Background

As one of the most successful vertebrate colonizers in subterranean habitats, cavefishes attract interest because of their unusual regressive features, such as the rudimentary eyes and scales, and loss of pigmentation. As possible compensation, some constructive traits have evolved, such as elongated appendages and non-visual sensory systems [1, 2]. Nevertheless, biologists have long puzzled over these troglomorphic characters [3], and the study of their relationship to their environment has been largely ignored [4]. Although recent work on a traditional model, Astyanax mexicanus, revealed some important advances [3], especially in aspects of eye loss [5] and identifying candidate genes underlying quantitative trait loci (QTL) [6], there is no whole genomic data available to unravel the evolution and general adaptation to dark subterranean life among other groups of distinct but closely-related cavefishes.

Sinocyclocheilus (Cypriniformes: Cyprinidae) is endemic to China’s massive southwestern karst area, the northern part of which abuts the eastern part of the Qinghai-Tibetan Plateau. This genus serves as an emerging cavefish model for its high species diversity and the phenotypic variation which has evolved in karst river systems. In a previous Science letter in 2013 [7], we reported a global hotspot of biodiversity and the threats to it through a survey of cave species in Southwestern China, with an illustration of a Sinocyclocheilus cavefish, S. rhinocerous, which was collected from Yunnan province. The present study deals with whole genome and transcriptome sequencing of three Sinocyclocheilus species (Fig. 1): the surface-dwelling S. grahami (Sg); the semi-cave-dwelling S. rhinocerous (Sr); and the cave-restricted S. anshuiensis (Sa). These species were carefully chosen as representatives of three key nodes on the path to obligate cave life, although almost all Sinocyclocheilus species show some cave-related traits and habits [8]. We investigate whether whole genomic data would provide some clues about genetic adaptations to subterranean habitats, and whether the strict co-occurrence of the karstic landscape and Sinocyclocheilus species might serve to reveal aspects of the biogeographic history of the uplift of the Qinghai-Tibetan Plateau.

Fig. 1
figure 1

Comparison of biological traits, habitat and basic genomic information for the three sequenced Sinocyclocheilus species. They are representative of surface-dwelling (Sg), semi-cave-dwelling (Sr) and cave-restricted (Sa) species, respectively. Please note the regressive characters in the adult Sa, such as loss of eyes, little scale covering and translucent skin

Full size image

Results and discussion

Genome assembly, assessment and annotation

High-quality genomic DNA was extracted from muscle tissues of the three Sinocyclocheilus species, which were collected from Yunnan (Sg and Sr) and Guangxi (Sa) provinces in China (Additional file 1: Figure S1). A series of sequencing libraries (250 bp to 20 kb) were constructed and applied in a whole-genome shotgun sequencing strategy, and a total of 313.3, 174.0 and 188.2 Gb of raw data were obtained for the Sg, Sr and Sa fishes, respectively (Additional file 2: Table S1). The assembled genome sizes are approximately 1.7 Gb (1.75 Gb for Sg, 1.73 Gb for Sr and 1.68 Gb for Sa), and the calculated contig N50 and scaffold N50 values are 17–29 Kb and 0.9–1.3 Mb, respectively (Fig. 1; Additional file 2: Table S3). The quality of the three genome assemblies was evaluated (Additional file 2: Tables S5 and S9, Additional file 3: Figures S6 and S7), and our assessment confirmed that all were high in quality and could be used for further comparative analyses. In addition, we employed a standard annotation pipeline to predict gene sets, resulting in approximately 40,000 genes in the three fish genomes (42,109 for Sg, 40,333 for Sr and 40,470 for Sa; see more details in Additional file 2: Table S15 and Additional file 4: Figure S14), which were double that of most diploid species (Additional file 4: Figure S13). The further Hox gene distribution analyses in three Sinocyclocheilus species (Additional file 4: Figure S11) and genome alignment between zebrafish and Sg (Additional file 4: Figure S10) provided more evidence to support the tetraploid nature of Sinocyclocheilus.

