Introduction

Axolotls (Ambystoma mexicanum) possess remarkable regenerative abilities, enabling the regeneration of entire limbs after amputation. Among tetrapods (four-limbed vertebrates), only urodele amphibians retain the lifelong ability to regenerate fully developed limbs. Investigating the molecular mechanisms underlying axolotl limb regeneration could provide valuable insights for advancing regenerative medicine in humans, potentially leading to new therapies for tissue repair and organ regeneration.

The limb regeneration process can be divided into two stages: the blastema (regenerating limb primordium) induction process and the limb patterning process. The induction of a blastema, which contains highly proliferative multipotent and unipotent cells, depends on the presence of nerves at the injured region. When a denervated limb is amputated, a blastema is not induced (Todd, 1823). The blastema induction process is followed by the limb patterning process. In limb development, positional values are thought to be established along three-dimensional axes— anteroposterior, dorsoventral, and proximodistal—based on positional information. For example, WNT7A, secreted from dorsal ectoderm, acts as dorsal positional information to induce Lmx1b expression, which acts as dorsal positional value, in underlying mesenchymal cells (Riddle et al., 1995; Vogel et al., 1995; Cygan et al., 1997; Chen et al., 1998; Chen, 2002). Similarly, SHH is known to serve as posterior positional information, regulating anteroposterior patterning (Riddle et al., 1993). For successful limb patterning in limb regeneration, interactions among anterior, posterior, dorsal, and ventral cells within a blastema have been considered essential. The importance of these interpositional interactions was investigated in creating double-half limbs. Amputating the double-half limbs can provide a blastema in a “one-positional-value-missing” state. For example, when a double-dorsal limb is amputated, the resulting blastema lacks ventral positional value. Such blastemas form hypomorphic, spike-like structures or fail to regenerate entirely (Bryant, 1976; Bryant and Baca, 1978; Burton et al., 1986). It is noteworthy that structural patterns along the anteroposterior axis have been lost, even though both anterior and posterior cells should be present in a blastema induced from double-dorsal and double-ventral limb. Additional experiments involving ectopic contact between opposite regions (e.g., anterior-posterior or dorsal-ventral) of a blastema further support the critical role of interpositional interactions in the limb patterning process, as such interactions can lead to ectopic limb formation (Iten and Bryant, 1975; Bryant and Iten, 1976; Maden, 1980; Stocum, 1982). These findings suggest that specific signals from all four positional domains must be integrated for successful limb patterning, such that the absence of any one of them leads to failure. In the present study, we define these position-specific signals required for a blastema to form a limb as “positional cues.” We distinguish “positional cues” from “positional information” to clarify their distinct roles: positional information refers to signals that induce positional values within recipient cells, whereas positional cues do not necessarily induce positional values and may act downstream of positional values. For example, WNT7A from the dorsal ectoderm in amniote limb development functions as dorsal positional information because it induces the dorsal positional values within the underlying mesenchyme. In contrast, FGF8 from the anterior region in axolotl limb regeneration functions as the anterior positional cue because it can substitute for anterior tissues in the limb patterning process (Nacu et al., 2016). Of course, these definitions are not mutually exclusive. One important feature of positional cue molecules is that they can substitute for tissues of one orientation in limb regeneration. To further understand limb regeneration, the molecular basis of the integration of the positional cues in limb patterning process should be elucidated.

The accessory limb model (ALM), a non-amputation experimental system, provides a valuable approach for studying axolotl limb regeneration, particularly the role of interpositional interactions (Endo et al., 2015). In this model, a blastema is induced through skin wounding combined with nerve deviation (Endo et al., 2004, Satoh et al., 2007). When ALM surgery is performed on the anterior side, the induced blastema (ALM blastema) lacks posterior-derived cells and is unable to form a properly patterned limb. This limitation can be overcome by supplying posterior-derived cells via a skin graft from the posterior side, enabling the successful regeneration of a patterned limb. Similarly, an ALM blastema can be induced in a position-specific manner along the limb axes. In this case, the induced ALM blastema will lack cells from the opposite side. This condition allows the investigation of blastema induction and limb patterning by selectively removing cells from specific orientations of the limb. Analyzing the characteristics and developmental limitations of these ALM blastemas provides valuable insights into the positional cues that regulate axolotl limb regeneration.

The molecular basis of the interpositional interactions within a blastema in the limb patterning process has been partially elucidated (Nacu et al., 2016). In a normal blastema, Fgf8 and Shh are expressed at the anterior and the posterior regions, respectively. A previous study demonstrated that FGF8 and SHH proteins can substitute for anterior and posterior tissues, respectively, to form a patterned limb. Moreover, these proteins function as mutually inductive signals, maintaining each other’s expression. These findings suggest that FGF8 and SHH serve as the anterior and posterior positional cues, respectively. In contrast, the molecular basis of dorsal and ventral positional cues remains poorly understood. Previous studies have shown that retinoic acid supplementation can convert the dorsal positional value to the ventral, enabling patterned limb formation even in the absence of ventral tissues (Vieira et al., 2019). However, the ventral positional cue remains uncertain, as retinoic acid induces ventral positional values rather than acting as the ventral positional cue itself, suggesting the existence of another, downstream ventral positional cue. Consequently, the mechanisms underlying dorsal and ventral positional cues, as well as their integration with anterior and posterior cues, remain unclear.

In the present study, we investigated the roles of dorsal and ventral positional cues in limb patterning. We confirmed the necessity of dorsoventral tissue contact for limb patterning using the ALM. We found that the induction of Shh expression was dependent on the co-existence of dorsal and ventral cells in axolotl limb regeneration. Furthermore, we identified Wnt10b and Fgf2 as dorsal and ventral positional cues, respectively, by RNA-seq analysis, and confirmed that these genes mediate Shh expression in axolotl blastemas. Our results suggest that the crucial role of the dorsal and ventral positional cues in the limb patterning process is to induce Shh expression, thereby enabling anteroposterior interaction. This model provides a framework for understanding the integration of the four positional cues during axolotl limb regeneration.

