Newt Hoxa13 has an essential and predominant role in digit formation during development and regeneration

The 5′Hox genes play crucial roles in pattern formation and growth along the proximal-distal and anterior-posterior axes during limb development (for a review, see Zákány and Duboule, 2007). The expression patterns and phenotypes of spontaneous and gene knockout mutants support a model in which Hox9/10, Hox11 and Hox12/13 paralogs specify the identity of the stylopod, zeugopod and autopod, respectively, during limb development (Davis et al., 1995; Fromental-Ramain et al., 1996a,b; Wellik and Capecchi, 2003; Zákány and Duboule, 2007). In the case of the autopod, Hoxa13 and Hoxd13 presumably specify the distal autopod, because double-mutant mice lacking Hoxa13 and Hoxd13 lose all their digits and metacarpal/tarsal bones (Fromental-Ramain et al., 1996b). These genes also regulate the development of the autopod in human limbs, because disorders of digit formation, such as hypodactyly, have been linked to human mutations in HOXA13 and HOXD13 (for a review, see Lappin et al., 2006).

Unlike in mammals, there is no direct genetic evidence that the 5′Hox genes are essential for limb development in non-mammalian tetrapod vertebrates. Instead of genetic studies, the expression patterns of 5′Hox genes have been examined during limb development in various non-mammalian tetrapods, such as chickens, frogs and axolotls. Although there are differences between species, these overall expression patterns are similar to those of mice (Bickelmann et al., 2018; Gardiner et al., 1995; Satoh et al., 2006; Torok et al., 1998; Wagner et al., 1999; Woltering et al., 2019), suggesting that the basic functions of 5′Hox genes are conserved in tetrapod limb development. However, their conservation is not conclusive because no functional analyses, such as by using genetics, have been performed. In addition, it is still unknown what functions of the 5′Hox genes are specific for which species and how their specificity is related to the distinct morphology and functions of limbs in various taxa.

In addition to limb development in non-mammalian tetrapods, it is still unknown whether Hox functions are essential for limb regeneration. Furthermore, if they are essential, it is not known whether Hox genes also have regeneration-specific functions. These issues are strongly related to how similar regeneration is to development, and to what is specific about regeneration. To investigate the roles of Hox genes in limb regeneration and their similarity or specificity, Hox expression patterns during limb regeneration have been examined in axolotls. Their expression in blastemas is similar to that in developing limb buds, except for the simultaneous expression of Hoxa13 and Hoxa9 within 24 h after amputation (Gardiner et al., 1995; Torok et al., 1998). However, the early expression pattern was not supported by analysis using antibodies against Hox proteins (Roensch et al., 2013). In any case, Hox genes are expected to have crucial functions in limb regeneration, and regeneration-specific functions have not been excluded.

One major reason why genetic analyses have not been performed in non-mammalian tetrapods has been the lack of reverse genetic techniques other than in rodents. In particular, gene disruption had not been achieved in animals that can regenerate their limbs. However, genome editing using CRISPR/Cas9 has been established as an efficient tool for gene disruption in many species and has enabled functional studies in many organisms. In addition, several studies have reported that CRISPR/Cas9 is effective in amphibians such as axolotls and Xenopus (for example, Flowers et al., 2014; Sakane et al., 2018).

We developed a molecular genetic system for Iberian ribbed newts (Pleurodeles waltl), which have the remarkable ability to regenerate many tissues, including limbs (Hayashi et al., 2013). We recently established a highly efficient gene knockout system using CRISPR/Cas9 that enables us to mutate a target gene in the whole body and almost all alleles (more than 99%) in F0 animals (Suzuki et al., 2018), probably owing to the longer time for the first cleavage in P. waltl fertilized eggs. We can also disrupt multiple genes in F0 animals by the injection of multiple guide (g)RNAs and analyze their functions without crossing newts. We also performed a comprehensive analysis of the P. waltl transcriptome, and established gene models for almost all protein-coding genes in P. waltl, including Hox genes (Matsunami et al., 2019).

In the present study, we analyzed the functions of Hox13 paralogs during limb development and regeneration in newts using our CRISPR/Cas9 system. Newts have four Hox13 paralogs (Hoxa13, Hoxb13, Hoxc13 and Hoxd13).We produced single, double, triple and quadruple Hox13 paralog mutants. We showed that Hox13 genes are essential for digit formation during development, as in mice. In addition, unlike in mice, Hoxa13 plays a predominant role in digit formation. Finally, Hox13 genes are also required for digit formation in limb regeneration. These results indicate that the general function of Hox13 genes is conserved between limb development and regeneration, and across tetrapod taxa. In addition, the predominance of Hoxa13 in newts has significant implications for the evolution of appendages.

A comprehensive analysis of the P. waltl transcriptome revealed that P. waltl has at least three Hox13 paralogs (Hoxa13, Hoxc13 and Hoxd13), and transcripts of these three genes were identified in the limb blastemas. We also found transcripts showing high homology with Hoxb13 of the palmate newt (Lissotriton helveticus, GenBank: DQ158059) and axolotl (Ambystoma mexicanum, GenBank: AF298184.1) in P. waltl transcriptome data (Matsunami et al., 2019).

