What snakes and caecilians have in common? Molecular interaction units and the independent origins of similar morphotypes in Tetrapoda

Abstract

Developmental pathways encompass transcription factors and cis-regulatory elements that interact as transcription factor-regulatory element (TF-RE) units. Independent origins of similar phenotypes likely involve changes in different parts of these units, a hypothesis promisingly tested addressing the evolution of the rib-associated lumbar (RAL) morphotype that characterizes emblematic animals such as snakes and elephants. Previous investigation in these lineages identified a polymorphism in the Homology region 1 [H1] enhancer of the Myogenic factor-5 [Myf5], which interacts with HOX10 proteins to modulate rib development. Here we address the evolution of TF-RE units focusing on independent origins of RAL morphotypes. We compiled an extensive database for H1-Myf5 and HOX10 sequences with two goals: (i) evaluate if the enhancer polymorphism is present in amphibians exhibiting the RAL morphotype and (ii) test a hypothesis of enhanced evolutionary flexibility mediated by TF-RE units, according to which independent origins of the RAL morphotype might involve changes in either component of the interaction unit. We identified the H1-Myf5 polymorphism in lineages that diverged around 340 Ma, including Lissamphibia. Independent origins of the RAL morphotype in Tetrapoda involved sequence variation in either component of the TF-RE unit, confirming that different changes may similarly affect the phenotypic outcome of a given developmental pathway.

1. Introduction

The evolution of developmental pathways has been addressed in the context of independent origins of similar phenotypes in lineages as diverse as plants (C4 photosynthesis [1,2]; flower evolution and development [3,4]) and vertebrates (antifreeze proteins in Arctic and Antarctic fish [5]; echolocation in bats and toothed whales [6]; and loss of flight in birds [7]). Developmental pathways comprise signalling cascades that integrate information among components often settled on interactions between transcription factors and cis-regulatory elements [8–10], establishing transcription factor-regulatory element (TF-RE) units. Given that changes in different parts of these units could equally affect the developmental signalling cascade, similar phenotypes might evolve in phylogenetically distant species through changes either in the transcription factor, the cis-regulatory element, or both. This proposition challenges strict classifications of evolutionary processes as parallelism (homoplastic traits that independently evolved through the same genetic bases) or convergence (similar phenotypes derived from different genetic mechanisms) because similar phenotypes may actually evolve through different changes in the same TF-RE unit that equally affect the phenotypic outcome of a given developmental pathway (see [11–14] for discussions related to homoplastic traits). Moreover, it admits a mechanism that might confer evolutionary flexibility to developmental interactions and eventually facilitate recurrent evolution of the same phenotype in different lineages. Validation of this hypothesis presumes empirical evidence derived, for example, from molecular information of TF-RE units involved in developmental pathways associated with homoplastic phenotypes.

Independent origins of ribs in the lumbar region of the tetrapod axial skeleton provide an example of homoplasy that has been evaluated from a developmental evolution perspective (see [15]). Morphologically discrete regions characterize the axial skeleton of most vertebrate species, being the lumbar region recognized by the absence of ribs [16–18] (see also figure 1a). Interactions between transcription factors (especially HOX proteins) and cis-regulatory elements establish the transition from thoracic to lumbar regions [15–20], as schematized in figure 1b. Two HOX paralogues (HOX10 and HOX6) play a key role for rib development—they bind to the Homology region 1 [H1] enhancer and regulate the expression of the Myogenic factor-5 [Myf5], a gene responsible for the signalling cascade of rib development [19,20]. Rib development is activated in the presence of a HOXB6/Paired box (PAX3) complex, while the expression of HOX10 paralogues (HOXA10, HOXC10, HOXD10; see [21]) in the lumbar region inhibits rib formation [15–17,21]. This signalling cascade corresponds to a TF-RE unit hereafter referred to as HOX10-H1-Myf5 (figure 1b). Lineages of emblematic animals such as snakes, elephants and manatees evolved phenotypes in which the lumbar region corresponds to a posterior extension of the thorax (see also figure 1a), exhibiting vertebrae with associated ribs in this region (rib-associated lumbar phenotype, hereafter referred to as RAL morphotype) despite the apparent conservation of HOX10 expression patterns [15,22]. In these animals, the genetic mechanism underlying rib development in a region where HOX10 is expressed resides on a polymorphism in one transcription factor-binding site (TFBS) of the H1-Myf5 enhancer that blocks interaction between HOX10 and this cis-regulatory element without affecting the enhancer interactions with the HOX6-PAX3 complex [15] (figure 1b). The RAL morphotype might occur in specific lineages also due to different axial gene expression patterns, rather than changes in components of the HOX10-H1-Myf5 interaction unit. The presence of ribs associated with vertebrae along the entire trunk may represent the ancestral condition of tetrapods, perhaps even older, given the recent evidence for a hox-based axial skeletal regionalization in the cartilaginous fish Leucoraja erinacea [23]. The expression domains of Hox10 and Hox11 seem to define the transition to the lumbar-sacral axial region [16,19], despite the still scarce evidence from non-model organisms. In caecilians and snakes, Hox10 expression invades the rib-forming somites, a pattern that differs from that described for the whiptail lizard Aspidoscelis uniparens [22,24]. Transgenic experiments have empirically demonstrated the functional implications of the snake polymorphism in H1-Myf5 [15], and in the present study, we expand the phylogenetic coverage of this discussion for all tetrapods.

