A high-quality de novo genome assembly of one swamp eel (Monopterus albus) strain with PacBio and Hi-C sequencing data

Abstract

The swamp eel (Monopterus albus) is one economically important fish in China and South-Eastern Asia and a good model species to study sex inversion. There are different genetic lineages and multiple local strains of swamp eel in China, and one local strain of M. albus with deep yellow and big spots has been selected for consecutive selective breeding due to superiority in growth rate and fecundity. A high-quality reference genome of the swamp eel would be a very useful resource for future selective breeding program. In the present study, we applied PacBio single-molecule sequencing technique (SMRT) and the high-throughput chromosome conformation capture (Hi-C) technologies to assemble the M. albus genome. A 799 Mb genome was obtained with the contig N50 length of 2.4 Mb and scaffold N50 length of 67.24 Mb, indicating 110-fold and ∼31.87-fold improvement compared to the earlier released assembly (∼22.24 Kb and 2.11 Mb, respectively). Aided with Hi‐C data, a total of 750 contigs were reliably assembled into 12 chromosomes. Using 22,373 protein-coding genes annotated here, the phylogenetic relationships of the swamp eel with other teleosts showed that swamp eel separated from the common ancestor of Zig-zag eel ∼49.9 million years ago, and 769 gene families were found expanded, which are mainly enriched in the immune system, sensory system, and transport and catabolism. This highly accurate, chromosome-level reference genome of M. albus obtained in this work will be used for the development of genome-scale selective breeding.

Introduction

The swamp eel (Fishbase ID: 4663; NCBI Taxonomy ID: 43700), Monopterus albus, is an economically important freshwater fish species in China, Japan, and Southeast Asia countries (Khanh and Ngan 2010; Nhan et al. 2019). As M. albus has a sex transition process during its life (Liu 1944), and it is becoming an emerging model species in development, genetics, and evolution (Cheng et al. 2003). Due to high nutritional level and impressive economic value of this species, the annual production of swamp eel has increased rapidly and exceeded 319,000 tons in China in 2018 (Zhang 2019), and its artificial propagation has been studied since last century in China (Guan et al. 1996) and corresponding techniques have been succeeded and promoted in Hubei, Sichuan provinces in China recently. Four or five different genetic lineages were recognized based on mitochondrial control region (Cai et al. 2013; Liang et al. 2016), and different local strains with different body colors (yellow, cyan, and grey) were cultivated in China (Yang et al. 2009; Wu et al. 2014). Due to the superiority in growth rate and fecundity (Chen et al. 2009; Yang et al. 2009), we have carried out selective breeding of one local strain of swamp eels (deep yellow and big spots). Though a genome of M. albus using second-generation sequencing technology (Illumina HiSeq 2000 plat-form) was published, which only covers 81.3% (Zhao et al. 2018), a chromosome-level, highly accurate reference genome for M. albus is still lacking, which hindering genome-scale genetic breeding for sustainable aquaculture of this species. Compared to the second-generation sequencing technologies, which constructs a complete genome by assembling numerous small sequence reads, thus leaving many gaps, third-generation sequencing technologies produce long reads and enable the production of a genome with a high level of completeness. Here, we report the chromosome-level genome assembly of M. albus using PacBio long-read sequencing and Hi-C technology. The assembly resulted in excellent continuity at the contig and scaffold levels, which facilitated subsequent investigations on genomic-based breeding studies targeting its economic traits.

Materials and methods

Ethics statement

All experiments in the present study were approved by the Animal Care and Use Committee of Yangtze River Fisheries Research Institute, Chinese Academy of Fishery Sciences (CAFS), China.

Sample collection and sequencing

The local strain (deep yellow and big spots) of M. albus was bred and cultivated at the Yangtze River Fisheries Research Institute (Jingzhou, Hubei Province, China). The F₁ full-sibling families were established from breeding population by brother–sister mating, and the F₂ generation full-sibling families were obtained from the F₁ generation with brother–sister mating. The larvas were hatched in plastic incubation boxes and fed with red worms, and then they were transferred to netcages in traditional ponds about 6–7 days post-hatching. The juvenile and adult fish were fed with artificial diets. Two 1-year-old healthy female F₂ individuals were used to collect sample tissues (Figure 1). The fish was immediately dissected after anesthesia with MS-222. White muscle and liver tissues of one fish were collected for whole genome sequencing and Hi-C library construction, respectively. Six different tissues (white muscle, liver, spleen, intestine, pituitary, and ovary) of another fish were collected and used for transcriptome sequencing. Genomic DNA from muscle tissue was extracted using the standard phenol/chloroform extraction method for DNA sequencing library construction. The integrity of the genomic DNA molecules was checked using agarose gel electrophoresis. Both the Illumina NovaSeq platform and the PacBio SEQUEL II platform were applied for genomic sequencing to generate short and long genomic reads, respectively. For the Illumina NovaSeq platform (San Diego, CA, USA), a pair-end library was constructed with an insert size of 350 base pairs (bp) according to the protocol provided by the manufacture. For the PacBio SEQUEL II platform, 10 μg genomic DNA was used for 20 kb SMRTbell library construction according to the manufacturer’s protocol (Pacific Biosciences). Liver tissue collected from the same individual was used for Hi-C library construction. One gram of fresh liver was first fixed using formaldehyde with a final concentration of 1%. The fixed tissue was then homogenized with tissue lysis, digested with the restriction enzyme (MboI), in situ labeled with a biotinylated residue, and end-repaired. Then the Hi-C libraries were quantified and sequenced using the Illumina NovaSeq 6000 instrument (Illumina).