We first examined the mRNA expression of all four Hox13 paralogs (Hoxa13, Hoxb13 Hoxc13 and Hoxd13) by RT-PCR in the forelimb (stage 40, the stage when the fourth digit is initially formed) and forelimb blastemas (two-digit stage) (Fig. S1). The expression of Hoxa13, Hoxc13 and Hoxd13, but not Hoxb13, was detected in both developing limbs and blastemas (Fig. S1). The data from blastemas are consistent with those in P. waltl transcriptome data (Matsunami et al., 2019).

Next, we analyzed the expression patterns of Hoxa13, Hoxc13 and Hoxd13 in developing forelimbs using whole-mount in situ hybridization (WISH) to detect the spatiotemporal expression patterns of these genes. Forelimb buds are observed from stage 33 (st33) embryos (Shi, and Boucaut, 1995). At this stage, the expression of Hoxa13 and Hoxc13, but not that of Hoxd13, could be detected in the whole limb bud mesenchyme (Fig. 1A,C). Whereas Hoxa13 expression was gradually restricted to the distal region, Hoxc13 expression was still detected in the whole limb buds except the cartilage primordia (humerus, radius, ulna and digits) (Fig. 1A, st36-38). Hoxd13 expression was first observed at the distal posterior region at st36 (Fig. 1A, st36), and the pattern was maintained at later stages (st37 and 38). Hoxd13 expression was not detected in the distal anterior margin corresponding to the digit I-forming region at st37 and 38 (Fig. 1A,C). The expression patterns of Hoxa13 and Hoxd13 were similar to those of the axolotl, another urodele amphibian (Bickelmann et al., 2018; Gardiner et al., 1995; Roensch et al., 2013; Woltering et al., 2019).

We next examined the expression patterns of Hoxa13, Hoxc13 and Hoxd13 in forelimb blastemas (late blastema and three-digit stages) using in situ hybridization on sections. Of the three genes, only Hoxa13 expression could be clearly detected at the late blastema stage. Its expression was observed in the distal mesenchymal region (Fig. 1B, arrowheads; see also Fig. 1C). Hoxa13 expression was also detected in the digit-forming region at the three-digit stage. Weak Hoxd13 expression could be detected in the posterior region of the autopod at the three-digit stage (Fig. 1B, arrowheads; see also Fig. 1C). Hoxc13 expression was indistinguishable from the background signal in all stages (Fig. 1B). The spatiotemporal expression patterns of Hoxa13 and Hoxd13 were very similar to those in developing limbs (Fig. 1A) and also to those in regenerating limbs of axolotls (Gardiner et al., 1995; Roensch et al., 2013).

We disrupted all Hox13 paralogs using a CRISPR/Cas9 system to investigate the function of Hox13 genes in limb development and regeneration. Although Hoxa13, Hoxc13 and Hoxd13, but not Hoxb13, were expressed in the forelimbs during development and regeneration (Fig. 1, Fig. S1), it cannot be excluded that Hoxb13 starts to be expressed and compensates for Hox13 functions in knockout (KO) animals in which Hox13 genes other than Hoxb13 were disrupted.

In order to produce single, double, triple and quadruple compound mutants, we designed nine gRNAs (Fig. 2, Table 1). All gRNAs except G25 recognize homeobox sequences. The G25 target sequence is located around the splicing acceptor of exon 2 in Hoxd13. Most of the gRNAs recognize only one paralog; however, G22 and G27 recognize three and four paralogs, respectively, by use of common sequences in Hox13 homeoboxes (Fig. 2). To avoid the misreading of phenotypes due to off-target effects, each Hox13 paralog gene was targeted by two to five different gRNAs (Fig. 2, Table 1). The ribonucleoprotein complexes (RNPs), containing Cas9 and a gRNA or multiple gRNAs, were injected into fertilized eggs. Table 1 shows groups of F0 animals (hereafter referred to as crispants). For example, only Hoxa13 was targeted in a crispant groups (a-1 and a-2). The gRNAs used were different between a-1 and a-2 (Table 1). In ac and ad crispants, Hoxa13/Hoxc13 and Hoxa13/Hoxd13 were doubly targeted, respectively. Similarly, in acd and abcd crispants, Hoxa13/Hoxc13/Hoxd13 and all four paralogs were multiply targeted, respectively (Table 1).

We obtained genomic DNAs from the tail fins of crispants at 2 months post-fertilization (mpf) and performed amplicon sequencing by next-generation sequencing (NGS) for genotyping. Genomic DNAs of Hox13 genes of 2-16 newts in each crispant group were investigated (Tables S1-S5). In total, 14 target sites, 51 crispants and 107 amplicons were analyzed. All target sites were mutated with high efficiency (mean±s.d., 97.4±2.3% in a total of 14 target sites; 96.7±5.4% in a total of 107 amplicons) (Table S1). Mutations by all gRNAs except G25 yielded deletions and/or insertions causing either frameshift mutations or deletions of amino acid residues in the homeodomains (Tables S2-S5). Because G25 also caused a deletion around the splicing acceptor of exon 2 in Hoxd13 (Table S3, ad-1), the mutations might have caused splicing abnormalities. Thus, the genotyping strongly suggested that target Hox13 genes were disrupted in almost all Hox13 crispants.

