Draft Genome of White-blotched River Stingray Provides Novel Clues for Niche Adaptation and Skeleton Formation

Abstract

The white-blotched river stingray (Potamotrygon leopoldi) is a cartilaginous fish native to the Xingu River, a tributary of the Amazon River system. As a rare freshwater-dwelling cartilaginous fish in the Potamotrygonidae family in which no member has the genome sequencing information available, P. leopoldi provides the evolutionary details in fish phylogeny, niche adaptation, and skeleton formation. In this study, we present its draft genome of 4.11 Gb comprising 16,227 contigs and 13,238 scaffolds, with contig N50 of 3937 kb and scaffold N50 of 5675 kb in size. Our analysis shows that P. leopoldi is a slow-evolving fish that diverged from elephant sharks about 96 million years ago. Moreover, two gene families related to the immune system (immunoglobulin heavy constant delta genes and T-cell receptor alpha/delta variable genes) exhibit expansion in P. leopoldi only. We also identified the Hox gene clusters in P. leopoldi and discovered that seven Hox genes shared by five representative fish species are missing in P. leopoldi. The RNA sequencing data from P. leopoldi and other three fish species demonstrate that fishes have a more diversified tissue expression spectrum when compared to mammals. Our functional studies suggest that lack of the gc gene encoding vitamin D-binding protein in cartilaginous fishes (both P. leopoldi and Callorhinchus milii) could partly explain the absence of hard bone in their endoskeleton. Overall, this genome resource provides new insights into the niche adaptation, body plan, and skeleton formation of P. leopoldi, as well as the genome evolution in cartilaginous fishes.

Introduction

The transition of jawless to jawed vertebrates lays a foundation for the evolution of vertebrates, which was accompanied by many morphological and phenotypic innovations, especially jaws and the adaptive immune system [1]. As ancient jawed vertebrates, the fish constitutes a highly diverse and evolutionarily successful class found in both marine and freshwater habitats [2]. The jawed fishes belong to two clades, the cartilaginous fishes (Chondrichthyes) and bony vertebrates (Osteichthyes), which diverged about 450 million years ago (MYA) [3]. Cartilaginous fishes are the most basal group of living fishes, which contain about 1000 living species [4]. Except for a few published genome information of Chondrichthyes [5], [6], [7], too few genetic data in this important taxonomic position are available for scientists to further study the evolution of chordates and the origin of hard bone formation.

The white-blotched river stingray (Potamotrygon leopoldi), also known as Xingu River ray, is a freshwater cartilaginous fish native to the Xingu River basin in Brazil [8]. The Xingu River is a geographical part of the Amazon River basin, which was inundated by sea during the Pleistocene Epoch. The ancestor of this stingray experienced the transition from marine to freshwater environment. P. leopoldi belongs to the family Potamotrygonidae in the order Myliobatiformes composed of a group of cartilaginous fishes most-closely related to sharks [9]. The species under the family Potamotrygonidae all live in the tropical and subtropical regions of South America [9]. Unlike the freshwater stingrays in Africa, Asia, and Australia, which belong to the family Dasyatidae, most Potamotrygonidae species live strictly in freshwater, whereas most Dasyatidae species are saltwater dwellers [10], [11]. Except a few widespread members, most river stingrays typically reside in and are confined to a single river basin [10]. For its unique appearance (e.g., white spots on black skin) and distinct behavior (e.g., swimming-maneuvering capabilities), P. leopoldi becomes a pricy pet fish popular in home- and office-based aquaria. Till now, no fish species from the family Potamotrygonidae has been extensively studied at the genome level. Whole-genome data of a Potamotrygonidae member and its comparative analysis with other available fish genomes might help us further reveal the evolutionary features unique to Potamotrygonidae and provide insights into the ancestral state of gnathostome-specific morphological characters and physiological systems. Therefore, P. leopoldi provides an excellent model for studying evolution and niche adaptation of freshwater cartilaginous fishes.

In this study, we assembled a 4.11-Gb genome of a male stingray, P. leopoldi, using the whole-genome shotgun (WGS) approach and based on a raw data collection with a total of 370.97× genome coverage, generated from Pacific Biosciences (PacBio) single molecule real time sequencing (SMRT), Illumina HiSeq2000, and 10X Genomics sequencing platforms. We subsequently compared its genome to other five representative fish genomes (Cyprinus carpio, Lepisosteus oculatus, Latimeria chalumnae, Danio rerio, and Callorhinchus milii) and one chordate genome (Branchiostoma floridae), to capture its unique evolutionary features and molecular basis. Our results indicate that P. leopoldi is one of the slowest-evolving fish species, even within the cartilaginous fish lineage. The transcriptomic data, obtained from six tissues of P. leopoldi, shed further lights into highly diversified gene expression profiles among fish lineages, as opposed to the highly coordinated gene expression among mammalian tissues. The knockdown experiment in the fish model reveals the possible genetic foundation for the divergence of hard and cartilaginous skeleton formations. Together, our results start from the P. leopoldi genome sequencing to the experimental model, providing novel clues for niche adaptation and skeleton formation in the evolutionary history of fish.

Results

P. leopoldi with a draft genome assembly of 4.11 Gb

We constructed a total of 17 sequencing libraries using genomic DNA extracted from a male P. leopoldi, and acquired raw data from three sequencing platforms, PacBio SMRT, Illumina HiSeq2000, and 10X Genomics, with coverages of 61.61×, 200.95×, and 108.41×, respectively (Table S1). After stringent filtering and redundancy checking, 1628.63-Gb sequence data were used for a scaffold-based de novo genome assembly. The initial combined assembly was based on the data from PacBio long reads. Illumina paired-end data and 10X Genomics data were used for error correction. The DNA composition of the assembled contigs was of 41.97% GC content (Table S2). The genome size estimated by K-mer analysis using Illumina paired-end data was about 4.11 Gb in size with 0.79% heterozygosity (Table S3). The read-to-genome alignment rate is 98.48% with a coverage of 98.74% in the assembly (Table S4). The final assembly consisted of 16,227 contigs and 13,238 scaffolds, with a contig N50 of 3937 kb and a scaffold N50 of 5675 kb in size (Table S5). The completeness of the P. leopoldi genome was estimated to achieve 91.8% (3081/3354) coverage using the Benchmarking Universal Single-Copy Orthologs (BUSCO) method. We mapped 248 core eukaryotic genes (CEGs) to the scaffolds, and 90% of them are found in the predicted exons, based on BLAST-like alignment tool (BLAT) scores. Additionally, Merqury gave the accuracy in consensus base calling with 99.9% (Q30) for the genome assembly (Table 1). Taken together, the BUSCO results, CEG results, and mapping quality indicate that our genome assembly is highly accurate and nearly complete.

Table 1.