Figure 1.

(a) Whole skeleton CT-scans of two Lissamphibia lineages that evolved the RAL morphotype: Caudata (Aneides lugubris; image from Morphosource) and Gymnophiona (Caecilia tentaculata; image provided by Dr. Emma Sherratt). Two additional lineages are included as reference for other trunk morphotypes: Anura (clear sacral region indicated by a red arrow; Megaelosia goeldii, image from Morphosource) and Squamata (lizard exhibiting a lumbar region, indicated by a red arrow, characterized by the absence of ribs; Ceratophora stoddartii, image from Morphosource). (b) Model for the HOX10-H1-Myf5 interaction unit in embryos developing a lumbar region where ribs are absent (top) and in lineages that develop the RAL morphotype (bottom). The transcription factor HOX10 (green) usually binds to the H1 region of the enhancer (purple), repressing the expression of Myf5 (yellow), a gene that regulates rib growth, while HOX6-PAX3 (blue and pink, respectively) complex activates the expression. In lineages that evolved the RAL morphotype, mutations in either component of the interaction unit block the HOX-H1 association, and the gene Myf5 remains expressed, producing ribs despite HOX10 expression.

In addition to the lineages aforementioned (snakes, elephants and manatees), the RAL morphotype also evolved in tetrapod lineages external to Amniota (e.g. Gymnophiona [caecilians] and Caudata [salamanders]; figure 1a). A hypothesis that remains to be tested presumes that changes in the HOX10-H1-Myf5 interaction unit might underlie the independent origins of a RAL morphotype not only in Lepidosauria and Mammalia, but also in a lineage as distant as Lissamphibia. In this study, we postulate a mechanism of enhanced evolutionary flexibility in developmental systems settled on variation in either component of a TF-RE unit that could facilitate recurrent evolution of similar phenotypes in specific lineages. We take advantage of current empirical knowledge on how HOX proteins interact with the H1-Myf5 enhancer during rib development to test the hypothesis that the same phenotype might have evolved through sequence changes located in either component of the TF-RE unit (i.e. the transcription factor [HOX10], the cis-regulatory element [H1-Myf5] or both). We performed an extensive search for the H1-Myf5 enhancer in genomes available in public repositories to evaluate how distantly sequence convergence can be detected within Tetrapoda. To our knowledge, this is the first time Lissamphibia is evaluated in the context of HOX10-H1-Myf5 evolution. In order to compile an expanded dataset, we also sequenced H1-Myf5 in additional species of lizards and snakes. After detecting convergence in the polymorphism of the H1-Myf5 cis-regulatory element among lineages that diverged more than 340 Ma, we assembled a dataset with sequences for the region of HOX10 that binds to the TFBS at H1-Myf5 (characterized by [15]) to evaluate the evolution of the HOX10-H1-Myf5 unit in Tetrapoda. As predicted, independent origins of RAL morphotypes involved sequence variation in different components of the TF-RE unit (either the transcription factor HOX10 or the enhancer H1-Myf5), confirming that different changes may similarly affect the phenotypic outcome of a given developmental pathway.

