Comparative genomics sheds new light on the convergent evolution of infrared vision in snakes

Abstract

Infrared vision is a highly specialized sensory system that evolved independently in three clades of snakes. Apparently, convergent evolution occurred in the transient receptor potential ankyrin 1 (TRPA1) proteins of infrared-sensing snakes. However, this gene can only explain how infrared signals are received, and not the transduction and processing of those signals. We sequenced the genome of Xenopeltis unicolor, a key outgroup species of pythons, and performed a genome-wide analysis of convergence between two clades of infrared-sensing snakes. Our results revealed pervasive molecular adaptation in pathways associated with neural development and other functions, with parallel selection on loci associated with trigeminal nerve structural organization. In addition, we found evidence of convergent amino acid substitutions in a set of genes, including TRPA1 and TRPM2. The analysis also identified convergent accelerated evolution in non-coding elements near 12 genes involved in facial nerve structural organization and optic nerve development. Thus, convergent evolution occurred across multiple dimensions of infrared vision in vipers and pythons, as well as amino acid substitutions, non-coding elements, genes and functions. These changes enabled independent groups of snakes to develop and use infrared vision.

1. Introduction

Convergent evolution, a central topic in evolutionary biology, involves adaptive change outside of shared ancestry and with the exclusion of non-adaptive noise. Analyses of molecular convergence have been widely used to reveal genetic mechanisms underlying various convergent phenotypes, such as echolocation among bats and whales [1], obligate scavenging in vultures [2], flight loss in birds [3], craniofacial morphologies in the Tasmanian tiger and the eutherian grey wolf [4], aquatic adaptation in marine mammals [5] and underground environment adaptation in subterranean mammals [6,7]. Canonical genetic and genomic methods for detecting convergence have previously focused on identifying convergent amino acid substitutions [8,9]. However, recent studies revealed that non-coding regions are more likely to evolve convergently than coding regions as protein-coding genes are pleiotropic [3,4]. Thus, convergence seems to occur at multiple levels of organization [10,11].

Infrared vision in nocturnal snakes enables them to use infrared radiation to detect and prey on warm-blooded animals [12]. Infrared vision, a remarkable evolutionary innovation in some snakes, originated independently in three clades of snakes: Crotalinae (pit vipers), Pythonidae (pythons) and Boidae (boas) [13]. These snakes can detect infrared radiation through thermotransduction and with high sensitivity to temperature fluctuations as low as 0.001℃ [12]. Infrared signals are initially received and transformed into nerve impulses by the pit organ. Vipers have loreal pits located between each eye and nostril (figure 1a), while pythons possess labial pit organs distributed across the snout (figure 1b). Instead of a pit organ, infrared boas only possess infrared-sensitive receptors in labial scales.

Figure 1.

Anatomy of two kinds of infrared-sensing snakes. (a) Lateral view of Kaulback’s lance-headed pit viper (Protobothrops kaulbacki). (b) Lateral view of Angolan python (Python anchietae). (c) Diagram of the infrared vision system of pit vipers. (d) Diagram of the infrared vision system of pythons. Snake photos by Xiao-Long Liu and http://www.boapython.ch/.

The pit organ is a recently derived organ in infrared-sensing snakes and is composed of a thin membrane innervated by three branches of primary afferent nerve fibres (one ramus ophthalmicus branch and two ramus maxillary branches) originating at the trigeminal ganglion (TG) (figure 1c,d) [14]. Pit-bearing snakes have enlarged TGs compared with those of mammals while snakes without pits, such as Vipera ammodytes, lack this structure [12,15]. Through TG neuron fibres, electric signals are transduced ipsilaterally to the nucleus of the lateral descending trigeminal tract (LTTD) [16,17]. This shared pathway of infrared signal transduction in infrared-sensing snakes is independent from the canonical trigeminal sensory system found in other snakes and other animals [15]. From there, infrared signals are transmitted to the optic tectum (OT), which contains neurons that infrared stimuli can activate [18]. In the OT, infrared information is integrated with visual information to generate a thermal image [19].

Genetically, several genes have been reported with or suspected of functional association with infrared sensation in snakes. For example, transient receptor potential ankyrin 1 (TRPA1) is a heat receptor that is highly expressed in neuron fibres embedded in the pit organ membrane. Evidence of convergent evolution among infrared-sensing snakes was reported in TRPA1, consisting of three convergent sites [20]. KCNK4 is a potassium channel gene previously identified as highly derived in infrared-sensing pit vipers and provisionally associated with infrared perception through regulation of the resting membrane potential and nerve excitability [21].

Despite progress in molecular mechanistic studies of snake infrared vision, limited research exists on the convergent evolution of infrared vision among different clades of snakes at the genome-wide scale. Furthermore, little is known about the genetic differences underlying the specific nervous system structures found in infrared-sensing snakes, such as the OT and trigeminal nerve. One major reason for this is that a robust phylogeny for analysis of convergence is not available among currently published snake genomes. A recent analysis [21] examined the infrared vision-associated genes in pit vipers, but it was limited by the availability of genomic resources (only exome data were available) during their identification of candidate genes associated with infrared vision in pit vipers.

