Retinoic acid signaling controls the formation, proliferation and survival of the blastema during adult zebrafish fin regeneration

Fish and amphibians have the ability to regenerate appendages that are lost or injured. Following amputation, the lost appendage regrows through a proliferation-dependent process known as epimorphic regeneration that involves three successive stages: wound healing, blastema formation, and regenerative outgrowth and repatterning. Despite great progress in recent years (Brockes and Kumar, 2008; Tanaka and Reddien, 2011), the underlying molecular mechanisms are still insufficiently understood. The retinoic acid (RA) signaling pathway (Theodosiou et al., 2010) has a long history in the study of vertebrate appendage regeneration. Treatment of regenerating amphibian limbs with excess RA causes patterning defects and a respecification of positional information (Maden and Hind, 2003; Maden, 1982; Maden, 1983; Niazi and Saxena, 1978). A role for RA has therefore been invoked in proximal-distal patterning. However, reliable loss-of-function experiments to verify the endogenous role(s) of RA in limb regeneration have never been performed.

Owing to the advantages offered by genetic screens, transgenesis and chemical genetics, fin regeneration in adult zebrafish has received exceptional attention (Iovine, 2007). The adult caudal fin consists of bony fin rays that are connected to each other by soft interray tissue (Akimenko et al., 2003; Becerra et al., 1983). Each fin ray is composed of two facing, concave hemirays that surround a core of fibroblasts, osteoblasts, pigment cells, nerves and blood vessels. The RA receptor rarga has been found to be strongly expressed in the adult fin blastema (White et al., 1994) and gene expression profiles of regenerating larval and adult fins have identified aldh1a2 (raldh2), which encodes the major enzyme for embryonic RA synthesis, as highly expressed (Mathew et al., 2009). In zebrafish larvae, repair of the caudal fin fold after amputation has been shown to depend on RA signaling (Mathew et al., 2009). However, larval fin folds are different from adult fins in many respects and it is unclear whether the signaling mechanisms driving larval regeneration apply to the adult blastema. Taken together, despite more than three decades of research into the effects of exogenous RA on regenerating amphibian limbs, evidence for a functional involvement of RA signaling in regenerating appendages of adult vertebrates is still missing. In this study, we demonstrate that RA signaling is essential for adult fin regeneration and provide mechanistic insights into a function for RA signaling in wound epidermis formation and in the generation and maintenance of the blastema.

Zebrafish strains of Konstanz wild types and the transgenic lines [Tg(hsp70I:dn-fgfr1)]^pd1 (Lee et al., 2005), [Tg(hsp70l:dn-zrar-egfp)]^pd18 (Kikuchi et al., 2011) and [Tg(hsp70I:cyp26a1)]^kn1 were reared and staged at 28.5°C according to Kimmel et al. (Kimmel et al., 1995). Transgenic strains were analyzed as heterozygotes; wild-type siblings served as controls. Fish that were 3-14 months old were used for regeneration experiments. Caudal fins were amputated along the dorsoventral axis, intersecting the median rays approximately halfway. Fish were allowed to regenerate for various times at 27-28°C.

To construct the hsp70I:cyp26a1 transgene (Kikuchi et al., 2011), egfp from phsp70I:egfp (Halloran et al., 2000) was replaced by zebrafish cyp26a1 (NM_131146) and the entire cassette was inserted into the I-SceI backbone vector (Thermes et al., 2002). Plasmid DNA was injected together with I-SceI meganuclease (NEB) into one-cell stage embryos to create germline transgenic founder fish.

Embryos were heat shocked at 38°C for 1 hour. Heat shock of adult fish was performed once daily by transferring fish from 27-28°C water to 33-34°C water for 30 minutes and subsequently to 38°C water for 1 hour.

For bromodeoxyuridine (BrdU, Sigma) and all-trans RA (Sigma) treatments, fish were injected intraperitoneally with 30 μl solution. BrdU, at 2.5 mg/ml in PBS, was injected 6 hours (during blastema formation) or 30 minutes (during regenerative outgrowth) prior to fixation. RA was injected at 1 mM in 1% DMSO/PBS. Control fish were injected with an equivalent concentration of DMSO/PBS. Fins of BrdU-treated fish were fixed in 4% PFA in PBS, washed in 0.3% Triton X-100 in PBS (PBTx) and DNA was denatured with 2M HCl for 20 minutes at 37°C. Fins were washed, then incubated with mouse anti-BrdU antibody (1:50, Sigma) and subsequently with goat anti-mouse Alexa Fluor 568 antibody (1:800, Molecular Probes). Cryosections were counterstained with DAPI.