Phylogenetic relationships and divergence time

The collision between India and Asia in the early Cenozoic may have been the largest-ever orogenic event in the Earth’s history [9]. It not only triggered the extraordinary uplift of the Qinghai-Tibet Plateau [10], but also profoundly influenced the Asian climate through the genesis of the Asian monsoon [11], the development of large-scale drainage patterns [12], and especially, the speciation and biodiversity of the organisms living on and below the Plateau (e.g. [9, 13]). Because Sinocyclocheilus, the most species-rich cyprinid genus, is endemic to the massive southwestern karst area (the northern part of which abuts the eastern part of the Qinghai-Tibetan Plateau), it has been proposed that the development of this diversity was correlated with the cyclic uplift and planation of the Qinghai-Tibet Plateau [14, 15]. Both the phylogenetic analyses of single-copy orthologous genes and of the mitochondrial DNA datasets (Additional file 5: Figures S15 and S16) supported this correlation, recovering the same relationships for the three Sinocyclocheilus species studied here (Fig. 2a). The generated chronogram revealed that Sr emerged first from the other two species at 26.3 Ma, and Sg divided from Sa at 17.5 Ma. These divergence times are similar to those obtained for other speciation events of congeners inferred from a more comprehensive study of Cyprininae phylogeny [16]. These splitting events occurred during the initial and final phases, respectively, of the second tectonic uplift of the Qinghai-Tibet Plateau (25–17 Ma), a time when evidence shows that the limestone region they lived in had been experiencing large-scale karst development [17]. The frequent capture and isolation of the subterranean river systems along with the karst development were regarded as the key reasons for speciation in this genus [15], although the complete picture will remain unknown until more species have been examined. The ancient separation revealed here might be one of the reasons to explain the high species diversity within a relatively narrow distribution area (about 270,000 km2 [15]) (Additional file 1: Figure S1).

Fig. 2
figure 2

Phylogenomic analysis and demographic histories of the three Sinocyclocheilus species. a Phylogenomic relationships inferred from 3,181 orthologous genes of the three Sinocyclocheilus and six other teleost species (Homo sapiens was the outgroup), with the branch lengths scaled to estimated divergence times (numbers in blue show median and range values). The numbers besides the branch indicated expanded (green) and contracted (red) gene families since the split from a most recent common ancestor (MRCA). b Demographic histories were reconstructed using the pairwise sequentially Markovian coalescent (PSMC) model. The uplift process of the Qinghai-Tibet Plateau since 3.6 Ma was obtained from a published paper [11] and other significant environmental events, such as the atmospheric surface air temperature and Eurasian ice volume for the past 3 Ma, were taken from the National Centers for Environmental Information (NCEI; http://www.ncdc.noaa.gov/). The time range of three rounds of intense uplift (Qingzang, Kunhuang and Gonghe Movement) is highlighted in lilac. All three species have similar patterns of decrease under the Qingzang and Kunhuang Movement, but a subsequent divergent trend under the Gonghe Movement

Full size image

Population history

A reconstruction of the population demography of the three Sinocyclocheilus species has revealed a similar start but a subsequent divergent trend from 104 to 107 years ago (Fig. 2b). It seems that the population demography has a greater correlation with the uplifting of the Qinghai-Tibet Plateau than other significant environmental events, such as the Eurasian ice volume or the atmospheric surface air temperature. In a relatively short period of time (from 3.0 to 0.5 Ma), all three species underwent two rounds of population decline, which occurred following two intense uplift phases in the third tectonic uplift of the Qinghai-Tibet Plateau [17, 18]. These two phases of movements at Qingzang (3.6–1.7 Ma) and Kunhuang (1.1–0.6 Ma) uplifted the Plateau from an average of <1,000 m up to 4,000 m, followed by the intensification of the Asian monsoon and increased precipitation [18], in addition to the large-scale development of glaciers [17]. The patterns of coincident population declines in the three species are believed to be a response of a common background following the intense uplift of the Plateau that may have been unfavorable to them (although it is still not clear which kind of environmental changes contributed the most). However, two events of unusual population expansion were also detected after these events. One was recognized in Sr during the interglacial period (0.5–0.15 Ma), a period during which Sg and Sa kept their low populations, which hints at the presence of extensive subterranean water systems in the Sr-native Luoping Basin at this time. Another was observed for Sg from 0.023 Ma when its main distribution area, the paleo-lake Dianchi, was coincidentally documented from geological sediments to have been some three times larger than at present [19]. Range expansion along new river channels might accompany Plateau uplift, thereby providing some fish the opportunity to increase their population sizes as reported in the typical Plateau schizothoracine fish Schizopygopsis pylzovi [20] and Gymnocypris chilianensis [21].

