Genomic adaptations for tail-length evolution in arboreal snakes

Abstract

Adaptation to arboreal environments requires overcoming gravitational constraints, driving repeated morphological innovations across snake lineages. Among these, elongated tails represent a key adaptation that enhances branch-gripping ability, yet the genomic changes underlying this trait remain poorly understood. Here, ancestral state reconstruction revealed that arboreality evolved independently in multiple snake clades, with tail elongation as a recurrent morphological adaptation. To investigate its genetic underpinnings, we generated a high-quality, chromosome-level genome assembly for the green cat snake (Boiga cyanea) and performed comparative analyses with the Asian vine snake (Ahaetulla prasina). We identified accelerated evolution in genes associated with somite specification, a critical process for axial elongation, and detected positive selection in key somitogenesis regulators, including HES7 and TBX18. Notably, LOXL3, which contributes to somite boundary formation, exhibited a conserved amino acid substitution in both arboreal lineages. In addition, convergent divergence of conserved nonexonic elements (CNEs) was observed in genomic regions linked to the GDF11-LIN28-HOX13 pathway, which governs the axial-to-tail transition. Functional assays confirmed that divergence in these CNEs alters regulatory activity, potentially modulating gene expression within critical developmental pathways. Collectively, our findings establish a genomic framework for snake axial elongation, highlighting how arboreal specialization shaped tail length evolution.

Introduction

Forested environments are defined by vertically stratified and spatially fragmented niches that provide access to resources and shelters while facilitating predator avoidance (Byrnes and Spence 2011; Taverne et al. 2018; Wu et al. 2022). However, the structural complexity of these habitats imposes major biomechanical challenges for arboreal locomotion. Navigating narrow, compliant, and variably oriented supports demands precise postural control and effective substrate engagement (Cartmill 1985; Taverne et al. 2018; Young 2023). These constraints have driven convergent morphological adaptations across vertebrate lineages, including the evolution of adhesion toepads in anurans and lizards, grasping forelimbs in certain mammals, and elongated, mass-distributing tails that facilitate balance and stability during arboreal movement (Irschick et al. 2006; Böhmer et al. 2018; Taverne et al. 2018; Russell et al. 2019; Wu et al. 2022). Tail elongation is particularly prevalent among arboreal and semi-arboreal quadrupedal mammals, such as rodents, primates, leopards, and tree shrews (Larson and Stern 2006; Young et al. 2015; Mincer and Russo 2020; Yuan et al. 2023), in which increased tail length enhances dynamic stability during canopy locomotion (Mincer and Russo 2020; Hager and Hoekstra 2021; Kingsley et al. 2024).

Among vertebrates, snakes exhibit an extreme form of axial elongation and limb reduction, enabling occupation of diverse ecological niches, including marine, fossorial, terrestrial, and arboreal systems (Peng et al. 2020, 2023; Yan et al. 2022; Tang et al. 2023). Notably, the transition to an arboreal lifestyle has occurred independently across multiple snake clades, spanning both basal (e.g. Boidae and Pythonidae) and derived (e.g. Viperidae and Colubridae) lineages, accounting for approximately 17% of the 4,000 extant species (Harrington et al. 2018; Uetz et al. 2025). In the absence of limbs, arboreal snakes rely heavily on axial elongation and prehensile tails to achieve secure anchorage and maneuverability along branches (Guimarães et al. 2014; Sheehy et al. 2016; Harrington et al. 2018). Comparative morphometric analyses have consistently revealed significantly greater relative tail lengths in arboreal snakes compared to nonarboreal snakes (Sheehy et al. 2016). Furthermore, morphological differences between arboreal and nonarboreal vipers are modulated by evolutionary signatures (Alencar et al. 2017). These differences likely reflect strong ecological selection on postcranial body plan architecture, potentially modulated by clade-specific developmental constraints.

Axial structure in vertebrates is progressively established during embryogenesis via the sequential formation of paraxial mesoderm into somites, which subsequently differentiate into vertebral elements along the anterior-posterior axis (Sun et al. 2024). This process is regulated by the segmentation clock, a molecular oscillator that controls periodic gene expression to produce bilaterally paired somites (Aulehla and Pourquié 2008). In snakes, the segmentation clock operates at an accelerated rate, approximately four-fold faster than in mice or lizards, resulting in rapid and extended somite production (Gomez et al. 2008; Aires et al. 2019). Posterior axial extension, including caudal vertebral formation, is driven by the proliferation and migration of progenitor cells within the tail bud, which continuously contribute to the elongation of the posterior body axis. Tail morphology in vertebrates is tightly regulated by a genetic network comprising GDF11, LIN28, and HOX13 genes. Functional studies have demonstrated that loss of LIN28A leads to marked tail shortening, while its overactivation increases the number of caudal vertebrae (Aires et al. 2019; Robinton et al. 2019). In snakes, HOX13 display weak and transient expression in the tail bud region, potentially contributing to tail elongation (Young et al. 2009; Di-Poi et al. 2010). These findings indicate that modifications in the regulatory architecture of somitogenesis contribute to the development of elongated caudal vertebrae. Such genetic changes likely supported the evolution of longer tails in snakes, facilitating arboreal adaptation. Nonetheless, the genomic mechanisms remain insufficiently resolved.

In this study, integrated morphological and genomic approaches were applied to investigate the molecular mechanisms underlying tail elongation associated with arboreal adaptation in snakes. Morphometric analyses revealed that arboreal snakes exhibit significantly elongated tails compared to nonarboreal counterparts, primarily driven by an increased number of caudal vertebrae. Ancestral state reconstruction indicated that arboreality evolved independently across multiple snake lineages, while diversification analyses showed no significant correlation between arboreality and speciation rates, suggesting that arboreality represents an adaptive trait rather than a driver of lineage diversification. To explore the genetic basis of this morphological innovation, we subsequently generated a high-quality, chromosome-level genome assembly for the arboreal snake Boiga cyanea (contig N50 = 158 Mb). Combined with the genome of the distantly related arboreal snake Ahaetulla prasina, comparative genomic analyses revealed evolutionary signatures in genes involved in the FGF and NOTCH signaling pathways, such as FGFR1 and HES7, which collectively drive somitogenesis during development. Additionally, convergently divergent conserved nonexonic elements (dCNEs) were found to be associated with the key GDF11-LIN28-HOX13 axis, which orchestrates the termination of the tail during axial patterning. Luciferase assays of these dCNEs revealed significant regulatory variation between arboreal and nonarboreal groups, suggesting that cis-regulatory modifications may contribute to altered axial patterning associated with tail elongation. Together, these findings provide insight into the molecular basis of the axial morphological innovation associated with arboreal specialization in snakes.