For cryosectioning, fins were embedded in 1.5% agar/5% sucrose in PBS. Embedded fins were saturated in 30% sucrose and subsequently frozen in Tissue-Tek O.C.T. Compound (Sakura) in liquid nitrogen. Longitudinal sections were cut at 18 μm.

Digoxigenin-labeled RNA antisense probes were synthesized from cDNA templates: egr2a (krox20) (Oxtoby and Jowett, 1993), myoD (myod1 – Zebrafish Information Network) (Weinberg et al., 1996), aldh1a2 (Grandel et al., 2002), rarga (Joore et al., 1994), rdh10a (ImaGenes, IRAKp961E15293Q) and rdh10b (ImaGenes, IRBOp991E024D). In situ hybridization of whole fins and embryos was performed as previously described (Poss et al., 2000a) with minor modifications. For in situ hybridization on cryosections, proteinase K treatment was replaced by permeabilization in PBTx for 30 minutes prior to prehybridization. Stained whole-mounts and sections were cleared in ethanol. Whole-mounts were transferred into 75% glycerol in PBS for documentation.

For TUNEL staining on cryosections, fins were fixed and processed as for in situ hybridization on sections. Sections were permeabilized in PBTx and equilibrated with terminal deoxynucleotidyl transferase (TdT) buffer [200 mM potassium cacodylate, 25 mM Tris, 0.05% (v/v) Triton X-100, 1 mM CoCl₂, pH 7.2]. The buffer was subsequently replaced with TdT buffer containing 0.5 μM fluorescein-12-dUTP, 40 μM dTTP and 0.02 units/μl TdT (all Fermentas). Slides were incubated at 37°C for 3 hours and washed in PBT. Sections were blocked in 0.5% Blocking Reagent (Roche) in PBT and incubated with sheep anti-fluorescein-AP antibody (1:2000, Roche). The staining reaction was carried out as for in situ hybridization.

For quantification of BrdU-labeled and TUNEL-labeled cells, two to six representative sections per fin from 5-11 fish per group were used. Labeled cells were counted within 100 μm proximal to the amputation plane in the epidermis and inside an area of 50 × 100 μm in the mesenchyme at 32 hours post-amputation (hpa). For quantification at 3 and 4 days post-amputation (dpa), cells were counted distal to the amputation plane and calculated per 500 μm regenerate length. To determine growth in regenerating fins of heat-shocked RA-injected hsp70:dn-fgr1 fish, fins were photographed at 70 hpa (before the first heat shock and first RA injection) and 5 dpa. The length of the regenerate (from the amputation plane to the distal tip) was measured using AxioVision software (Carl Zeiss). Growth between the two time points was calculated for each fish (n=10-12 fish per group). Statistical significance was calculated using Student’s t-test.

Fins were fixed in 4% PFA in PBS, transferred to methanol and stored at –20°C. Fins were rehydrated prior to cryosectioning. Sections were stained in Mayer’s Hematoxylin Solution (Sigma) for 3-5 minutes, washed in water and cleared in 0.37% HCl in 70% ethanol for 5-10 seconds.

For RNA extraction at 0, 6 and 10 hpa, tissue within 1 mm proximal to the amputation plane was harvested. At 73 hpa, tissue distal to the amputation plane was harvested. Each sample was prepared from 4-11 fins. Total RNA was extracted with Trizol reagent (Invitrogen) and treated with DNase I (Fermentas). Equal amounts of total RNA from each sample were reverse transcribed with SuperScript III reverse transcriptase (Invitrogen) using oligo(dT) primers. For all samples, reverse transcriptase-negative controls were included to verify the purity of the samples. Quantitative real-time PCR (qPCR) was performed using a C1000 thermal cycler combined with a CFX96 real-time PCR detection system (Bio-Rad) and Maxima SYBR Green qPCR Master Mix (Fermentas). Primers are listed in supplementary material Table S1. qPCR reactions for each sample and each target gene were performed in triplicate. Three samples were used for each gene at 0, 6 and 10 hpa; two to three samples were used for each gene at 73 hpa. qPCR data were analyzed using CFX Manager software (Bio-Rad). Expression levels at 0 and 6 hpa in wild-type fins were normalized to ef1a levels (normalization to gapdh levels produced very similar results). Expression levels at 73 hpa were normalized to actb1 levels (normalization to ef1a levels produced very similar results). Expression levels at 0 and 10 hpa under altered RA levels were normalized to the input RNA amount by performing a RiboGreen assay (Invitrogen) for exact RNA quantification. This technique was used because normalization to different reference genes gave conflicting results. Statistical significance was calculated using Student’s t-test.