To investigate the function of Hox13 genes in limb development, we analyzed the phenotypes of the forelimbs of Hox13 crispants (Fig. 3, Fig. S2). All forelimbs of uninjected siblings (wild-type animals, n=206) had normal limb structures (Fig. 3A, Wt). Bone staining of limbs in all groups showed that the stylopod and zeugopod bones formed and appeared to be normal (Fig. 3A, Fig. S2). Whole autopod structures of all newts in c and d (d-1 and d-2) crispant groups were normal (Fig. 3A, Fig. S2A). The autopods of all newts in all other crispant groups showed abnormalities both in the digits and in the carpal elements. The phenotypes of digits and carpal elements are described below.

Almost all newts in all crispant groups other than c and d lacked apparent digit structures (Fig. 3A, Fig. S2A), although a few crispants had one to three digits. Some crispants had cartilaginous phalange-like projections (for example, Fig. 3B, a-2, arrowheads). Therefore, we classified the digit phenotypes into five groups: normal, normal structure; class 1, three or two digits; class 2, one digit; class 3, one digital phalange-like projection; class 4, no digit structure (neither phalanges nor metacarpal bones) (Fig. 4A; for class 3 and 4, also see a-2 and ad-3/acd, respectively, in Fig. 3B). The digits of almost all forelimbs of a (a-1 and a-2) crispants showed class 3 and 4 phenotypes (97.2% in a total of 36 forelimbs of a-1 and 80.0% in a total of 15 forelimbs of a-2) (Fig. 4B). Phenotypes of ad (ad-1, ad-2 and ad-3) crispants were more severe than those of a (a-1 and a-2) because the percentages of class 4 (no digit structure) increased significantly (P<0.05 in all combinations, Fisher's exact tests) (Fig. 4B). Phenotypes of abcd (abcd-1 and abcd-2) crispants appeared to be slightly more severe than those of ad (Fig. 4B), because there were no class 1 and 2 forelimbs. ac and acd crispants appeared to show similar phenotypes to a crispants (a-1 and a-2) and ad crispants (ad-1, ad-2 and ad-3), respectively (Fig. 4B). These results showed that disruption of Hoxd13 had a synergistic effect on the Hoxa13 single mutation, whereas that of Hoxc13 did not on the Hoxa13 single or Hoxa13/Hoxd13 double mutations for digit formation. Observation of forelimb development in class 4 showed the elongation of limb buds, but no digits were formed (for abcd-1 and abcd-2, Fig. 4C).

The morphology of carpal elements in class 3 and 4 forelimbs (almost all crispants in all crispant groups other than c and d) were also abnormal, and the numbers of carpal elements decreased compared with wild-type animals, c and d-1 crispants (Fig. 3). Although the morphology was abnormal, the presence of carpal elements, unlike digits, in abcd-1 and abcd-2 crispants (Fig. 3A, Fig. S2A), in which all Hox13 paralogs were disrupted, indicated that carpal elements can develop without any Hox13 genes. They also had different phenotypes from the digits; the numbers of carpal elements seemed not to differ among a-2, ac and ad-3 crispants, whereas those of acd and abcd crispants appeared to be fewer than those of a-2, ac and ad-3 crispants (Fig. 3, Fig. S2; for examples, see a-2, ad-3 and acd in Fig. 3B).

Taken together, we conclude that newt Hox13 genes have essential roles in digit formation and are necessary for the normal development of carpal elements. In particular, Hoxa13 has a predominant function in digit formation. And whereas Hoxc13 and Hoxd13 are not indispensable for digit formation, Hoxd13 has some functions in digit formation that probably complement Hoxa13 function, even if only slightly.

Next, we investigated the expression patterns of several genes to examine their contribution to limb development in Hox13 crispants. The expression of Shh, a key regulator of anterior-posterior patterning, is regulated by Hoxa and Hoxd genes (Kmita et al., 2005), and is activated directly by Hox13 (Capellini et al., 2006; Galli et al., 2010). Its expression was detected in the posterior region of forelimb buds of a-1 and ad-2 crispants. Although the patterns were similar to that of wild-type newts, the intensity decreased (Fig. 5A), suggesting regulation of Shh by Hox13 in the newt limb. The expression of Hoxd11 was observed in the middle region of the limb buds in wild-type animals at st35. However, the expression pattern slightly shifted to the distal region in a-1 crispants (Fig. 5A). In ad-2 crispants, the expression shifted more distally than that in a-1 crispants (Fig. 5A). Shifts of Hoxd11 expression have been reported in the limb buds of Hoxa13 KO mice (Bastida et al., 2020). In addition, a shift and expansion to the distal region of Hoxa11 expression have been observed in Hoxa13 single and Hoxa13/d13 double KO mice, respectively (Sheth et al., 2014). The shifts of Hoxd11 expression in newts suggested the reduction of the presumptive autopod region or ectopic expression of Hoxd11 in presumptive autopod cells. If the former is the case, then the zeugopod was in the most distal region. We also examined Hoxd13 expression in a-1 crispants. Unexpectedly, Hoxd13 expression was not detected in a-1 crispants (Fig. 5A). qRT-PCR analysis also showed a significant decrease of Hoxd13 in a-1 crispants (Fig. 5B).