Merqury metrics for drafts of the P. leopoldi genome

Metric	Genome
Consensus quality value	37.53
Assembly error rate (%)	< 0.01
K-mer completeness (%)	96.51

More than 90% of P. leopoldi genes have known functions

The P. leopoldi genome assembly contains more than 71% repetitive content based on de novo and sequence homology analyses (Figure S1; Tables S6 and S7), which is much higher than that of white shark (58.5%) [12]. A total of 23,240 protein-coding genes were predicted with high-confidence by combining de novo and homologous gene prediction methods with the transcriptomic data and orthologs from other fish genomes (Table S8). This gene number seems higher than that of elephant shark but lower than that in white shark [1], [12]. We assigned preliminary functions to 23,030 (99.1%) protein-coding genes using BLASTp against protein databases including Swiss-Prot, Non-Redundant Protein Sequence Database (NR), Kyoto Encyclopedia of Genes and Genomes (KEGG), and InterPro (Figure S2; Table S9), and also assigned Gene Ontology (GO) terms to 21,040 (90.5%) protein-coding genes (Table S9). In addition, the orthologs of these protein-coding genes were also analyzed in other fish species (Table S10). Moreover, non-coding genes including microRNAs (miRNAs), transfer RNAs (tRNAs), ribosomal RNAs (rRNAs), and small nuclearRNAs (snRNAs) are also identified accordingly (Table S11).

P. leopoldi split from bony fish about 381 MYA and has the slowest evolutionary rate among fish species

A phylogenomic tree of P. leopoldi and 24 other fish species was constructed using 212 one-to-one orthologous genes with 48,202 amino acid sites, with B. floridae (a chordate) as the outgroup. A jawless fish, sea lamprey (Petromyzon marinus), was basal to cartilaginous and bony fishes, and there was a conspicuous split between cartilaginous and bony fishes (Figure 1). Both maximum likelihood and Bayesian trees showed exactly the same topology (Figures S3 and S4). P. leopoldi was grouped with C. milii, forming the Chondrichthyes clade, whereas 22 other bony fishes were clustered as the Osteichthyes clade (Figure S5). This result is consistent with the traditional taxonomic classification of fishes. A MCMCtree-based divergence time estimation indicated that the split of the other fish species from the class Cyclostomata (P. marinus) occurred ∼ 533 MYA (Figure 1, Figure S6), and the splits between Chondrichthyes and Osteichthyes and between the superorder Batoidea (P. leopoldi) and Selachimorpha (C. milii) are ∼ 381 MYA and ∼ 96 MYA, respectively.

Figure 1.

Phylogeny of white-blotched river stingray and other 25 selected species. The phylogenomic tree was constructed with FastTree and MRBAYES using 212 one-to-one orthologous genes. B. flo (amphioxus) was used to root the tree. The 25 fishes are P. mar (Petromyzon marinus; sea lamprey), C. mil (Callorhinchus milii; elephant shark), P. leo (Potamotrygon leopoldi; white-blotched river stingray), L. cha (Latimeria chalumnae; coelacanth), L. ocu (Lepisosteus oculatu; spotted gar), I. pun (Ictalurus punctatus; channel catfish), A. mex (Astyanax mexicanus; blind cave fish), C. car (Cyprinus carpio; common carp), D. rer (Danio rerio; zebrafish), E. luc (Esox lucius; northern pike), S. sal (Salmo salar; Atlantic salmon), O. kis (Oncorhynchus kisutch; coho salmon), O. myk (Oncorhynchus mykiss; rainbow trout), G. mor (Gadus morhua; Atlantic cod), G. acu (Gasterosteus aculeatus; three-spined stickleback), T. nig (Tetraodon nigroviridis; green spotted puffer), T. rub (Takifugu rubripes; Japanese puffer), C. sem (Cynoglossus semilaevis; tongue sole), P. oli (Paralichthys olivaceus; Japanese flounder), O. nil (Oreochromis niloticus; Nile tilapia), O. lat (Oryzias latipes; rice fish), N. fur (Nothobranchius furzeri; turquoise killifish), X. mac (Xiphophorus maculatus; platyfish), P. ret (Poecilia reticulate; guppy), and P. for (Poecilia formosa; Amazon molly). Magenta indicates the Chondrichthyes lineage and sky blue indicates the Osteichthyes lineage. Divergence time is shown in million years. MYA, million years ago.

We also calculated the evolutionary rates as total substitution rate per site for each species, using the same set of orthologous genes as in our phylogenomic analysis (Table 2). P. leopoldi had the lowest amino acid substitution rate among 26 species. Tajima’s relative rate tests confirmed that its evolutionary rate was significantly slower than those of B. floridae, P. marinus, L. chalumnae, and D. rerio, but similar to the other cartilaginous fish, such as C. milii (Table S12). However, Tajima’s relative rate tests also showed that L. oculatus exhibited a slower evolutionary rate than P. leopoldi when using amphioxus or sea lamprey as the outgroup. Thus, the results above suggest that P. leopoldi is one of the slowest evolving fishes.

Table 2.

Amino acid substitution rate in 26 fish species

Species	Number of substitutions per site
Amphioxus	0.408705
Sea lamprey	0.277635
Elephant shark	0.252635
White-blotched river stingray	0.250005
Coelacanth	0.265845
Spotted gar	0.250345
Channel catfish	0.338655
Blind cave fish	0.364285
Zebrafish	0.329255
Common carp	0.353285
Northern pike	0.329085
Atlantic salmon	0.321695
Rainbow trout	0.328565
Coho salmon	0.335825
Atlantic cod	0.381235
Three-spined stickleback	0.363065
Green spotted puffer	0.410095
Japanese puffer	0.394355
Nile tilapia	0.358895
Rice fish	0.407615
Turquoise killifish	0.388785
Platyfish	0.388865
Amazon molly	0.397335
Guppy	0.393995
Tongue sole	0.391035
Japanese flounder	0.359955

Specific genes/expanded gene families found in P. leopoldi

To explore the evolutionary features of P. leopoldi, we performed an orthology analysis, by comparing its protein-coding genes with those of B. floridae, C. milii, L. chalumnae, L. oculatus, C. carpio, and D. rerio (Figure 2A). After filtering the genes encoding proteins shorter than 100 amino acids and having low sequence complexity from the predicted 23,240 protein-coding gene pool, we kept a total of 18,894 stingray protein-coding genes for orthology analysis. Among these kept P. leopoldi genes, 12,219, 4347, and 292 of them are chordate orthologs (shared with those of B. floridae), bony fish orthologs (shared with those of 5 other bony fishes), and cartilaginous fish orthologs (shared with those of C. milii), respectively. In addition, 2036 of them are P. leopoldi-specific genes (no homologous relationship with the other six species). Notably, C. carpio has four rounds of genome duplication and possesses a very large number of protein-coding genes (55,756) [13].

Figure 2.