A previous microarray analysis has shown that expression of the RA-synthesizing enzyme aldh1a2 is upregulated in regenerating caudal fins of adult zebrafish at 24 hours post-amputation (hpa) (Mathew et al., 2009). In order to better understand the spatial and temporal expression of aldh1a2 and other RA pathway components during blastema formation we performed gene expression studies and found that aldh1a2 expression is upregulated within 6 hpa (Fig. 1A). aldh1a2 transcripts were detected within approximately one segment length proximal to the amputation plane in the ray and interray mesenchyme (Fig. 1B,C; data not shown), whereas the most distal mesenchyme is initially (at 18 hpa) devoid of aldh1a2. RA synthesis through aldh1a2 requires a reliable source of retinaldehyde. Accordingly, we found that expression of retinol dehydrogenase 10b (rdh10b) is induced after amputation (Fig. 1A). rarga has been shown to be expressed in the mature blastema (White et al., 1994). We investigated rarga expression in the fin stump and detected 1.7-fold higher expression of rarga at 6 hpa as compared with 0 hpa (Fig. 1A). Together, our expression analyses shows that fin amputation induces upregulation of essential components of the RA pathway.

During blastema formation, which occurs at ∼12-48 hpa, cells of the ray mesenchyme within one to two segment lengths proximal to the amputation plane start to proliferate and migrate distally to form the blastema. Expression of aldh1a2 in this region suggests that blastema formation requires high levels of RA at the local origin of blastema cells. To investigate the consequences of impaired RA signaling for fin regeneration we developed a transgenic zebrafish line that allows heat shock-inducible degradation of endogenous RA. The Tg(hsp70I:cyp26a1)^kn1 strain, referred to hereafter as hsp70:cyp26a1, harbors zebrafish cyp26a1, which encodes an RA-degrading enzyme, driven by the heat-inducible zebrafish hsp70 promoter (Halloran et al., 2000). We found that induction of the transgene during embryogenesis results in strong, ubiquitous cyp26a1 expression (not shown) and phenocopies the effects of complete loss of RA signaling. A brief heat shock at 6 hours post-fertilization (hpf) caused specific and organ-wide developmental defects that are known hallmarks of impaired RA signaling (supplementary material Fig. S1) (Begemann et al., 2001; Begemann et al., 2004; Gibert et al., 2006; Grandel et al., 2002). When adult hsp70:cyp26a1 fish were exposed to a heat shock during fin regeneration, strong cyp26a1 expression could be detected in the whole fin (not shown). Thus, the hsp70:cyp26a1 line is a reliable tool to interfere with RA signaling in embryos and adult fish.

To test whether RA signaling is required for blastema formation, we applied daily heat shocks to adult hsp70:cyp26a1 fish starting with the first heat shock 2 hours before fin amputation. This treatment caused a complete and early block to fin regeneration (22/34 fish), whereas regeneration was unperturbed in heat-shocked wild-type fish (39/39 fish) (Fig. 2A). In addition, this effect is reversible: removing the heat-shock treatment resulted in normal blastema formation and complete fin regeneration. To determine the cellular nature of regenerative failure, we examined Hematoxylin-stained fin sections of heat-shocked wild-type and hsp70:cyp26a1 fish at 45 hpa. Whereas wild-type regenerates displayed a well-developed blastema between the amputation plane and a multilayered wound epidermis, hsp70:cyp26a1 fins exhibited a complete absence of blastema cells (Fig. 2B). Several layers of epithelial cells sealed the wound in hsp70:cyp26a1 fish, indicating normal re-epithelialization of the stump surface. However, cells of the basal epidermal layer did not adopt their typical cuboidal shape (Fig. 2C). Extracellular matrix remodeling and disorganization of the stump mesenchyme adjacent to the amputation site are an early response prior to blastema formation. Interestingly, disorganized stump mesenchyme proximal to the wound site was also observed in hsp70:cyp26a1 regenerates (Fig. 2D).

To confirm the absence of blastema cells in hsp70:cyp26a1 fish we examined the expression of fgf20a and msxb, two markers that are strongly expressed in blastema cells (Akimenko et al., 1995; Whitehead et al., 2005). Neither gene could be detected in heat-shocked hsp70:cyp26a1 fins at 32 hpa (fgf20a, 6/7 fins; msxb, 3/5 fins) and 48 hpa (msxb, 6/8 fins) (supplementary material Fig. S2). The absence of a distinct basal epidermal layer in hsp70:cyp26a1 fish suggests that the initial specification of the wound epidermis is affected. lef1, which marks the basal epidermal layer and the distal blastema (Poss et al., 2000b), could not be detected in hsp70:cyp26a1 fins at 46 hpa (3/4 fins) (supplementary material Fig. S2). Furthermore, lef1 expression was also absent at an earlier time point (at 32 hpa; 5/8 fins) demonstrating that lef1 expression is not initiated in the absence of RA signaling.