These results suggest that Hox13 genes regulate the expression of Shh and are required for the generation or maintenance of the normal autopod region in newt limb buds, and that Hoxa13 strongly regulates the expression level and pattern of Hoxd13.

Our analyses above showed that newt Hoxd13 is not indispensable for limb development, and we could not obtain any clear data showing its functions in the limb development of newts. We had expected to infer the functions of Hoxd13 by analyzing the phenotypes of a crispants; however, this was not possible because the Hoxd13 expression level was strongly decreased in a crispants (Fig. 5). Therefore, we analyzed its functions directly by the induction of Hoxd13 expression in limb buds of a crispants, in which almost no digit structure was formed.

For this purpose, we induced the expression of both Hoxd13 and GFP or only GFP (control) by heat-shock treatment in a-1 crispants (Fig. 6A). Because Hoxd13 expression was initiated at st35-36 in the forelimb (Fig. 1A), we started heat shock at st34-35 (Fig. 6B). We examined only the limbs in which GFP florescence was observed after the heat shock.

Almost all forelimbs of control a-1 crispants in which only GFP expression was induced showed class 3 or 4 phenotypes (one digital phalange-like projection or no digit structure) (Fig. 6C,D). In contrast, after induction of Hoxd13 expression in the forelimbs, limbs showing class 1 or 2 phenotypes (one to three digits) appeared and accounted for 40% of all limbs (Fig. 6C,D). The difference between only GFP and Hoxd13 expression experiments was significant (Fisher's exact test, P=0.02).

These results showed that newt Hoxd13 has functions in digit formation, but the function seemed to disappear or be very low in a crispants, probably because its expression was downregulated. These results also showed that knockout of Hoxa13 can be rescued by the paralogous gene Hoxd13.

Next, we examined Hox13 functions in limb regeneration. Hox13 crispants grew and metamorphosed, although some crispants died. Notably, abcd-1 and abcd-2 showed lower survival rates than other crispant groups, and only two animals survived in each group for unknown reasons. The forelimbs of uninjected siblings (wild type) and some surviving crispants after metamorphosis (3 mpf) were amputated at the proximal region in the stylopod. The autopod cells in Hox13 crispants other than in c and d groups became abnormal during development. Therefore, regeneration by amputation at the autopod would be from cells that were already abnormal and would not be appropriate for studying the function of Hox13 genes in limb regeneration. However, because the stylopods of Hox13 crispants were considered to be normal (Fig. 3, Fig. S2), and all distal regions were reconstructed from dedifferentiated stylopod cells by amputation at the stylopod, we thought it was feasible to examine the functions of Hox13 genes in limb regeneration by amputation at the stylopod. The forelimbs of class 3 or 4 were amputated in a and ac crispants. In the case of ad, acd and abcd crispants, the forelimbs of class 4 were analyzed (Fig. 3, Fig. S2).

The external appearance and bone staining showed that almost all regenerates in wild-type animals and c and d crispants had normal morphology (Fig. 7A, Fig. S3). A small number of these animals, including wild types, showed full or partial defects in the digits, mainly in digit III, probably as a result of errors in regeneration (class 1 in Fig. 7C). Almost all regenerates of ad, acd and abcd crispants showed class 4 phenotypes similar to development (Figs 4B, 7C). During limb regeneration of the class 4 regenerates, no apparent phenotypes were observed before the notch stage compared with wild types (Fig. S4, 1-18 days post-amputation; Movie 2); however, neither the notch structure nor digit formation was observed thereafter (Fig. S4, 21-39 days post-amputation; compare Movie 2 with Movie 1). Interestingly, phenotypes in a crispants became slightly weaker than those from initial limb development. The majority of forelimbs in class 3 or 4 (80% and 60% in a-1 and a-2, respectively) moved to class 2 (Fig. 7C). These regenerates in class 2 had one digit-like structure with one or two phalange bones that could be stained with Alizarin Red (for example, Fig. 7B, a-2, arrows; compare with Fig. 3B, a-2, arrowheads). Similar structures were also observed in ac crispants (Fig. 7A, ac, arrows). The numbers and morphology of carpal elements did not show apparent changes compared with development in any group (Figs 3, 7A,B).

Overall, these results showed that Hox13 genes are essential for digit formation in limb regeneration. As with development, Hoxa13 had a predominant function in digit formation during regeneration.

Finally, in order to confirm the predominant role of Hoxa13 in digit formation, we examined whether Hox13 germline mutants showed the same phenotypes as in crispants.

We generated Hoxa13 and Hoxc13 mutants by intercrossing F1 animals (Hoxa13^+/Δ40, c13^+/Δ6×Hoxa13^+/Δ40, c13^+/Δ6; for details, see ‘Materials and methods’). The sequence of an Hoxa13 mutant allele showed the deletion of 40 nucleotides in the homeobox, indicating that the allele was null because of a large deletion and frameshift (Fig. 8A).

The developed forelimbs of Hoxa13 mutants (a13^Δ40/Δ40) showed mainly class 3 or 4 phenotypes (for an example of class 3, Fig. 8B). Similar results were also shown for other alleles (a13^{Δ40/Δ∼1000}, Fig. S5; for details, see ‘Materials and methods’). The distribution pattern of the class of all Hoxa13 mutants (a13^Δ40/Δ40 and a13^{Δ40/Δ∼1000}) was similar to that of a crispants (Figs 4B, 8C). By contrast, no abnormalities were observed in the digit formation of Hoxc13^Δ6/Δ6 newts (Fig. 8B,C).