Gene orthology and family analyses for seven species. A. Gene orthology comparison among B. floridae (amphioxus), C. milii (elephant shark), P. leopoldi (white-blotched river stingray), L. chalumnae (coelacanth), L. oculatus (spotted gar), C. carpio (common carp), and D. rerio (zebrafish). B. Expansion and contraction of the IGHD gene family among seven species. C. Expansion and contraction of the TRAV/TRDV gene family among seven species. Number in red indicates the number of expanded genes; number in green indicates the number of contracted genes; number in the parenthesis indicates the total number of expanded and contracted genes. IGHD, immunoglobulin heavy constant delta; TRAV/TRDV, T-cell receptor alpha/delta variable.

Next, we examined gene family expansions and contractions across seven selected species using a total of 205,728 protein-coding genes. According to their homologous relationships, these genes could be classified into 14,818 gene families, and 7830 out of these gene families experienced multiple expansion or contraction events in one or several selected species. A total of 56 gene families were significantly expanded and 14 families were significantly contracted in P. leopoldi only. GO analysis showed that the expanded and contracted gene families in  P. leopoldi had different biological emphases (Figure S7). The expanded gene families were primarily related to the immune system like defense response to virus (Bonferroni corrected P < 0.05), whereas the contracted gene families were more enriched with cellular component such as crystallin and lectin (Bonferroni corrected P < 0.05). Among the P. leopoldi gene families that experienced expansion, two of them, immunoglobulin heavy constant delta (IGHD) gene family and T-cell receptor alpha/delta variable (TRAV/TRDV) gene family, are important constituents of the vertebrate immune system (Figure 2B and C). Interestingly, their expansions were P. leopoldi-specific and not found in elephant shark, suggesting their possible roles in freshwater niche adaption. They were also under purifying selection tested with the branch model in PAML (data not shown).

Eleven Hox genes are missing in P. leopoldi Hox gene clusters

The distinct body shape of P. leopoldi is assumed to have genetic basis, attributable to its Hox gene clusters that exhibit striking spatial collinearity and drive morphologic diversification of almost all metazoans. Due to the contribution of whole-genome duplication events among vertebrates (especially in teleost fishes) and lineage-specific secondary losses, the number of Hox gene clusters or genes varies greatly among vertebrates [14]. As shown in Figure 3, Hox gene clusters range from one in cephalochordate B. floridae to 13 in C. carpio in number [15], [16], and there are 33 Hox genes belonging to four putative Hox clusters (A, B, C, and D) in P. leopoldi within single scaffolds. Therefore, P. leopoldi retains the majority of 2R Hox cluster duplicates. Compared with C. milii, P. leopoldi possesses fewer Hox genes, especially in the HoxC cluster. Totally, P. leopoldi lacks HoxA4, HoxA5, and HoxA6 in the HoxA cluster, HoxB6 in the HoxB cluster, HoxC1, HoxC3, HoxC4, HoxC5, HoxC12, and HoxC13 in the HoxC cluster, and HoxD4 in the HoxD cluster. This difference in Hox gene composition may contribute greatly to stingray’s body shape differences compared with C. milii. Moreover, there are seven Hox genes lost in P. leopoldi but present in all other fishes, including HoxA5, HoxB6, HoxC3, HoxC4, HoxC5, HoxC13, and HoxD4. We thus assume that such Hox gene diversity between P. leopoldi and other fish species may genetically explain its specific body morphology.

Figure 3.

Comparison of Hox gene clusters among seven fish species. The Hox genes were identified from B. floridae (amphioxus), C. milii (elephant shark), P. leopoldi (white-blotched river stingray), L. chalumnae (coelacanth),  L. oculatus (spotted gar), C. carpio (common carp), and D. rerio (zebrafish).

Diversified tissue expression profiles in P. leopoldi

To document its tissue-associated genes and evaluate their possible functions, we acquired RNA sequencing (RNA-seq) data from six P. leopoldi tissues. First, we identified each tissue’s differentially expressed genes (DEGs) by comparing its expression profile with the rest five ones. After normalization based on transcripts per kilobase million (TPM), 3559, 4482, 1806, 2347, 1703, and 2328 DEGs were identified in the blood, brain, heart, liver, muscle, and skin of P. leopoldi, respectively. Among these DEGs, 87, 2281, 184, 537, 388, and 641 genes were up-regulated, whereas 3472, 2201, 1622, 1810, 1315, and 1687 were down-regulated in the blood, brain, heart, liver, muscle, and skin, respectively. Up-regulated DEGs were all tissue-specific, and no shared up-regulated genes were found in the brain, heart, liver, muscle, and skin (Figure 4A). GO analysis showed that each tissue’s up-regulated DEGs were faithful to their tissue’s function (Table 3). In the six tissues of P. leopoldi, down-regulated DEGs shared many overlapping parts among six tissues, and exhibited a similar expression background of antibiotics biosynthesis, catalytic activity, and metabolic process (Figure S8). Compared to C. milii, L. oculatus, and D. rerio, about one third of the up-regulated DEGs in each tissue were P. leopoldi-specific (38/87 in blood, 738/2281 in brain, 65/184 in heart, 180/537 in liver, 132/388 in muscle, and 212/641 in skin), and thus their possible functions remain to be discovered in stingray (Figure 4B).

Figure 4.

Transcriptome analyses ofP.leopoldiand other three fishes. A. Venn diagram flower plot of up-regulated genes in six tissues of P. leopoldi. B. Venn diagram flower plot of up-regulated genes specific in P. leopoldi compared to C. milii, L. oculatus, and D. rerio. C. PCA of the expression levels of 3738 one-to-one orthologous genes in the brain, heart, liver, and muscle of P. leopoldi, C. milii, L. oculatus, and D. rerio. PC, principal component; PCA, principal component analysis.

Table 3.