Variations of eye structures and related genetic and expression responses

Cavefishes often display regressive features, such as the degeneration of the eyes, which often develop normally during embryogenesis but subsequently arrest, degenerate or sink beneath the skin [22]. In the three adult Sinocyclocheilus fishes in this study, Sg has normal-sized eyes with a lens and retinal structures; Sr has small eyes with a reduced lens and retinal cell density; and Sa has lost its external eyeballs and lens, and the retina has degenerated to become merely disorganized and indistinguishable cellular layers (Fig. 3). The recently published A. mexicanus genome identified many specific candidate genes under eye-related QTL, and revealed expression differences in several development stages [6]. In our study, we analyzed the genetic changes of these existing crystallin and opsin genes and also presented copy number variations among cave-dwelling and surface-dwelling fishes. Our comparative genomic data suggested that several opsin genes, including Lws2 (long wavelength-sensitive), Rh2-1 and Rh2-2 (middle wavelength-sensitive), have been lost in Sinocyclocheilus, and that Rh2-4 has been lost in Sa (Additional file 2: Table S26). In dim conditions, long wavelength light is rapidly attenuated [23], and it seems rational that the long and middle wavelength-sensitive genes were lost specifically in Sa. The results from transcriptomic analyses of eyes also unsurprisingly showed significantly lower expression levels of most visual opsin genes in Sa when compared with Sg and Sr (Additional file 2: Table S27). Development and maintenance of photoreceptors requires a series of transcriptional factors. As expected, nine important transcriptional factors, including Crx, Nrl, Otx2, Otx5, Nr2e3, Ggca1A, Gnat1, Gnat2 and Rorb, were significantly down-regulated in Sa (Additional file 2: Table S27 and Additional file 6: Figure S25). These results are consistent with those found for Astyanax cavefishes [24] and for other Sinocyclocheilus species [25, 26], which supports the hypothesis that down-regulation of rhodopsin might play a critical role in eye degradation of cavefishes [25, 26].

Fig. 3
figure 3

Comparison of retinal structures among the three Sinocyclocheilus species. Phenotypes and H&E stained sections of eyes from top to bottom are those in a Sg, b Sr and c Sa, respectively. GCL, ganglion cell layer; INL, inner nuclear layer; IPL, inner plexiform layer; IS, inner segment; ONL, outer nuclear layer; OPL, outer plexiform layer; OS, outer segment; RCL, relict cell layer; RPE, retinal pigmented epithelium

Full size image

By screening crystallin genes in the three Sinocyclocheilus genomes, we found that copy numbers of most genes were lower than those of the diploid zebrafish (Additional file 2: Table S26 and Additional file 6: Figure S28). Transcriptomic data further demonstrated that the expression levels of most crystallin genes were maintained at high levels in Sg, but were not expressed in Sa (Additional file 2: Table S27). We also observed that several crystallin genes in Sa (such as Cryball1, Crygm2d12 and Crygm7) have evolved into pseudogenes due to existing premature stop codons (Additional file 2: Table S25E). Previous studies in Astyanax found that the expression of crystallin genes were down-regulated in the development of cavefish lens [27], suggesting that the lens plays a critical role in promoting cell survival in the development of eyes [28]. In particular, β- and γ-crystallins play a pivotal role in retinal tissue remodeling and repair, and also strongly enhance axon regeneration of retinal ganglion cells [29]. Hence, the lack, or down-regulation, of crystallin gene expression in Sa supports its phenotype of a visual system in which both the lens and retina have degenerated. Interestingly, we also found that Sa has two copies of Hsp90α1.1 and Hsp90a1.2 (heat shock protein 90α) genes, while both Sg and Sr have only one. Meanwhile, the expression levels of Hsp90α in Sa eyes were much higher than those in Sg and Sr (Additional file 2: Table S27). These observations provide further evidence to support a novel role of Hsp90α in lens apoptosis and eye degeneration of cavefishes [30].