The failure of blastema formation might be a consequence of defects in wound healing and in the formation of a proper wound epidermis. We showed that this is not the case by inhibiting RA signaling after wound healing had taken place, applying the first heat shock at 24 hpa. Regeneration was completely blocked in 13 out of 21 hsp70:cyp26a1 fish (supplementary material Fig. S3). In summary, neither blastema cells nor the basal epidermal layer is formed or specified correctly in the absence of RA signaling.

During blastema formation, formerly quiescent cells of the ray mesenchyme start to proliferate and migrate towards the amputation plane. To understand why blastema formation fails in the absence of RA signaling, we assayed cell proliferation in heat-shocked hsp70:cyp26a1 fish at 32 hpa. We found a dramatic decrease in proliferating stump cells (Fig. 3A,B). Similar results were obtained for the transgenic strain hsp70:dn-zrar, in which heat-shock treatment induces expression of a dominant-negative zebrafish retinoic acid receptor alpha (rara) (Kikuchi et al., 2011) (Fig. 3C), providing independent evidence for the requirement of RA signaling for blastema formation. The lack of proliferating cells suggested that proliferation was either not induced or that cycling cells underwent cell death. To discriminate between these possibilities, we compared the number of dying cells between wild-type and hsp70:cyp26a1 stumps at 32 hpa. Since we did not observe enhanced cell death in hsp70:cyp26a1 stumps, neither in the mesenchyme nor in the epidermis (supplementary material Fig. S4), we conclude that induction of cell proliferation in the ray mesenchyme fails in the absence of RA signaling.

RA signaling might be sufficient to induce cell cycle entry of blastema progenitor cells, as cells of the ray mesenchyme become exposed to high levels of RA as a consequence of fin amputation. Previous studies have shown that RA treatment can cause mispatterning in regenerating fins, slow down regeneration or even block blastema formation (Géraudie et al., 1995; White et al., 1994). However, because these effects might have been caused by increased cell death, especially in the wound epidermis (Géraudie and Ferretti, 1997), we developed an RA treatment regime that efficiently enhances RA signaling in the regenerating fin, but does not induce cell death. We found that intraperitoneal (IP) injection of 1 mM RA dissolved in a low concentration of DMSO does not induce cell death during blastema formation and regenerative outgrowth, even if injected every 12 hours for several days (supplementary material Fig. S5B; data not shown). Increased RA signaling in the regenerate should result in decreased aldh1a2 and rdh10a transcript levels and enhanced cyp26a1 levels, as has been shown for embryonic development (Dobbs-McAuliffe et al., 2004; Hu et al., 2008). Accordingly, we detected an autoregulatory component of RA signaling in the regenerating caudal fin 4 hours after IP injection of 1 mM RA (supplementary material Fig. S5A).

We tested the effect of exogenous RA on proliferation of the ray mesenchyme during blastema formation. We injected RA every 12 hours, with the first injection directly after fin amputation, and assayed cell proliferation at 32 hpa. Mesenchymal proliferation was significantly increased in RA-treated fish (Fig. 3D), demonstrating that RA signaling is not only required for cell cycle entry but is also sufficient to increase the proliferation of stump cells. These findings clearly show that the previously reported negative effects of RA on fin regeneration were secondary effects caused by enhanced cell death.