Results

Dorsal-ventral tissue contact is required for limb regeneration in the ALM

We performed the ALM experiment on Ambystoma mexicanum to investigate whether dorsoventral tissue contact, as well as anteroposterior tissue contact, is required for limb regeneration (Fig. 1). The skin was removed from one side (anterior, posterior, dorsal, or ventral) of a limb, and the large nerve fibers running along the center of a limb (Nervus medianus and N. ulnaris, Fig. 1A) were dissected and rerouted to the wounded region. As mentioned before, we supplied the cells from the contralateral side of a limb as a skin graft in experimental conditions to ensure the presence of sufficient cells for patterned limb formation. We conducted eight types of ALM blastema inductions, which can be categorized into two groups: basic ALM blastemas and skin-grafted ALM blastemas. The basic ALM blastemas included anteriorly induced ALM blastema (AntBL), posteriorly induced ALM blastema (PostBL), dorsally induced ALM blastema (DorBL), and ventrally induced ALM blastema (VentBL), all of which should lack cells from the contralateral side. The skin-grafted ALM blastemas included AntBL with posterior skin grafting (AntBL+P), PostBL with anterior skin grafting (PostBL+A), DorBL with ventral skin grafting (DorBL+V), and VentBL with dorsal skin grafting (VentBL+D); these ALM blastemas should contain cells with all anteroposterior and dorsoventral orientations. For the ALM experiment, we defined the two large blood vessels running on the anterior and posterior sides at the stylopod level (Vena cephalica and V. basilica, Fig. 1A) as the anterior and posterior dorsoventral borders, respectively. At 10 days post-surgery (dps), an ALM blastema was observed in all experimental groups (Fig. 1B‒I). Consistent with previous studies (Endo et al., 2004; Nacu et al., 2016), limbs with multiple digits were formed from AntBL+P and PostBL+A (Table 1, Fig. 1F, G), whereas AntBL and PostBL either regressed or formed bump structures (Table 1, Fig. 1B, C). Similarly, DorBL+V and VentBL+D formed limbs (Table 1, Fig. 1H, I), whereas most DorBL and VentBL either regressed or formed bump structures (Table 1, Fig. 1D, E). Notably, only 1 out of 12 DorBL formed a limb, which may have resulted from ventral tissue contamination, as the rerouted nerves originally ran along the ventral side of the humerus. In contrast, none of 22 VentBL formed multi-digit limbs. These results suggest that both dorsal and ventral tissues, as well as anterior and posterior tissues, are required for successful limb formation.

Figure 1.

Table 1.

Gene expression patterns in ALM blastemas

To elucidate the characteristics of the ALM blastemas that typically fail to form limbs (AntBL, PostBL, DorBL, and VentBL), we analyzed gene expression patterns at 10 dps (Fig. 2). Lmx1b, a gene necessary and sufficient for establishing the dorsal identity in mesenchymal cells in developing limb buds (Vogel et al., 1995; Chen et al., 1998; Chen, 2002), was used as a dorsal marker because we previously reported that Lmx1b is exclusively expressed in dorsal-derived cells and can be activated in the ALM experiment (Iwata et al., 2020; Yamamoto et al., 2022). The expression patterns of Fgf8, Shh, and Lmx1b were investigated using in situ hybridization (ISH). In AntBL and PostBL, Lmx1b expression was restricted to the dorsal half of the blastemas (Fig. 2B, G; n=5/5 for both), suggesting the presence of both dorsal and ventral tissues. These expression patterns correspond to the anatomically defined dorsoventral borders (Fig. 1A). In contrast, Lmx1b was expressed throughout the entire region of DorBL (Fig. 2L, n=6/6), whereas no Lmx1b expression was detected in VentBL (Fig. 2Q, n=6/6), suggesting the absence of ventral tissue in DorBL and dorsal tissue in VentBL. Consistent with previous studies, Fgf8 expression was observed in AntBL (Fig. 2C, n=4/5), but not in PostBL (Fig. 2H, n=5/5, Nacu et al., 2016). Conversely, Shh expression was detected in PostBL (Fig. 2I, n=5/5), but not in AntBL (Fig. 2D, n=5/5). In DorBL and VentBL, Shh expression was largely absent (Fig. 2N, n = 5/6; Fig. 2S, n = 6/6), whereas Fgf8 expression was present in both (Fig. 2M, n = 4/6; Fig. 2R, n = 4/6). In DorBL, Shh expression was observed in only 1 out of 6 samples, possibly due to ventral tissue contamination, as described above. These results suggest that Fgf8 expression is independent of the co-existence of dorsal and ventral tissues, whereas Shh expression appears to require their presence.

Figure 2.

Induction of Shh expression requires both dorsal and ventral cells

The absence of Shh expression in most DorBL and VentBL, combined with its consistent expression in all PostBL—which contain both Lmx1b-positive and Lmx1b-negative regions—suggests that the induction of Shh expression depends on the presence of both dorsal and ventral cells (Fig. 2). To test this hypothesis, we performed cell-tracing experiments using GFP-expressing skin from transgenic animals (Fig. 3). We induced VentBL in wild-type animals and then grafted the posterior half of the dorsal skin (VentBL+PD_gfp) or posterior half of the ventral skin (VentBL+PV_gfp) from GFP transgenic animals (Fig. 3A). Similarly, DorBL was induced in wild-type animals, and the posterior half of the dorsal skin (DorBL+PD_gfp) or the posterior half of the ventral skin (DorBL+PV_gfp) from GFP transgenic animals was grafted (Fig. 3A). GFP-positive mesenchymal cells derived from the posterior skin were expected to express Shh if provided with a suitable environment, because posteriorly derived cells are known to have the competency to express Shh in a blastema (Iwata et al., 2020). If the induction of Shh expression depends on the co-existence of dorsal and ventral cells, Shh expression in the GFP-positive cells should be observed in VentBL+PD_gfp and DorBL+PV_gfp but not in DorBL+PD_gfp or VentBL+PV_gfp. Samples were collected at 10 dps, and the expression of Shh and Lmx1b was analyzed using ISH. In all four experimental groups, Lmx1b expression patterns in the GFP-negative regions were consistent with those in DorBL and VentBL (Fig. 3D, F, H, J). In DorBL+PD_gfp and VentBL+PD_gfp, Lmx1b expression was observed in GFP-positive mesenchymal cells, whereas GFP-positive cells in DorBL+PV_gfp and VentBL+PV_gfp were Lmx1b-negative (Fig. 3C‒F). These Lmx1b expression patterns suggest that dorsoventral tissue contact was present in VentBL+PD_gfp and DorBL+pV_gfp but absent in DorBL+PD_gfp and VentBL+PV_gfp. In the GFP-positive cells derived from PD_gfp skin, Shh expression was observed in VentBL+PD_gfp (Fig. 3C, n=3/3) but not in DorBL+PD_gfp (Fig. 3E, n=3/3). Similarly, in the GFP-positive cells derived from PV_gfp skin, Shh expression was observed in DorBL+PV_gfp (Fig3I, n=3/6) but not in VentBL+PV_gfp (Fig. 3G, n=7/7). In addition, Shh expression was observed in some GFP-negative mesenchymal cells in samples where Shh expression was observed in GFP-positive cells. These results suggest that the induction of Shh expression in posteriorly derived cells requires the co-existence of dorsal and ventral cells.