Different mechanisms of albinism compared with Astyanax

In order to check the mechanisms of loss of pigmentation in Sa skin, we compared melanogenesis-related genes in the three Sinocyclocheilus genomes. In Astyanax, the albinism of some cave populations is caused by the deletions of Oca2 gene exon regions, which disturbs the upstream steps of the melanin synthesis pathway [31]. Although similar deletions of Oca2 genes were not found in either of the two copies in Sa (Additional file 6: Figure S21), nor were any identical mutations specifically present in Sa and Astyanax cave populations, some other new mutations were identified (Additional file 6: Figure S21), and the transcriptome analysis performed on the skin of Sg, Sr and Sa indicated that the expression of Oca2 gene in Sa was the lowest (Additional file 2: Table S20). In the melanin synthesis pathway, several enzymes work downstream, such as Tyr (tyrosinase), a key rate-limiting enzyme, and Tyrp1 (tyrosinase-related protein 1) [32]. More interestingly, we found that Tyr has an amino acid mutation (G420R) in Sa (Additional file 6: Figure S18), which was identical to that identified in Caucasian human patients (G419R , http://www.ifpcs.org/albinism/oca1mut.html) [33, 34]. Meanwhile, the expression levels of Sa genes in the melanogenesis pathway in the skin transcriptomes were the lowest, especially Tyr and Tyrp1, with significantly lower expression levels in Sa compared with Sg (Additional file 2: Table S20). It seems that similar phenotypes in Sinocyclocheilus cavefish might evolve by different mechanisms from Astyanax. Furthermore, we found that the Mpv17 protein in Sa had a deletion in the signal region compared with zebrafish, Sg and Sr (Additional file 6: Figure S22), and the expression level of Mpv17 in Sa is the lowest (Additional file 2: Table S20). A previous study shows that the deletion in Mpv17 can cause the tra mutant phenotype in zebrafish, and cause a loss (or strong reduction) of iridophores throughout larval and adult stages [35]. This study also pointed out that differentiated iridophores were required for the accumulation and maintenance of melanophores during pigment pattern formation [35], and a parallel study showed that the interaction between iridophores and other chromatophores is critical in the stripe formation of zebrafish [36]. Thus we infer that the deletion of Mpv17 might cause the loss of iridophores, which affected the formation of melanophores, and consequently played a role in the albinism of Sa.

Gene mutation and loss in scale degeneration

A previous study has indicated that mutations in the ectodysplasin-A receptor (Edar) encoding locus can lead to complete scale loss in fish such as medaka [37]. For this reason, two copies of Edar gene (named as Edar1 and Edar2, respectively) in the three Sinocyclocheilus genomes were identified and checked. Interestingly, one of the proteins, Edar1, has deletions in the signal peptide and partial extracellular regions in all three Sinocyclocheilus species, which may lead to functional changes or loss in this copy (Additional file 6: Figure S20). For the other protein copy, Edar2, only Sa has the signal peptide region and partial extracellular regions totally deleted when compared with Sg and Sr (Additional file 6: Figure S20). This deletion in Sa may lead to a functional disorder in guiding the Edar protein transfer across the membrane, thus generating fewer scales at the skin surface of Sa (Additional file 7: Figure S29). Coincidentally, two important genes, Lamb3 and Col7a, were also lost from the Sa genome, which may cause a defect in the anchoring between the epidermis and dermis, resulting in friction and skin fragility [38, 39] in the scale covering.