We also obtained Hoxd13 germline mutants by intercrossing F1 animals (Hoxd13^+/Δ15×Hoxd13^+/Δ13; for details, see ‘Materials and methods’). No Hoxd13^Δ13/Δ15 newts showed abnormalities in digit formation (Fig. S6). In addition, we obtained Hoxd13 germline mutants with other alleles by intercrossing d-2 crispants (F0). All Hoxd13 alleles were considered to be null in these F1 animals because of a deletion in the sequence coding homeodomain helix III, which binds to target DNAs (Fig. 8A). Moreover, Hoxd13^Δ6-1 and Hoxd13^Δ6-2 alleles had an abnormal stop codon and a deletion of codons coding for DNA-recognition residues in the Hox13-type homeodomain, respectively (Zhang et al., 2011) (Fig. 8A). These Hoxd13 mutant newts showed no abnormalities in digit formation (Fig. 8B,C).

The Hoxa13 mutants were also examined for their limb regeneration phenotypes. The regenerated forelimbs of Hoxa13^Δ40/Δ40 newts showed class 2-4 phenotypes (Fig. 8D), as were observed in a crispants (Fig. 7C).

These results are consistent with those in the crispants (Figs 3, 4, 7; Figs S2, S3), indicating that genome editing in newts is extremely useful. Based on all these results, we concluded that newt Hoxa13 plays a predominant role in digit formation in limb development and in regeneration.

Our analysis of Hox13 genes in newt limb development revealed two important findings. First, the role of Hox13 in digit formation during development is essential and conserved in both newts and mice (Fig. 3) (Fromental-Ramain et al., 1996b). Second, among the newt Hox13 paralogs, Hoxa13 in particular plays a predominant role in digit formation. The evidence is that (1) a crispants and Hoxa13 germline mutants lacked all digit structures other than one phalange-like piece; (2) c crispants, Hoxc13 germline mutants, d crispants and Hoxd13 germline mutants show no phenotypes; and (3) Hoxb13 expression was not detected in the developing limbs (Figs 3, 4, 8; Figs S1, S2, S6). In contrast, mouse Hoxa13 and Hoxd13 have overlapping and mutually redundant functions, because major parts of digit structures remain in Hoxa13 or Hoxd13 single KO mice, although each has a specific phenotype such as the lack of digit I in the Hoxa13 KO mouse (Dollé et al., 1993; Fromental-Ramain et al., 1996b). What makes newts and mice different? Their differences might not result from the protein functions of Hoxd13 for digit formation, because the induction of newt Hoxd13 expression caused digit formation (Fig. 6).

We speculate that there are two major differences between mice and newts. One is the expression pattern of Hoxd13 in the urodeles. Newt Hoxd13 expression at least started at a later stage than that of Hoxa13, and the pattern in newts and axolotls is restricted to the distal posterior region, whereas Hoxa13 was expressed broadly in the future autopod region, as in mice (Figs 1, 9A) (Bickelmann et al., 2018; Gardiner et al., 1995; Roensch et al., 2013; Satoh et al., 2006; Woltering et al., 2019). Conversely, in mice, Hoxa13 and Hoxd13 expression overlaps in a broad area of the future autopod region (Bastida et al., 2020; Dollé et al., 1993; Zákány et al., 2004; for a review, see Zákány and Duboule, 2007) (Fig. 9A), which enables the redundant function of the two genes. The other reason is that newt Hoxd13 expression is strongly dependent on Hoxa13 in the limb bud (Fig. 5). In mice, similar regulation is observed only in digit I (Bastida et al., 2020; Sheth et al., 2014). As a result of these expression patterns and the dependency on Hoxa13, mouse Hoxd13 can form digits II-V independently of Hoxa13, whereas the independent function of newt Hoxd13 is largely limited.

What causes the urodele-specific expression pattern of Hoxd13? It has been suggested that enhancer sharing has been disrupted by the expansion of the intergenic regions between Hoxd11 and Hoxd13 in axolotls (Meyer et al., 2021). This expansion may have resulted in a urodele-specific expression pattern of Hoxd13 and the predominant role of Hoxa13 in newt digit formation.

It is also interesting why there is a difference in the dependency of Hoxd13 on Hoxa13 between newts and mice. As described above, mouse Hoxa13 regulates Hoxd13 only in the future digit I region (Bastida et al., 2020; Sheth et al., 2014), whereas it regulates the newt gene in the whole autopod region (Fig. 5). Mouse Hoxa13 regulates Hoxd13 expression by attenuating Gli3 transcription directly (Bastida et al., 2020). However, it is still unknown why this regulation is limited only to the future digit I region and not limited in newts. It may be that other genes, such as 5′Hox, compensate for Hoxa13 regulation of Hoxd13 in the mouse autopod in regions other than that of future digit I, but not in the whole autopod of newts. Therefore, determining the expression and functions of these genes is important for understanding the differences between newts and mice. It is also intriguing how the dependency of Hoxd13 on Hoxa13 has diverged during evolution.