To understand the molecular consequences of early upregulation of RA levels in the stump, we examined the effects of altered RA signaling on key pathways involved in blastema and wound epidermis formation. Fgf and Wnt/β-catenin signaling have been shown to be required for blastema formation and subsequent blastema proliferation. Within the first few hours after amputation, expression of fgf20a and wnt10a is strongly upregulated. Moreover, homozygous mutants in fgf20a (dob) fail to form a blastema and show an abnormal wound epidermis (Lee et al., 2005; Poss et al., 2000a; Stoick-Cooper et al., 2007; Whitehead et al., 2005). By contrast, the ligand responsible for activation of the Wnt/β-catenin signaling pathway during fin regeneration remains a matter of speculation. We examined whether induction of fgf20a and Wnt ligand expression in the fin stump is regulated by RA signaling by performing loss- and gain-of-function experiments. Amputation-induced fgf20a upregulation was clearly diminished in heat-shocked hsp70:cyp26a1 fish at 10 hpa, whereas upregulation of wnt10a and wnt10b was unaffected (Fig. 4A). Because not all hsp70:cyp26a1 fish respond equally well to the heat-shock treatment, the reduction in fgf20a expression observed might be underrepresentative. Furthermore, overexpression of cyp26a1 results in strong downregulation of baseline fgf20a expression in unamputated fins (Fig. 4A), suggesting that reduced upregulation of fgf20a in hsp70:cyp26a1 fins is not a secondary effect of impaired fin regeneration, but reflects a more direct requirement for RA signaling for fgf20a expression. Exogenous RA does not induce upregulation of fgf20a expression in unamputated fins and we did not detect increased upregulation of fgf20a in RA-treated fins at 10 hpa (Fig. 4B). Interestingly, RA treatment resulted in decreased upregulation of wnt10a at 10 hpa, but further increased regeneration-induced wnt10b upregulation. Strikingly, we found that exogenous RA is sufficient to induce wnt10b upregulation in unamputated fins.

Although re-epithelialization of the stump surface does not require RA signaling, formation of the basal epidermal layer depends on RA signaling. Chablais and Jazwinska (Chablais and Jazwinska, 2010) reported a fundamental role for Igf signaling in the formation of the basal epidermal layer via paracrine activation of Igfr in the wound epidermis. Igf2b is produced and secreted from cells of the stump mesenchyme, demonstrating the importance of interactions between the mesenchyme and the forming wound epidermis. The lack of the basal epidermal layer in RA-deficient fish might therefore be caused by a reduction in paracrine signals derived from the stump mesenchyme. To test this, we examined the effects of altered RA signaling on early igf2b expression in the fin stump. Loss of RA in unamputated fins had no effect on baseline igf2b expression, whereas regeneration-induced igf2b upregulation was reduced in heat-shocked hsp70:cyp26a1 fins (Fig. 4A). RA treatment further increased igf2b expression at 10 hpa and even induced igf2b upregulation in unamputated fins (Fig. 4B). Thus, RA is required for the strong induction of igf2b during regeneration and is sufficient to induce igf2b in the unamputated fin, indicating that impaired wound epidermis formation in hsp70:cyp26a1 fish is caused by impaired igf2b induction. Together, these findings demonstrate that RA signaling stimulates wnt10b expression and is crucial for amputation-induced fgf20a and igf2b expression in the fin stump.

Following the formation of a blastema aldh1a2, rarga and rdh10a are strongly expressed in the blastema (supplementary material Fig. S6), whereas rdh10b transcripts could not be detected. Since RA signaling is necessary for cell proliferation during blastema formation, we tested the effect of inhibiting RA signaling on proliferation of the mature blastema. We applied a single heat shock at 72 hpa and examined proliferation 8 hours later. Our analysis revealed a significant reduction of BrdU-positive cells in hsp70:cyp26a1 regenerates, both in the mesenchyme and epidermis (Fig. 5A,B). msxb is strongly expressed in blastema cells and has been shown to be required for blastema proliferation (Akimenko et al., 1995; Nechiporuk and Keating, 2002; Thummel et al., 2006). To test whether the observed decrease in proliferation is reflected in altered msxb expression, and to detect early expression changes, we examined msxb expression by qPCR at 73 hpa immediately after a single heat shock. We found that msxb transcripts were reduced to half their normal levels in hsp70:cyp26a1 regenerates (Fig. 5C). Thus, RA signaling is not only required for blastema formation but also for subsequent blastema proliferation.

Daily heat shocks starting at 72 hpa resulted in a reversible and robust block to regenerative outgrowth in hsp70:cyp26a1 fish (10/10 fish) and a loss of already regenerated tissue (Fig. 6A,C). Remarkably, within the first 10 hours after heat shock, the blastema in hsp70:cyp26a1 regenerates turned opaque (10/10 fish) (Fig. 6B), suggesting extensive cell death. In support of this, we found a high number of TUNEL-labeled mesenchymal cells in hsp70:cyp26a1 regenerates 8 hours after heat shock (Fig. 7A,B). Importantly, most TUNEL-positive cells were restricted to the distal half of the mesenchyme, indicating that undifferentiated blastema cells are unable to survive in the absence of RA signaling. Very similar results were obtained with the hsp70:dn-zrar line (supplementary material Fig. S7), demonstrating that the massive cell death in hsp70:cyp26a1 fish is caused by impaired RA signaling, rather than by non-specific effects of cyp26a1 overexpression.

