermal regeneration. However, the process of perfect skin reformation during the epimorphic regeneration was not strikingly different from that of mammalian skin repair and renewal. It is thought that this is also true for the trunk and other skin regions of zebrafish, and therefore the difference of regenerative abilities between fish and mammalian skin is possibly due to the proliferative ability of the basal stem cells. Elucidation of the control mechanism for self-renewal of basal stem cells should therefore open a way to developing scar-less regeneration or rejuvenation in mammalian skin.

Wild-type zebrafish (Danio rerio) strain, which was originally derived from the Tubingen strain and has been maintained in our facility for more than 10 years by inbreeding, and all transgenic lines were kept in a recirculating water system with a 14 h day/10 h night photoperiod at 28.5°C. Tg(krt4p:gal4) (Wada et al., 2013) and Tg(UAS:GFP) (Asakawa et al., 2008) were generous gifts from Hironori Wada  (Kitasato University, Japan) and Koichi Kawakami (National Institute of Genetics, Mishima, Japan), respectively. All animal procedures were approved by the Animal Care and Use Committee at the Tokyo Institute of Technology. All the animals were handled according to the Animal Research Guidelines of the Tokyo Institute of Technology. All surgical procedures were performed under tricaine (3-aminobenzoic acid ethyl ester) anaesthesia, and every effort was made to minimize suffering. Experiments on adult regeneration were performed using 3- to 12-month-old adult zebrafish.

Fin amputation was performed under anaesthesia with 0.002% tricaine (3-aminobenzoic acid ethyl ester, Sigma-Aldrich) and the caudal fins were cut in a straight line using a scalpel. For the single-cell tracking experiment shown in Fig. 5, fins were amputated at a position 1 mm distal from the root of the fin. For other experiments, fins were amputated at the middle position of the central fin ray.

BAC recombineering and generation of fn1b transgenic fish were performed as described previously (Ando et al., 2017). The egfp or creERt2 cassettes for BAC recombineering were amplified by PCR using the following primers (lowercase letters indicate the regions homologous with the fn1b gene): fn1b egfp fw, 5′-agccgaaatacagtcaaagccagaagctgctctccataacgcgggtgaaaATGGTGAGCAAGGGCGAGGA-3′; fn1b egfp rv, 5′-ggcttgtcttttacttttccctgcggattgtggcatgcagtggacagTCAGAAGAACTCGTCAAGAAGGC-3′; fn1b creERt2 fw, 5′-cgaaatacagtcaaagccagaagctgctctccataacgcgggtgaaaATGTCCAATTTACTGACCGTACA-3′; and fn1b creERt2 rv, 5′-ggcttgtcttttacttttccctgcggattgtggcatgcagtggacagatcCACAGGATCAAGAGCACCCG-3′.

By crossing the founder F0 zebrafish with each other, or with wild-type zebrafish, zebrafish lines expressing EGFP were obtained. As all of these lines displayed indistinguishable levels of EGFP expression, one line was selected to establish Tg(fn1b:egfp). Tg(fn1b:creERt2) was first screened for EGFP expression in the lens and then tested for Cre recombination by treating with TAM in the adult amputated fin. One F0 line that had a distribution of Cre-labelled cells consistent with that of Tg(fn1b:egfp) was identified, and was established as the Cre-expressing line.

Tg(krt4:mchrerry-t2a-creERt2) was generated by replacing the heat-shock promoter of pTol2(hsp70l:mcherry-t2a- creERt2) (Yoshinari et al., 2012) with the krt4 promoter using the SfiI sites. The plasmid DNAs at a concentration 25 ng/μl were injected, along with 25 ng/μl transposase mRNA, into fertilized eggs at the one-cell stage.

Whole-mount in situ hybridization was performed according to a standard protocol (Thisse and Thisse, 2008). For RNA probe generation, a region of the fn1b-coding sequence was amplified by PCR using the following primers: fn1b fw, 5′-GCATGTGTCATGGACCGCAC-3′; and fn1b rv: 5′-AGCGTACAGTCAAGGTATTG-3′.

The PCR product was cloned into pCR4 vector (Clontech). The egfp probe was synthesized from pCS2-egfp, which harbours the egfp sequence in the pCS2 vector.

Immunostaining was performed as described previously (Shibata et al., 2016). Anti-E-cadherin antibody (BD Transduction Laboratories) was used at 500 ng/ml. The anti-GFP antibody (Nacalai Tesque) was used at a 1:1000 dilution. Fins were mounted in 80% glycerol containing 25 mg/ml triethylenediamine (DABCO, Nacalai Tesque) as an anti-fading reagent. Images were obtained using a confocal microscope. Cell death was detected by TUNEL staining, as described previously (Hasegawa et al., 2017).

The Tg strains carrying fn1b:creERt2 or krt4:creERt2 were crossed with the Tg(Olactb:loxp-dsred2-loxp-egfp) strain (Yoshinari et al., 2012) to generate the double Tg zebrafish. Cre-loxP recombination was induced by treating the double Tg strains with either 4-OH TAM (Sigma-Aldrich) or ICI (Sigma-Aldrich) in fish water (0.3% artificial sea salt, 0.0001% Methylene Blue). The conditions for Cre-loxP recombination are indicated in the respective figure legends.

The number and area of the epidermal cell colonies that were derived from the Cre-labelled RE cells (Fig. 4C,D) were quantified using ImageJ software (ver1.49; https://imagej.nih.gov/ij/). To measure colony area, fluorescent images were obtained using a stereoscopic microscope and were binarized by applying a colour threshold (Red-Green-Blue colour model) to select only the areas appearing in green and analysed using the ‘analyse particles’ command in ImageJ software.

For live imaging of the Tg strains, zebrafish were anaesthetized with 0.002% tricaine. Fins were embedded in 1.5% low melting point agarose gel and observed under a 40× water-immersion objective lens using a confocal laser scanning microscope (LSM780, Carl Zeiss). For single cell tracking, confocal images were captured at intervals of 2 days.

Proliferating cells were detected by EdU labelling using the Click-iT EdU Imaging Kit (Thermo Fisher Scientific), as described previously (Shibata et al., 2016). Only the dorsal half of the fins was amputated at the mid position of the central fin ray, while the ventral half of the fin was used as an uncut control. Zebrafish were incubated in a solution containing 50 µM EdU in fish water for 6 h at 28.5°C before fin fixation. After EdU detection, samples were counterstained with DAPI [0.1 µg/ml in PBS containing 0.1% Triton X-100 (PBTx)], washed with PBTx and mounted in 80% glycerol containing DABCO. The third fin rays from the dorsal (regenerated) or ventral sides (uncut) were used for scoring EdU⁺ cells. The epidermal or non-epidermal identities of EdU⁺ cells were determined on the captured confocal z-stack images by using the ZEN microscope software (Carl Zeiss).

Clutch-mates were randomized into different treatment groups for each experiment. All experiments were performed with at least two biological replicates, using the appropriate number of samples for each replicate. Sample sizes are indicated in each figure legend. For expression patterns, at least five zebrafish were examined. Statistical analyses were performed using Microsoft Excel 2013. All data are displayed as mean±s.e.m. Sample sizes, statistical tests and P values are indicated in the respective figure legends. Student's t-tests (two-tailed) were applied when normality and equal variance tests were passed (Figs 5D, 6F and 7G).

We thank Hironori Wada (Kitasato University) for providing the Tg(krt4p:gal4) line and Koichi Kawakami (National Institute of Genetics, Mishima) for the Tg(UAS:GFP) line.