2. Results

(a) . Sequence convergence in regulatory elements: H1-Myf5 enhancer

Analyses performed using a 229-species dataset identified the same polymorphism in the H1-Myf5 enhancer previously described in Lepidosauria (snakes), Mammalia (Afrotheria) also in Lissamphibia (caecilians and salamander; figure 2; electronic supplementary material, S1). These lineages evolved a RAL morphotype (lineages characterized by the RAL morphotype are referred in further analyses as ‘foreground’; see also figure 1a for examples) and were compared to the typical morphotype of tetrapods (ribs absent in the lumbar region, referred here as ‘background’). Despite detecting this polymorphism also in one salamander, we chose to remove Caudata from subsequent analyses due to genome singularities (e.g. large size and ambiguous annotation) of the only representative sequenced so far.

Figure 2.

Ancestral reconstructions of polymorphisms identified in the H1-Myf5 enhancer (H1), and in amino acid sites of HOXA10 (HA) and HOXC10 (HC). The first column corresponds to the polymorphism in position 20 of H1. Dark red, light red and black circles represent the bases thymine [T], cytosine [C] and adenine [A], respectively. The second, third and fourth columns correspond to molecular patterns identified in the amino acid sites 1, 44 and 48 of HOXA10. In the second column, light brown, yellow and dark brown circles represent the amino acids asparagine [N], threonine [T] and serine [S], respectively. In the third, serine [S], threonine [T] and alanine [A] are represented, respectively, by dark green, light green and light grey. The fourth column represents threonine [T] and serine [S] by dark blue and light blue, respectively. The fifth column shows mutations identified in HOXC10, representing by dark purple the amino acid serine [S] and by light purple threonine [T]. Dated phylogeny for Tetrapoda was retrieved from TTOL [25]. Figure created using the packages ‘ape’ [26], ‘protr’ [27] and ‘RColorBrewer’ (https://cran.r-project.org/web/packages/RColorBrewer/index.html). For a better visualization, we only included a few representatives for each clade. Species in bold indicate those having the RAL morphotype.

Molecular evolution analyses in the H1-Myf5 enhancer suggested significant associations between the foreground lineages and shifts in the evolutionary rate of H1 (p = 0.002). We calculated the Bayes factor (BF) value of the correlation for each site to the phenotypic trait (i.e. RAL morphotype) and identified significant associations between the RAL morphotype and the sites 9, 20, 24 and 34 of the sequence alignment (BF > 2, figure 3; electronic supplementary material, S2). The strongest association was identified in the twentieth site of the alignment (BF = 172.6), which corresponds to the second of seven sites that belong to the TFBS region that interacts with HOX10 (figure 3; electronic supplementary material, S1). The nucleotide of this site is a thymine [T] in most tetrapod species and also estimated in the ancestral sequence, but in Serpentes, it has been replaced by a cytosine [C] (figure 2). The same polymorphism has been also identified in the mammalian lineage Afrotheria (figure 2 and electronic supplementary material, S1), represented here by nine species (five species of Paenungulata and four of Afroinsectiphilia).

Figure 3.

Ancestral sequences of H1-Myf5 enhancer (H1) and the transcription factors HOXA10, HOXC10 and HOXD10. The amino acid sites 43, 45, 46 and 67 (dark red letters) have specific amino acids for each paralogue of HOX10 proteins. In the enhancer H1, a black box indicates the TFBS that interacts with the HOX protein. In HOX10 proteins, a black box indicates the HD adjacent to two motifs (M1 and M2). Green intensity indicates the (BF) values of the trait-dependent evolutionary rate shifts in each site/amino acid of H1 and the three HOX10 proteins, calculated using TraitRateProp [28]: light green = sites with no significant values (BF < 2), intermediate green = significant values (BF > 2), dark green = the site with the highest significant value (BF = 172.6).

We also provide in this study the first report of the H1-Myf5 polymorphism in tetrapod lineages external to Amniota (figure 2; electronic supplementary material, S1). Specifically, we identified the same nucleotide substitution in the caecilian Rhinatrema bivittatum. In two other caecilians, Geotrypetes seraphini and Microcaecilia unicolor, there is a difference in the same site but the substitution resulted in the presence of an adenine [A], instead of thymine [T] or cytosine [C]. Another sequence particularity observed in the species Microcaecilia unicolor at the interaction region of the HOX10-H1-Myf5 unit consists of the occurrence of a cytosine [C], instead of a thymine [T], in the sixth site of the HOX10-binding region in H1-Myf5 (figure 2; electronic supplementary material, S1).