It is important to know whether and to what extent the predominant role of Hoxa13 is related to the developmental pattern, morphology and functions of newt limbs, especially digits. Swapping of Hoxa13 and Hoxd13 genes between newts and mice or changing the Hoxd13 expression pattern of one to that of the other are ideal experiments and would provide insights regarding the biological significance of the role of Hoxa13 in newts.

To what extent regeneration is similar to development and which events are specific for development or regeneration have long been topics of discussion. In limbs, several factors have been shown to be regeneration specific. For example, signaling from neurons is necessary for blastema growth in limb regeneration (for a review, see Stocum, 2017), but not for limb bud growth in development. Although previous studies have suggested that the pattern formation of limbs is conducted in a similar manner during development and regeneration (for a review, see, Nacu and Tanaka, 2011), whether the same molecular system is used for both cannot be concluded, mainly because no functional analyses have been performed. 5′Hox genes are speculated to act in the same molecular system for patterning during limb development and regeneration.

Our results indicate that newt Hox13 genes are essential for digit formation during limb regeneration. This is the first finding not only of a Hox gene but also of any gene having the same function both in development and regeneration. It is worth noting that Hoxa13 plays a predominant role in digit formation during limb regeneration as well as in development, showing that the general roles of Hox13 paralogs are conserved between the two processes.

We found no phenotypes that suggested regeneration-specific functions of Hox13, except for the minor phenotypes observed in a crispants (Fig. 7, Fig. S3). Phenotypes in a crispants became slightly weaker than those in limb development. Because this phenotypic change was not confirmed in a few of the Hoxa13 germline mutants that we could analyze (Fig. 8D), the change needs to be investigated by using more germline mutants in the future.

The intercalation model theorizes a mechanism unique to limb regeneration but not to development (McCusker et al., 2015; Stocum, 2017). In one of the models for the initiation of urodele limb regeneration, the distalization at the tip of the stump, which follows limb amputation, triggers limb regeneration (Maden, 1977; for review, see Stocum, 2017). Here, the simultaneous expression of Hoxa13 and Hoxa9 within 24 h after amputation (Gardiner et al., 1995; Torok et al., 1998) is suggestive of early distalization. However, the expression pattern was not supported by analysis using antibodies against Hox proteins (Roensch et al., 2013). Our data here do not support the intercalation model in which Hox13 plays indispensable roles, such as in distalization at the tip of the stump, because limb regeneration itself progressed without Hox13 functions. However, our results cannot exclude the intercalation model based on other molecules, such as FGFs (Makanae and Satoh, 2018). Further detailed analyses of the expression and function of various genes required for the intercalation model are necessary to test it. Our studies have shown that a CRISPR/Cas9 system in newts is extremely useful for these analyses.

In the evolution of limbs from ancestral fish fins, the acquisition of the autopod with digits is a crucial event. In primitive tetrapods, such as Acanthostega, the limbs were polydactyl (Coates, 1994), and a pentadactyl state was only stabilized in later tetrapods. 5′Hox genes are considered to play important roles in this evolutionary process (the fin-to-limb transition and changes in the dactyl state), and changes in the expression patterns and activities of these genes may be involved (for reviews, see Leite-Castro et al., 2016; Paço and Freitas, 2018; Schneider and Shubin, 2013; Tanaka, 2016; Woltering and Duboule, 2010). However, most studies on the role of 5′Hox genes in evolution using present-day animals have analyzed expression patterns, and functional analyses using genetics, such as KO experiments, have only been performed in mice and zebrafish. Of note, there have been no functional studies on amphibians, which were the first terrestrial tetrapods with digits. The present study is important because it helps fill this gap.

Our results and those of several other studies have led us to consider the following hypotheses (Fig. 9B). (1) The function of Hoxd13 in the formation of the distal structures of appendages such as digits was not strong from fishes to amphibians. (2) However, the function of Hoxa13 was consistently strong from fishes to amniotes.

The evidence for the above is as follows. First, KO studies of hoxa13 and hoxd13 in zebrafish have shown that zebrafish hoxa13, but not hoxd13, is necessary to form the fin ray, which is considered to share a common developmental history with the digits (Nakamura et al., 2016). Second, when only Hoxa13 or Hoxa11/a13 are active in mice, Acanthostega-like truncated digits with polydactyly are observed (Zákány et al., 1997). As an opposing phenotype, oligodactyly is observed in cases when only Hoxd genes are active. These results suggest that Hoxa13, but not Hoxd13, clearly functions in the polydactyly of tetrapods such as Acanthostega. Finally, in newts, our analysis showed that Hoxa13, but not Hoxd13, has essential functions in digit formation (Hoxa13 predominance).

Analyses of expression patterns and functions of 5′Hoxd genes in mice and chickens (Yokouchi et al., 1991; Zákány and Duboule, 2007) suggest that late 5′Hoxd expression has important functions in digit formation in amniotes. Furthermore, a comparison of the expression patterns and enhancers mainly in mice and fishes provides a model in which quantitative and spatiotemporal changes in late 5′Hoxd expression, as a result of modification or acquisition of late enhancers, may establish the autopodial area within the digital skeleton during fin-to-limb transition (for reviews, see Leite-Castro et al., 2016; Paço and Freitas, 2018; Schneider and Shubin, 2013; Tanaka, 2016; Woltering and Duboule, 2010). However, because in this model Hoxd13 has main functions for digit formation among 5′Hoxd genes, and the predominant function of Hoxd13 is necessary for digit acquisition, our hypotheses described above challenge this model.