To help close this gap, we report a reference-quality genome of the sunbeam snake Xenopeltis unicolor, the closest known non-infrared relative of the infrared-sensing snake Python bivittatus. This addition, along with existing resources available for infrared-sensing vipers and related non-infrared outgroups, forms a robust model for assessing convergent evolution of infrared vision. Our analyses detect signals of convergence in multiple dimensions of sequence evolution, including changes in amino acid sequences, gene sequences, gene function and non-coding regions. Independent interrogation of these multifaceted changes among pythons and pit vipers yields insights into the drivers underlying the evolution of infrared sensing in snakes.

2. Material and methods

(a). Sampling, library construction and sequencing

A male specimen of the sunbeam snake (X. unicolor) was collected, and the DNA extracted from it was sent for genome and transcriptome sequencing. To obtain sufficient high-quality DNA, fresh liver tissue of X. unicolor was first ground into powder using liquid nitrogen. The DNA extraction process was conducted using the Blood and Cell Culture DNA mini kit (Qiagen, Hilden, Germany). Subsequently, the sequencing library was prepared for PacBio (Menlo Park, CA, USA), Illumina HiSeq X-ten (libraries with short insert sizes of 350 bp for 2 × 150 bp paired-end sequencing; San Diego, CA, USA), and high-throughput chromosome conformation capture (Hi-C) sequencing. In addition, total RNA was extracted from five different tissues/organs of the specimen, including the brain, liver, kidney, heart and muscle, using the Trizol RNA extraction and isolation kit (Invitrogen, Carlsbad, CA, USA). The extracted RNA was quantified using the Nanodrop 2000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA) and then prepared for sequencing on the Illumina HiSeq X-ten platform.

(b). Data filtering

Any Illumina short read that contained more than 10% unknown bases or more than 50% low-quality bases, along with its paired-end read, was discarded. In addition, Illumina reads were trimmed of adaptors, and duplicated reads were removed. The filtering of short reads was conducted using custom Perl scripts. Subreads of PacBio long reads were retained for further analysis. For Hi-C reads, the HiC-Pro software (v. 2.10.0) [22] was used to filter out low-quality reads.

(c). Genome assembly

The genome of the sunbeam snake was newly assembled following methods outlined in a previous study [23]. Initially, high-quality PacBio reads were assembled using the ‘correct-then-assemble’ method from the NextDenovo v. 2.4.0 package (https://github.com/Nextomics/NextDenovo), with the options ‘read_cutoff = 1 k, seed_cutoff = 32 k, blocksize = 3 g’. Next, the raw contigs were corrected using NextPolish v. 1.3.1 [24], which used PacBio sequences and Illumina paired-end reads, using the ‘best’ algorithm module. Subsequently, Hi-C read pairs were mapped onto the raw contigs using Bowtie2 [25] with the ‘single-ended’ model. The HiCUP pipeline (v. 0.8.0) [26] was then used to remove the invalid self-ligated and unligated fragments.

Valid interaction pairs were used to calculate the linkage frequency among all contigs based on an agglomerative hierarchical clustering algorithm [27]. Using the linkage suggested by Hi-C signal density, we clustered linked contigs based on a preset number of partitions (2n = 36). Subsequently, PacBio reads were mapped to raw assembly using the Minimap2 package [28]. The optimal mapped reads of each linked contig group were used to conduct local assembly. The parameter ‘seed_cutoff’ for each local assembly was used for the calculation of results using the command ‘seq_stat’ for each assembly. These local assembled reads were again assembled by using NextDenovo and then polished by NextPolish to obtain the optimized contigs. Based on the linkage information, restriction enzyme site and string graph formulation, the optimized contigs were anchored to the chromosome-scale genome using ALLHiC. The placement and orientation errors that exhibited obvious discrete chromatin interaction patterns were manually adjusted [27].

(d). Genome characteristics estimation

All filtered Illumina reads were used to estimate the genome size of the sunbeam snake using the k-mer method. The k-value was set to 17, and the frequency distribution of 17-mer depth was calculated. The genome size was calculated using the following formula: genome size = total k-mer number/peak k-mer frequency depth. This estimated genome size was then used to assess the integrity of the assembly.

(e). Transposable elements annotation

Transposable elements (TEs) were identified using a combination of homology searching and ab initio prediction methods. For homology-based prediction, we used RepeatMasker [29] and RepeatProteinMask to search against the Repbase TE library using default parameters. For ab initio prediction, we first built a reference repeat library based on the results from LTR FINDER [30], PILER [31] and RepeatScout [32] with default parameters. Subsequently, we used RepeatMasker to search against this library. In addition, the software package Tandem Repeats Finder [33] was used for the identification of tandem repeats within the genome.

(f). Gene annotation

In the genome prediction of X. unicolor, the following approaches were used.