Possible hearing loss in Sa

The hearing of cavefishes is interesting but less studied than the other functions discussed above. We found the deletion of Mpv17 gene, which we mentioned in the section on albinism, may also have some influence in hearing. A previous study reported that Mpv17-deficient mice suffered from degeneration of cochlea and loss of sensorineural hearing at 2 months old [40]. Another gene, Ush2a, has also changed in Sa, especially two amino acid sites, R334S [41, 42] and V382A [43] (referred to the human sites), which may affect splicing and the termination codon (Additional file 6: Figure S19). It has been proved that the encoded protein of Ush is found in the basement membrane and may be related to the development of the inner ear [44, 45]. These mutations in Sa may cause its sensorineural deafness. There are no similar changes in Sg and Sr (Additional file 6: Figure S22). Reconstructions of the saccular otolith morphology using synchrotron X-ray microtomography among these three species show that the ventral surface of this otolith in Sa is seriously aberrant, and the degree of corrosion has distinctly increased in the following order: Sg < Sr < Sa (Fig. 4). Both the anatomy of the swim bladder (Additional file 7: Figure S30) and the numbers of neuromasts and scales of the trunk lateral line system (Additional file 7: Figure S31) also indicate these hearing-related organs in Sa have different degrees of weakness. Our data together indicated that the cave-restricted species Sa might have reduced hearing, which could be similar with that demonstrated in amblyopsid cavefishes [46]. Furthermore, a comparison of the distribution of neuromasts on the head (Fig. 4) also suggested that the response to vibration of these three species was Sg > Sr > Sa, which is different from the general pattern in Astyanax [47].

Fig. 4
figure 4

I The distributions of superficial neuromasts on the head and II morphology of saccular otolith in the inner ear among the three Sinocyclocheilus species. The superficial neuromasts after DASPEI staining from the plates I-(a–c), I-(d–f) and I-(g–i) represent Sg, Sr and Sa, respectively. The photos from left to right show the lateral view, dorsal view and ventral view. These figures show that the numbers of neuromasts in the adult fishes decline in the following order: Sg > Sr > Sa. The morphology of the saccular otoliths was reconstructed based on synchrotron X-ray microtomography. The plates II-(a–c), II-(d–f) and II-(g–i) represent Sg, Sr and Sa, respectively. The photos from left to right show the location of saccular otoliths in the inner ear, the dorsal view and ventral view of its morphology. The ventral of saccular otolith in Sa is seriously aberrant, with a deep and expanded central pit, encircled by another lateral sulcus. The degree of corrosion increase is in the following order: Sg < Sr < Sa. Scale bar: 1 mm

Full size image

Different immune responses to specific habitats

The immune activities of Sa may be lower than its epigean counterparts because it probably lives in a less diverse microbial environment. The fact that Sa is more susceptible to disease in captivity might support this inference (unpublished data). We found relatively fewer copies of immune genes in Sa when compared with those in Sg and Sr (Additional file 2: Table S22). However, one important innate immune group, the Tlr (toll-like receptor) gene family, showed some degree of expansion in Sa (Additional file 2: Table S23), through the duplication of Tlr8 and Tlr18 in the Sa genome. This suggests that these three species may have evolved differential immune activities for innate immunity and adaptive immunity according to their different habitats. Interestingly, the semi-cave-dwelling Sr had substantially more copy numbers of immune genes than Sg and Sa (Additional file 2: Table S22), which may be a result of an adaptation to heterogeneous elements between epigean and hypogean habitats, as found in the amphibious mudskippers [48].

Lack of diurnal rhythms

Previous research reported that some cavefishes lack diurnal rhythms when living in perpetual darkness [49]. For example, cave populations of A. mexicanus have a phenotype of reduced sleep in comparison to their surface relatives [50]. We found that both the two copies of Skp1 in Sa had deletions in the N-terminal end of its protein (Additional file 6: Figure S24). Since Skp1 is one of the components of the Skp1-Cul1-Fbxl3 (SCF) protein complex, and the SCF complex is most relevant in the mammalian clock mechanism [51], the deletions in Skp1 might lead to some dysfunction in SCF, which suggests weaker light rhythms in Sa. Meanwhile, the transcriptomic analysis of the eye demonstrated that expression levels of the rhythm pathway genes decrease in the order Sg > Sr > Sa (Additional file 2: Table S28 and Additional file 6: Figure S26).

Low fecundity in Sa

Although low fecundity is often assumed to be normal in cave species [4], there is little empirical evidence, and the different fecundity levels between surface and cave forms sometimes seem to be habitat plastic (e.g. [52]). We know of no study on the fecundity of the cave-dwelling Sinocyclocheilus species. Our analysis of the absolute fecundity (number of mature eggs) of Sg and Sa (2,402.9 ± 881.9 in Sg [53] vs. 143 ± 116 in Sa (count from four specimens with mature eggs in this study)) indicated that the fecundity in cave Sinocyclocheilus species is much less than surface congeners. Interestingly, one related gene, Creb3l4, was found to have been lost in the Sa genome. It has been reported that Creb3l4 can regulate the expression of genes required for germ cell survival, although it is insufficient to disrupt the normal fertility in mice [54].