GO enrichment analysis of up-regulated tissue-specific genes inP. leopoldi

Category	GO term	P value
Blood
GOTERM_BP_DIRECT	Erythrocyte development	6.62 × 10−6
GOTERM_BP_DIRECT	Embryonic hemopoiesis	3.41 × 10−5
GOTERM_BP_DIRECT	Erythrocyte differentiation	6.48 × 10−5
GOTERM_BP_DIRECT	Hemopoiesis	7.09 × 10−4
Brain
GOTERM_BP_DIRECT	Central nervous system development	8.02 × 10−4
INTERPRO	Neurotransmitter-gated ion-channel	0.001797
GOTERM_BP_DIRECT	Ganglioside biosynthetic process	0.002529
GOTERM_BP_DIRECT	Neuropeptide signaling pathway	0.012035
GOTERM_BP_DIRECT	Neural crest cell migration	0.016101
GOTERM_BP_DIRECT	Hindbrain development	0.027042
Heart
GOTERM_BP_DIRECT	Cardiac muscle cell differentiation	2.46 × 10−5
GOTERM_BP_DIRECT	Heart morphogenesis	6.17 × 10−4
GOTERM_BP_DIRECT	Heart looping	9.05 × 10−4
GOTERM_BP_DIRECT	Heart development	0.005143
Liver
GOTERM_BP_DIRECT	Liver development	7.39 × 10−4
GOTERM_MF_DIRECT	Oxidoreductase activity	1.54 × 10−11
KEGG_PATHWAY	Glycine, serine and threonine metabolism	0.007413
KEGG_PATHWAY	Fatty acid degradation	8.55 × 10−5
Muscle
GOTERM_BP_DIRECT	Muscle organ development	1.76 × 10−5
GOTERM_BP_DIRECT	Skeletal muscle tissue development	7.32 × 10−5
GOTERM_BP_DIRECT	Myofibril assembly	2.78 × 10−4
INTERPRO	Myogenic basic muscle-specific protein	4.65 × 10−4
GOTERM_BP_DIRECT	Skeletal muscle cell differentiation	0.00223
UP_KEYWORDS	Myogenesis	0.003216
Skin
GOTERM_BP_DIRECT	Ectodermal placode formation	0.068466
GOTERM_CC_DIRECT	Melanosome	0.015904

Next, we compared the expression profiles of 3738 one-to-one orthologs in four tissues (brain, heart, liver, and muscle) between P. leopoldi, C. milii, L. oculatus, and D. rerio, and performed the principal component analysis (PCA) to investigate the expression patterns for the four tissues across the four fishes. The PCA result showed a scattered expression pattern for each of the four tissues across four fishes (Figure 4C). Our data were separated neither by tissue nor by species. In mammals, expression analyses have shown that the same tissues from different species tend to cluster together [17], proposing that the same tissues in different mammals usually perform similar physiological functions. In fishes, the diverged expression pattern of the same tissues suggests a probably much more diversified physiology or a much longer evolutionary history, both of which are not mutually exclusive.

P. leopoldi lacks the gc gene for hard skeleton formation

As a cartilaginous fish, P. leopoldi already has a complex skeleton structure providing support for its body and internal organs. How hard skeleton emerged in vertebrates and what is the genetic basis for bone formation remain to be investigated. We systematically compared the bone formation-related gene families between cartilaginous and bony fish genomes. First, we examined the bone morphogenetic protein (BMP) gene family and BMP receptor (BMPR) genes among seven selected species. It is known that both BMPs and their receptors play an essential role in skeleton formation. Generally, BMP genes are classified into eight different clusters according to their phylogenetic relationships (Figure 5A). Cartilaginous and bony fishes have the representative genes in all eight clusters. P. leopoldi does not have the bmp15 gene, whereas C. milii has. Compared with fish species, B. floridae does not have the bmp3, bmp9, bmp10, bmp15, and bmp16 genes. As to BMPR genes [18], both types I and II are present in both cartilaginous and bony fishes, but B. floridae misses BMPR type II (Table S13). Our analyses suggest that BMPs and their receptors are less likely to answer the question whether the skeleton is made of cartilage or hard bone, because both cartilaginous and bony fishes have all representative BMP and BMPR genes. Second, we examined the presence or absence of other bone formation-related genes between cartilaginous and bony fishes. After scrutinizing the candidate gene list involved in skeleton formation from six fish species, we found that the gc gene, encoding vitamin D-binding protein (VDBP), is present in bony fishes but absent in cartilaginous fishes. L. chalumnae, L. oculatus, C. carpio, and D. rerio have 2, 1, 3, and 1 copies of gc, respectively. We hypothesize that VDBP may be involved in the bone formation process for bony fishes. D. rerio is a widely used model organism for studies of bone development and formation [19]. To verify the possible function of the gc gene in bone formation, we designed two small guide RNAs (sgRNAs), gc-e4 and gc-e8, against the exon 4 and exon 8 of the gc gene, respectively [20], using  a sgRNA against the enhanced green fluorescent protein (EGFP) gene as a scrambled control. As shown in Figure 5B and C, embryos injected with control ribonucleoproteins (RNPs) displayed normal bone formation, whereas embryos injected with RNPs against the gc gene displayed incomplete craniofacial skeleton mineralization. The results support our hypothesis that the gc gene is involved in hard skeleton formation.

Figure 5.

Analyses of bone formation-related genes among seven selected species. A. Phylogenetic analysis of BMP gene family among seven selected species. B. Alizarin Red S staining of 6-dpf control zebrafish. WT represents embryos without injection; EGFP represents embryos injected with RNPs against EGFP. C. Alizarin Red S staining of 6-dpf gc-knockdown zebrafish. gc-e4 represents embryos injected with RNPs against exon 4 of the gc gene; gc-e8 represents embryos injected with RNPs against exon 8 of the gc gene. WT, wild-type; dpf, days post-fertilization; BMP, bone morphogenetic protein; RNP, ribonucleoprotein; EGFP, enhanced green fluorescent protein.

Discussion

The evolutionary features of the genome and Hox gene clusters of P. leopoldi

The genomic data of P. leopoldi provide us an information bonanza for understanding fish evolution, especially the split between cartilaginous and bony fishes. Our phylogenomic analysis showed that Chondrichthyes and Osteichthyes were two parallel monophyletic groups and the split between them can be dated back to 381 MYA, which is less than the estimation of 450 MYA based on mitochondrial genome [3]. The discrepancy between our result and the mitogenomic estimation could be caused by the different evolutionary rates of nuclear genome and mitochondrial genome. Mitochondrion has a particularly high mutation rate and a much diversified evolutionary spectrum across species [21]. The non-unified estimations of split time can be explained by different data sources. Generally, the evolutionary rate of elasmobranchs is much lower than that in mammals [22], [23], and the Chondrichthyes also shows a slower evolutionary rate than the Osteichthyes. In this study, P. leopoldi, C. milii, and two basal bony fish species (L. chalumnae and L. oculatus) evolved much slower than the other fish species (Figure 1). These four fish species also did not experience the third round of genome duplication (3R) which happened in the ray-finned fish lineage [24], [25]. This result supports the hypothesis that the 3R might promote the teleost in a higher rate of sequence evolution [1], [26], [27], and suggests that cartilaginous and ancient bony fishes might have been well adapted to their niches. In P. leopoldi, C. milii, L. chalumnae, and L. oculatu, the substratum (the number of genes) for evolution to work on was limited and therefore slow evolution is expected. Although P. leopoldi genome was not assembled at the chromosomal level, the completeness of the genome has achieved 91.8% coverage estimated by the BUSCO methoed. P. leopoldi has four Hox gene clusters containing 33 genes, the smallest number of Hox genes among six fish species analyzed in our study (Figure 3). C. milii, L. chalumnae, L. oculatus, C. carpio, and D. rerio have 45, 42, 34, 62, and 49 Hox genes, respectively. Moreover, P. leopoldi lacks HoxA5, HoxB6, HoxC3, HoxC4, HoxC5, HoxC13, and HoxD4 presented in the other five fish species. P. leopoldi has a disc-like body, whereas most fish species usually have a streamline body. The Hox gene difference between P. leopoldi and the other fish species may help to explain why P. leopoldi has a simple body design with large pectoral fins and a whip-like tail. However, our study does not provide the answer for which Hox gene contributes to what morphological alteration in stingray’s body plan.