Fgf, Wnt/β-catenin and Activin-βA, among other pathways, have been shown to positively regulate blastema proliferation (Jazwinska et al., 2007; Poss et al., 2000a; Stoick-Cooper et al., 2007). Cell death caused by inhibition of these pathways has not been reported. Thus, RA might be an essential part of the mechanism promoting survival of the blastema. Bcl2 family proteins are essential regulators of various cell death mechanisms, including apoptosis, necrosis and autophagy (Yip and Reed, 2008). bcl2 overexpression has been shown to inhibit cell death induced by many stimuli, including growth factor deprivation. Bcl2 is therefore a major candidate for RA-mediated protection from cell death in blastema cells and possibly other stress-induced states. Because bcl2 expression in the regenerating fin had not been investigated, we analyzed bcl2 transcript levels. qPCR analysis revealed 2.5-fold higher levels of bcl2 transcripts at 6 hpa compared with uninjured fins (0 hpa) (Fig. 7C), indicating that protection against cell death is enhanced in the regenerating fin. To test whether bcl2 expression is regulated by RA signaling, we performed loss- and gain-of-function experiments. We applied a single heat shock to wild-type and hsp70:cyp26a1 fish at 72 hpa and compared expression immediately following the heat shock. Transcript levels were significantly reduced in hsp70:cyp26a1 regenerates (Fig. 7D). We next increased RA signaling by IP injection of RA and examined bcl2 expression 4 hours later. Expression was 1.75-fold higher in RA-injected than in vehicle-injected fish (Fig. 7D). These results reveal a strong RA-mediated pro-survival mechanism in blastema cells that is mediated by upregulation of bcl2 expression.

To gain insights into the gene network underlying blastema proliferation we investigated regulatory interactions between RA, Fgf and Wnt/β-catenin at 73 hpa using a qPCR approach. Expression of the Fgf target mkp3 (dusp6 – Zebrafish Information Network) and the Wnt/β-catenin target axin2 was strongly diminished at the end of a single heat shock in hsp70:cyp26a1 regenerates (Fig. 8A). Conversely, treating fish for 4 hours with RA caused upregulation of axin2. Expression of the Fgf target mkp3 remained unchanged under these conditions, most likely owing to the short duration of the treatment.

We then investigated which Fgf and Wnt ligands might mediate the positive effect of RA. fgf20a showed increased expression in RA-injected fish and decreased following cyp26a1 overexpression. Expression of both wnt10 paralogs was unchanged in hsp70:cyp26a1 fins. However, RA treatment resulted in downregulation of wnt10a expression and in a striking upregulation of wnt10b. Thus, it is likely that wnt10b, either alone or together with other Wnts, mediates the positive effect of RA signaling on the Wnt/β-catenin pathway in the mature blastema. Wnt/β-catenin signaling has previously been shown to be required for Fgf signaling during fin regeneration and has therefore been suggested to act upstream of Fgf activation (Stoick-Cooper et al., 2007). To test this, we made use of the transgenic hsp70:dn-fgfr1 strain (Lee et al., 2005), which allows heat shock-induced inhibition of Fgf signaling. We applied a single heat shock and compared expression levels directly at the end of the heat shock. Expression of mkp3 was strongly reduced in hsp70:dn-fgfr1 regenerates, demonstrating the efficiency of Fgf signaling inhibition. Remarkably, both axin2 and aldh1a2 transcript levels were reduced in hsp70:dn-fgfr1 regenerates (Fig. 8A). Thus, Wnt/β-catenin signaling and RA synthesis are positively regulated by Fgf signaling. Together, these findings demonstrate that RA, Fgf and Wnt/β-catenin signaling regulate each other in a positive reciprocal manner, rather than following an epistatic hierarchy during fin regeneration (Fig. 8B).

If this model is correct, enhancing RA signaling in heat-shocked hsp70:dn-fgfr1 fish should not rescue blastema proliferation and regenerative outgrowth. To examine this, we blocked Fgf signaling in hsp70:dn-fgfr1 fish for 2 days during regenerative outgrowth by applying the first heat shock at 72 hpa and simultaneously enhanced RA signaling through daily RA IP injections. We found an increase in tissue distal to the amputation plane by ∼400 μm in vehicle- and RA-injected heat-shocked wild-type fish (supplementary material Fig. S8). By contrast, RA injection into hsp70:dn-fgfr1 fish failed to rescue regenerative outgrowth. Moreover, there was a decrease in already regenerated tissue in hsp70:dn-fgfr1 fish that suggests that the interactions we identified between the RA and Fgf pathways are also employed to promote cell survival.