(i). Homology-based prediction

We used a five-step process for homology-based gene prediction. First, we aligned protein sequences from an anole lizard, Burmese python, chicken and tiger rattlesnake to the masked genome assemblies using genblastg, which used tblastn hits to define gene models of high quality [34]. This step yielded the raw gene models. Next, we extracted candidate gene regions from the filtered and extended gene models. In addition, these candidate gene regions were subjected to a blast search against a query protein database to identify the best match. Subsequently, using GeneWise [35] and taking the best match for each peptide, we constructed the final gene models. Finally, we retained the gene models with the highest score for each candidate gene region.

(ii). RNA-sequence-based prediction

For annotation based on transcripts, we used raw RNA-sequence (RNA-seq) data obtained from five different tissues/organs (brain, liver, kidney, heart and muscle), which were cleaned with trimmomatic −0.36 [36], and de novo assembled using Trinity [37]. The resulting transcripts were further trimmed using seqclean to remove vectors, adaptors and primers. Subsequently, the tool Launch_PASA_pipeline.pl in PASA [38] was used to map clean transcripts to the sunbeam snake genome. Gene models were extracted using the ‘pasa_asmbls_to_training_set.dbi’ tool. Finally, tophat [39] was used to map RNA-seq reads to the repeat-masked genome, and the intron hints were obtained using bam2hints in augustus −3.2.3 [40].

(iii). Ab initio prediction

To train the hidden Markov model (HMM) parameters, we followed a standard procedure for both Augustus and SNAP [41]. High-quality gene models obtained from step (b) were used for this purpose. In the case of Augustus, we ran it with a hints file, which can greatly enhance the accuracy of gene prediction.

(iv). Integration of gene models

The gene models obtained from the three aforementioned gene prediction approaches were subsequently integrated using EVidenceModeler [42].

(g). Orthologue definition and alignment

To identify putative orthologous genes, we used the reciprocal best-hit approach [43]. The longest protein sequences from the Burmese python were used as reference sequences to perform reciprocal blast analysis with the protein sequences of other snakes. The pairs with the best hits were considered as orthologues. The code used in this study was deposited at https://github.com/JinfengChen/Scripts (last accessed 5 August, 2022). Subsequently, the orthologous genes were aligned and cleaned using Guidance with the following parameters: ‘--program GUIDANCE2 --msaProgram CLUSTALW --seqType codon --bootstraps 100 --seqCutoff 0.99 --colCutoff 0.99’ [44].

(h). Genomic convergence analysis

To test for genomic convergence between python and pit vipers, a total of 9186 orthologous genes were analysed. Protein sequences were aligned following the aforementioned procedure. We examined molecular convergence using two approaches: (i) the method proposed by Zou and Zhang [8]. A site was assumed as a convergent site if amino acids of a focused node at that site were identical to each other, but different from their most recent ancestral amino acids. We reconstructed the amino acid sequences of internal nodes using CODEML in PAML [45]. Herein, we did not compare the number of observed convergent sites with the neutral expectations derived from the JTT-f gene model, as the resulting genes would be filtered out using the second method. (ii) The conserved convergent cite (CCS) method only used highly conservative sites where all species with or without convergent phenotype are invariant and identical with the state of the outgroup [9].

(i). Selection analysis

Branch-site model and branch model in the PAML package were used to identify positive selection. The preferred standard test for positive selection is the branch-site Model A [46]. Model A categorized sites into one of four categories: ω0 < 1 (purifying selection); ω1 = 1 (neutral evolution); ω2a > 1 (positive selection in the foreground, purifying in the background) and ω2b > 1 (positive selection in the foreground and neutral in the background). In the respective null model, ω in site classes 2a and 2b could not exceed 1 in the foreground branch, thus bounding the null at neutral evolution. We compared how well each model fits the data using the likelihood ratio test with the significance evaluated by a χ² test under one degree of freedom. To identify rapidly evolving genes (REGs), we used the branch model, in which the alternative model allowed different rates for different branches and the null model assigned the same ratio to all branches. Selection tests were run on gene alignments that contained at least one pit-bearing snake, four outgroup species and a minimum of five total taxa. To avoid masking parallel signatures of adaptation, when testing the Burmese python, pit vipers were removed from each alignment, and vice versa when testing pit vipers, the Burmese python was excluded from the analyses. The tree topologies used for selection are shown in figure 2b.

Figure 2.

Convergent amino acid substitutions of genes related to heat response. (a). Diagram of pit organ and secondary structure of TRPA1, dots and numbers indicating positions of convergent sites. (b) Evolutionary rate of TRPA1. Numbers on each branch were the ω values (the ratio of non-synonymous to synonymous substitutions) estimated by the free ratio model in CODEML. (c) Diagram of convergent amino acid substitution sites in infrared-sensing snakes identified using Zhang and Kumar’s test and CCS method. Numbers at the top of each amino acid alignment indicate the position of this site.

(j). Gene ontology enrichment analysis

For each infrared-sensing snake lineage, the significant results of branch site model and branch model were pooled and gene ontology (GO) enrichment for positively selected genes was tested with the KOBAS pipeline. Chicken (gga) was chosen as the background species owing to the better characterization of the chicken genome in comparison to other reptiles. Fisher’s exact test was used to determine the significance (p < 0.05).