The expansion of two immune-related gene families in P. leopoldi

Our gene family analysis showed that 56 gene families were significantly expanded in P. leopoldi. Many of them encode protiens involved in immune and stress response, such as heat shock proteins (HSPs), cytochrome P450 (CYP450), and inhibitors of apoptosis (IAPs). Two expanded gene families, IGHD and TRAV/TRDV, were found to be directly related to the immune system (Figure 2). Immunoglobulins are membrane-bound or secreted glycoproteins produced by B lymphocytes. Immunoglobulin D (IgD) is made up of two heavy chains of the delta class, and two light chains. IgD is present in species ranging from fish to mammal, suggests that IgD has important immunological functions. IgD has been reported to be able to bind to basophils and mast cells and induce antimicrobial factors, which contributes to immune surveillance and inflammation under pathological conditions [28]. T-cell receptor alpha (TCRα) recognizes peptides that are bound to major histocompatibility complex (MHC) molecules, and TCR delta (TCRδ) recognizes antigens directly, both of which are considered as a bridge between the innate and adaptive immune system. TCRα and TCRδ cDNA sequences have been identified in the nurse shark [29], as well as in other vertebrates. The existence of TCRs in sharks suggests that the adaptive immune system evolved in cartilaginous fish has the same fundamental major components as those existing in vertebrates such as humans [30]. Our studies of TCR expansion in P. leopoldi help elucidate that the V genes of TCRs have evolved for 500 million years, and indicate their role in the diversification of the pre-immune repertoire. The natural habitat of P. leopoldi is Xingu River basin, and the phylogenetic relationship of stingrays shows that the family Potamotrygonidae evolved from the sea-dwelling ancestors [31]. The fact that stingrays can transit from marine environment to pathogen rich freshwater suggests that stingrays should have a complex immune system. The expansion of immune system-related genes (Figure 2) suggests that P. leopoldi could require specific immune responses for pathogens and antigens in new niches, although there has been no evidence that these two gene families are directly involved in freshwater habitat adaptation. Usually, genes related to immune response are evolutionarily active to coup with the constant-evolving pathogens [32]. When a gene family expanded through duplication event, some duplicated genes would be free from their function constraint. The old genes performed the normal function, and the new one would go through the neo-functionalization or sub-functionalization process [33]. Thus, their selection pressure tends to be positive, as well. However, our selective pressure analysis showed that these two gene families were under negative selection, a sign of function constraint. It indicates that their functions were rapidly fixed after the gene family expansion. One possible explanation for this phenomenon is that P. leopoldi fast adapted to its new niche in Xingu River and its immune system had to rapidly deal with new enemy after its ancestor migrated from the Atlantic Ocean to the Amazon River system. Furthermore, the fact that P. leopoldi no longer migrated to the other niches renders the functions of these expanded genes important for its survival in order to deal with fixed microbial enemy. Thus, these two expanded gene families were negatively selected. Surely, the argument mentioned above is our hypothesis and needs to be verified in the other fish species whose ancestor migrated to river from ocean.

The gene expression silhouette of P. leopoldi tissues

Expression profile comparison among six P. leopoldi tissues demonstrates that its brain has a much wider expression spectrum than the other tissues (Figure 4), and the result is consistent with that of mammalian organs. In mammals, the central nervous system has more specifically expressed genes than the heart and liver [34]. Compared with mammals, there are more species-specific genes expressed in P. leopoldi in each of the analyzed tissues [17]. The brain-specific modules are enriched with genes involved in typical processes as for central nervous system development (Benjamini–Hochberg corrected P < 0.05), and thus define common neural tissue functions. Furthermore, the same tissues from different fish species exhibited a much more diversified expression pattern than that observed in mammals. The earliest mammal appeared around 225 MYA, and our analysis showed that fishes emerged at least 533 MYA [35]. As fishes have a much longer evolutionary history than mammals, a greater divergence for fishes at both expression and sequence levels is expected. We are only able to identify 3738 one-to-one orthologous genes among white-blotched river stingray, elephant shark, coelacanth, and zebrafish, which is much less than 5636 amniotic one-to-one orthologous genes identified among nine endothermic species in a previous study [17]. Therefore, the long fish evolutionary history may explain the non-unified expression profiles of their tissues at least in part. Our analysis of tissue transcriptomes from all representative fish lineages refines previous hypotheses and provides a new viewpoint for the evolution of chordate tissue functions.

Absence of VDBP in P. leopoldi

Skeleton is an essential part of vertebrate body, which can be made of hard bone or cartilage only. For bone, two cell types, osteoblast and osteocyte, contribute to the formation and mineralization, but cartilage has only one cell type called chondrocyte [18], [19]. Therefore, the emergence of hard skeleton in vertebrates must have engaged complex cellular processes and multiple genetic inventions. We searched the P. leopoldi gene inventory for the ones known to be involved in bone formation in osteichthyans. All gene families involved in skeleton formation seem to be present, except the gc gene which encodes VDBP. Our analyses of BMP and BMPR genes show that both cartilaginous and bony fishes have all representative members of these two gene families (Figure 5A; Table S13). The slight difference in BMP and BMPR genes between the two fish lineages is insufficient to explain the major departures in their skeletons. One major difference between bones and cartilages is whether calcium phosphate is present in their extracellular matrix or not. A previous study has reported that secreted calcium-binding phosphoproteins (SCPP) are involved in the bone formation in D. rerio [1]. The analysis of C. milii genome has reported the lack of genes encoding SCPP in cartilaginous fishes, which explains the absence of bone in their endoskeleton [1]. Our study further demonstrates that VDBP may also take part in the bone formation process (Figure 5B and C). Vitamin D promotes the absorption of calcium through the intestines [36]. Together, our result and that of a previous study both suggest that genes responsible for calcium metabolism are essential for the hard skeleton formation in bony fishes. Because cartilaginous and bony fishes evolved in parallel, several crucial genes alone may sufficiently exclude the hard skeleton from cartilaginous fishes, albeit under ongoing studies.