If our hypotheses are correct, then the following scenario is possible. Some of the mechanisms required for autopod formation were already being used in fin morphogenesis (e.g. Hoxa13 function for fin-ray formation). Then, the separation of the expression regions of Hoxa11 and Hoxa13, which were insufficient in fishes, and the acquisition of a sufficient autopodial area, enabled the formation of an endochondral skeleton by Hoxa13. This was the first autopod in tetrapods and the initial stage of tetrapod digit morphogenesis. Digit formation by Hoxa13 alone is unstable, as seen in the polydactyl state in several ancestors of amphibians. Extant amphibians still strongly retain this unstable state, and polydactyly can also be seen in some extant amphibians (Hayashi et al., 2015). Then, the stronger function of Hoxd13 could have established the pentadactyl state in amniotes. It is still possible that newts may have independently acquired Hoxa13 predominance, and further studies will be important to show Hoxa13 predominance in other amphibians and sarcopterygians, and the function and expression of other 5′Hox genes in newts.

The present study and previous studies have shown that the Hox13 genes are essential for the formation of distal appendages (fin rays and digits) in mammals, amphibians and fish. Our studies also show that our newt-based analysis system can cover the gap between zebrafish and mice and is a new tool for investigating the roles of Hox genes in the development, regeneration and evolution of various organs.

In this study, we used Iberian ribbed newts (Pleurodeles waltl) that were raised in our laboratory. The animals were reared as described previously (Hayashi et al., 2013). All procedures were carried out in accordance with the Institutional Animal Care and Use Committee of Tottori University (Tottori, Japan) and the national guidelines of the Ministry of Education, Culture, Sports, Science and Technology of Japan. Staging was performed as described in a previous study (Shi and Boucaut, 1995).

gRNAs were designed using CRISPR-direct (Naito et al., 2015). Positions of the targets and the sequences are shown in Fig. 2 and Table S6, respectively. No gRNAs showed high identity with any other gene bodies according to P. waltl comprehensive transcriptome data, including the sequences of any other homeobox genes (Matsunami et al., 2019). At least two nucleotides were unmatched among the 20 nucleotides of the protospacer sequences, which further had PAM sequences at the 5′ or 3′ end. The target sequences of G22 and G26 had PvuII and SfoI cleavage sites, respectively (Table S6). The synthetic tracrRNA, gRNA and Cas9 protein were obtained from Integrated DNA Technologies (IDT). The tracrRNA and gRNA were annealed, and then RNPs, containing Cas9, were produced in accordance with the manufacturer's instructions just before injection. Microinjection of RNPs was performed based on our previous reports (Hayashi et al., 2019; Suzuki et al., 2018).

We outcrossed acd crispants with wild-type newts. Because sufficient fertility was not observed in class 3 and 4 crispants, we used class 1 crispants. Double monoallelic F1 newts (Hoxa13^+/Δ40, Hoxc13^+/Δ6) were obtained and were intercrossed to obtain single homozygous biallelic mutants (Hoxa13^Δ40/Δ40 or Hoxc13^Δ6/Δ6). We also obtained Hoxa13^{Δ40/Δ∼1000} mutants by intercrossing Hoxa13^+/Δ40 and Hoxa13^+/Δ∼1000 animals. The Hoxa13^+/Δ∼1000 newts were generated from outcrossing ad-2 crispants with wild-type newts. The Hoxa13^Δ∼1000 allele had a large deletion (∼1000 bp) around the G24 target region (Fig. S5). Hoxd13^Δ13/Δ15 mutants were obtained by intercrossing Hoxd13^+/Δ13 and Hoxd13^+/Δ15 animals, which were also generated from outcrossing ad-2 crispants with wild-type newts. The sequences of Hoxa13^Δ40, Hoxc13^Δ6, Hoxd13^Δ13 and Hoxd13^Δ15 alleles are shown in Fig. 8A and Fig. S6. Genotyping of F2 animals was performed using PvuII (Hoxa13) or SfoI (Hoxd13) digestion of amplicons. The recognition sequence of the enzyme was mutated in these alleles (Fig. 8A, Fig. S6). For PCR amplification, the same primers as for NGS analysis, but without the barcoded overhang adapter sequence, were used (Table S7). In addition, we obtained Hoxd13^Δ6-1/Δ6-2 and Hoxd13^Δ6-1/Δ6-3 mutants by intercrossing d-2 crispants (F0). The sequences of Hoxd13^Δ6-1, Hoxd13^Δ6-2 and Hoxd13^Δ6-3 alleles are shown in Fig. 8A. For genotyping, see ‘Genotyping using NGS’ below.