The non-canonical Wnt pathway has been shown to act as a negative modulator of fin regeneration (Stoick-Cooper et al., 2007). Overexpression of wnt5b inhibits fin regeneration, whereas wnt5b loss-of-function accelerates regeneration. Interestingly, wnt5b expression is positively regulated by Fgf, Igf and Wnt/β-catenin signaling (Fig. 8A) (Chablais and Jazwinska, 2010; Lee et al., 2009; Stoick-Cooper et al., 2007), indicating that a negative-feedback mechanism modulates the overall rate of regeneration through non-canonical Wnt signaling. Because of the observed positive regulation between RA and the Fgf and Wnt/β-catenin pathways, we expected reduced wnt5b expression in hsp70:cyp26a1 regenerates. However, expression was mainly unaffected (Fig. 8A). Moreover, we found a decrease in wnt5b expression in RA-injected fish. Our findings show that the non-canonical Wnt pathway is negatively regulated by RA, indicating that RA signaling counteracts the negative-feedback loop that is activated by Fgf and Wnt/β-catenin signaling. Moreover, the interactions between RA, Fgf and Wnt/β-catenin in the mature blastema correlate with those observed for blastema formation.

In this study we have identified fundamental roles of RA signaling in adult fin regeneration, findings that contribute to a more thorough understanding of the molecular events underlying the development and maintenance of the regeneration blastema. Amputation of the caudal fin results in the activation of as yet unidentified signals that initiate blastema formation in the stump tissue. Strong aldh1a2 expression is rapidly induced in the stump mesenchyme, indicating that cells that will give rise to the blastema become exposed to high RA levels within the first few hours after amputation. Rapid induction of aldh1a2 is also a hallmark of zebrafish heart regeneration, where it is expressed in both the epicardium and endocardium, highlighting the importance of RA signaling in promoting cell division during the injury response of a variety of tissues (Kikuchi et al., 2011; Lepilina et al., 2006).

Activation of RA signaling in the fin stump is required for the strong amputation-induced fgf20a expression and promotes induction of wnt10b expression. Thus, RA, Fgf and Wnt/β-catenin signaling constitute parts of a signaling network that controls blastema formation. After re-epithelialization of the wound, mesenchymal cells in the stump start to proliferate and migrate distally to form the blastema. Loss of RA signaling blocks entry of the blastema precursors into the cell cycle. Conversely, exogenous RA is sufficient to promote mesenchymal proliferation, suggesting that RA signaling confers mitogenic activity during blastema formation. In addition, it might control an earlier step in blastema formation that is preconditional for cell cycle entry.

Another early and essential step in regeneration is the formation of a specialized wound epidermis. Although the initial sealing of the stump surface does not require RA signaling, specification of the basal epidermal layer fails in the absence of RA. Igf signaling has been shown to regulate proper wound epidermis formation via paracrine activation of Igf receptors in the forming wound epidermis (Chablais and Jazwinska, 2010). We show that RA signaling is required and probably sufficient to induce igf2b expression in mesenchymal cells underlying the wound epidermis. The lack of the basal epidermal layer in RA-deficient regenerates might therefore be caused by a reduction in paracrine signals derived from the stump mesenchyme.

Once the blastema has formed, RA signaling remains highly active in the mature blastema to ensure proliferation and survival. Besides a possible, more direct mitogenic function, RA signaling indirectly regulates blastema proliferation through integrating signals that either stimulate or inhibit proliferation (Fig. 8B). RA antagonizes the inhibitory effect of non-canonical Wnt signaling and, in parallel, promotes this same pathway via stimulation of Fgf and Wnt/β-catenin signaling. We propose that this complex regulatory network ensures tight control over blastema proliferation. Moreover, our findings demonstrate that RA, Fgf and Wnt/β-catenin signaling cooperate through mutually stimulatory interactions to regulate blastema proliferation, rather than acting in a fixed epistatic hierarchy. It remains to be shown whether this also holds true for the initiation of regeneration in the fin stump.