Figure 3.

Limb formation in the absence of dorsoventral tissue contact

DorBL and VentBL failed to form limbs (Fig. 1D, E). Shh expression was absent in these blastemas, whereas Fgf8 was expressed (Fig. 2). Results from the cell-tracing experiments suggest that Shh expression requires the co-existence of dorsal and ventral cells (Fig. 3). Based on these findings, we hypothesized that while the co-existence of dorsal and ventral cells is necessary to induce Shh expression, it is not directly required for limb patterning if SHH proteins are externally provided. To test this hypothesis, we overexpressed Shh in DorBL and VentBL (Fig. 4A, B). As a positive control, Shh was overexpressed in AntBL, where Fgf8 was expressed but Shh expression was absent (Fig. 2C, D), as a previous study has shown that SHH supplementation to AntBL can induce limb patterning (Nacu et al., 2016). As a result, Shh-electroporated AntBL, DorBL, and VentBL formed limbs (n=10/19, 6/18, and 8/14, respectively, Fig. 4C‒G). In contrast, GFP-electroporated DorBL and VentBL (negative controls) failed to form limbs (n=7/7 and 6/6, respectively). We further analyzed the anatomy of the induced limbs (Fig. 4H‒K). We found that limbs formed from DorBL and VentBL exhibited symmetric structures along the dorsoventral axis, suggesting double-dorsal and double-ventral structures, respectively (Fig. 4J, K). In contrast, limbs formed from AntBL, which contained both Lmx1b-positive and Lmx1b-negative regions (Fig. 2B), exhibited asymmetric structures similar to those of normal limbs (Fig. 4H, I). To evaluate the symmetry of the limbs, we applied a machine learning-based method using ilastic. Each pixel in the images was classified into five classes (Classes 1 to 5, Fig. 4L). In this process, we annotated regions of background (Class 1), cartilage (Class 2), muscle (Class 3), other connective tissue (Class 4), and epidermis (Class 5) as training data for classification. Using these annotated regions as references, pixels were automatically classified into the five classes. Each class was assumed to primarily represent these tissues and regions with similar characteristics. Symmetry scores were then calculated for each class individually and for the combined set of all classes (Fig. 4M‒R, see Materials and Methods). The analysis revealed that symmetry scores for Classes 2 and 3, which encompass cartilage and muscle, and scores for the combined set of all classes were significantly higher in limbs formed from DorBL and VentBL compared to intact limbs (Fig. 4N, O, R). In contrast, the differences in symmetry scores for Classes 1, 4, and 5 in these limbs were relatively small (Fig. 4M, P, Q), likely because the axis of symmetry was manually set to maximize the external shape as symmetrically as possible. No significant differences in symmetry scores across all classes were observed between intact limbs and limbs formed from AntBL. These results suggest that limbs formed from DorBL and VentBL exhibit symmetric internal structures compared to normal limbs. Therefore, SHH supplementation appears to compensate for the lack of co-existence of dorsal and ventral cells in limb patterning, without inducing new dorsal or ventral identities. Thus, we conclude that the co-existence of both dorsal and ventral cells is critical for inducing Shh expression, which in turn is essential for limb patterning.

Figure 4.

The molecular basis of the dorsal and ventral positional cues

To investigate the molecular basis of dorsal and ventral positional cues, we performed RNA-seq analysis on DorBL and VentBL and identified differentially expressed genes (DEGs) between the two groups (p<0.05, Fig. 5A). Among the DEGs, we specifically focused on genes annotated as “intercellular signaling molecules” in PANTHER and identified 21 genes (Fig 5B). In this analysis, we found that 5 genes were expressed at higher levels in DorBL and 16 genes were expressed at higher levels in VentBL (Fig. 5B). Notably, Wnt4, Wnt10b, Fgf2, Fgf7, and Tgfb2, which encode secreted proteins, belong to the WNT, FGF, and TGFB families, which play key roles in major signaling pathways regulating limb developmental processes. To examine whether these genes regulate the induction of Shh expression in ALM blastemas, we overexpressed each gene in either DorBL or VentBL. As a result, Shh expression was observed in Wnt10b-electroporated VentBL (n=3/3) and Fgf2-electroporated DorBL (n=5/7, Fig. 5C). In contrast, Shh expression was not detected in Wnt10b-electroporated DorBL (n=4/4) or Fgf2-electroporated VentBL (n=3/3). Similarly, Shh expression was not detected in Fgf7- or Tgfb2-electroporated DorBL (n=3/3 for both) or in Wnt4-electroporated VentBL (n=4/4). To confirm Shh induction by FGF2 in DorBL, we supplied commercially obtained mouse FGF2 proteins into DorBL via bead grafting and observed higher Shh expression in DorBL with FGF2-soaked bead compared to DorBL with control DDW-soaked bead (Fig. 5D). Next, to confirm Shh induction by WNT signaling in VentBL, we treated VentBL with 1 μM 6-bromoindirubin-3-oxime (GSK-3 Inhibitor, BIO). As a result, Shh expression was relatively higher in BIO-treated VentBL compared to DMSO-treated control VentBL (Fig. 5E). Additionally, symmetric limbs were formed in Wnt10b-electroporated VentBL (n=3/12), BIO-treated VentBL (n=3/8), and in Fgf2-electroporated DorBL (n=5/10), being consistent with the results of Shh overexpression (Fig. 5F‒H, Fig. S1). These findings suggest that WNT10B, expressed highly in dorsal blastema cells, and FGF2, expressed highly in ventral blastema cells, function as the dorsal and ventral positional cues, respectively, to induce Shh expression, which subsequently supports limb patterning.