To generate the plasmids hs:GFP and hs:Hoxd13/GFP, 0.6-kb genomic sequences upstream of the P. waltl hsp70 and P. waltl Hoxd13 open reading frame (ORF) fragments were amplified from wild-type P. waltl genomic DNA and cDNA using PCR with primers (Table S7). Heat-shock promoter (Xenopus laevis-type heat-shock promoter) sequences of the original plasmid (hsp70:gene-P2A-EGFP/I-SceI, a kind gift from Dr Yokoyama of Hirosaki University, Japan) were exchanged with the P. waltl-type (0.6-kb) using KpnI and BamHI sites. To ensure that the G24 gRNA targeting region of the Hoxd13 ORF was not cleaved by the RNP, containing G24 gRNA, the PAM sequence was replaced with another sequence in which the codon was unchanged. In hs:Hoxd13/GFP, the T2A sequence was inserted at the 3′ site of the Hoxd13 ORF (the stop codon was deleted) using the In-Fusion cloning system (Takara Bio). A FLAG tag was also inserted at the 5′ site of the Hoxd13 ORF using PCR. To generate hs:GFP or hs:Hoxd13/GFP transgenic newts (F0) in which Hoxa13 is disrupted, the I-SceI plasmid solution (300 ng of plasmids) digested with I-SceI (11.5 units, NEB) and an RNP containing G24 gRNA were mixed at a ratio of 1:1. The amount of total injection solution and plasmid was 18.4 nl and 138 pg, respectively, per egg. After injection, eggs were kept at 12°C for 3-6 h to delay the cleavages, and then kept at 15-16°C overnight. Injected embryos were incubated at 18°C until 16 days post-fertilization (dpf) to avoid leakage expression from the introduced constructs. At 16 dpf (st34-35), heat shock was performed by incubating embryos at 37°C for 1 h. Embryos were repeatedly heat shocked every 5 days until 41 dpf. After 16 dpf, embryos were raised at 25-26°C.

Genomic DNA was extracted from tail fins after 2 mpf. The target region was amplified from the lysates using KAPA HiFi enzyme (for NGS analysis, Roche Diagnostics) or KOD Fx Neo (TOYOBO) with primer sets (Table S7). Amplicons were obtained by primers containing barcoded overhang adaptor sequences according to a 16S Metagenomic Sequencing Library Preparation Kit and purified using AMPure XP (Beckman Coulter). All amplicons were mixed and subjected to an Illumina MiSeq run (paired-end 300; Macrogen). Sequencing data were analyzed according to previous studies (Iida et al., 2020; Suzuki et al., 2018).

For conventional RT-PCR (Fig. S1), three forelimb blastema, six limb buds and three tails at st40 were used for sampling of mRNA. For quantitative RT-PCR (Fig. 5B), a-1 crispants and uninjected wild-type embryos at st38 were used. Four mixtures were analyzed for each genotype (a-1 and wild type). Each mixture contained six forelimb buds from three animals. In both RT-PCR analyses, RNA was extracted using the ReliaPrep RNA Tissue Miniprep System (Promega) and cDNA was synthesized using a PrimeScript II First Strand cDNA Synthesis Kit (Takara Bio). Other methods were performed as described in a previous study (Toyoda et al., 2003). Gapdh was used as the endogenous control. Primers are shown in Table S7.

Limb amputation and bone staining were performed as described previously (Koriyama et al., 2018). The forelimbs were amputated at the most proximal region in the stylopod. The developed and regenerated limbs were fixed in 10% formalin solution at 4°C for at least 2 days and stained with Alizarin Red and Alcian Blue.

The developing and regenerating limbs were observed with a stereoscopic microscope (Leica, MZFL III), and images, including limbs before and after bone staining, were acquired with a microscopy camera (Nikon, DS-Ri2).

For in situ hybridization, fragments for probes were amplified by PCR using cDNA generated from P. waltl embryos (st27). Primers are listed in Table S7. The gene coding sequences for P. waltl were obtained from iNewt (Matsunami et al., 2019) (http://www.nibb.ac.jp/imori/main/). Each fragment was cloned using the pGEM-T easy vector system (A1360, Promega). Digoxygenin-labeled RNA probes were synthesized by T7 or SP6 RNA polymerase (Roche). WISH for amphibians was performed as described previously (Purushothaman et al., 2019), except for bleaching (6% H₂O₂, 4°C overnight) and permeabilization with 20 µg/ml Proteinase K (FUJIFILM Wako Chemicals) at room temperature for 30 min (FUJIFILM Wako Chemicals). Section in situ hybridization was performed as described previously (Kragl et al., 2013) with the following modifications. Before hybridization, sections were permeabilized with 1 µg/ml proteinase K at 37°C for 15 min, and alkaline phosphatase activity was visualized by NBT/BCIP. For serial frozen sections, samples after fixation with MEMFA (0.1 M MOPS pH 7.4, 2 mM EGTA, 1 mM MgSO₄·7H₂O and 3.7% formaldehyde) were treated with 30% sucrose/PBS, embedded in OCT compound (Sakura Finetek Japan), and were serially sectioned at 20-µm thickness.

Experimental data were analyzed using Welch's two-sample t-test and Fisher's exact test. Values of P<0.05 were considered statistically significant.

The authors wish to thank Dr Hitoshi Yokoyama (Hirosaki University, Hirosaki, Japan) for critical reading of the article and the kind gift of the hsp70:gene-P2A-EGFP/I-SceI vector, and Ms Akiko Adachi and Messrs Hiroshi Onishi and Yuuki Senoue (Tottori University, Yonago, Japan) for their technical assistance and newt raising.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200282.