Using the sampling available, we were able to reconstruct five events of transitions [T → C] and one transversion [C → A], as illustrated in figure 2. The higher frequency of transitions (purine-to-purine or pyrimidine-to-pyrimidine) than transversions (purine-to-pyrimidine or vice versa) has been already described in the literature, and usually the ratio of transition to transversion exceeds what would be expected if all substitutions occurred at the same rate (to be 0.5, see [29]). Such a trend is explained by the similar chemical structure between purines (double rings) and between pyrimidines (single ring) [30–32].

(b) . Sequence evolution in transcription factors: the HOX10-complex

We compiled a dataset of HOX10 amino acid sequences from several tetrapod species in order to evaluate molecular patterns in the HOX10-complex (HOXA10 = 223 species; HOXC10 = 200 species; HOXD10 = 207 species; see electronic supplementary material, S3). Combined action of these proteins plays a key role in the patterning of lumbar vertebrae (see [21] for a summary), and we aimed to identify specific patterns likely related to the origins of a RAL morphotype in tetrapods. The structure of HOX proteins comprises a homeodomain (HD) flanked by two conserved motifs (M1 and M2), and both M1 and HD are required for the rib-repressing HOX10 activity [33]. In the assembled dataset, we identified specific patterns in each HOX10 protein: HOXD10 is extremely conserved in Tetrapoda (few polymorphisms were detected, electronic supplementary material, S3), while conspicuous amino acid changes in HOXA10 evolved in different lineages (electronic supplementary material, S3). Shifts in evolutionary rates of proteins from the HOX10-complex apparently were not associated with the origins of the RAL morphotype. Although not finding in the studied lineages an amino acid pattern related to this phenotypic trait, in the ancestral reconstructions, we identified that all snakes have a different amino acid in the first site of the M1 region in HOXA10 (figure 2; electronic supplementary material, S3). Specifically, the ancestral reconstruction analysis suggested a substitution of the ancestral amino acid Asparagine [N] by a Threonine [T] in the early evolutionary history of Serpentes (figure 2).

The DNA sequences of HoxA10 were used to test if the foreground lineages evolved under specific selection regimes (electronic supplementary material, S4). First, we used CodeML (CML) branch models in PAML [34] to test for differences in selective regimes between foreground and background lineages (electronic supplementary material, S2). The highest values of ω in HoxA10 were associated with Serpentes (ω = 0.0235), although ω differences between the foreground lineages (ω1 [Gymnophiona] = 0.0001, ω2 [Serpentes] = 0.0235; ω2 [Afrotheria] = 0.0001) and the background species (ωB = 0.0059) were not significant (p = 0.106). First, we used aBSREL [35] and found no evidence of episodic diversifying selection in the tree branches (p > 0.05, electronic supplementary material, S2). We also tested for relaxed selection using RELAX [36], and the analysis did not detect any signal of relaxed or intensified selection in a given lineage (p = 0.572). Finally, we used codeML branch models in PAML [34] to test for differences in selective regimes between foreground and background lineages (electronic supplementary material, S2). The highest values of ω in HOXA10 were associated with Serpentes (ω = 0.0235), although ω differences between the foreground lineages (ω1 [Gymnophiona] = 0.0001, ω2 [Serpentes] = 0.0235; ω2 [Afrotheria] = 0.0001) and the background species (ωB = 0.0059) were not significant (p = 0.106). Finally, we evaluated selection in specific sites of HoxA10 using the CML branch-site models (BS, [34]) and the HyPhy Contrast-FEL (CF; [37]), and results converged by not detecting differences between the null and alternative models (electronic supplementary material, S2).

(c) . Evolutionary mismatches between HOX10 and H1-Myf5: enhanced flexibility in transcription factor-regulatory element units may facilitate recurrent evolution of rib-associated lumbar morphotype