Mature blastema cells are in a fast-cycling state. It is an open and interesting question how such a highly proliferative cell population can be maintained in an adult animal. We found that expression of the pro-survival gene bcl2 is upregulated within the first few hours after fin amputation. Importantly, blocking RA signaling by overexpression of cyp26a1 or dn-zrar in the mature blastema results in downregulation of bcl2 expression, followed by massive cell death. Our data indicate that blastema cells receive protection against cell death through increased levels of the anti-apoptotic factor Bcl2, the expression of which is positively regulated by RA signaling. We therefore suggest that blastema cells possess a strong RA-mediated pro-survival mechanism that allows maintenance of the blastema in the adult fish and provides robustness to environmental perturbation. Interestingly, RA signaling inhibition during blastema formation does not cause cell death in the stump mesenchyme, nor does manipulation of RA signaling during the first few hours after amputation affect upregulation of bcl2 expression (data not shown). These findings indicate that only mature blastema cells require RA signaling for enhanced bcl2 expression and survival.

Exogenous RA has previously been shown to adversely affect fin regeneration and even block blastema formation (Géraudie et al., 1995; White et al., 1994), but the underlying cellular mechanisms have not been analyzed. A follow-up study by Géraudie and Ferretti found that incubation of zebrafish in high RA concentrations or IP injection of RA, when diluted in high DMSO concentrations, causes enhanced cell death in the wound epidermis and blastema (Géraudie and Ferretti, 1997). We have developed an RA treatment regime that does not induce cell death; rather, it demonstrably enhances RA signaling and positively influences fin regeneration. Thus, the reported negative effects of RA treatment (Géraudie et al., 1995; White et al., 1994) are very likely secondary effects caused by enhanced cell death.

An important question is whether this spectrum of specific functions of RA in zebrafish blastema development is at work in other regeneration-competent vertebrates. In support of this view, endogenous RA has been detected in salamander blastemas (Brockes, 1992; Scadding and Maden, 1994; Viviano et al., 1995). Furthermore, axolotl limb regeneration can be blocked by applying aldehyde dehydrogenase inhibitors with broad target specificity (Maden, 1998), suggesting that RA signaling fulfils similar roles in fin and limb regeneration.

In regenerating salamander limbs, RA is thought to be synthesized in the wound epidermis (Viviano et al., 1995); however, the expression of genes encoding RA-synthesizing enzymes has not been investigated so far. Although we cannot exclude the possibility that RA might also be produced by epidermal cells in the regenerating fin, the zebrafish aldh1a2 and rdh10a expression patterns strongly suggest that RA production is restricted to the stump mesenchyme and blastema. Interestingly, a study by McEwan et al. (McEwan et al., 2011) has shown that expression of the Xenopus aldh1a2 ortholog is not upregulated in regenerating tadpole hindlimbs. However, expression is retained from development in proximal cells bordering the body wall, indicating that an RA source would be available for regeneration. Investigating the sources of RA in regenerating limbs of adult urodele amphibians might reveal important differences in RA distribution between the regenerating limbs of adult and larval amphibians as well as between regenerating limbs and fins. Thus, although the roles of RA in regeneration might be conserved, its sources might have diverged during evolution.

In amphibians, treatment of regenerating limbs with RA results in a phenomenon known as ‘super-regeneration’, in which additional limb structures are regenerated that would normally be found proximal to the amputation plane (Maden and Hind, 2003; Maden, 1982; Maden, 1983; Niazi and Saxena, 1978). RA has therefore been proposed to act as a morphogen responsible for a gradient of positional information along the proximal-distal (P-D) axis during limb regeneration. It has been demonstrated that the P-D pattern of limb segments in mouse and chicken is specified during embryonic development through a balance between proximal RA and distal Fgf activity (Cooper et al., 2011; Roselló-Díez et al., 2011). These studies support a patterning function of RA signaling in vertebrate limbs. Provided that loss-of-function experiments can show that this also holds true for limb regeneration, super-regeneration would be the expected outcome, reflecting the earlier patterning role of RA in limb development. So far, investigations in teleosts into a putative role for RA signaling in P-D patterning of the regenerating caudal fin have proved extremely difficult owing to the lack of reliable readouts of P-D patterning and insufficient knowledge concerning the processes that set up the caudal fin P-D axis. Although fin rays branch dichotomously, the mechanisms controlling branching are unknown and might not be appropriate readouts of a hypothetical P-D patterning gradient. The findings reported here are the first to be supported by loss- and gain-of-function experiments that propagate the idea that RA signaling is an essential component of vertebrate appendage regeneration. The mechanistic framework provided here should inform and advance future research to help uncover the function of RA during the repatterning phase of regeneration and to understand its roles from an evolutionary perspective.

We thank Prof. Axel Meyer, in whose laboratory the work was carried out, for continuous support; M. Hammerschmidt for hsp70:dn-fgfr1 fish; K. Poss for hsp70:dn-zrar fish; I. May for fish care; and D. Leo and A. Offergeld for assistance in screening transgenic fish.