(k). Conserved non-exonic elements analysis

To identify conserved non-exonic elements (CNEs), we followed the procedures described by Zou et al. [2]. To perform the analysis, we used a custom Python script, which was deposited at GitHub (https://github.com/BIGtigr/GenomicPipelines/tree/master/CNE). Pairwise alignment between nine genomes and the reference genome of the anole lizard (AnoCar2.0) was conducted using the LASTZ program (https://github.com/lastz/lastz). All genomes used here were unmasked. Subsequently, the 10-way multiple-alignment was generated by Multiz [47]. Based on the whole-genome alignment, we extracted fourfold degenerate sites and estimated a neutral phylogenetic model (non-conserved model) using phyloFit [48].

The expected substitution rate of conserved elements relative to neutrality (rho) was estimated using phastCons [48] with the option of ‘–target-coverage 0.25–expected-length 20–estimate_rho’ and we ran separately on non-overlapping 10⁻⁷ bp windows of the input alignment. Conserved models for each window were combined by phyloBoot [48] and then used for initial conserved element prediction. Exon regions were excluded from the highly conserved elements to generate CNEs, using the command ‘subtract’ in BEDTools (https://bedtools.readthedocs.io/en/latest/, last accessed 15 December, 2020).

We mainly focused on CNEs located in the intergenic regions, in introns and within the 10 kb upstream or downstream flanking regions of genes. Trees with branch lengths for each CNE were generated by ‘baseml’ in PAML. Convergent evolutionary rate shifts for CNEs were detected by RERconverge, an R package that can test for associations between genes’ relative rates and traits of interest in a phylogeny [10]. Motifs in CNEs were searched using meme−5.5.3 with parameters of ‘--verbosity 1 --bgfile --nrdb-- --thresh 1.0E−4’ [49].

3. Results

(a). Reference-quality genome of the sunbeam snake

We sequenced a sunbeam snake (X. unicolor) through the integration of 57.41 Gb PacBio high-fidelity (HiFi) long-read sequences, 182.07 Gb Hi-C data and 94.74 Gb Illumina paired-end sequences (electronic supplementary material, tables S1–S3). The genome size was estimated to be 1.26 Gb based on the k-mer spectrum (electronic supplementary material, figure S1). We applied an improved assembly method that used Hi–C interaction pairs to cluster HiFi sequences that possess potential linkages and avoid any erroneous overlap caused by long-distance repetitive sequences during string graph assembly [23]. This resulted in a 1.52 Gb genome with contig N50 length of 99.3 Mb and scaffold N50 length of 205.56 Mb, making it one of the most contiguous snake genomes currently published (electronic supplementary material, table S4).

Based on the karyotype of the sunbeam snake (2n = 36), a total of 1.50 Gb (98.37%) assembled genome sequences were anchored onto 18 chromosomes (electronic supplementary material, table S5). Our assembled genome exhibited a high degree of completeness, as evidenced by the coverage of 99.93% short-reads across 99.78% of the genome (electronic supplementary material, table S6), and a recovery percentage of 93.6% of BUSCOs (Benchmarking Universal Single-Copy Orthologues, electronic supplementary material, table S7 and figure S2) [50] in 3354 conserved vertebrate genes from vertebrata_odb10, and 97.58% coverage of the 248 Ultraconserved CEGs (Core Eukaryotic Genes) (electronic supplementary material, table S8). Furthermore, we used a reference-free, k-mer-based approach and estimated an assembly quality value (QV) of 50.9691, exceeding the Vertebrate Genome Project standard of QV40 [51,52].

Taken together, we generated a reference-quality snake genome that possesses good contiguity, completeness and accuracy. Subsequently, we predicted 507.29 Mb (33.25% of the total) of TEs in the sunbeam snake genome, with the highest proportion (23.65%) in LINEs (long interspersed nuclear elements). By integrating homology and ab initio-based prediction, and supporting with transcriptomic evidence, we identified 19 656 protein-coding gene models. We compared repeat contents in 10 reptile genomes and found that DNA/hAT-Tip100 was significantly expanded in three infrared-sensing species (p = 0.0202, Fisher’s exact test, electronic supplementary material, figure S3).

(b). Convergence of amino acid substitutions

Convergence of amino acids refers to substitutions in multiple independent evolutionary lineages that result in the same amino acid residue, which is one of the most common types of molecular convergence [8]. Using the reciprocal best-hit (rbh) method, we generated sequence data containing 9186 orthologous genes from 10 reptile species with a high-quality and well-annotated genome, including three infrared-sensing snakes (Python bivittatus from Pythonidae, Protobothrops mucrosquamatus and Crotalus tigris from Viperidae), six non-infrared-sensing snakes (Pantherophis guttatus, Thamnophis sirtalis, Naja naja, Pseudonaja textilis, Notechis scutatus and X. unicolor), and an anole lizard (Anolis carolinensis) as an out-group (figure 2b, electronic supplementary material, table S9).