We assembled evidence from molecular sequences to evaluate the hypothesis that changes in different components of a given interaction unit (TF-RE unit) might be associated with the recurrent evolution of similar phenotypes. Ancestral reconstructions for the sequences of H1-Myf5 enhancer and the respective interaction regions in HOX10 proteins provide evidence that recurrent evolution of the RAL morphotype in Tetrapoda involved changes in at least one—but not always the same—component of the TF-RE unit (figure 2; electronic supplementary material, S1 and S3). Moreover, the polymorphism postulated by Guerreiro et al. [15] as being responsible for this peculiar phenotype apparently is not a fundamental condition for rib development in the lumbar region of snakes (electronic supplementary material, S1). Our dataset provides evidence for two reversals in the H1-Myf5 polymorphism in Serpentes (figure 2): in the lineage represented by Vipera berus and also in Elapidae, the polymorphism reversed to the ancestral condition of a thymine [T]. The species V. berus is the only Viperinae representative in our dataset, but other species from the family Viperidae do not share the same pattern. Trunk morphology in vipers is similar to the phenotype that characterizes Serpentes, and sequences from all snake species we evaluated also have the substitution from asparagine [N] to threonine [T] in the first amino acid of the M1 region in HOXA10 (figure 2; electronic supplementary material, S3). Therefore, even if the reversals in vipers and elapids recovered a TFBS for HOXA10 in the enhancer H1-Myf5, the RAL morphotype probably remains expressed in the lineage due to the sequence patterns in HOXA10 that seem exclusive to snakes (electronic supplementary material, S3 and S4) and probably modify the TF-RE interaction unit.

3. Discussion

Developmental evolution explains the astonishing diversity of forms observed among living organisms, and the identification of evolutionary patterns in TF-RE interaction units contributes to a deep understanding of how developmental processes are involved in the recurrent evolution of specific phenotypes. Here we focus on independent origins of the trunk phenotype that characterizes emblematic animals such as snakes, elephants, manatees and caecilians, which is characterized by the presence of ribs in the lumbar region (RAL morphotype) and involved changes in the HOX10-H1-Myf5 interaction unit that regulates rib development in the axial skeleton of tetrapods. We compiled an extensive dataset for components of the HOX10-H1-Myf5 interaction unit (sequences for transcription factors of the HOX10-complex and the H1-Myf5 enhancer), and confirmed that changes in different components of the same TF-RE unit are associated with the evolution of similar phenotypes in lineages as distant as Lissamphibia and Mammalia, which diverged before 340 Ma (see [38,39]). These results imply that variations in different components of the same TF-RE units might confer enhanced flexibility to developmental systems, a mechanism that likely facilitates recurrent evolution of similar phenotypes in specific lineages. The sequence patterns identified here also pave de way for functional studies to experimentally explore the HOX10-H1-Myf5 interaction.

Previous studies in snakes, manatees and elephants identified a polymorphism in the H1-Myf5 enhancer that impedes the transcription factor HOX10 to interact with this regulatory element [15]. We expanded the phylogenetic coverage for this discussion and performed ancestral reconstructions that suggest a more ancient origin of the polymorphism in H1-Myf5. By including Lissamphibia for the first time in HOX10-H1-Myf5 analyses, we detected molecular convergence in the regulatory region H1-Myf5 among tetrapod lineages that diverged more than 340 Ma (following ages by [38,39]). Although we are not able to precisely estimate when in the last 310 million years the polymorphism has evolved in Caudata (following ages by [40]) because we have only one salamander species, it is possible to postulate a minimum age greater than 180–200 Ma [25] if other salamander species also have the polymorphism in H1-Myf5. Data from Gymnophiona suggest a similar interval (190–310 Ma) [40]. These results indicate a much earlier period for the first origin of this polymorphism than that of snakes (between 164 Ma and 82 Ma) [41] and Afrotheria (84 to 82 Ma) [42,43]. Accordingly, the polymorphism in H1-Myf5 might have first evolved in Lissamphibia and then more recently re-evolved in processes of molecular convergence in Serpentes and Afrotheria.

The evolution of a polymorphism in the H1-Myf5 enhancer in snakes has been interpreted as a key mechanism for the origin of a RAL morphotype (see [15]). Genetic manipulations reproducing this mutation in mice produce an extension of the rib cage [15], and in caecilians and salamanders, we presume an association between the RAL morphotype and the H1-Myf5 polymorphism similar to that described in snakes. The continuity of a RAL morphotype along the evolutionary history of Serpentes, however, does not seem conditional to the presence of this polymorphism in H1-Myf5. Although all snake lineages have the foreground morphotype (RAL region), our analyses including additional snake species not considered in previous studies suggest two reversals of this polymorphism in Serpentes (family Elapidae and the species Vipera berus). By integrating data from different components of the HOX10-H1-Myf5 unit, we identified that the RAL morphotype seems to persist in snakes if any part of the TF-RE interaction unit is modified. Specifically, in snakes, we detected sequence differences in the Motif 1 (M1) region of HOXA10, which binds to the TFBS of H1-Myf5. The HOX10 proteins inhibit rib formation, and the region M1 in HOXA10 is essential for the HOXA10-H1 interaction that modulates Myf5 expression and inhibits rib development, being crucial for establishing a ribless morphology in the lumbar region [15,17,33]. Our data challenges the assumption that occurrence of the RAL morphotype in Tetrapoda is conditional to mutations in the regulatory element H1 and provides evidence for associations between this phenotype and sequence variation in either component of the HOX10-H1-Myf5 unit. We acknowledge that the evolution of different trunk morphotypes among vertebrates might also involve other mechanisms, including differences in the expression limits of Hox10 genes as described in the whiptail lizard [22,24].