Conclusion

In this study, we report the assembly and analysis of a draft genome of P. leopoldi, a cartilaginous freshwater fish. Because cartilaginous fishes constitute a critical outgroup for understanding the evolution and diversity of bony vertebrates, the whole-genome analysis shows that the P. leopoldi genome is evolving significantly slower than other vertebrates. The transcriptomic data shed lights into highly diversified gene expression profiles among fish lineages, as opposed to the coordinated gene expression among mammalian tissues. The expansion of immune-related gene families IGHD and TRAV/TRDV suggests that the diversification of the pre-immune repertoire in cartilaginous fishes could play a role in the evolution of an adaptive immune system. Our study further demonstrates that VDBP may partly explain the absence of hard bone in their endoskeleton. Together, our results starting from the P. leopoldi genome to the experimental model provide novel clues for niche adaptation and skeleton formation in the evolutionary history of fishes.

Materials and methods

P. leopoldi sample

A mature male P. leopoldi individual was acquired from an aquarium in China in January, 2018. It was the descendant of captive P. leopoldi breeding population. The fish was killed in a humane way, and the experimental procedure was performed in accordance with the guidelines of the Animal Care Committee at the Institute of Neuroscience, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences. Its skin, heart, blood, muscle, liver, and brain were used for DNA and RNA preparation and sequencing library construction.

Genome sequencing and assembly

The genomic DNA of P. leopoldi was sequenced by WGS strategy. Based on the genome features, three different lengths (230 bp, 350 bp, and 450 bp) of DNA inserts were produced. The Illumina HiSeq2000 platform (San Diageo, CA) was used to sequence these reads by the paired-end sequencing method with the read length of 150 bp in order to capture the whole-genome data. A total of 17 DNA libraries were constructed, and the total amount of sequencing data was 882.2 Gb with a coverage of 200.95×. The PacBio SMRT platform yielding an average read length of 20 kb (Menlo Park, CA) was also used to generate 270.51-Gb data, equivalent to a genome coverage of 61.61×. Additionally, a 10X Genomics library was constructed, coupled with the Illumina sequencing platform in a read length of 150 bp, yielding 475.93-Gb data, equivalent to a genome coverage of 108.41×. PacBio long reads were utilized to perform de novo assembly. Around 31 million subreads were used for the assembly with FALCON (v0.3.0) to generate contigs [37]. Primary contigs were polished using Quiver5. The scaffolds were built based on 10X Genomics data. Sequence data were generated using the 10X Genomics GemCode platform (Pleasanton, CA), and the error-corrected contigs were used as input for scaffolding to obtain the primary assembly. After scaffolding, shotgun sequences were used to close gaps between contigs. Paired-end clean reads from the Illumina platform were aligned to the assembly with BWA [38]. Contigs or scaffolds shorter than 10 kb were excluded from the analysis to avoid spurious misassembly. Gaps in contigs and scaffolds were closed with subreads. To survey the characteristics of the genome, a total of ∼ 140-Gb next-generation sequencing data equivalent to a genome coverage of 33× were generated. Adaptor sequences, polymerase chain reaction (PCR) duplicates, and low-quality sequences were removed from the raw data to generate high-quality sequences. K-mer statistics of the high-quality sequences were calculated by Jellyfish (v2.2.7) with the parameters of “-G 2 -m 17” [39]. GenomeScope 2.0 (https://github.com/tbenavi1/genomescope2.0) [40] was used to estimate the size, heterozygosity rate, and repeat content of the P. leopoldi genome. Finally, the completeness of the assembly was assessed through BUSCO analysis (v5.2.1; https://busco.ezlab.org/) and CEG analysis (https://korflab.ucdavis.edu/Datasets/cegma/).

Genome annotation

Repeat elements were annotated with both homology annotation and de novo prediction. RepeatMasker and RepeatProteinMask (https://www.repeatmasker.org/RepeatProteinMask.html) were used to search the assembled genome against Repbase for known repeat elements [41], [42]. LTR-FINDER and RepeatModeler (https://www.repeatmasker.org/RepeatModeler.html) were used to de novo develop repeat element library [43]. After the library was established, RepeatMasker was further used to detect species-specific repeat elements. Tandem repeats were also searched with Tandem Repeats Finder in the assembled P. leopoldi genome [44]. Overlapping transposable elements belonging to the same type of repeats were integrated together.

Protein-coding genes were predicted through combination of de novo annotation, homology annotation, and transcriptome-based annotation. AUGUSTUS, GlimmerHMM, SNAP (https://github.com/KorfLab/SNAP), Geneid, and GENSCAN (https://hollywood.mit.edu/GENSCAN.html) software packages were used to de novo predict protein-coding genes in the P. leopoldi genome [45], [46], [47]. For homology annotation, the protein sequences from Japanese puffer (Takifugu rubripes), rice fish (Oryzias latipes), Nile tilapia (Oreochromis niloticus), Atlantic cod (Gadus morhua), elephant shark (C. milii), green spotted puffer (Tetraodon nigroviridis), zebrafish (D. rerio), amphioxus (B. floridae), three-spined stickleback (Gasterosteus aculeatus), and coelacanth (L. chalumnae) were used to search the homologous genes in P. leopoldi genome with BLAST and GeneWise [48], [49]. For transcriptome-based annotation, the RNA-seq reads from P. leopoldi skin, heart, blood, muscle, liver, and brain were mapped and assembled with PASA and Cufflinks [50], [51]. EVidenceModeler (EVM) was employed to integrate the gene sets from three annotation methods into a complete and non-redundant gene set [52]. Finally, PASA was used to correct the EVM annotation result with untranslated region (UTR) and alternative splicing information.

Function annotation was performed through comparing the annotated protein-coding genes with the known protein banks. The final gene set was blasted against four common protein banks, Swiss-Prot (https://www.uniprot.org/), NR (https://www.ncbi.nlm.nih.gov/protein), KEGG (https://www.genome.jp/kegg/), and InterPro (https://www.ebi.ac.uk/interpro/). InterProScan was used to integrate the functional results from four protein banks [53].

The tRNA genes were identified by tRNAscan-SE software with eukaryote parameters [54]. The rRNA fragments were predicted by aligning to whale shark and C. milii template rRNA sequences using BLASTN at E-value of 1 × 10^–10 [55]. The miRNA and snRNA genes were predicted using Infernal software by searching against the Rfam database (release 9.1) [56].

Orthology analysis

We first compiled the complete proteomes of 25 fish and one chordate genomes. The proteome data of 13 selected organisms, including blind cave fish (Astyanax mexicanus), zebrafish (D. rerio), Atlantic cod (G. morhua), three-spined stickleback (G. aculeatus), coelacanth (L. chalumnae), spotted gar (L. oculatus), Nile tilapia (O. niloticus), rice fish (O. latipes), sea lamprey (P. marinus), Amazon molly (Poecilia formosa), Japanese puffer (T. rubripes), green spotted puffer (T. nigroviridis), and platyfish (Xiphophorus maculatus), were obtained from the Ensembl database (release 83, https://www.ensembl.org/). The proteome data of other 11 selected organisms, including elephant shark (C. milii), tongue sole (Cynoglossus semilaevis), common carp (C. carpio), northern pike (Esox lucius), channel catfish (Ictalurus punctatus), turquoise killifish (Nothobranchius furzeri), coho salmon (Oncorhynchus kisutch), rainbow trout (Oncorhynchus mykiss), Japanese flounder (Paralichthys olivaceus), guppy (Poecilia reticulate), and Atlantic salmon (Salmo salar), were downloaded from the National Center for Biotechnology Information (NCBI) database (https://www.ncbi.nlm.nih.gov/). The proteome of B. floridae was download from JGI Genome Portal (https://genome.jgi.doe.gov/portal/). Proteins shorter than 100 amino acids were discarded and, for alternatively spliced genes, only the longest splice variant of each gene was retained.