Figure 5.

Limb formation without nerve deviation and ventral skin grafting

A previous study demonstrated that a cocktail of BMP2, FGF2, and FGF8 can substitute for nerve deviation in the ALM experiment on the anterior region, suggesting that nerves contribute to blastema induction by supplying these proteins (Makanae et al., 2014). We found that FGF2 can also substitute for ventral tissue in DorBL (Fig. 5), suggesting that FGF2 serves not only as part of the nerve factors in blastema induction but also as the ventral positional cue in limb patterning. To investigate whether supplementation with BMP2, FGF2, and FGF8 to the dorsal region could substitute for both nerve deviation and ventral skin grafting, we performed experiments involving gelatin beads soaked in these proteins (Fig. 6). The dorsal or ventral skin was removed at the zeugopod level, and a BMP2+FGF2+FGF8-soaked gelatin bead was grafted at 3 dps without nerve deviation. Blastema induction was observed in both experimental groups (Fig. 6A, B, upper panels). We investigated gene expression patterns in the induced blastemas and found that both Fgf8 and Shh were expressed in blastemas induced at the dorsal site (n=8/8), whereas Shh expression was absent in ventral blastemas (n=5/5), although Fgf8 expression was detected in most cases (n=4/5, Fig. 6C, D, F, G). Lmx1b expression patterns corresponded to those observed in DorBL and VentBL (Fig. 6E, H, n=8/8 and 5/5, respectively). Furthermore, limb patterning was observed in the dorsal groups (n=12/20) but not in the ventral groups (n=17/17, Fig. 6A, B, lower panels). These findings demonstrate that a straightforward procedure involving dorsal skin wounding and supplementation with BMP2, FGF2, and FGF8 is sufficient to induce accessory limb formation. Remarkably, this method also enabled the induction of multiple limbs within the same limb. Grafting BMP2+FGF2+FGF8-soaked beads to multiple dorsal sites resulted in the formation of multiple limbs along the proximodistal axis (Fig. 6I‒L). These results refine our understanding of the role of FGF2 in the ALM, particularly its critical function on the dorsal side.

Figure 6.

Discussion

Dorsal and ventral positional cues are required for the induction of Shh expression

In axolotl limb regeneration, cells derived from all four positional origins—anterior, posterior, dorsal, and ventral—are required for an induced blastema to form a limb. This was confirmed through the ALM experiments (Fig. 1). The gene expression patterns in these ALM blastemas revealed distinct molecular characteristics for each group (Fig. 2). Fgf8 expression in AntBL and Shh expression in PostBL are consistent with a previous report (Nacu et al., 2016). The absence of SHH in AntBL and of FGF8 in PostBL likely accounts for their failure to form limbs. Lmx1b expression in the dorsal half of both AntBL and PostBL suggests that both dorsal and ventral tissues co-existed in these blastemas (Fig. 2B, G). In contrast, DorBL and VentBL exhibited distinct Lmx1b expression patterns, suggesting that most cells in DorBL and VentBL are derived from dorsal or ventral origins, respectively (Fig. 2L, Q). A possible explanation for the lack of Shh expression in DorBL and VentBL, despite the expression of Fgf8 (Fig. 2M, N, R, S), is that the induction of Shh expression may depend on the co-existence of dorsal and ventral cells, although Fgf8 can be expressed independently of the co-existence of dorsal and ventral cells. This idea is supported by the cell-tracing experiment (Fig. 3). In these assays, the posteriorly derived cells expressed Shh only when both dorsal and ventral cells were present. These results strongly suggest that the co-existence of dorsal and ventral cells is necessary for inducing Shh expression. Consequently, one of the essential roles of the dorsal and ventral positional cues in limb patterning would be inducing Shh expression, which facilitates the anteroposterior interaction.

Our findings clarify the hierarchical relationship between dorsal and ventral positional cues and anteroposterior interaction. Our data demonstrate that dorsal and ventral cells are essential for inducing Shh expression in a regenerating blastema. Furthermore, we showed that DorBL and VentBL, which typically fail to form a limb, could form patterned limbs with ectopic Shh expression, even in the absence of dorsal and ventral cell co-existence (Fig. 4). Furthermore, such limbs exhibited dorsally or ventrally symmetric structures (Fig. 4F, G, J‒R), highlighting that dorsoventral patterning depends on the cell origin. Thus, we concluded that dorsal and ventral positional cues are essential for patterned limb formation via inducing Shh expression.