Enhancement of taste

Taste buds are enhanced in some cavefishes, such as Astyanax [55]. We applied the taste-related gene sequences of zebrafish to BLAST to the three Sinocyclocheilus genomes, and unexpectedly found that one taste receptor gene, the Tast1r2-1, was significantly expanded (almost fourfold) and one important transcription factor, the Prox1 gene, had threefold copies in these three species compared with zebrafish (Additional file 8: Table S29). However, if these expansions were correlated to the overall enhancement of taste in the three Sinocyclocheilus species (benthic and cave-preferred) we still need further testing. Some taste receptor genes, such as Tas1r1 and Tas2r200-2, were specifically duplicated in the Sa genome (vs. Sg and Sr) (Additional file 8: Table S29), which suggests a further improvement in the sense of taste in the cave-restricted Sa. A preliminary study of the distribution of taste buds within the jaws of these three species also indicated that their numbers increase in a sequence from Sg < Sr < Sa (Additional file 7: Figure S32).

Conclusion

This paper focused on comparative genomic and transcriptomic studies of three Sinocyclocheilus species, representative of surface-dwelling (Sg), semi-cave-dwelling (Sr) and cave-restricted (Sa) species. We found that speciation and population dynamics of these fishes are closely related to the uplifting stages of the Qinghai-Tibetan Plateau. Comparative genome analysis revealed many genetic changes, such as gene loss, pseudogenes, mutations and down-regulation, which were associated with regressive features (eye degeneration, albinism, rudimentary scales and low fecundity), and some gene expansions suggest that constructive features (such as the sense of taste) have evolved in caves (a few supplementary notes can be seen in Additional file 9: Note S6). The most important genetic changes in the cave-restricted Sa are summarized in Fig. 5. Among them, some changes are similar to those in other cavefishes (e.g. A. mexicanus) from very different parts of the world (such as the down-regulation of some rhodopsin-related genes), but some are not (such as without vs. with the exon deletion of Oca2 gene in Sa vs. some A. mexicanus cave populations). Although the incidence of gene losses and expansions in cave-restricted Sa corresponds to the distinct phenotype variations, the changes seem to be fixed in its genome rather than in other mechanisms. The genetic changes found at the genome level in this study, although needing further functional confirmation, still give some good clues to understanding the mysteries of cave adaptation.

Fig. 5
figure 5

Summary of the most important genetic changes in the cave-restricted Sa. The main results are outlined as follows: Lws2, Rh2-1, Rh2-2 and Rh2-4 are lost in Sa. Several crystallin genes, including Crygmx in the Sr and Cryball1, Crygm2d2, Crygm7 and Crygmx in Sa, have evolved into pseudogenes. Sa has two Hsp90α genes while Sg and Sr have only one; meanwhile, the expression of Hsp90α in Sa eyes is higher than that in Sg and Sr. Mpv17 has a deletion in the signal region in the Sa genome. Ush2a has two amino acid changes, i.e. R334S and V382A. Tyr has a nucleotide mutation (G420R) in one copy of the Sa genome. Two copies of Edar gene in Sa represent deletions, and Lamb3 and Col7a were lost. Two copies of Skp1 protein in Sa have deletions in the N-terminal end. Prox1 and Tast1r2-1 are under expansions in the three Sinocyclocheilus species genomes, and Tas1r1 and Tas2r200-2 are specifically duplicated in the Sa genome. Red, gene loss; green, gene expansion; purple, pseudogene; orange, mutation or deletion

Full size image

In the history of cave biology research, people have often been confused by the plethora of views and terms, such as disuse, preadaptation, opportunism, compensation, regressive evolution, etc. Adaptations to cave living are complicated because there is no morphological ‘archetype’ for cave animals, and the characters show a highly diverse mix [4]. There is still a long way to go before this work can be applied to general evolutionary theory. As the first report on cavefish genomes among distinct species in Sinocyclocheilus, our work provides not only insights into genetic mechanisms of cave adaptation, but is also a fundamental resource for better understanding of cavefish biology.