Results

Evolutionary dynamics of arboreality in snakes

Four morphological traits, including total length (TOL), tail length (TL), number of ventral scales, and number of subcaudal scales, were quantified to derive relative tail length (RTL) and the relative caudal vertebrae ratio (RCV) (Fig. 1a). Morphological measurements and habitats were obtained from 323 individuals representing 110 species across 15 families (Table S1). Ancestral habitat reconstruction indicated that arboreal ecology evolved independently across multiple snake lineages, particularly within Colubridae, encompassing canonical arboreal species such as Chrysopelea ornata, Ahaetulla prasina, and Boiga cyanea (Fig. 1b, Figure S1a-b). Ecological transition analyses revealed that the most common shift occurred from terrestrial to terrestrial/arboreal habitats, followed by direct transitions to arboreality (Fig. 1c). In contrast, transitions originating from aquatic habitats were rare, suggesting that terrestrial niches formed the basis of the evolution of arboreal specialization. To evaluate whether arboreality influenced diversification rates, Hidden State-dependent Speciation and Extinction (HiSSE) models were applied to a 244-species phylogeny (Figure S2). The best-fit model was a character-independent framework with four hidden states (CID-4, Akaike information criterion weight (AICw) = 0.66) (Table S3), indicating that diversification is better explained by unobserved variables rather than ecological state. Although arboreal species displayed a higher mean net diversification rate than nonarboreal counterparts (Figure S3), phylogenetic comparative analyses revealed no significant association between habitat and diversification rate (Tables S4 and S5). These results demonstrate that arboreality in snakes has evolved convergently across phylogenetically distant lineages but does not constitute a primary driver of lineage diversification.

Figure 1.

Evolutionary dynamics and morphological comparison of arboreality in snakes. (a) Schematic of four morphological measurements in snakes: total length (TOL), tail length (TL), number of ventral scales (VS), and number of subcaudal scales (SC). Position of the anal plate indicates the boundary between the trunk and tail. (b) Ancestral state reconstruction for habitat types across 22 snake families. (c) Estimated transition rates between habitat types inferred from the best-fitting model using 100 stochastic character maps. (d) Comparative analysis of relative tail length (RTL, above) and relative caudal vertebrae ratio (RCV, below) across four habitat types. Representative species for each habitat type: Ahaetulla prasina (arboreal), Python molurus (arboreal/terrestrial), Naja atra (terrestrial), and Hypsiscopus plumbea (aquatic). Box plots show distribution for each group, with sample sizes (n) indicated. Significance (Welch's ANOVA with Games-Howell post hoc test) is marked for comparisons between arboreal type and others. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Arboreal snakes display pronounced tail elongation

Incorporation of RTL data from Sheehy et al. (2016) revealed that arboreal and arboreal/terrestrial species possessed significantly higher RTL values than aquatic or terrestrial species (Fig. 1d, Figure S4a, Table S6). RTL was also significantly higher in arboreal species than in arboreal/terrestrial snakes (Figure S4a), suggesting that tail elongation functions as a key morphological adaptation associated with arboreal specialization. Comparison of RCV across ecological groups further demonstrated that arboreal snakes exhibited markedly elevated RCV values relative to the other three groups (Fig. 1d, Figure S4b, Table S7). Notably, RTL and RCV were strongly positively correlated (Spearman's R = 0.91, P = 1.12E-123) (Figure S5), demonstrating that tail elongation in snakes is primarily driven by an increase in caudal vertebral number.

Chromosome-level assembly and chromosome evolution

To investigate the genetic basis of tail elongation in arboreal snakes, a high-quality reference genome for the representative arboreal species B. cyanea was assembled by integrating 156.54 Gb of PacBio CLR long-reads, 51.21 Gb of PacBio HiFi long-reads, 130.57 Gb of Illumina paired-end data, and 244.64 Gb of Hi-C reads from a male individual (Fig. 2a, Tables S8 and S9). Long reads were used for initial genome construction, followed by iterative polishing using the Illumina short reads. The final assembly spanned 1.75 Gb, comparable to other published snake genomes (Figure S6). Scaffolds were anchored to 18 chromosomes ranging in size from 12 Mb to 371 Mb (Fig. 2b, Table S10). The genome exhibited high continuity and completeness (contig N50 = 158 Mb) (Table S11), recovering 98.1% of complete BUSCO genes (Table S12). A total of 20,038 protein-coding genes were annotated, 96.47% of which were functionally annotated using public databases. Mean gene number, coding sequence (CDS), and exon length were comparable to those reported in other publicly available snake genomes (Table S13). Repetitive elements accounted for 59.15% of the genome, comprising 2.23% nontransposable elements (non-TEs) and 56.92% transposable elements (TEs) (Table S14), a proportion consistent with other snake genomes (Figure S7). The genome exhibited marked expansion of long interspersed nuclear elements (LINEs) and long terminal repeat (LTR) retrotransposons, with divergence levels ranging from 0.15 to 0.20 from their respective consensus sequences (Figure S8). Synteny analysis across nine chromosome-level snake genomes revealed a high degree of collinearity (Fig. 2c), reflecting broad chromosomal conservation, consistent with previous reports (Yan et al. 2022; Peng et al. 2023). Using the B. cyanea genome as a reference, 567 evolutionary breakpoint regions (EBRs) were identified, primarily distributed across chromosomes 1, 2, 4, and 5 (Figure S9). Previous studies have shown that EBRs frequently demarcate areas of elevated structural plasticity (Larkin et al. 2009; Álvarez-González et al. 2022). In line with these findings, our results suggest that these regions may be important in generating new genetic variation within focal lineages. Together, these genomic data provide the foundation for subsequent analysis.

Figure 2.

Genome assembly of Boiga cyanea. (a) Circos plot of B. cyanea genome. From outermost to innermost ring: 18 pseudochromosomes (scale: 4 Mb per tick); gene density on forward (blue) and reverse (yellow) strands; GC content (40% to 60%); and transposable element (TE) occupancy (0% to 100%). A photograph of B. cyanea is shown in the center. (b) A Hi-C interaction heatmap supporting 18-chromosome assembly, with color intensity reflecting contact frequency. (c) Chromosome synteny comparison among nine snake genomes.