Orthologous protein groups were determined by OrthoFinder (v2.3.3) with blast search and default parameters [57]. This procedure led to 2792 orthologous groups with at least one representative protein from aforementioned 25 species plus P. leopoldi. To maximize orthology, these orthologous groups were filtered with an in-house Perl script to provide a subset group that contained strict one-to-one orthologous proteins from each species. We eventually retained 212 one-to-one orthologous groups for phylogenomic analysis.

Phylogenomic analysis and estimation of divergence time

Protein sequences in each of the 212 one-to-one orthologous groups were aligned using MUSCLE and ClustalW, and the resulting alignments were combined by M-Coffee to produce the multiple sequence alignments (MSA) [58], [59], [60]. We used an in-house Perl script to remove gaps, and the final MSA contained 48,202 amino acid sites. FastTree 2.1 was used to construct the maximum likelihood tree for the final MSA with Jones–Taylor–Thornton (JTT) and category mixture model (CAT) models [61]. MRBAYES 3.2.6 was used to construct the Bayesian tree for the final MSA with JTT and invgamma models [62]. We ran the Markov chain Monte Carlo (MCMC) algorithm for 500,000 generations with 4 chains. Bayesian trees were sampled every 100 generations, and the first 25% of trees were excluded from the analysis as burn-in. The Bayesian tree was summarized after the average standard deviation of split frequencies below 0.01. RAxML was used to constructed a maximum likelihood tree using the JTT model with gamma distribution [63]. Then, MCMCtree was used to predict the divergence time of 26 species [64]. The intervals of the divergence time between different species were obtained from the TimeTree database [65].

Identification of Hox gene clusters

Forty-nine unique Hox genes obtained from a previous study were used as queries to conduct BLAST search (threshold of E-value 1 × 10⁻²⁰) against seven species, including P. leopoldi, C. carpio, L. oculatus, L. chalumnae, D. rerio, C. milii, and B. floridae, respectively [16], [48]. The Hidden Markov mode (HMM) profile for the homeodomain was used to identify the potential homeodomain containing genes from these genomes with HMMER 3.2.1 (https://hmmer.janelia.org/) [66], as well. All of the obtained genes were further validated using SMART database to determine whether the protein sequences contain homeodomains [67].

Gene family expansion and contraction estimation and selective pressure analysis

Gene family expansion and contraction were analyzed by CAFE (v3.1) using the same seven species in the identification of Hox gene clusters [68]. After filtering the genes encoding proteins shorter than 100 amino acids and with low sequence complexity, we collected a total of 205,728 genes from seven selected species. OrthoFinder (v2.3.3) was used to classify these genes into orthologous groups based on their sequence similarity. Each orthologous group is actually a gene family. We calculated the probability of each orthologous group by 10,000 Monte Carlo random samplings and estimated the lambda value based on the maximum likelihood model, which represents the rate of expansion and contraction of each gene family. A branch with P < 0.05 was considered to have gene amplification and contraction over evolutionary time scales.

PAML was used to estimate the selective pressure of selected genes. Both branch and branch-site mode were applied to detected the selective pressure in P. leopoldi [64]. PAL2NAL was utilized for alignment nucleotide sequences based on protein alignment [69]. The likelihood ratio tests (LRTs) of M1a vs. M2a were employed to examine the selective pressure of each site among selected genes.

Transcriptome analysis

Total RNA from six P. leopoldi tissues (skin, heart, blood, muscle, liver, and brain) was prepared using the Qiagen RNeasy Kit (Catalog No. 75142, Qiagen, Düsseldorf, Germany) according to the manufacturer’s instructions, and RNA-seq libraries were constructed according to a standard protocol for the Illumina novaseq6000 sequencing platform (San Diageo, CA) with 100-bp paired-end reads. The reads were aligned onto the assembled P. leopoldi reference genome with STAR [70].

Other RNA-seq data of the brain, heart, muscle, and liver from three fish species, L. oculatus, D. rerio, and C. milii, were retrieved from the NBCI Sequence Read Archive (SRA). The raw RNA-seq data were filtered using Trimmomatic 0.32 to generate clean reads [71]. Per base sequence qualities of filtered fastq files were checked with FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). The Ensembl genome of each species was used as a reference genome, and filtered reads were aligned onto the references using STAR [70].

RSEM was used to quantify expressed genes into TPM values [72]. R package DEGseq was used to detect DEGs [73]. One-to-one orthologous genes between P. leopoldi and D. rerio and web-based DAVID Bioinformatics Resources were used for GO annotation [74].

Examination of bone formation-related gene families

The BMP and BMPR genes of zebrafish were used as query to search the BMP and BMPR gene families in P. leopoldi, C. carpio, L. oculatus, L. chalumnae, C. milii, and B. floridae. Other bone formation-related gene families were examined as follows. The zebrafish orthologous genes were used to annotate each gene family. The gene family with the keyword of “bone”, “calcium”, or “vitamin D” was kept as the bone formation-related candidate gene family for further examination. These bone formation-related candidate gene families were manually checked with literature evidence in order to find the target genes for knockdown experiment.

gc gene knockdown in zebrafish

D. rerio (AB strain) was provided by an in-house D. rerio Core Facility (CAS Center for Excellence in Molecular Cell Science), and all experimental protocols were approved by the Institutional Animal Care and Use Committee. Two sgRNAs (gc-e4: 5′-GCTCAATGCCTGGATGCTTGGT-3′; gc-e8: 5′-TCGGTTTGGATTCATCGCAGGT-3′) were designed to target the sequences in the exon 4 and exon 8 of the gc gene in zebrafish, respectively, based on the Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) design website CCTop (https://crispr.cos.uni-heidelberg.de/index.html) and CHOPCHOP (https://chopchop.cbu.uib.no/). DNA templates for sgRNAs were produced by annealing and elongating a forward primer containing T7 promoter, guide sequence, and a reverse primer encoding the standard chimeric sgRNA scaffold [75] (Table S14). DNA templates were purified, and sgRNAs were in vitro synthesized and purified. Cas9 RNP complexes were prepared with Cas9 protein and sgRNAs as previously described [76]. The RNPs were injected into one-cell-stage D. rerio embryos. Each embryo was injected with a 1-nl mix containing ∼ 5 μM Cas9 and 1 μg/μl (31 μM) sgRNA.