Methods

Genome sequencing and assembly

Three female adult Sinocyclocheilus fishes, collected from Yunnan province (Sg and Sr) and Guangxi province (Sa) of China, were used for sequencing. The research protocol and treatment of experimental fishes have been reviewed and approved by the internal review board of the Kunming Institute of Zoology, Chinese Academy of Sciences (approval ID: SYDW-2014020). High-quality genomic DNA was extracted from muscle tissues using Puregene Tissue Core Kit A (Qiagen, MD, USA) for construction of libraries with different inserted sizes (250 bp to 20 kb) (Additional file 2: Table S1). In total, 25 paired-end libraries (11 for Sg, 7 for Sr and 7 for Sa, respectively) were generated with the Illumina standard operating procedure. HiSeq 2000 shotgun sequencing and assembling by SOAPdenovo assembler were performed as previously reported [56]. Artificial and low-quality reads were filtered first and then sequencing errors with 17-mer frequency lower than four were collected using a method described in a previous study [57]. Next, 88.05-, 48.31- and 60.54-fold coverage of Sinocyclocheilus genomes were used for assembly (Additional file 2: Table S2). In addition, we estimated genome sizes of the three Sinocyclocheilus fishes using the 17-mer depth frequency distribution method: G (Genome size) = K-mer_num/Peak_depth. The estimated genome sizes are 1.79, 1.89 and 1.81 Gb, respectively (Additional file 2: Table S4). Subsequently, the filtered reads were assembled into contigs and scaffolds with SOAPdenovo [58] and gaps were fulfilled with GapCloser [58]. Finally, a primary assessment was performed on the genome assemblies (Additional file 9: Note S1).

Genome annotation

We performed homology-based predictions by running RepeatMasker (version 3.3.0) [59] against the RepBase [60], and identified repeat sequences at DNA and protein levels using TE library (version 16.10) and RepeatProteinMask. We searched the de novo prediction build repeat library using RepeatModeler (version 1.0.5) and generated TE results with classification information for each repeat family by running RepeatMasker on the genome sequences subsequently. Tandem repeats were also searched using Tandem Repeats Finder (version 4.04) [61]. The protein coding genes were obtained using a combination of the de novo method, homology-based gene prediction and RNA-Seq data (Additional file 9: Note S2). All predicted gene evidence was integrated by GLEAN [62] to get non-redundant data [63].

Transcriptome analysis

RNA was isolated for sequencing from four tissues (eye, skin, liver and ovary) of Sg, Sr and Sa, respectively. We applied an in-house C++ program to filter raw reads and then obtain high-quality reads (Additional file 9: Note S3). All of the clean RNA-Seq reads were mapped onto the corresponding reference genomes (Sg, Sr and Sa) using TopHat (version 2.0.4) [64]. According to the mapped results, transcripts were constructed using Cufflinks (version 2.0.0) [65]. The de novo transcriptomes of the four tissues were assembled by Trinity with filtered reads from each tissue separately into contigs and scaffolds. Trinity contains Inchworm, Chrysalis and Butterfly, which were employed sequentially to process large volumes of RNA-Seq reads.