Phylogenetic reconstruction and gene family expansion

Phylogenetic analysis was conducted using two representative arboreal snakes (B. cyanea and A. prasina), eight nonarboreal snakes, and three lizard outgroups (Table S15). A total of 8, 059 single-copy orthologous genes were identified and used to construct a maximum-likelihood phylogeny (Figure S10). The resulting tree showed that B. cyanea and A. prasina belong to distinct evolutionary lineages, supporting independent origins of arboreal adaptation in these species (Fig. 3a). Divergence time estimates, based on seven fossil calibration points (Table S16), indicated that B. cyanea diverged approximately 23.5 million years ago (Mya), while A. prasina diverged around 35.5 Mya (Figure S11). Gene family expansion analysis based on 18,727 orthologous groups revealed 15 significantly expanded gene families in the B. cyanea genome (Fig. 3a). These gene families were significantly enriched in 58 GO terms (Table S17), with the most pronounced signal observed for olfactory receptor activity (P = 1.58E-55). In A. prasina, eight significantly expanded gene families were identified, with enrichment in 57 GO terms (Table S18), with olfactory receptor activity representing the top-ranked signal (P = 2.17E–05). These findings suggest that enhanced olfactory function may be a shared characteristic associated with arboreal adaptation. This aligns with previous findings that the olfactory system has undergone adaptive modifications to support habitat specialization in snakes (Peng et al. 2022).

Figure 3.

Adaptive evolution of coding genes in arboreal snakes. (a) Phylogeny of 10 snakes and three lizard outgroups, scaled by divergence time. Green and red numbers indicate significantly expanded (+) and contracted (−) gene families, respectively. Bar plots show RCV values across 10 snakes, colored by four habitat types. (b) Convergent amino acid substitution in LOXL3 shared by A. prasina and B. cyanea, with affected sites highlighted in red. (c) Three-dimensional structure of B. cyanea LOXL3 protein, with substitution site marked by an arrow. (d) Common rapidly evolving genes (REGs), with shading intensity indicating P value. Axes represent ω values in B. cyanea and A. prasina. Genes related to somite development are labeled. (e) Significantly enriched GO terms for common REGs. Colors indicate -log (P) and dot size represents gene count.

Adaptive evolution of coding genes in arboreal snakes

The RCV of snake species across the four habitat types was markedly elevated in the two arboreal species, with A. prasina exhibiting a caudal vertebral count approaching parity with its trunk vertebrae number (Fig. 3a, Table S19). To investigate potential genetic convergence underlying this trait, amino acid replacement analysis (Xu et al. 2017) was performed between B. cyanea and A. prasina. This analysis identified 275 genes exhibiting convergent amino acid substitutions (Table S20). Among them, LOXL3, a gene expressed at the ends of myofibers and essential for somite boundary formation (Kraft-Sheleg et al. 2016), harbored a shared substitution (P205L) in both arboreal snakes (Fig. 3b). Structural modeling indicated that this substitution is located in the random coil region, which is adjacent to a β-sheets region (Fig. 3c). A previous study has shown that mutations in the random coil increase the binding free energy of residues within it (Du et al. 2021). This suggests that LOXL3 mutations may affect protein function or stability in arboreal snakes.

A genome-wide scan for shared rapidly evolving genes (REGs) in B. cyanea and A. prasina identified 1,105 candidates (Fig. 3d, Table S21), significantly enriched in 620 GO terms (Fig. 3e, Table S22). FGF7, the top-ranking gene under selection (P = 1.10E-05), is known to regulate mesodermal organization and somitic cell fate in zebrafish embryos (Yin et al. 2018). TEAD1, a transcriptional effector of the Hippo signaling pathway, also exhibited a significant signal of selection (P = 1.21E-04) and has been implicated in mesoderm specification in human pluripotent stem cells (Pagliari et al. 2021), suggesting a potential role in paraxial mesoderm development and axial patterning. Enriched functional categories included terms related to structural and physiological adaptations, such as positive regulation of blood vessel endothelial cell migration, response to gravity, and skeletal muscle satellite cell commitment (Fig. 3e). Notably, several terms directly associated with somitogenesis were detected, including somite specification, somite rostral/caudal axis specification, and paraxial mesoderm development. Among these genes, MEOX1, a transcriptional factor involved in somite morphogenesis, patterning, and differentiation, was identified (Mankoo et al. 2003). In humans, mutations in MEOX1 result in congenital syndromes characterized by cervical vertebral fusion (Bayrakli et al. 2013), highlighting its role in axial morphogenesis. FGFR1, a receptor within the FGF signaling pathway, was also detected. It is expressed in the presomitic mesoderm (PSM) and at the determination front, where it regulates somite boundary formation (Dubrulle et al. 2001). Functional disruption of FGFR1 leads to defective somitogenesis, characterized by segmentation defects in paraxial mesoderm (Yamaguchi et al. 1992). Additional shared REGs with established developmental roles included TBX6 and TCF15 (Fig. 3d). TBX6 is essential for paraxial mesoderm specification and somite patterning (Chapman et al. 1996; White et al. 2003; Yasuhiko et al. 2006), with zebrafish mutants exhibiting defective tail morphogenesis due to loss of somite integrity (Morrow et al. 2017). In mice, disruption of TCF15, which is required for normal segmentation, results in caudal agenesis and abnormal somite formation (Burgess et al. 1996). These findings suggest that conserved selective pressures targeted segmentation-associated genes during arboreal axial trait evolution.