WNT10B and FGF2 as dorsal and ventral positional cue molecules

We identified Wnt10b and Fgf2 as candidate genes providing the dorsal and ventral positional cues required to induce Shh expression, respectively (Fig. 5). DEG analysis between DorBL and VentBL revealed higher Wnt10b expression in DorBL and higher Fgf2 expression in VentBL (Fig. 5A, B). Supplementation with WNT10B in VentBL and FGF2 in DorBL induced Shh expression and subsequent limb patterning (Fig. 5C‒G). Wnt4 expression was also elevated in DorBL, while Fgf7 and Tgfb2 were upregulated in VentBL. However, we did not observe Shh induction in the Fgf7- or Tgfb2-electroporated DorBL or in the Wnt4-electroporated VentBL. FGF7 is known to be capable of inducing the apical ectodermal ridge (AER) in chick limb development (Yonei-Tamura et al., 1999), but little is known about its role in limb regeneration. In newt limb regeneration, KGFR, which acts as FGF7 receptor, is expressed in the basal layer of the wound epithelium, while FGFR1, which acts as FGF2 receptor, is primarily expressed in the mesenchyme (Poulin et al., 1993). This differential expression pattern suggests that FGF2 and FGF7 target distinct cell populations, potentially explaining the differences in their ability to induce Shh expression. The difference in the ability of Wnt10b and Wnt4 to induce Shh expression in VentBL may be explained by their distinct signaling pathways. WNT10B and WNT4 are typically classified as canonical and non-canonical WNTs, respectively. We observed Shh induction in BIO-treated VentBL, indicating that the canonical WNT signaling pathway regulates Shh expression. These results support the idea that canonical, but not non-canonical, WNT signaling contributes to Shh induction. Regarding Tgfb2, TGF-β signaling has been implicated in blastema induction during salamander limb regeneration (Sader and Roy., 2022; Lévesque et al., 2007). However, little is known about the relationship between TGF-β signaling and Shh induction in limb development and regeneration, suggesting that TGF-β signaling does not play a primary role in Shh induction. Consistent with this, we did not observe Shh induction in the Tgfb2-electroporated DorBL, as described above. This suggests that TGFB2 signaling may be involved in dorsoventral regulation independent of Shh induction. Taken together, among the candidate genes we identified, WNT10B via the canonical WNT pathway and FGF2 via FGFR1 appear to regulate Shh induction and the subsequent patterning process.

WNT signaling may be the key for the dorsal properties in limb formation. It has been reported that canonical Wnt/β-catenin signaling plays essential roles in limb regeneration among vertebrates, including axolotls (Kawakami et al., 2006; Lovely et al., 2022). In the present study, we demonstrated that introducing Wnt10b or BIO treatment could induce Shh expression in VentBL, facilitating limb patterning. In mouse limb development, Wnt10b is expressed in the apical ectodermal ridge (AER, Witte et al., 2009), and mutations in Wnt10b are associated with Split-Hand/Foot Malformation (Ugur and Tolun, 2008; Al Ghamdi et al., 2020; Bilal et al., 2023).

However, there is no evidence suggesting that Wnt10b contributes to dorsal specificity during limb development. This raises questions about whether the function of Wnt10b is unique to axolotls or whether its orthologs in other species might share functional redundancy with other WNT genes. Further studies are needed to clarify this point. Notably, WNT10B utilizes the canonical Wnt signaling pathway, similar to WNT7A, a well-known WNT family gene critical for dorsoventral limb patterning in amniotes. In amniotes, Wnt7a is expressed in the dorsal ectoderm of developing limb buds and induces Lmx1b expression in dorsal mesenchyme, thereby acting as positional information to establish dorsal characteristics (Riddle et al., 1995; Cygan et al., 1997; Chen, 2002). In axolotls, dorsal-specific Wnt7a expression has not been confirmed. Our RNA-seq analysis showed that there was almost no significant difference in Wnt7a expression levels between DorBL and VentBL (log₂ (normalized counts) = 5.97, log₂ (fold change) = -0.810, p = 0.401), consistent with previous studies (Shimokawa et al., 2013). These results suggest that Wnt7a does not have a dorsal-specific function, at least in axolotl limb regeneration/development. While WNT10B could function as a dorsal positional cue, WNT10B is unlikely to induce dorsal identity, as ectopic Lmx1b expression was not observed in Wnt10b-introduced VentBL, which formed double-ventral limbs (Fig. 5F‒H). This indicates that Wnt10b in axolotl limb regeneration does not simply replace the function of Wnt7a in amniote limb development. Nevertheless, the involvement of canonical WNT signaling in dorsal function and limb morphogenesis remains an important area for further investigation.

We identified FGF2 as the ventral positional cue, which plays a crucial role in limb patterning process in axolotl limb regeneration. This aligns with previous studies showing that Fgf2 expression correlates with limb regeneration in salamanders (Giampaoli et al., 2003). Fgf2 expression was relatively high in VentBL, and Fgf2-introduced DorBL formed patterned limbs (Fig. 5). These results indicate that FGF2 supplementation can substitute for the presence of ventral cells in the limb patterning process. It is noteworthy that FGF2 application does not appear to induce ventral identity, as the limbs formed from Fgf2-introduced DorBL exhibited dorsally symmetric limb structures. The finding that FGF signaling functions in dorsoventral control is not a well-known mechanism across species studied to date. Whether this represents an axolotl-specific regulatory mechanism or a broader function of FGF signaling in limb morphogenesis remains to be determined.

Our findings suggest that although WNT10B and FGF2 act as dorsal and ventral positional cues, they do not affect dorsal or ventral identities. In amniote limb development, WNT7A functions as dorsal “positional information” by inducing dorsal identity through Lmx1b expression. In contrast, in axolotls, supplementation with WNT10B in VentBL or FGF2 in DorBL did not affect the expression patterns of Lmx1b (Fig. 5C). Moreover, the limbs formed from such DorBL and VentBL exhibited dorsally and ventrally symmetric structures (Fig. 5F‒H). These findings suggest that dorsoventral identities in axolotls are not affected by WNT10B or FGF2 supplementation. Thus, in the present study, we used the term “positional cue” to distinguish this role from “positional information,” which provides positional identities. We previously reported that the expression patterns of Lmx1b in axolotl limb regeneration are likely to depend on the positional origins of cells (Iwata et al., 2020; Yamamoto et al., 2022). The identities of cells along the dorsoventral axis may be controlled by their positional memory and determined before the initiation of limb regeneration.

The present study revealed that the dorsal and ventral positional cue molecules WNT10B and FGF2 regulate Shh expression during limb patterning in axolotls. These findings highlight a degree of conservation between axolotl limb regeneration and amniote limb development. In amniote limb development, FGF and WNT signaling pathways are key regulators of Shh expression. In developing amniote limb buds, Fgf2 is expressed in the ectoderm, including AER and adjacent mesoderm, mediating distal outgrowth. FGFs, including FGF2, from AER are known to maintain Shh expression (Laufer et al., 1994; Niswander et al., 1994; Yang and Niswander, 1995; Li et al., 1996). Similarly, WNT7A regulates Shh expression in amniote limb development, and loss of WNT7A function results in reduced Shh expression in the zone of polarizing activity (ZPA) and the deletion of posterior structures (Parr and McMahon, 1995). Thus, our findings provide new insights into the conserved and divergent roles of dorsal-ventral signaling in regulating Shh expression, furthering our understanding of limb morphogenesis.