Transcription factors are expressed in several regions during embryo development, and HOXA10 is involved in processes as diverse as limb patterning [44–46], accessory sex organ development [47–49] and the inhibition of rib development in the lumbar region [15–17,33,50]. Given the pleiotropic effects expected from mutations in coding regions of transcription factors involved in multiple functions (see [51–54]), it seemed unlikely that the mutations identified in the snake HOXA10 reflect relaxed selection in lineages that evolved the RAL morphotype. Even if this protein has lost its function of inhibiting rib development in snakes that evolved the H1-Myf5 polymorphism, the transcription factor should remain functional in other regions of the embryo. In fact, our analyses do not support differences between the null model (neutral evolution) and alternative models admitting relaxed selection in any lineage or site in the HoxA10 dataset. The polymorphism in H1-Myf5 may have allowed non-synonymous substitutions in HoxA10 without impairing fitness, since the HOX10-H1-Myf5 binding was already blocked. In this case, a new mutation returning to the same ancestral base in the regulatory element may have improved the binding with HOX6, while the modification of HOX10, occurring much earlier in the evolutionary history of this lineage, might have assumed the role of maintaining a RAL morphotype.

Our study provides evidence for a polymorphism apparently recurrent among lineages that diverged 340 Ma and independently evolved a RAL morphotype, addressing a mechanism of enhanced flexibility in developmental systems granted by changes in different components of the same TF-RE unit—the HOX10-H1-Myf5—that may similarly affect the functional outcome of this signalling pathway. We postulate that this mechanism of enhanced flexibility facilitates independent origins of similar phenotypes, such as the RAL morphotype observed in Lissamphibia, Lepidosauria and Mammalia. We recognize that polymorphism analyses may be challenged by the possibility of sequencing error, so we focus on patterns shared by more than one species from the same lineage. Our dataset is the largest one so far analysed for the HOX10-H1-Myf5 unit, but it comprises nearly 1% of the 20 000 species of Amniota so far described, a percentage even less representative for non-mammalian lineages. Such limitations, however, did not impair our ability to detect molecular convergence among some of the most phylogenetically distant lineages within Tetrapoda and also enabled corroboration of a hypothesis for HOX10-H1-Myf5 evolution in the context of independent origins of the RAL morphotype. Additional sequencing of Hox10 and H1-Myf5 (as well as other genes involved in the rib development regulatory cascade) and in vivo experiments assure a deeper understanding of how TF-RE units evolve in the context of independent origins of similar phenotypes. We postulate a mechanism of enhanced flexibility in developmental systems settled on variation in either component of a TF-RE unit acting as a facilitator of recurrent evolution of similar phenotypes that might be considered in discussions distinguishing parallelism from convergence (see [11–14]), given that similar phenotypes may evolve through different changes in the same TF-RE unit that likewise affect the phenotypic outcome of a given developmental pathway.

4. Material and methods

(a) . Data for the enhancer H1-Myf5

In order to evaluate evolutionary patterns in the H1 regulatory region of the gene Myf5 (H1-Myf5), we used BLAST [55–57] implemented in NCBI (https://blast.ncbi.nlm.nih.gov/Blast.cgi) and ENSEMBL (https://www.ensembl.org/Multi/Tools/Blast) databases. Sequences with at least 90% of identity and the smallest e-values (less than 1 × 10⁻¹⁰) were retrieved. We also obtained tissue and DNA samples for seven Squamata species available at the Herpetological Collection of Ribeirão Preto (CHRP; University of São Paulo, Brazil) or donated to the laboratory. These consisted of two snake species (Agkistrodon contortrix and Coluber constrictor) and five lizards (Alopoglossus carinicaudatus, Ameivula ocellifera, Brasiliscincus heathi, Phyllopezus lutzae and Tropidurus torquatus). In most analyses performed we did not consider the sequence of the salamander Ambystoma mexicanum because this genome is distinctive among tetrapods in features that include a large size and a high variability [58]. The complete list of studied species with associated information is provided at the electronic supplementary material, S2.