Lineage-specific evolution of somitogenesis-related genes in arboreal snakes

A total of 307 positively selected genes (PSGs) and 749 REGs were detected in B. cyanea, while 372 PSGs and 776 REGs were identified in A. prasina (Fig. 4a, Tables S23–S28). Analyses revealed enrichment in terms associated with somite development and key developmental signaling pathways (Tables S25 and S28). In B. cyanea, genes were enriched in axial mesoderm morphogenesis, somite rostral/caudal axis specification, and specification of axis polarity (Fig. 4b, Table S25). In A. prasina, genes were enriched in fibroblast growth factor receptor signaling and paraxial mesoderm development (Fig. 4b, Table S28). MEOX1 was identified as both a top-ranked PSG (P = 1.78E-06) in B. cyanea and a REG (P = 9.90E-05) in A. prasina, indicating strong lineage-specific selective signals in both arboreal species. Additional genes associated with axial development were also detected. Notably, HES7, a core component of the segmentation clock (Harima and Kageyama 2013), showed evidence of positive selection in the B. cyanea lineage (Fig. 4a). HOXD3 and HOXC4, both essential for anterior-posterior axis patterning and axial elongation in vertebrates (Di-Poi et al. 2010; Denans et al. 2015), were identified as REGs in B. cyanea and A. prasina, respectively. Three T-box genes (TBX6, TBX18, and TBXT) were also identified as REGs. Among them, TBXT is a key regulator of axial skeleton extension (Schulte-Merker et al. 1994), and its disruption through coding or regulatory mutations is known to produce tail truncation phenotypes (Yuan et al. 2023; Xia et al. 2024). In B. cyanea, TBX18 was identified as a REG with three species-specific amino acid substitutions (Fig. 4c). In mouse embryos, loss of TBX18 disrupts anterior-posterior polarity in the presomitic mesoderm, leading to defective segmentation of the axial skeleton (Bussen et al. 2004). These demonstrated that genes related to somitogenesis would also be influenced by lineage-specific evolutionary signals, which would further shape the characteristic body axis of species.

Figure 4.

Lineage-specific evolutionary signals in arboreal snakes. (a) Intersections of positively selected genes (PSGs) and rapidly evolving genes (REGs) in A. prasina and B. cyanea. UpSet plot shows number of genes unique to or shared between each set, with bars below indicating set sizes. (b) Representative significantly enriched GO terms associated with body plan development identified from each lineage-specific gene set. (c) TBX18 shows three amino acid replacements specific to B. cyanea and under accelerated evolution.

Convergent divergence of conserved nonexonic elements involved in the GDF11-LIN28-HOX13 network

Conserved nonexonic elements (CNEs) typically function as cis-regulatory elements to regulate the expression of target genes. Whole genome-alignments identified 139,474 CNEs, among which 479 exhibited convergent sequence divergence in the two arboreal lineages (Table S29). These diverged CNEs (dCNEs) were then mapped to genes located within 300 kb upstream or downstream, resulting in the identification of 2,489 associated genes. Functional enrichment analysis identified 225 significantly overrepresented GO terms (Table S30), including a strong signal in anterior/posterior pattern specification (P = 1.12E-09). Notably, FGF4 and FGF8 were located near two top-ranked dCNEs (CNE53850-FGF4, P = 6.31E-04, CNE8208_1-FGF8, P = 1.75E-03). These genes form the posterior-to-anterior signaling gradient that constitutes the wavefront regulating the spatial and temporal control of somite formation during vertebrate embryogenesis (Naiche et al. 2011). Several genes associated with segmentation were also identified (Fig. 5a). LFNG, a key component of the segmentation clock, functions as a cyclic gene regulating the timing of somite formation through oscillatory gene expression (Cole et al. 2002; Oates et al. 2012). DLL1, a downstream target of NOTCH signaling, plays a critical role in establishing somite anteroposterior polarity and maintaining synchronized oscillations during somitogenesis in mice (Miao and Pourquié 2024).

Figure 5.

Convergent divergence of CNEs. (a) Genes associated with dCNEs located within 300 kb. X-axis indicates number of convergent dCNEs near a gene; Y-axis indicates total number of CNEs. Genes associated with somite development are indicated in orange; key regulators are highlighted in red. (b) Schematic of gene regulatory network governing tail bud development during somitogenesis. Expression patterns in the trunk (OCT4 and GDF11) and tail bud (GDF11, LIN28, and HOX13) are color-coded. The embryo is created with BioRender.com. (c) Comparisons of regulatory activity among arboreal (A. pra and B. cya) and nonarboreal snakes (A. dia and X. uni) detected by dual-luciferase assays. Data are presented as mean ± standard error of four individual experiments. P value was calculated using Student's t-test (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001). (d) Schematic of genetic networks and pathways involved in somitogenesis in snake embryos. Genes identified in the genomic analysis are marked with color-coded stars.

Tail formation in vertebrates is governed by a conserved regulatory network comprising GDF11, LIN28, and HOX13 (Aires et al. 2019). In this pathway, GDF11 signaling activates HOX13 expression, which, in turn, suppresses LIN28 activity, thereby controlling the proliferative capacity of axial progenitor cells following the trunk-to-tail transition (Aires et al. 2019; Robinton et al. 2019). Notably, several dCNEs were found in proximity to genes within this regulatory module (Fig. 5b), suggesting that cis-regulatory modifications in arboreal snakes may contribute to altered expression of genes involved in tail development. To further validate our findings, luciferase assays were performed using constructs from two arboreal snakes (B. cyanea and A. prasina) and two nonarboreal snakes (Argyrophis diardii and Xenopeltis unicolor). Our results demonstrated that four of the five dCNEs exhibited significantly different regulatory activities between the arboreal and nonarboreal groups (Fig. 5c, Figure S12). Notably, two dCNEs (CNE31620-GDF11 and CNE5361-LIN28) showed significant differences in both B. cyanea and A. prasina compared to the nonarboreal outgroups. However, two dCNEs located near the HOX13 genes (CNE5794-HOXB13 and CNE34227-HOXC13) displayed significant regulatory differences specifically in one specific foreground lineage (either B. cyanea or A. prasina) and the background groups. These findings suggested a model in which the divergence of conserved nonexonic regulatory elements contribute to modifications in the genetic architecture of somite patterning and tail elongation, facilitating adaptation to arboreal environments.