To identify convergent substitutions in two clades of infrared-sensing snakes, two methods were applied complementarily. Genes identified in both methods were regarded here as convergently evolved. First, we searched for putative convergent sites at which amino acids were the same in two infrared-sensing clades but differed from that of their respective most recent ancestral node based on a published topology (figure 2b) [8,53]. This method identified a total of 1059 genes with convergent sites. In addition, we used the CCS method and identified a set of 836 genes. In total, 698 genes with 910 sites were recovered by both methods (electronic supplementary material, table S10).

Of the convergent genes among infrared-sensing snakes, TRPA1 was found to contain nine convergent substitutions (figure 2a, electronic supplementary material, figure S4). In infrared-sensing snakes, TRPA1 was expressed in trigeminal fibres that innervate the pit organ and respond primarily to infrared stimuli (figure 2a) [20]. Besides TRPA1, we newly identified four convergent loci (ACADM, BAG3, NOS1 and TRPM2) that respond to temperature stimulus (electronic supplementary material, table S28). Two convergent sites were identified within TRPM2 (figure 2b), a thermosensory ion channel gene that drives responses of sensory neurons to increases in temperature [54]. Two other convergent loci (SLITRK6 and TMEM126A) associated with cranial nerve development (figure 2b), and six loci (ANOS1, CANX, FA2H, MAP3K13, MYO9A and PTPN11) play roles in axon guidance (figure 3). SLITRK6, a transmembrane protein gene, was shown to be of great importance for the development of normal vision [55]. TMEM126A encodes a transmembrane mitochondrial protein and was reported to be relevant to non-syndromic autosomal-recessive optic neuropathies [56].

Figure 3.

Transduction path of infrared signals and associated genes. Genes are grouped by functional category and listed in their standard abbreviations. Colours correspond to which analyses recovered that gene.

We then mapped the 698 convergent genes to Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways based on the KOBAS pipeline. We determined that 349 GO terms and 84 pathways were significantly enriched (electronic supplementary material, table S11, Fisher’s exact test, p < 0.05), including the functional categories neurogenesis (GO:0022008, p = 0.0104), mitochondrion organization (GO:0007005, p = 0.0119), optic nerve development (GO:0021554, p = 0.0183), axon initial segment (GO:0043194, p = 0.0284), face morphogenesis (GO:0060325, p = 0.0388) and animal organ morphogenesis (GO:0009887, p = 0.0451).

(c). Convergence of gene selection pressure

To investigate gene-level convergence, we performed episodic positive selection tests on each branch of the focal lineages that led to infrared-sensing snakes. Genes indicating positive selection or accelerated evolution in both clades of infrared-sensing snakes, but not in closely related non-infrared-sensing snakes, were recovered as convergent. We analysed 16 621 gene trees for pythons and 16 295 for vipers based on orthologues and constrained each tree to the resolved phylogeny (figure 2b). When performing branch and branch-site model tests in PAML on a focal infrared-sensing snake, we removed other infrared lineages from the background set of taxa. For each focal infrared-sensing snake clade, we also assessed selective pressures on the nearest branch leading to non-infrared-sensing snakes as a control. Genes that were also under positive selection on neighbouring non-infrared branches were filtered from the gene sets identified in infrared clades. Finally, we identified 370 positively selected genes (PSGs, electronic supplementary material, table S12) and 681 REGs (electronic supplementary material, table S13) in the infrared-sensing python branch. For the ancestral branch of two infrared-sensing vipers (Protobothrops mucrosquamatus and Crotalus tigri), our analyses yielded 344 PSGs (electronic supplementary material, table S14) and 473 REGs (electronic supplementary material, table S15).

Of these genes, 52 were both recovered in the infrared-sensing python and viper branches (electronic supplementary material, table S16) and retained as convergent genes. Among these convergent genes, TRPA1 was identified as a PSG in the python branch and a REG in both python and pit viper branches. We further estimated the ω values (d_N/d_S) on the branches for TRPA1 using a free ratio model in PAML and found that TRPA1 had undergone accelerated evolution twice: once across the entire branch of vipers, including their ancestral branch (ω = 1.07–1.28, figure 2b), and once at the ancestral branch of the python and sunbeam snake (ω = 2.234, figure 2b).

Among other genes identified with convergent positive selection, two (HELZ2 and PRPF4) are involved in vision, while five (IFT122, Rbms3, Spag17, TOP3B and POGZ) are related to neural development, two (LHX3 and TRIO) are associated with axons, three (IFT122, LHX3 and TRIO) are related to neuron projection and five (SCN8A, TCIRG1, SLIT2, VAX2 and EPHB1) function in the process of optic nerve development. HELZ2 was reported as a candidate gene responsible for inherited optic atrophy [57]. PRPF4 was reported to be a splicing factor that also correlated with defects in vision, photoreceptor morphology and retinal gene expression [58]. TOP3B was shown to play a crucial role in synaptic formation, and deficient TOP3B zebrafish embryos have defects in spinal motor nerves from the trunk and visual neural pathways from the head [59].