The integration mechanism of the four positional cues

In the present study, we reveal the molecular hierarchy of the four positional cues— anterior, posterior, dorsal, and ventral—in axolotl limb regeneration and explain how these positional cues are integrated to form a patterned limb (Fig 7). In our proposed model, following amputation, nerves first trigger blastema induction by secreting nerve factors, such as BMP and FGFs (Makanae et al., 2014, Satoh et al., 2016). These nerve factors stimulate connective tissue cells, including dermal cells, to generate multipotent mesenchymal blastemal cells (Muneoka et al., 1986; Kragl et al., 2009; Hirata et al., 2010; Gerber et al., 2018). Simultaneously, Wnt10b and Fgf2 are expressed at higher levels in dorsal and the ventral blastemal cells, respectively. In the next phase, the co-existence of WNT10B and FGF2 induces Shh expression in the posterior region of the blastema. During this phase, Fgf8 is expressed in the anterior mesenchyme independently of the co-existence of dorsal and ventral cells (Fig. 7A). These expression patterns of Wnt10b, Fgf2, Shh, and Fgf8 may be determined by the positional memory of cells (Otsuki et al., 2021). In the subsequent phase, the anteroposterior interaction, mediated by FGF8 and SHH, supports distal outgrowth to form a limb. This model explains previous observations in studies on double-half limbs and ALM blastemas (AntBL, PostBL, DorBL, and VentBL), which typically fail to form a limb. In blastemas induced from double-anterior and double-posterior limbs, or AntBL (Fig. 7B) and PostBL (Fig. 7C), SHH or FGF8 proteins are likely absent due to the lack of anteriorly or posteriorly derived cells. Similarly, in blastemas induced by amputating double-dorsal and double-ventral limbs, or DorBL (Fig. 7D) and VentBL (Fig. 7E), FGF2 or WNT10B proteins are likely absent or insufficient because of the lack of ventrally or dorsally derived cells. This results in the absence of SHH, even if posteriorly derived cells are present. In all these cases, the absence of either FGF8 or SHH disrupts the anteroposterior interaction, leading to failure in limb patterning. We concluded that the requirement for cells derived from all four positional origins is underpinned by this integration model. In this model, dorsal and ventral positional cues activate the posterior positional cue, enabling mutual interaction with the anterior cue. This interplay ensures proper anteroposterior interaction and complete limb patterning.

Figure 7.

Materials and Methods

Animal procedures

Axolotls between 4 and 15 cm from snout to tail tip were housed in tap water at 22℃. Both forelimbs and hind limbs were used for surgical procedures without distinction. Transgenic axolotls were obtained from the Ambystoma Genetic Stock Center (http://www.ambystoma.org/genetic-stock-center). Animals were anesthetized in 0.1% MS222 (Sigma-Aldrich) solution before surgical procedures.

In the ALM experiments, skin was peeled from the anterior, posterior, dorsal, or ventral side of a limb at the stylopod level, so that the skin of the opposite side was not injured. Thus, the size of the injured area depended on the limb size. Then, thick bundles of nerve trunks running the center of a limb was deviated the injured region. For the cell-tracing experiment, a piece of posterior half of the dorsal or ventral skin was obtained from GFP-expressing transgenic animals and grafted to the ALM region.

Protein-soaked beads for grafting were prepared as previously described (Makanae et al., 2014; Kashimoto et al., 2023). In brief, air-dried gelatin beads were allowed to swell in stock solution (1 μg/μL) prepared following the manufacturer’s instructions. Equal amounts of proteins were used when formulating the combination protein mixture. Beads were soaked in the mixture of proteins (Bmp2 [mouse], Fgf2 [mouse], and Fgf8 [human/mouse]; R&D Systems) overnight at 4℃. For control experiments, gelatin beads were soaked in DDW. Before grafting, dorsal or ventral skin was peeled because full-thickness skin inhibits limb regeneration (Thornton, 1962; Mescher, 1976; Tsai, 2020). The beads were grafted under the wounded epidermis at 3 dps. The limbs were fixed 10 days after grafting. The details of grafting procedures are as previously described (Makanae et al., 2014; Kashimoto et al., 2023).

To analyze skeletal patterns, induced limbs were stained with alcian blue and alizarin red. Samples were fixed in 100% ethanol for 1 day at room temperature, then stained with Alcian blue solution (Wako, pH 2.0) in 20% acetic acid (Nacalai Tesque) with 80% ethanol solution at 37°C overnight. Then, samples were washed in tap water several times and fixed in 10% Formaldehyde Neutral Buffer Solution (Nacalai Tesque) for 1 day. Samples were then stained with Alizarin red S (Nacalai Tesque) in the solution (4% KOH:10% Formaldehyde Neutral Buffer Solution = 2:3) at room temperature for 1 day. Finally, samples were placed in graded glycerol with 4% KOH for clearing.

For BIO treatment, 10 mM stock solution of 6-bromoindirubin-3-oxime (BIO, Selleck, S7198) dissolved in DMSO was stored in the dark at 4°C. Axolotls soon after surgery were raised in tap water with BIO solution (experimental) or with the same amount of solvent, DMSO (control), until 14 dps for qPCR and 30 dps for phenotype analysis. The containers including water and axolotls were kept in the dark during BIO or DMSO treatment. For BIO treatment, axolotls between 4 and 5 cm from snout to tail tip were used.

Sectioning and histological staining

Samples were fixed in 4% PFA/PBS overnight at room temperature before sectioning. The fixed samples were embedded in O.C.T. compound (Sakura) following 30% sucrose/PBS treatment for approximately 12 h at 4℃. Frozen sections of 14 μm thickness were prepared using Leica CM1850 (Nussloch). The sections were well dried under an air dryer and kept at -80℃ until use.