Evolutionary analysis

(1) Gene family cluster: we defined gene families using TreeFam (http://www.treefam.org) among the three Sinocyclocheilus fishes (Sg, Sr and Sa) and seven other vertebrates, including fugu (Takifugu rubripes), green spotted puffer (Tetraodon nigroviridis), three-spined stickleback (Gasterosteus aculeatus), Atlantic cod (Gadus morhua), medaka (Oryzias latipes), zebrafish (Danio rerio) and human (Homo sapiens). A total of 17,883 gene families and 210 single-copy gene families were identified. The numbers of orthologous genes across the ten species were counted (Additional file 2: Table S19 and Additional file 4: Figure S13) and plotted in a Venn diagram (Additional file 4: Figure S14). (2) Phylogenetic analysis: phylogenetic relationships were established using 3,181 single-copy orthologous genes shared among nine teleost fish genomes (Additional file 5: Figure S15) using maximum likelihood (ML) method in PhyML [66, 67] and Bayesian inference method in MrBayes [68]. An additional dataset of six mitochondrial gene sequences were also used to reconstruct the phylogenetic trees (Additional file 5: Figure S16). (3) Divergence time estimation: the divergence times were estimated using the mcmctree [69] in PAML [70] and recalculated using the multidivtime [71, 72] program, and all were calibrated by five fossil records [73] (Additional file 5: Figure S17 and Additional file 9: Note S4). (4) Demographic history: the distribution of time to TMRCA (the most recent common ancestor) between two alleles in an individual can be related to the history of population size fluctuation. The population size histories of Sg, Sr and Sa were inferred using the pairwise sequentially Markovian coalescent (PSMC) model [74] on heterozygous sites with the generation time (g = 1 year) (according to artificial breeding of Sg) and the mutation rate (m = 3.51 × 10−9 per year per nucleotide) [75]. Reconstructed population history was plotted for Sg, Sr and Sa separately using gnuplot (version 4.4) [76]. In addition, we obtained the atmospheric surface air temperature (°C) and Eurasian ice volume (m sea level equivalent) data for the past 3 million years from the NCEI (http://www.ncdc.noaa.gov/). (5) Gene family contraction and expansion: we performed Sinocyclocheilus lineage-specific expansion and contraction analysis using the CAFE program [77]. Based on random birth and death model [78], a global parameter λ was estimated using maximum likelihood. Comparing each branch and their ancestor branch, we calculated a conditional P value and marked families with a P value of less than 0.05 as a significant change [79], which means it underwent contraction or expansion during evolution. These families were then subjected to GO/KEGG/IPR enrichment analyses along each Sinocyclocheilus lineage, respectively (Additional file 2: Table S25). More details are shown in Additional file 8: Table S29. (6) Genes with accelerated evolutionary rate: positive Darwinian selection at the DNA sequence level has been tested by estimating the ratio (ω) of nonsynonymous nucleotide substitutions (dN) to synonymous nucleotide substitutions (dS) between ortholog genes [80]. A branch-site model was used to search for the positive selection genes (PSG) [81]. After obtaining the Sinocyclocheilus PSG list (Additional file 10: Table S30), we converted it to the corresponding human orthologs as the input against a background of human genes [80] using the DAVID Functional Annotation [82] tool. (7) Evolution of Hox clusters: to define Hox genes in the three Sinocyclocheilus genomes, the Hox genes of zebrafish were downloaded from the ensemble as cross-references. The Hox gene numbers (Additional file 2: Table S18) and the order along the scaffolds (Additional file 4: Figure S11) in Sinocyclocheilus indicated that these fishes are indeed tetraploids when compared to diploid zebrafish. (8) Loss of Sa-specific gene families: in order to identify Sa-specific gene family loss, we extracted gene families that have no member in Sa while more than zero in the other nine species. The lost gene family list is included in Additional file 2: Table S24.

Morphological comparison

(1) Paraffin sections of the eyes of three Sinocyclocheilus species: histological studies on eyes were performed from paraffin sections and hematoxylin and eosin (H&E) staining. (2) Immunocytochemistry of taste buds: this analysis focused on distributions of taste buds on upper and lower jaws, using a primary antibody (rabbit against calretinin, labeling entire receptor cells within taste buds) according to a standard protocol (Additional file 9: Note S5). (3) Anatomy of gas bladder, absolute fecundity and other measurements: all samples for anatomy and measurements were from specimens immersed in 90 % ethanol. (4) Synchrotron X-ray microtomography of the saccular otolith: this experiment was performed at BL13W1 beamline at the Shanghai Synchrotron Radiation Facility (Shanghai, China); the slices were reconstructed using FBP algorithm, and 3D renderings were created and manipulated in VGStudio 2.1 software.

Data availability

The Whole Genome Shotgun projects have been deposited at GenBank under accession numbers LCYQ00000000 (Sg), LAVF00000000 (Sr) and LAVE00000000 (Sa), which are the same versions described in this paper. The RNA-Seq data from four tissues (eye, skin, liver and ovary) have been deposited at GenBank under accession numbers SRS1179797–SRS1179800 (Sg), SRS1179996–SRS1179999 (Sr) and SRS1180000–SRS1180003 (Sa).