Applying the homology-based method, we successfully predicted 17 963 gene models in the genome of boas (Boa constrictor). We also examined the evolutionary events that occurred in the boas genome by using the same strategy. In total, we identified 576 PSGs (electronic supplementary material, table S17) and 555 REGs (electronic supplementary material, table S18). Among these genes, seven (ALPK3, CDH17, HSD17B4, IL17RC, JAK1, PCYOX1 and TRPA1) were shared by three clades of infrared-sensing snakes, 52 were shared by python and boas (electronic supplementary material, table S19) and 42 were shared by pit vipers and boas (electronic supplementary material, table S20).

(d). Convergence of function

Convergence of function can occur even when different genes are involved [11,60]. To identify this, we used the KOBAS pipeline to perform functional enrichment analyses separately on the python and viper branches using all PSGs and REGs recovered in each (929 genes for the python and 781 for pit vipers). Enriched functional categories shared by the python and pit viper branches were retained as putative convergence at the functional level. We found that 355 GO terms and 100 pathways were significantly enriched for the python (electronic supplementary material, table S21, Fisher’s exact test, p < 0.05), while 327 GO terms and 129 pathways were significantly enriched for the pit vipers (electronic supplementary material, table S22; p < 0.05). In total, 101 GO terms or pathways (electronic supplementary material, table S23), including neuron migration (GO:0001764), toll-like receptor signalling pathway (GO:0002224), extracellular matrix (ECM) receptor interaction (gga04512), synapse (GO:0045202) and ATPase activity (GO:0016887) were shared by both two clades of infrared-sensing snake.

We also searched for terms that were significantly enriched (p < 0.05) in one infrared-sensing snake clade but less significant (p < 0.1) in the other and found 70 such terms (electronic supplementary material, table S23) including axon (GO:0030424) and glutamatergic synapse (GO:0098978). Neuronal migration is a fundamental process that affects the final allocation of neurons in the nervous system, establishing the basis for the subsequent wiring of neural circuitry [61]. Axons are cables for the transmission of action potentials [62], and axons in the skin on the lateral side of the head are a fundamental component of sensory input [63]. To investigate convergences involving the same function but different genes, we removed genes that occurred in both branches and obtained 78 terms (electronic supplementary material, table S24), including the terms neuron migration (GO:0001764) and synapse (GO:0045202).

(e). Convergence of conserved non-coding elements

Regulatory regions are more likely than protein-coding genes in common pathways to underlie convergent phenotypes, as they may be subject to less pleiotropic constraint [3,64,65]. We identified 60 963 conserved CNEs (≥30 bp in length) based on 10-way whole-genome alignments (WGAs) of reptiles. Using RERconverge, a package that computes the gene-specific rates on branches of trees for detecting molecular convergence, we found 2193 out of 60 963 CNEs evolved convergently in two focal clades of infrared-sensing snakes. We annotated the function by attributing each CNE to its closest gene within 50 Kb. This yielded 997 potentially regulated genes, of which 21 genes with evidence for an excess of nearby convergent accelerated CNEs (figure 4), 42 were also found to contain loci with convergent amino acid substitutions (electronic supplementary material, table S25) and 58 were found to be positively selected (electronic supplementary material, table S26). To further screen for regulatory elements at greater distances, we extended the gene distance from 50 Kb to 1 Mb. This newly yielded 367 potential regulated genes, of which 15 genes demonstrated an excess of nearby convergent accelerated CNEs (electronic supplementary material, figure S5).

Figure 4.

Genes with evidence for an excess of nearby (within 50 Kb) convergent accelerated CNEs (conserved non-exonic elements). X-axis indicates the number of convergent CNEs around a gene while the y-axis indicates the total number of CNEs around this gene (log10-transformed). The top five genes within each of the seven groups (x-axis) with the smallest total number of CNEs were empirically regarded as genes with an excess of convergent accelerated elements. Genes with an excess of convergent elements are shown in red. Genes with more than one convergent element are plotted.

Among these genes, seven (NFIA, NFIB, NRP1, NRP2, PLXNA3, POU4F1 and Sema3A) are functionally associated with facial nerve and trigeminal nerve structural organization and five (DCC, NKX2-2, GLI3, CACNA1C and EXT1) are associated functionally with optic nerve development. DCC showed an excess of nearby convergent accelerated CNEs, with four convergent CNEs around it (figure 5a, electronic supplementary material, figure S6). DCC, a single-pass transmembrane protein, plays a key role in axon guidance and optic nerve development (figure 5b) [66,67].

Figure 5.

Convergent evolution of CNEs in infrared-sensing snakes. (a) Two convergent CNEs around DCC in infrared-sensing snakes. Each plot is labelled with the CNE locus (scaffold, start and end). Infrared branches are highlighted in red. Dots at the bottom represent the internal branches. THASI, Thamnophis sirtalis; PANGU, Pantherophis guttatus; NAJNA, Naja naja; PSETE, Pseudonaja textilis; NOTSC, Notechis scutatus; PROMU, Protobothrops mucrosquamatus; CROTI, Crotalus tigris; PYTBI, Python bivittatus; XENUN, X. unicolor; ANOCA, Anolis carolinensis. (b) Regulatory network of axon guidance in snakes. Genes with convergent CNEs are coloured in red.