Standard hematoxylin and eosin (HE) staining and Masson’s trichrome staining were used for histological analysis. To visualize cartilage formation, Alcian blue staining was performed before HE staining. In brief, sections were washed in tap water several times to remove the O.C.T. compound. Then, the sections were stained with Alcian blue (Wako, pH 2.0), and then HE staining was performed. Trichrome stain (Masson) kit (Sigma, HT15-1KT) was used for trichrome staining. The stained sections were mounted using Softmount (Wako).

In situ hybridization

In situ hybridization was performed as described previously (Yamamoto et al., 2022). For probe synthesis, target genes cloned on pTAC-2 plasmid (BioDynamics Lab Inc.) were amplified by PCR using M13 primers. PCR fragments were purified and used as an RNA probe template. RNA probe synthesis was performed with Sp6 RNA polymerase (Takara) for 3 h, and RNA was hydrolyzed for 11 min. The sections were washed in PBT to remove the O.C.T. compound, treated with proteinase K (10 μg/mL) (Invitrogen)/PBT for 20 min at room temperature, washed in PBT, treated with 4% PFA/PBS for 20 min at room temperature, washed in PBT, and then probes were hybridized at 62.5℃for approximately 18 h. The sections were washed in wash buffer 1 (formamide:H₂O:20× SSC [3M NaCl:0.3M sodium citrate, pH 5.0] = 2:1:1), and then in wash buffer 2 (formamide:H₂O:20× SSC = 5:1:4). The samples were then incubated with anti-digoxigenin-AP Fab fragments (Sigma-Aldrich, 1/1000) for 2 h at room temperature. Samples were stained with BCIP (Nacalai Tesque) and NBT (Nacalai Tesque) in alkaline phosphatase buffer (0.1M NaCl, 0.1M Tris-HCl [pH9.5], 0.1% Tween20) for 18 h at room temperature after washing in TBST.

Immunofluorescence

Immunofluorescence on sections was carried out based on a previous report (Yamamoto et al., 2022). The antibodies were as follows: anti-GFP (MBL, #598, 1/500), anti-acetylated alpha tubulin (Santa Cruz, #sc-23950, 1/1000), anti-mouse IgG Alexa 488 (Invitrogen, #A11017, 1/1000), anti-rabbit IgG Alexa 488 (Invitrogen, #A21206, 1/500). Nuclei were stained with Hoechst (Nacalai Tesque), and images were captured using an Olympus BX51 system.

RNA-Seq analysis

Total RNA was extracted from DorBL and VentBL at 10 dps using TriPure reagent (Roche), following the manufacturer’s instructions. Three biological replicates were prepared for both samples. 150 bp paired-end RNA-seq reads were obtained under contract to Rhelixa (Tokyo, Japan), using Illumina Nova Seq 6000 and SMART-Seq HT Plus Kit (#R400449). RNA-seq data were analyzed on Galaxy (https://usegalaxy.eu/root) as follows. Sequence reads were trimmed and the quality was filtered by Trimmomatic v0.39 (Bolger et al., 2014) with the following parameters (LEADING:20, TRAILING:20, SLIDINGWINDOW:4:15, MINLEN:36). Axolotl genome data (AmexG_v6.0-DD) and annotation data (AmexT_v47-AmexG_v6.0-DD.gtf) were obtained from AXOLOTL-OMICS (https://www.axolotl-omics.org/) (Schloissnig et al., 2021). Mapping to the Axolotl genome (AmexG_v6.0-DD) was performed with HISAT2 v2.2.1 (Kim et al., 2015). Count data were obtained with featureCounts v2.0.8 (Liao et al., 2013) on the basis of the Axolotl gene model (AmexT_v47-AmexG_v6.0-DD.gtf). DEGs were identified with DESeq2 v2.11.40.8 (Love et al., 2014) (p < 0.05). Among identified DEGs (DorBL > VentBL; 762 genes, VentBL > DorBL; 513 genes), genes annotated as “intercellular signaling molecule” were explored on PANTHER (https://www.pantherdb.org/) (Thomas et al., 2003), and 21 genes (DorBL > VentBL; 5 genes, VentBL > DorBL; 16 genes) were identified as candidate genes of dorsal and ventral positional cues, as shown below:

DorBL > VentBL:

WNT4 (AMEX60DD052091), WNT10B (AMEX60DD029981), SEMA3A (AMEX60DD023165), LECT2 (AMEX60DD041044), ANGPTL5 (AMEX60DD049512).

VentBL > DorBL:

FCN1 (AMEX60DD050926), CXCL14 (AMEX60DD028973), DNER (AMEX60DD002010), FGF7 (AMEX60DD003767), CXCL12 (AMEX60DD052412), CXCL8 (AMEX60DD044133), SEMA6A (AMEX60DD042813), ANGPTL2 (AMEX60DD050490), ANGPTL6 (AMEX60DD031854), FGL2 (AMEX60DD006387), VWDE (AMEX60DD022491), TGFB2 (AMEX60DD036126), FGF2 (AMEX60DD044865), EFNA4 (AMEX60DD014882), EFNA3 (AMEX60DD014867), ANGPTL1 (AMEX60DD018261).

Cloning candidate genes and preparation of construct vectors for overexpression

Among the identified candidate genes, we focused on Wnt10b, Wnt4, Fgf2, Fgf7, and Tgfb2. We cloned these genes in pTAC-2 plasmids from cDNA obtained from blastemas. The following primers were used:

Wnt10b Fwd:ATGGCCCACAGCTCACCCTCCGACACC

Wnt10b Rev:TCACTTGCACACATTCACCCATTCTGTG

Wnt4 Fwd:ATGGATGCTCACGAAAGCAGCGTATATC

Wnt4 Rev:TCACCGGCAGGTGTGCATTTCTACCAC

Fgf2 Fwd:ATGGCGGCGGGGAGCATCACCACCTTGC

Fgf2 Rev:TCAACTCTTGGCCGACATGGGAAGGAAAAG

Fgf7 Fwd:ATGCGCAGATGGGTGCTAGCTTGGATC

Fgf7 Rev:TCATGTGTTATTGGATATACGCATTGGA

TGFB2 Fwd:ATGAGATTACAATTACTGAGAAAAAAAAATG

TGFB2 Rev:TTAGCTGCACTTGCAAGATTTTACAATCA

We then inserted these genes to p2a-AcGFP-pCS2 vectors with In-Fusion® HD Cloning Kit (Clontech). The following primers were used for PCR before In-Fusion reaction:

pCS2 inverse Fwd:GCTACTAACTTCAGCCTGCTGAAGCAGG

pCS2 inverse Rev:CATCGATGGGATCCTGCAAAAAGAACAAGTAGCTT

Wnt10b Fwd:AGGATCCCATCGATGGCCCACAGCTCACCCTCCGA

Wnt10b Rev:GCTGAAGTTAGTAGCCTTGCACACATTCACCCA

Wnt4 Fwd:AGGATCCCATCGATGGATGCTCACGAAAGCAGCG

Wnt4 Rev:GCTGAAGTTAGTAGCCCGGCAGGTGTGCATTTCTA

Fgf2 Fwd:AGGATCCCATCGATGGCGGCGGGGAGCATCACCA

Fgf2 Rev:GCTGAAGTTAGTAGCACTCTTGGCCGACATGGGAA

Fgf7 Fwd:AGGATCCCATCGATGCGCAGATGGGTGCTAGCTTG

Fgf7 Rev:GCTGAAGTTAGTAGCTGTGTTATTGGATATACGCA

Tgfb2 Fwd:AGGATCCCATCGATGAGATTACAATTACTGAGAAA

Tgfb2 Rev:GCTGAAGTTAGTAGCGCTGCACTTGCAAGATTTTA

Finally, the following DNA constructs were obtained; Wnt10b-p2a-GFP (pCS2), Wnt4-p2a-GFP (pCS2), Fgf2-p2a-GFP (pCS2), Fgf7-p2a-GFP (pCS2), and Tgfb2-p2a-GFP (pCS2).

Electroporation

Each DNA construct was injected directly into the target region. Immediately after injection, electric pulses were applied (20 V, 50 ms pulse length, 950 ms interval, 10 times) with NEPA21(Nepa gene). The injected DNA constructs were as follows: pCS2-AcGFP, pCS2-Shh-p2a-AcGFP, Wnt10b-p2a-GFP (pCS2), Wnt4-p2a-GFP (pCS2), Fgf2-p2a-GFP (pCS2), Fgf7-p2a-GFP (pCS2), and Tgfb2-p2a-GFP (pCS2). All plasmids were purified using a Genopure Maxi kit (Roche). Electroporation was performed at -3, 4, and 7 dps and samples were fixed at 10 dps for in situ hybridization. To analyze the phenotype of 90 dps samples, electroporation was performed at -3, 4, 7, 10, 15, 20, 25, and 30 dps.

qRT-PCR

The procedures of qRT-PCR were previously described (Yamamoto et al., 2022). In brief, RT was performed using Prime Script II (Takara), and qPCR was performed using KAPA SYBR FAST qPCR Master Mix (Kapa Biosystems) and StepOne (ThermoFisher Scientific). Primers were as follows:

Ef-1α Fwd: AACATCGTGGTCATCGGCCAT

Ef-1α Rev: GGAGGTGCCAGTGATCATGTT

Shh Fwd: GCTCTGTGAAAGCAGAGAACTCG

Shh Rev: CGCTCCGTCTCTATCACGTAGAA

RNA for Fig. 5D was prepared from FGF2- or DDW-soaked bead grafted DorBL. Bead grafting was performed at 7 dps and RNA preparation was performed at 14 dps. RNA for Fig. 5E was prepared from 1μM BIO- or DMSO-treated VentBL at 14 dps. Ef-1α was used as the internal control. We conducted a two-tailed Welch’s t-test for statistical analysis.

Pixel classification and calculating symmetry scores

A machine learning-based method was applied for pixel classification using ilastic software (downloaded on https://www.ilastik.org/). The details and workflows were previously described (Berg et al., 2019). Each pixel in the images was classified into five classes. The regions of the background (Class 1), cartilage (Class 2), muscle (Class 3), other connective tissue (Class 4), and epidermis (Class 5) were annotated for training data. Using these annotated regions as references, pixels were automatically classified into the respective classes. In this process, the same training data were used for images obtained from the same limb. Then, symmetry scores were calculated for each class individually and for the combined set of all classes. In this process, the external shape of the section, which could bend randomly during sample fixation process, would affect the symmetry scores if the entire region were used. To focus on the symmetry of the anatomical patterns, images with 400 μm width were prepared (Fig. 4L, 5E, F). The symmetry scores of pixel-classified images were calculated using python. First, the center of the dorsal end and ventral end was set as the axis of symmetry. Then, one side of the image was flipped. Next, color masks were generated for both sides of the image. These masks identified pixels that matched the specified color. Pixels were considered to match if their RGB values were within the given tolerance range for all three channels. In pixel comparison, the masks for one side and the other side were compared to calculate the following pixels:

Matching pixels: The number of pixels that match in both sides for the specified color. Total pixels: The total number of pixels matching the color in either side.

The symmetry scores were computed as (Number of Matching Pixels)/(Total Pixels in Both Sides). The scores were obtained from 12 areas of each group. For statistical analysis, a two-tailed a two-tailed Welch’s t-test was used. In this analysis, each group was compared to the intact limb group because the intact limb should be set as a typical asymmetrical structure.

Additional information

Author contributions

Conceptualization: S.Y., A.S.; Methodology: S.Y., M.H., A.S; Validation: S.Y.; Formal analysis: S.Y.; Investigation: S.Y.; Resources: S.Y., S.F., A.O., A.S.; Data curation: S.Y.; Writing - original draft: S.Y.; Writing - review & editing: S.Y., S.F., A.O., M.H., A.S.; Visualization: S.Y., A.S.; Supervision: A.S.; Project administration: A.S.; Funding acquisition: A.S., S.Y.

Acknowledgements

We are grateful to R. Iwata and T. Satoh for supporting office work and animal housing. Animals were obtained through Hiroshima University Amphibian Research Center. This work is supported by the Japan Society for the Promotion of Science KAKENHI grant-in-aid for scientific research (B) (20H03264 and 24K02034 to A.S.) and by a Grant-in-Aid for Japan Society for the Promotion of Science fellows (24KJ1718 to S.Y.).