Discussion

Environmental transitions often drive profound morphological diversification among vertebrates, such as the evolution of piscivory in fish species (Collar et al. 2009), body size disparity in monitor lizards (Collar et al. 2011), and the development of adhesive toe pads in tree frogs (Wu et al. 2022). Arboreal habitats, while offering new ecological niches, impose considerable mechanical constraints on locomotion and stability, particularly for snakes, which lack appendages and rely entirely on axial movement. In this study, ancestral state reconstruction revealed that arboreality originated independently multiple times within the snake phylogeny, with the most frequent evolutionary transitions occurring from terrestrial to arboreal/terrestrial niches. These findings indicated that terrestrial habitats facilitated the ecological expansion of snakes into diverse niches, consistent with the prevailing view of a terrestrial origin for crown snakes (Apesteguia and Zaher 2006; Peng et al. 2023). Morphological analyses demonstrated that arboreal snakes possess significantly longer RTL (Fig. 1d), aligning with previous findings in snakes (Sheehy et al. 2016). Tail elongation in arboreal snakes provides critical stability by enabling anchorage around narrow, flexible branches during locomotion (Guimarães et al. 2014; Alencar et al. 2017; Harrington et al. 2018). Similar trends have been observed in other arboreal vertebrates, such as clouded leopards (Yuan et al. 2023) and deer mice (Kingsley et al. 2024), where elongated tails enhance climbing ability and maneuverability in structurally complex forest environments. Although morphological adaptations can arise rapidly and contribute to species diversification (Franchini et al. 2014), our results suggest that arboreality itself may not have acted as a primary driver of snake diversification. This conclusion is consistent with previous work in vipers, which demonstrated that arboreal specialization did not significantly affect species diversity or speciation rates, but instead constrained morphological evolution (Alencar et al. 2017). Comparable decoupling of morphological innovation and taxonomic diversification has been observed in other vertebrate clades (Bergmann and Irschick 2012; Lee et al. 2016). For instance, the evolution of endothermy in ray-finned fish species, while representing a major physiological advance, did not accelerate diversification due to the high energetic demands limiting its adaptive advantage under most ecological scenarios (Melendez-Vazquez et al. 2025). Collectively, these findings suggest that although arboreality may confer specific ecological benefits, it may also impose physiological or ecological trade-offs, including increased predation risk or metabolic costs.

Axial elongation in vertebrates arises through the segmentation of the PSM, which sequentially produces somites that later differentiate into vertebrae, patterning to form the trunk and tail regions (Sun et al. 2024). Our analyses identified signatures of accelerated evolution in multiple genes enriched in somitogenesis-related pathways within arboreal snake lineages (Fig. 3e and Fig. 4b). In vertebrates, the number of somites—and thus vertebrae—varies extensively, ranging from approximately 10 in frogs to 300 in snakes (Richardson et al. 1998; Gomez et al. 2008). Previous study in the corn snake have suggested that the increased number of vertebrae may result from delayed regression of the PSM, which prolongs the duration of somitogenesis and increases the number of cell divisions through modulation of key developmental signaling cascades, including the FGF and WNT pathways (Gomez et al. 2008). This study identified accelerated evolutionary rates in key developmental regulators within arboreal snake lineages, particularly FGFR1, a central transducer of FGF signaling, and multiple genes associated with the WNT signaling pathway. Two top-ranked dCNEs in A. prasina and B. cyanea were also identified flanking the FGF4 and FGF8 loci. T-box transcription factors are known to coordinate somitogenesis by shaping WNT and FGF signaling gradients, thereby defining epithelialization thresholds essential for somite boundary formation (Bussen et al. 2004; McDaniel et al. 2024). In both arboreal species, TBX6 and TBX18 exhibited signatures of rapid evolution (Fig. 4a). These patterns suggest that natural selection has targeted both regulatory and coding components of the PSM signaling pathways, contributing to modifications in somite morphogenesis associated with tail elongation in arboreal snakes (Fig. 5d).

Snakes exhibited an exceptionally elongated axial body plan, a trait underpinned by a dramatic increase in the number of vertebral segments. This expansion is determined by the segmentation clock, a molecular oscillator that drives somitogenesis through periodic waves of gene expression in the PSM (Sun et al. 2024). In snakes, this clock operates at a markedly accelerated speed—approximately four times faster than in mice—enabling the production of a higher number of somites within a comparable developmental timeframe (Gomez et al. 2008; Gomez and Pourquié 2009). The segmentation clock functions through evolutionarily conserved negative feedback loops in cycling gene modules, including LFNG-mediated NOTCH signaling and HES7-dependent transcriptional repression. These oscillatory systems coordinate the precise spatiotemporal patterning required for axial skeleton formation (Aulehla and Pourquié 2008; Krol et al. 2011; Sun et al. 2024). In the present study, positive selection signatures were detected in HES7 within the B. cyanea genome. Additionally, LFNG was found in proximity to a convergently diverged CNE in both B. cyanea and A. prasina (Fig. 5a). These findings suggest that evolutionary modifications in core components of the segmentation clock may influence the rate of somite formation, contributing to the axial elongation observed in snakes. In arboreal species, such changes may enhance the precision of somite segmentation, supporting the development of elongated tails optimized for stability and maneuverability in forest environments.

In vertebrates, tail formation is governed by an evolutionarily conserved regulatory network comprising GDF11, LIN28, and HOX13 genes, which coordinates the spatiotemporal transition from axial elongation to tail termination (Aires et al. 2018; Aires et al. 2019). Alterations in HOX13 expression have been shown to influence caudal morphology in deer mice, while its weak and transient expression in corn snake tail buds has been implicated in extended axial growth and tail elongation (Young et al. 2009; Di-Poi et al. 2010; Kingsley et al. 2024). In this study, convergent dCNEs were identified near GDF11, LIN28, HOXB13, and HOXC13, which are key components of the tail termination network (Fig. 5b). Luciferase assays of the regulatory activity of the dCNEs revealed significant differences between arboreal and nonarboreal groups (Fig. 5c). Notably, two dCNEs located near GDF11 and LIN28 consistently exhibited significant differences in regulatory activity in the two arboreal snakes (B. cyanea and A. prasina), while two dCNEs located near HOX13 genes showed species-specific regulatory differences. This suggests that the convergent evolution of tail elongation in arboreal snakes may be driven by divergence in regulatory elements of conserved genetic pathways, with species-specific mutations further shaping their distinct tail morphologies. Collectively, these results indicate that divergence in these regulatory elements may alter the timing or intensity of gene expression during late axial development, potentially leading to an increased number of vertebrae and consequent tail elongation. Such modifications are likely adaptations that enhance postural stability and anchorage in arboreal environments. Additionally, a dCNE was found near TBXT, a transcription factor essential for axial extension (Fig. 5a). Previous work in clouded leopards has linked changes in TBXT to tail elongation as an adaptation to forested habitats (Yuan et al. 2023). These genomic signatures across squamates and mammals suggest that tail elongation represents a recurrent, functionally integrated response to the demands of locomotion on three-dimensional substrates.

Arboreal locomotion also imposes physiological constraints, particularly in elongated species such as snakes, where vertical orientation increases gravitational disturbance in circulatory function (Martins et al. 2001; Sheehy et al. 2016). Prior studies have proposed that tail elongation in arboreal snakes may represent an adaptive response to gravity-induced cardiovascular challenges, particularly those affecting blood circulation (Sheehy et al. 2016). In this study, REGs shared by B. cyanea and A. prasina were significantly enriched in functional categories related to response to gravity and positive regulation of blood vessel endothelial cell migration (Fig. 3e). Furthermore, lineage-specific signals were detected in genes associated with regulation of blood vessel branching and blood vessel morphogenesis in B. cyanea and A. prasina, respectively (Fig. 4b). These patterns suggest that modifications in circulatory regulatory pathways may have evolved in concert with morphological changes, forming an integrated physiological system capable of maintaining isometric contractions during vertical climbing. While this interpretation is consistent with observed patterns, further investigation is required to resolve the mechanistic links between axial elongation and physiological adaptations in arboreal snakes.