NFIA also showed an excess of nearby convergently accelerated CNEs with five CNEs around it. NFIA plays an essential role in the emergence of A-fibre mechano-nociceptors in sensory ganglia [68]. Two convergent CNEs were identified around NRP1 and one convergent CNE near NRP2. NRP1 and NRP2 cooperate to guide cranial neural crest cells and position sensory neurons [69,70]. Sema3A is a diffusible guidance factor that induces cortical neurons expressing NRP1 (figure 5b). CNTNAP4 showed an excess of nearby CNEs with three convergent CNEs, belongs to the neurexin superfamily and has critical functions in neurological development and synaptic function.

Our functional enrichment analysis indicated that convergent CNE-associated genes were enriched for 230 GO terms and 25 KEGG pathways (electronic supplementary material, table S27) including axon guidance (GO:0007411), neuron projection (GO:0043005), neuronal cell body (GO:0043025), neuron migration (GO:0001764), neuron fate specification (GO:0048665), positive regulation of neuron projection development (GO:0010976) and negative regulation of neuron differentiation (GO:0045665).

4. Discussion

We present the first whole-genome evaluation of molecular and functional convergence between two lineages of infrared-sensing snakes. With the added robustness of a new non-infrared-sensing genome assembly, we used a variety of approaches to detect signals of convergence at multiple levels of genetic organization. Our analyses suggest that two of the repeated evolutionary origins of infrared vision in snakes have many common elements in their underlying architecture of genomic changes. Evidence of convergence between pythons and pit vipers exists in the amino acid substitutions of 910 sites in 698 genes, the accelerated evolutionary rates of 997 genes, the accelerated evolution of genes in particular functional categories and the conservation of certain non-coding elements. These loci, genes and functional categories involved with the evolution of infrared vision in snakes provide a rich set of targets to guide future functional studies of this complex trait.

Previously, whole-genome sequences of reasonable outgroups were available for pit vipers but not for pythons. Tu et al. [21] addressed this by supplementing their genomic analysis with whole-exome sequencing and focused on genetic changes happening in pit vipers specifically [21]. This provided a robust phylogeny, while limiting the scope of convergence they could detect to the exome and they highlight the importance of KCNK4. Compared with Tu et al. [21], we apply a genome-wide scanning strategy, enabling us to examine convergence in both coding and non-coding regions. In our results, KCNK4 is a REG in two tested vipers, but not in the python. Using a more taxonomically broad but genetically narrow approach, Peng et al. [71] explored convergence among three clades of infrared-sensing snakes (python, boas and vipers) focused on 234 target genes (heat-sensing-related and trigeminal development-related genes) [71]. They also examined the divergence of CNEs and found several CNEs around PMP22 and NFIB had diverged in snakes. Differing from Peng et al. [71], we elected to use a combination of methods and detected convergence associated with several different structures along the path of infrared perception at a genome-wide scale.

At the terminal trigeminal neurons that innervate in the pit membrane, TRPA1 serves as the primary transducer of heat (figure 2a). TRPA1 has been reported to experience positive selection in the ancestral branch of pit vipers [21]. TRPA1 has also been reported to evolve convergently among three infrared snakes and contained three convergent sites: L330M, Q391H and S434T [20]. In total, nine putatively convergent sites within TRPA1 were identified in our analysis (92, 151, 223, 331, 448, 639, 650, 877 and 918; figure 2a, electronic supplementary material, figure S4). Of these, five sites (92, 151, 223, 331 and 639) have been reported by two previous studies [72,73]. Thus, our study newly identifies four convergent sites (448, 650, 877 and 918) between pythons and pit vipers. In our sequence alignment, H₃₉₁ is present in both the out-group taxon of snakes (Anolis carolinensis) and the out-group taxon of pythons (X. unicolor), indicating that H₃₉₁ may have already occurred in the common ancestor of snakes. For S434T, only Crotalus tigris possess T₄₃₄ and the other nine species used in our analysis all have S₄₃₄. Thus, our results indicate that Q391H and S434T are not involved in the evolution of infrared vision. This highlights the importance of out-group taxa in detecting convergence.

Overall, our study further supports the significance of TRPA1 in the evolution of infrared vision and identifies possible key substitution sites within it. Convergent substitutions combined with positive selection are regarded as strong evidence for parallel adaptation [74]. Our analyses agree with Peng et al. [71], in that TRPA1 is positively selected in pythons and rapidly evolved in both pythons, pit vipers and boas. Furthermore, our analyses reveal that TRPA1 had rapidly evolved in the common ancestor of pythons and sunbeam snakes. This indicates that some changes may have occurred in the function of TRPA1 at the ancestral branch of the python and sunbeam snake. However, it remains unknown which site or group of sites gave rise to the functional changes of TRPA1. Further detailed functional assays are necessary to explore the mechanisms underlying the high heat sensitivity of TRPA1 in infrared-sensing snakes.