Genomic DNA was extracted from the tissue samples aforementioned using the DNeasy Tissue Kit (Qiagen), following the manufacturer's instructions. Fragments of H1-Myf5 were amplified by standard polymerase chain reaction (PCR) using two primers described by Guerreiro et al. [15]: ‘5-TGTTGCAGGTTACTTAGTTATAG-3′ and ‘5-TAAAATACTGCAGTGACTTCATTC-3′). The PCR was established using 40 µl of PCR Master Mix (Reddymix, Abgene Inc.), 4 µl of each primer and 2 µl of the DNA template and implemented as follows: one initial denaturation cycle of 10 min at 94 °C, followed by 31 cycles (denaturation at 94 °C, annealing at 54 °C, extension at 72 °C, 1 min at each temperature) and concluded by 10 min at 72 °C for final extension. After confirming that amplified fragments approximated the expected size (320 bp), PCR products were purified, cloned and sequenced. Specifically, we purified fragments from a 2% agarose gel using the QIAquick Gel Purification System (Qiagen) or Wizard SV Gel and PCR Clean-Up System (Promega) and then cloned the fragments using the pGEM-T vector system (Promega) in Escherichia coli thermo-competent cells. After double selection of E. coli colonies by antibiotic and staining, mini plasmid preparations (minipreps) were performed using the QIAprep Spin Miniprep Kit (Qiagen) or Wizard Plus SV Miniprep DNA Purification System (Promega). Sequencing reactions were performed in both directions (sensu and anti-sensu) using oligonucleotides T7 and SP6 (Promega), in an ABI3100 sequencer (Applied Biosystems).

Sequences of H1-Myf5 obtained in our laboratory were deposited in GenBank (accession numbers ON058269–ON058275). From online repositories, we filtered the dataset to include only one representative by genus. The final H1-Myf5 database encompassed sequences for 229 species of Tetrapoda and is available in the electronic supplementary material, S1. All sequences were aligned using MAFFT v7.305b [59].

(b) . Data for the transcription factor HOX10

We also evaluated molecular patterns in the complex HOX10 using orthologous sequences for HOXA10, HOXC10 and HOXD10 retrieved from NCBI and ENSEMBL. For repeated species, we selected the one with the highest number of amino acids using the package ‘protr’ [27] in R [60]. Foreground species were defined as the ones characterized by the RAL morphotype. For these, we manually blasted the missing Hox10 sequences against the NCBI database, and sequences with at least 90% of identity and the smallest e-values (less than 1 × 10⁻¹⁰) were retrieved. These sequences were translated using SeaView [61] and added to the dataset. We identified four well-conserved amino acids for sequences of each paralogue HOXA10, HOXC10 and HOXD10 (sites [46,48,49,62]) (figure 3). These conserved sites were used to confirm BLAST analyses, and sequences comprising any change in these four amino acids were interpreted as low-quality data and therefore excluded. We aligned sequences using the software MAFFT v7.305b [59]. The dataset was filtered to remove more than one species of the same genus. The final alignments for HOXA10, HOXC10 and HOXD10, comprising 223, 200 and 207 species, respectively, are available in the electronic supplementary material, S3.

(c) . Correlations between morphotypes and sequences from the HOX10-H1-Myf5 unit

Associations between the presence of a RAL morphotype and shifts of evolutionary rates in specific sites of the sequence alignment were tested using the TraitRateProp software [28]. We codified the species with the foreground morphotype (RAL morphotype) as ‘1’ and the others as ‘0’ and tested all four datasets (H1, HOXA10, HOXC10 and HOXD10). The dated phylogeny used was downloaded from the database TimeTree of Life (TTOL; [25]) for Tetrapoda. We trimmed the phylogeny species in order to match the molecular datasets (i.e. alignments).

(d) . Sequence ancestral reconstructions

We used the online version of IQ-Tree 1.6.12 [63,64] to select the following substitution models for our sequence datasets: H1-Myf5 – TN, HOXA10 – HIVb + G4, HOXC10 – LG + G4, HOXD10 – HIVb. Ancestral reconstructions of the H1 sequence were performed in the ‘baseML’ software and of HOX sequences in the ‘codeML’ software, both implemented in the PAML 4.9 h package [34].