The evolution of vertebrate body plans has profoundly shaped phenotypic diversity, driving the emergence of distinct morphological adaptations (Wiens et al. 2006; Ward and Mehta 2010; Collar et al. 2016; Bonett and Blair 2017). At the genetic level, elongation of the giraffe neck and reduction of the avian tail have both been linked to positive selection or mutations in core developmental pathways—namely WNT, FGF, and NOTCH—which regulate somite formation, skeletal growth, and tissue differentiation (Rashid et al. 2014; Agaba et al. 2016). In our study, key genetic signals and regulatory elements associated with somitogenesis were identified in arboreal snakes. These findings support a broader evolutionary paradigm in which recurrent modifications of genetic programs governing somite development enable remodeling of vertebrate body plans. Despite these insights, functional validation in nonmodel species such as snakes remains a major challenge. Recent advances in organoid technology have enabled functional studies of somitogenesis, with in vitro reconstruction of human somites revealing the dynamic properties of the segmentation clock (Diaz-Cuadros, et al. 2020) and organoid models recapitulating the process of somite formation (Yamanaka, et al. 2023). Future efforts should prioritize the development of versatile experimental systems, including in vivo functional assays and in vitro cultures derived from snake tissues, and integrate organoid-based systems with genomic and epigenomic profiling to directly assess the regulatory roles of candidate elements.

Materials and methods

Morphological measurements and statistical analyses

Morphological data were obtained from 323 individuals representing 110 species from 15 families (Table S1). Four continuous traits—total length (TOL), tail length (TL), ventral scale count (VC), and subcaudal scale count (SC)—were measured following standardized protocols (Zhao 2006), and one discrete ecological trait (habitat) was classified as terrestrial, aquatic, arboreal, or arboreal/terrestrial (Sheehy et al. 2016). Specifically, TOL was measured from snout to tail tip, and TL was measured from the posterior margin of the anal scale to tail tip. Ventral scales were defined as the enlarged midline scales anterior to the anal scale, and subcaudal scales were defined as those on the ventral aspect of the tail (Fig. 1a). TOL and TL were measured to the nearest 1 mm using a flexible measuring tape; scale counts were manually recorded based on positional criteria. As vertebral identity is established during somitogenesis, and ossification occurs subsequently in snakes (Tzika et al. 2023), ventral and subcaudal scale counts were used as proxies for trunk and caudal vertebrae number, respectively. Two derived ratios were computed: relative tail length (RTL, TL/TOL) and relative caudal vertebrae ratio (RCV, SC/VC). Additional RTL data for 226 species were compiled from Sheehy et al. (2016). Welch's analysis of variance (ANOVA) with Games-Howell post hoc comparisons was used to evaluate group differences, and Spearman's rank correlation was used to assess associations between RTL and RCV.

Ancestral state reconstruction

The snake phylogeny was obtained from Zaher et al. (2019). The dataset was filtered to retain only those with available RTL and habitat information, resulting in a final dataset of 244 snake species across 22 families. Ancestral state reconstruction of the continuous trait (RTL) was performed using the fastAnc function in R package Phytools v2.4.4 (Revell 2012), yielding maximum-likelihood estimates and associated confidence intervals. Results were visualized using the contMap function. Ancestral state reconstruction of the discrete trait (habitat) was performed using SIMMAP (Bollback 2006). Model selection was performed by fitting equal-rates (ER) and all-rates-different (ARD) models with fitDiscrete in geiger v2.0.11 (Pennell et al. 2014), with AICc used for model comparison. The ARD model showed superior fit (Table S2) and was used for all downstream analyses. A total of 100 stochastic character maps were generated using “make.simmap”, and posterior probabilities for ancestral nodes were estimated by averaging state frequencies across simulations. Transition frequencies were summarized across all simulations by calculating the mean number and standard deviation for each transition type using “countSimmap”. Transitions with nonzero mean frequencies were considered evolutionarily significant.

Trait-dependent diversification analysis

To further evaluate the evolutionary consequences of arboreality on snake diversification, state-dependent speciation and extinction (SSE) models were implemented using the hisse package (v2.1.13) (Beaulieu and O’Meara 2016), following Melendez-Vazquez et al. (2025). Habitat states were binarized as nonarboreal (state 0) or arboreal (state 1) and used to estimate turnover rates, extirpation rates, state transitions, and diversification rates. Five models were employed to estimate HiSSE rate parameters: a null BiSSE model with equal rates across states (Maddison et al. 2007); a BiSSE-like HiSSE model allowing independent parameter variation; a full HiSSE model incorporating differential diversification associated with arboreality and any hidden states; and CID-2/CID-4 models focusing on hidden state-dependent diversification and ignoring arboreality. Model fit was assessed using AIC, and the best-fitting model was retained for inference. Tip-specific diversification rates for each habitat state were visualized using boxplots generated in ggplot2 v3.5.1 (Ginestet 2011). Rate differences between arboreal and nonarboreal states were statistically compared using phylogenetic ANOVA in RRPP with Bonferroni correction (Adams and Collyer 2018; Collyer and Adams 2018).

Sampling, library construction, and genome sequencing

An adult male Boiga cyanea specimen was collected from Xishuangbanna, Yunnan, China. Genomic DNA was extracted from muscle and liver tissues using a standard cetyltrimethylammonium bromide protocol. For the long-read sequencing, genomic DNA was sheared using a g-TUBE device (Covaris, USA), then purified and concentrated using AMPure XP beads (Beckman Coulter, USA). SMRTbell libraries were prepared following the manufacturer's standard protocols (Pacific Biosciences, Palo Alto, USA) and sequenced on the PacBio Sequel II and Revio platforms. For short-read sequencing, a 150-bp paired-end library was constructed and sequenced on the Illumina NovaSeq 6,000 platform. Hi-C libraries were generated from the same individual. Muscle tissue was fixed with formaldehyde, digested with MboI enzyme, and subjected to end repair, biotin labeling, and ligation. Following reversal of crosslinks, DNA fragments were sheared, enriched, and sequenced on the DNBSEQ-T7 platform.