In addition, TRPM2 is a gene response to heat that encodes a thermally activated ion channel in somatosensory neurons (electronic supplementary material, table S28) [54]. Noticeably, TRPM2 is also expressed in the TG of pythons and rattlesnakes [20]. The temperature ranges for TRPM2 activation (35–45℃) [75] are close to the body temperature (36.8℃) of a mouse, the most preferred prey for infrared-sensing snakes [76]. Thus, TRPM2 may serve as the second infrared receptor on trigeminal nerve fibres.

Following the transduction of heat into electrical signals, signals are then transmitted along neuron fibres branching from the TG. These fibres are specific to infrared-sensing snakes and are the key components of the infrared vision system. The gene NFIB plays a role in the principal sensory nucleus of the trigeminal nerve and was reported to be diverged in snakes [71]. Our analyses indicate that besides CNEs around NFIB, 5 out of 86 CNEs around NFIA evolved convergently. Both NFIA and NFIB have been implicated to cooperate in late fetal forebrain development [77]. NFIA conditional knockout mice indicate a massive and largely selective absence of retinal amacrine cells (electronic supplementary material, table S28) [78]. Shedding further light on the underlying neural infrastructure, our analyses find evidence of a group of genes related to facial nerve structure, especially trigeminal nerve organization, demonstrating accelerated evolution in infrared-sensing snakes. This includes neuron navigator 2 (Nav2), which was first identified as an RA-responsive gene in human neuroblastoma cells (retinoic acid-induced in neuroblastoma 1, Rainb1), and is required for normal cranial nerve development in adults [79].

Our results indicate that regulatory elements might play important roles in the development of trigeminal nerves among infrared-sensing snakes. Among genes that neighbour convergent CNEs, DCC shows an excess of convergent CNEs in that 4 out of 32 total flanking CNEs are found to have evolved convergently. DCC functions in the process of axon guidance and optic nerve development, especially in the development of retinotectal synaptic connectivity [80]. Axons in DCC-deficient embryos are unable to exit into the optic nerve, causing hypoplasia of the optic nerve (electronic supplementary material, table S28) [81]. The importance of CNEs around DCC was also highlighted by the results of motif annotation that these four CNEs were found to contain a binding site of PBX2, a transcription factor of DCC [82]. This may indicate that DCC has evolved in its regulatory region to function in the genesis of infrared-sensing snake-specific TGs.

After passing through the TG, electrical signals reach the OT in the mid-brain and converge with input from other sensory modalities. The OT contains neurons that could respond to infrared signals and may have evolved modifications to process those signals from the pit organ. Accordingly, a new type of neuron exists at the OT of infrared-sensing snakes [18]. In agreement with this assertion, the functions of three genes (NRP1, NRP2 and Sema3A) with convergent CNEs closely correlate and are active in neural development. Sema3A induces the expression of NRP1 in cortical neurons and NRP1 cooperates with NRP2 to guide cranial neural crest cells and position sensory neurons [69]. Signals of Sema3A that are mediated by NRP1, are also involved in the process of tectal laminar formation in the OT [83]. The GO terms ‘brain development’, ‘toll-like receptor signalling pathway’, ‘peroxisome’ and ‘ECM-receptor interaction’ are enriched significantly in genes containing convergent sites. The toll-like receptor (TLR) signalling pathway is a term with evidence of functional convergence and is significantly over-represented in genes with convergent substitution sites. Besides the roles played in innate immune response, neuronal TLRs also serve to regulate neurite outgrowth and modulate synapse formation [84]. This indicates that TLRs may play an important role in the evolution of infrared vision.

Ethics

This work did not require ethical approval from a human subject or animal welfare committee.

Data accessibility

The whole-genome sequence data of Xenopeltis unicolor have been deposited in the Genome Warehouse in the BIG Data Center, Beijing Institute of Genomics (China National Center for Bioinformation), Chinese Academy of Sciences, under the accession number GWHETGS00000000.1 and BioProject number PRJCA025085. They are publicly accessible at [85]. All these data could also be accessed in GenBank under BioProject number PRJNA1113417. The scripts, orthologue sequences and pipelines used in this study have been deposited in the Dryad Digital Repository [86]. The other data are uploaded as supplementary material [87].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

D.Z.: conceptualization, formal analysis, software, writing—original draft, writing—review and editing; S.H.: writing— original draft, writing—review and editing; S.T.: formal analysis; F.K.K.: writing—review and editing; R.W.M.: writing—review and editing; H.A.D.: writing—review and editing; Y.Z.: conceptualization; P.-S.L.: data curation, formal analysis, project administration, writing—review and editing; J.-M.C.: conceptualization, funding acquisition, project administration, supervision, writing—original draft, 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 the National Natural Science Foundation of China (32200340, 31900323 and 32001222) and supported, in part, by the Beijing Nova Program (Z211100002121022 and 20230484446).