D. rerio larvae were processed for bone staining with Alizarin Red S using a modified protocol [77]. Briefly, larvae of 6 days post-fertilization (dpf) were fixed in 4% paraformaldehyde (w/v, pH 7.4) overnight at 4 °C, washed in 1% KOH for 5 min, and bleached in 3% H₂O₂/0.5% KOH for 40 min. All specimens were initially stained with 0.05% Alizarin Red S in 70% ethanol overnight, and soaked thoroughly with 25% glycerol/0.1% KOH, 50% glycerol/0.1% KOH, and 75% glycerol/0.1% KOH sequentially. All specimens were stored in 80% glycerol/H₂O.

Data availability

The raw sequencing data reported in this study have been deposited in the Genome Sequence Archive [78] at the National Genomics Data Center (NGDC), Beijing Institute of Genomics (BIG), Chinese Academy of Sciences (CAS) / China National Center for Bioinformation (CNCB) (GSA: CRA003264) , and are publicly accessible at https://ngdc.cncb.ac.cn/gsa. The whole-genome sequence data and its annotation file in GFF format in this study have been deposited in the Genome Warehouse [79] at the NGDC, BIG, CAS / CNCB (GWH: GWHAOTN00000000), and are publicly accessible at https://ngdc.cncb.ac.cn/gwh.

Competing interests

Xiang Zhang is the CEO of Shanghai Nanmulin Biotechnology Company Limited and pays for the sequencing of P. leopoldi. All the other authors have declared no competing interests.

CRediT authorship contribution statement

Jingqi Zhou: Methodology, Formal analysis, Visualization, Funding acquisition, Writing – original draft, Writing – review & editing. Ake Liu: Methodology, Formal analysis, Visualization, Data curation. Funan He: Methodology, Formal analysis, Visualization. Yunbin Zhang: Formal analysis, Validation. Libing Shen: Conceptualization, Writing – original draft, Writing – review & editing, Supervision. Jun Yu: Conceptualization, Writing – review & editing, Supervision. Xiang Zhang: Conceptualization, Resources, Funding acquisition, Writing – review & editing, Supervision. All authors have read and approved the final manuscript.

Handled by Kai Ye

Supplementary material

The following are the Supplementary data to this article:

Supplementary Figure S1.

The distribution of tandem repeat sequences in white-blotched river stingray genome X-axis is the divergence of stingray’s tandem repeat sequences to the TE sequences in the Repbase database. SINE, short interspersed element; LINE, long interspersed element; LTR, long terminal repeat; TE, transposon element.

Supplementary Figure S2.

Venn diagram of the function annotation for white-blotched river stingray’s protein-coding genes using four protein databases NR, Non-Redundant Protein Sequence Database; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Supplementary Figure S3.

The maximum likelihood tree of white-blotched river stingray and 25 selected species with statistical support values AMX for blind cave fish. DAR for zebrafish. GMO for Atlantic cod. GAC for three-spined stickleback. LAC for coelacanth. LOC for spotted gar. ONI for Nile tilapia. ORL for rice fish. PMA for sea lamprey. PFO for Amazon molly. TRU for Japanese puffer. TNI for green spotted puffer. XMA for platyfish. CMI for elephant shark. CSE for tongue sole. CCA for common carp. ELU for northern pike. IPU for channel catfish. NFU for turquoise killifish. OKI for coho salmon. OMY for rainbow trout. POL for Japanese flounder. PRE for guppy. SSA for Atlantic salmon. BRF for amphioxus. PLE for white-blotched river stingray. Branch length is proportional to the substitution rate.

Supplementary Figure S4.

The Bayesian tree of white-blotched river stingray and 25 selected species with posterior probability support AMX for blind cave fish. DAR for zebrafish. GMO for Atlantic cod. GAC for three-spined stickleback. LAC for coelacanth. LOC for spotted gar. ONI for Nile tilapia. ORL for rice fish. PMA for sea lamprey. PFO for Amazon molly. TRU for Japanese puffer. TNI for green spotted puffer. XMA for platyfish. CMI for elephant shark. CSE for tongue sole. CCA for common carp. ELU for northern pike. IPU for channel catfish. NFU for turquoise killifish. OKI for coho salmon. OMY for rainbow trout. POL for Japanese flounder. PRE for guppy. SSA for Atlantic salmon. BRF for amphioxus. PLE for white-blotched river stingray. Branch length is proportional to substitution rate.

Supplementary Figure S5.

The species tree of 25 selected fishes and one chordate

Supplementary Figure S6.

The MCMCtree for estimating the divergence time of white-blotched river stingray and 25 selected species AMX for blind cave fish. DAR for zebrafish. GMO for Atlantic cod. GAC for three-spined stickleback. LAC for coelacanth. LOC for spotted gar. ONI for Nile tilapia. ORL for rice fish. PMA for sea lamprey. PFO for Amazon molly. TRU for Japanese puffer. TNI for green spotted puffer. XMA for platyfish. CMI for elephant shark. CSE for tongue sole. CCA for common carp. ELU for northern pike. IPU for channel catfish. NFU for turquoise killifish. OKI for coho salmon. OMY for rainbow trout. POL for Japanese flounder. PRE for guppy. SSA for Atlantic salmon. BRF for amphioxus. PLE for white-blotched river stingray. The numbers adjacent to the nodes are divergence time which is shown in million years.

Supplementary Figure S7.

GO enrichment analyses for expanded and contracted gene families identified in white-blotched river stingray GO, Gene Ontology.

Supplementary Figure S8.

Upset plot of down-regulated genes in six tissues of white-blotched river stingray

Supplementary Table S1

Statistics of genome sequencing data of white-blotched river stingray

Supplementary Table S2

Base content statistics of white-blotched river stingray genome

Supplementary Table S3

Characteristic statistics of genome assembly

Supplementary Table S4

Reads coverage statistics of white-blotched river stingray genome

Supplementary Table S5

Genome assembly results of white-blotched river stingray

Supplementary Table S6

Statistical results of repeated sequences in white-blotched river stingray genome

Supplementary Table S7

Statistics of repeated sequence classification in white-blotched river stingray genome

Supplementary Table S8

Basic statistical information for gene structure prediction

Supplementary Table S9

The gene function annotation using different methods

Supplementary Table S10

Gene structures in eleven fish species

Supplementary Table S11

Statistics of non-coding RNAs in white-blotched river stingray genome

Supplementary Table S12

Evolutionary rate evaluated by Tajima’s rate tests among 26 species

Supplementary Table S13

The number of BMPs and BMP receptors in seven species

Supplementary Table S14

Primer sequences used to generate sgRNAs used in this study