Transcriptome sequencing

Heart, muscle, liver, lung, kidney, brain, eye, gonad, and spleen tissue samples were collected and pooled for transcriptome analysis to support genome annotation. Total RNA was extracted using a TRIzol reagent kit (Life Technologies, Carlsbad, CA, USA) and assessed for quality and concentration using a Nanodrop 2,000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA), Qubit 2.0 Fluorometer (Thermo Fisher Scientific), and Agilent 2,100 Bioanalyzer system (Agilent, USA). Sequencing libraries were constructed and sequenced on the Illumina NovaSeq 6,000 platform.

Chromosome-level assembly

Genome contigs were generated from PacBio HiFi reads using Hifiasm v0.19.9 (Cheng, et al. 2021). Contigs were subsequently corrected using PacBio sequences and Illumina paired-end reads using Nextpolish2 v0.2.0 (Hu et al. 2024), followed by further refinement with Hapo-G v1.3.7 (Aury and Istace 2021). Hi-C reads were aligned to the assembled contigs using Chromap v0.2.6-r490 (Zhang et al. 2021), and chromosomes were constructed by analyzing Hi-C contact maps with Yahs v1.2a.2 (Zhou et al. 2023). The final pseudochromosomes were generated based on the interaction matrix, which was manually adjusted using Juicebox (Durand et al. 2016). Assembly completeness was assessed using BUSCO v3.1.0 (Simão et al. 2015) against the vertebrata_odb10 database.

Genome annotation

For repeat annotation, transposable elements (TEs) were annotated using EDTA v2.0.1 (Ou, et al. 2019) and predicted with RepeatMasker v4.0.7 (http://www.repeatmasker.org/) against the RepBase transposable element library (http://www.repeatmasker.org/). Tandem repeats were detected using Tandem Repeats Finder (TRF) v4.09 (Benson 1999), and tRNAs were annotated using tRNAscan-SE v2.0.12 (Chan et al. 2021). Protein-coding genes were annotated through an integrative approach combining transcriptome-based and homology-based predictions, following repeat masking. RNA-seq reads were aligned with Hisat2 v2.2.1 (Kim et al. 2019) and assembled into transcripts using StringTie v2.2.1 (Pertea et al. 2015). For homology-based predictions, homologous protein sequences from closely related snake genomes were aligned to the assembled genome using Liftoff v1.6.3 (Shumate and Salzberg 2021) and GeMoMa v1.9 (Keilwagen et al. 2016). Final consensus gene models were generated by integrating all predictions with EvidenceModeler (EVM) v2.1.0 (Haas et al. 2008).

Gene functional annotation

Protein-coding genes were functionally annotated via BLASTp v2.15.0 (Altschul et al. 1990) searches against the NCBI nonredundant (NR, v20190401) and SwissProt (v 20200709) databases using a threshold E-value of ≤ 1e−5. Gene Ontology (GO) annotations were assigned using InterProScan v5.22-61.0 (https://www.ebi.ac.uk/interpro/), and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway annotation was performed using KoFamScan v3.0 (https://www.genome.jp/tools/kofamkoala/).

Chromosome synteny and evolutionary breakpoint region identification

Chromosomal evolution was analyzed across ten snake genomes, including Calamaria septentrionalis, Cylindrophis ruffus, Argyrophis diardii, Xenopeltis unicolor, Eryx tataricus, Psammodynastes pulverulentus, Ahaetulla prasina, Boiga cyanea, Hypsiscopus plumbea, and Pantherophis guttatus. Given that the genome of Hypsiscopus plumbea is assembled only at the scaffold level, synteny analysis was conducted using the remaining nine genomes with chromosome-level assemblies. Synteny analysis was assessed using JCVI v1.1.11 utility libraries for Python (https://github.com/tanghaibao/jcvi, accessed on 2 March 2022). Orthologous genes were identified using the jcvi.compara.catalog module, and syntenic blocks were assembled based on genomic coordinate anchor files. Block relationships were visualized using jcvi.graphics.karyotype. To further identify the evolutionary breakpoint regions, whole-genome alignments were constructed using LASTZ_D v1.03 (https://www.bx.psu.edu/∼rsharris/lastz/, parameters: K = 2,400, L = 3,000, Y = 3,000, H = 2,000, and the HoxD55 scoring matrix), using B. cyanea as the reference genome. Homology synteny blocks (HSBs) were defined using the makeBlocks function in DESCHRAMBLER at a resolution of 100 kb. Evolutionary breakpoint regions (EBRs) between two adjacent HSBs were identified following (Yan et al. 2022) and visualized using the R package RIdeogram v0.2.2 (https://github.com/tickingclock1992/rideogram).

Phylogenetic reconstruction and divergence time estimation

Ten snake species representing four habitat types (terrestrial, aquatic, terrestrial/arboreal, and arboreal) and three lizard outgroups were included for phylogenetic reconstruction (Table S15). RCV values for all snake species were calculated (Table S19). Orthologous single-copy genes were identified using OrthoFinder v2.5.4 (Emms and Kelly 2015). Four-fold degenerate sites from the aligned coding sequences of each gene family were extracted and concatenated for phylogenetic inference. Maximum-likelihood analysis was conducted using IQ-TREE v1.6.8 (Nguyen et al. 2015) with parameters: -alrt 2,000 -bb 2,000 -nt AUTO -bnni. Divergence times were estimated using seven fossil calibration points (Table S16) obtained from the TimeTree database (Kumar et al. 2022). Initial estimates were generated with r8s v1.71 (Sanderson 2003) using the Langley-Fitch (LF) method, followed by Bayesian relaxed molecular clock approach (−clock 2, –model 4) in MCMCTREE (PAML v4.9i) (Yang 2007) for refined divergence time inference.

Gene family expansion and contraction

Gene family evolution was assessed using CAFÉ v4.2.1 (Han et al. 2013) based on the phylogenetic topology. Gene families were identified by comparing the cluster size of each branch with the maximum-likelihood cluster size of the ancestral node leading to that branch. A larger ancestral node indicates gene family contraction and vice versa. The overall P value (family-wide P value in CAFE v4.2.1 based on MonteCarlo resampling procedure) was calculated for each branch and node. For each gene family with a significant overall P value (≤0.01), the exact P values were calculated using the Viterbi method. Gene families with both an overall P value and an exact P value ≤0.01 were defined as significant expansion or contraction.