(e) . Analyses of selection regimes

We identified specific amino acid mutations in the first site of the M1 region of HOXA10 in snakes and therefore decided to evaluate if sequence patterns in the gene that encodes this protein deviates from a model of neutral evolution. We first reduced our dataset preserving the taxonomic balance by filtering one representative species for each family, favouring those present in the largest number of datasets. As mammals were oversampled in our original dataset, we chose species at the order level, except for the foreground species, which we kept one for each family. We downloaded the DNA sequences for these species, available in the electronic supplementary material, S4. The final selection contained 32 species representing all lineages. In these analyses, we also used the phylogeny of the TTOL database. The tree topologies were fixed, and branch lengths were estimated using the ‘baseML’ program implemented in the PAML package [34] under the model UNREST, inferred by IQTree as previously described.

We evaluated if the foreground clades evolved under specific selection regimes using branch model and branch-site model. The ‘ete-evol’ tool of the ‘etetoolkit’ python environment [65] was used to produce the tree files with branches related to foreground clades marked. We evaluated the fit of branch models using three approaches: (i) aBSREL [35]; (ii) ‘RELAX’ framework [36]; both implemented in the Hyphy package 2.5.24 [66,67] and (iii) ‘codeML’ program of the PAML 4.9 h package [34]. The aBSREL searches for episodic diversifying selection on all branches, taking into account the site-level ω heterogeneity; however, it does not test for selection at specific sites. We chose the option that accounts for multiple nucleotide substitutions (aBSREL + Double + Triple). We also implemented ‘RELAX’ using default parameters. The results of aBSREL and RELAX were analysed and visualized in ‘HyPhy Vision’ [62]. Finally, in the ‘codeML’ program, we defined M0 as the one-ω ratio model (model = 0), where all branches have the same values of ω, while BM is the two (or more)-ω ratio model (model = 2) in which the program infers different ω for the background and foreground lineages. The ω ratio is a measure of natural selection acting on the protein, calculated by the ratio of non-synonymous and synonymous substitution rates [dN/dS] [34]. The ω values reveal the type of selection: purifying selection (ω less than 1); neutral evolution (ω = 1); positive selection (ω > 1). We set the ‘fix_blength’ parameter to 2 in order to fix the branch lengths previously estimated. The likelihood ratio test (LRT) and the p-values were calculated based on the log-likelihood (lnL) values and the number of parameters (np) in each model.

The branch-site models were evaluated using the following approaches, which estimate selection at specific sites: (i) contrast-FEL [37] implemented in the Hyphy package 2.5.24 [67] and (ii) PAML 4.9 h package [34]. Contrast-FEL tested for different selective regimes on sites of foreground clades, and we used the default parameters. To run PAML, we produced the control files using the ‘ete-evol’ tool, and ran the pre-configured branch-site models bsA, bsC and bsD. We also ran the site models M1 and M3 as null models. Again, the ‘fix_blength’ parameter was set to 2. The control files were used in the PAML 4.9 h package [34]. We calculated the LRT and the p-values to determine if the alternative model was better supported than the null model. We tested the alternative versus [versus] null pairs: bsA versus bsA1 (positive selection on sites in the foreground branches [68]), bsA versus M1 (relaxation on sites in the foreground branches [68]), bsC versus M1 (different omegas in foreground clades branches sites [69]), bsD versus M3 (different omegas in foreground clades branches sites [69,70]).

Data accessibility

Sequences of H1-Myf5 obtained in our laboratory were deposited in GenBank (accession nos. ON058269–ON058275).

The data are provided in the electronic supplementary material [71].

Authors' contributions

A.G.P.: conceptualization, data curation, formal analysis, investigation, methodology, software, validation, writing—original draft and writing—review and editing; M.B.G.: data curation, validation and writing—review and editing; T.K.: conceptualization, data curation, funding acquisition, investigation, project administration, resources, supervision, visualization, writing—original draft and writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This study was supported by FAPESP-Brazil fellowships to A.G.P. (grant no. 2019/21712-5) and M.B.G. (grant no. 2010/00447-7) and also FAPESP-Brazil Thematic Grants awarded to T.K. (grant nos. 2015/07650-6 and 2020/14780-1).