Convergent evolution in arboreal snakes

To detect convergent evolution at the amino acid level, single-copy orthologs from 13 species were analyzed, with B. cyanea and A. prasina designated as foreground branches. Ancestral amino acid states at each node were inferred using the JTT + gamma amino acid substitution model in the evolver module of PAML v4.9i (Yang 2007), based on empirical frequencies and phylogenetic topology. Sites were designated as convergent if the derived amino acid states were shared by both foreground branches but differed from the inferred ancestral state, which was also conserved across background branches. The reliability of ancestral inference was validated by comparing simulated and inferred ancestral characters. Protein structure prediction was conducted using AlphaFold 3 (Abramson et al. 2024).

Common rapidly evolved genes (REGs) in arboreal snakes

Common REGs were identified by applying the branch model in CODEML (PAML v4.9i), treating B. cyanea and A. prasina as foreground lineages. Gene alignments were generated using PRANK v.150803 (Löytynoja 2014) and trimmed with trimAl v1.4 (Capella-Gutiérrez et al. 2009) in automated1 mode. The branch model was applied to estimate the ω parameter (nonsynonymous (dN) to synonymous substitution (dS) ratio) for each gene, with a null model assuming equal ω across all branches (model = 0) and an alternative model allowing foreground-specific ω (model = 2). Likelihood ratio tests (LRTs) were used to assess significance, and genes with P < 0.05 were designated as rapidly evolving. Gene Ontology (GO) enrichment was performed across biological process (BP), molecular function (MF), and cellular component (CC) categories, using the clusterProfiler R package v4.0.1, with significance threshold set at P < 0.05. Enrichment terms showing strong signals or functionally relevant to body plan development were focused.

Identification of positively selected genes (PSGs) and REGs in two arboreal lineages

A subset of single-copy orthologs was analyzed to detect signatures of positive selection in B. cyanea and A. prasina. For each species, the branch-site model implemented in CODEML (PAML v4.9i) (Yang 2007) was used with the respective lineage assigned as the foreground branch. Under this model, the null hypothesis assumes that all sites across all branches evolve with ω fixed at 1, while the alternative hypothesis allows a subset of sites to evolve with ω > 1 specifically along the foreground branch. LRTs were applied to compare the two models, and P values were calculated based on chi-square distribution. Candidate PSGs were defined as genes with P < 0.05 and at least one positively selected site with a Bayes Empirical Bayes (BEB) posterior probability > 0.95. REG identification was performed using the branch model in PAML v4.9i (Yang 2007).

Analysis of CNE divergence

Multiple whole-genome alignments across ten snake species were constructed using LASTZ_D v1.03, with B. cyanea as the reference genome. Alignments were refined using AxtChains (http://www.soe.ucsc.edu/∼kent) and ChainNet (http://www.soe.ucsc.edu/∼kent), and low-quality regions were excluded prior to generating final alignments with Multiz v11.2 (Blanchette et al. 2004). Four-fold degenerate sites were extracted to estimate neutral branch length using phyloFit (parameters: -EM -precision HIGH -subst-mod REV) based on PHAST v1.3 packages (Hubisz et al. 2011). The expected substitution rates of conserved elements relative to neutrality were estimated using phastCons (parameters: expectedlength = 45, target coverage = 0.3, rho = 0.3). Regions overlapping with coding regions were filtered using Bedtools v2.25.0 (Quinlan and Hall 2010) and only CNEs ≥ 30 bp were retained following previously established criteria (Peng et al. 2020; Wang et al. 2023). For each CNE, local sequence identity between the extant node sequence and reconstructed ancestral sequence was calculated using alignments from PRANK v.150803 (parameters: -keep -showtree -showanc -prunetree -seed = 10) (Löytynoja 2014). Percent identity values were derived for each species relative to the ancestral sequence using Phytools (Revell 2012). The Forward Genomics branch method (parameter: -thresholdConserved = 0) (Prudent et al. 2016) was employed to identify CNEs that diverged significantly in the foreground species (B. cyanea and A. prasina) compared to all other species. This method considers phylogenetic relatedness and evolutionary rate differences between species to compute the significance of the association between sequence divergence and phenotype. Diverged CNEs (dCNEs) under the significance threshold of P < 0.005 were further annotated using previous criteria (Yan et al. 2022).

Dual-luciferase assays

Genomic fragments of conserved nonexonic elements (CNEs; Supplementary table S31) from representative arboreal (Boiga cyanea: B. cya, Ahaetulla prasina: A. pra) and nonarboreal snakes (Argyrophis diardii: A. dia, Xenopeltis unicolor: X. uni) were chemically synthesized and cloned into the pGL3.0 luciferase reporter plasmid. All constructs were verified by Sanger sequencing. HEK293 cells were seeded in 24-well plates prior to transfection. After 24 hours, cells were co-transfected with 500 ng of constructed pGL3.0 vectors and 50 ng of pRL-SV40 internal control plasmid using linear 25-kDa polyethylenimine (PEI). Cellular lysates were harvested 36 h post-transfection with passive lysis buffer. Luciferase activity was measured using a Dual-Luciferase Reporter Assay Kit (Vazyme Biotech, China) on a Synergy H1 microplate reader (BioTek, USA), with seven technical replicates per CNE. Firefly luciferase activity was normalized to Renilla luciferase activity, and statistical significance was assessed using Student's t-test.

Author contributions

J.T.L. conceived the study and designed the project. Z.W., W.W., F.S and J.L.R. conducted morphological analyses; Z.W., W.W., C.Y. and C.Y.T performed evolutionary analyses; W.Z. and W.W. wrote the original draft. Z.W, X.Z. and J.T.L contribute to the revision of the manuscript.

Supplementary material

Supplementary material is available at Molecular Biology and Evolution online.

Funding

This work was supported by the National Key Program of Research and Development, Ministry of Science and Technology of China (2023YFF1304800); National Natural Science Foundation of China (32400349, 32325011, 32220103004); China Postdoctoral Science Foundation (2023M743417); Sichuan Science and Technology Support Program (2024NSFSC1181); International Partnership Program of Chinese Academy of Sciences (071GJHZ2023041MI).

Data availability

Raw sequencing data and genome assemblies were deposited in the China National GeneBank Nucleotide Sequence Archive under accession number CNP0007757. The sequence alignments of TBX18 and LOXL3 have been deposited on Figshare through the following address: https://doi.org/10.6084/m9.figshare.31015771.