Figure 1.

Picture of the local strain of swamp eel (deep yellow and big spots) used in the genome sequencing and assembly.

We also performed RNA sequencing to generate transcriptome data on the Illumina NovaSeq platform for gene model prediction. TRIzol reagent (Invitrogen, USA) was used to separately extract RNA from collected six tissues (white muscle, liver, spleen, intestine, pituitary, and ovary). RNA purity was checked using the kaiaoK5500^®Spectrophotometer (Kaiao, Beijing, China), and RNA integrity and concentration were assessed using the RNA Nano 6000 Assay Kit of Bioanalyzer 2100 system (Agilent Technologies, CA, USA). Then, a total amount of 2 μg RNA per sample was mixed for RNA sequencing. Sequencing libraries were generated using NEBNext^® Ultra™ RNA Library Prep Kit for Illumina^® (#E7530L, NEB, USA) following the manufacturer’s recommendations and the libraries were sequenced on an Illumina platform and 150 bp paired-end reads were generated.

Genome size estimation and assembly

For the next-generation sequencing (NGS) short reads, the Kmer-based method (Liu et al. 2013) was used to perform genome survey analysis to estimate the genome size, heterozygosity and repeat content of the swamp eel genome. We countered the number of each 17-mer with Jellyfish (Marcais and Kingsford 2011), and the frequency distribution is plotted. The PacBio long reads were assembled with Canu package v1.8 (Koren et al. 2017)(corrected error rate = 0.045, cor out coverage = 40). The obtained draft assembly was then polished with two steps: The assembly was first polished with arrow (Chin et al. 2013) using long reads, and then was polished with Illumina short reads with Pilon (Walker et al. 2014). Lastly, redundant sequences were removed with Purge Haplotigs (Roach et al. 2018).

The reads from the Hi-C library sequencing were mapped to the polished swamp eel genome with Bowtie (v2.3.4.3) (Langmead and Salzberg 2012). We independently aligned the two read ends to the genome and only selected the read pairs for which both ends were uniquely aligned to the genome. The hiclib Python library (Imakaev et al. 2012) and a previously reported method (Servant et al. 2015) were applied to filter the Hi-C reads, and the interaction frequency was quantified and normalized among contigs. Based on the interaction matrix, Lachesis (Burton et al. 2013) with default parameters was then applied to anchor the contig to the chromosomes using an agglomerative hierarchical clustering method.

The old genome assembly (GCF_001952655.1) was aligned to our genome assembly by using minimap2 v2.14-r883(Li 2018)with the “asm5” preset, and the output results were plotted as dotplots using an R script called dotPlotly (https://github.com/tpoorten/dotPlotly) with default parameters. The completeness and accuracy were also assessed via short-read mapping and Benchmarking Universal Single-Copy Orthologs (BUSCO) analysis. The reads generated by Illumina NovaSeq platform were mapped to the M. albus genome assembly using BWA (version 0.7.17) (Li and Durbin 2009), and variant calling was performed with SAMTOOLS (Li et al. 2009). We queried the assembled swamp eel genome against the Actinopterygii (actinopterygii_odb9, containing 4584 highly conserved single-copy core Actinopterygian genes) by using BUSCO v3.0.2 (Simao et al. 2015). Besides, the completeness and consensus accuracy of our new assembly were also computed using Merqury version 1.0 (Rhie et al. 2020).

Gene model prediction and functional annotations

Repeat elements were annotated in the swamp eel before gene model annotation. We applied Tandem Repeat Finder (TRF) (Benson 1999), LTR_FINDER (Xu and Wang 2007), PILER (Edgar and Myers 2005), and RepeatScout (Price et al. 2005) for the ab initio prediction of repeat elements in the genome. Thereafter, RepeatMasker and RepeatProteinMask (v4.0.9) (http://www.repeatmasker.org, last accessed April 10, 2019) were used to search the genome sequences for known repeat elements, with the genome sequences used as queries against the Repbase database (version 23.08) (Jurka et al. 2005; Bao et al. 2015).

Gene annotation was performed on the repetitive-element-masked genome. A combined strategy of homology-based, ab initio and transcriptome-based gene prediction methods was used. For homology-based prediction, protein sequences of closely related fish, including Acanthochromis polyacanthus (GCF_002109545.1), Amphiprion ocellaris (GCF_002776465.1), Anabas testudineus (GCF_900324465.1), Astatotilapia calliptera (GCF_900246225.1), Mastacembelus armatus (GCF_900324485.1), and M. albus (GCF_001952655.1) were downloaded from National Center for Biotechnological Information (NCBI). Proteins from the closely related fish species were mapped to the swamp eel genome using TBLASTN (Gertz et al. 2006). The alignment hits were joined with Solar (Yu et al. 2006). Next, GeneWise (Birney et al. 2004) was used to predict the exact gene structure of the corresponding genomic regions. In addition, RNA-seq reads were directly mapped to the assembled genome to identify putative exon regions using TopHat package (v2.1.1) (Trapnell et al. 2009) and Cufflinks (v2.2.1) (Ghosh and Chan 2016). For ab initio gene prediction, AUGUSTUS v3.3.2 (Stanke et al. 2006) was first trained using the BUSCO annotation of core Actinopterygii genes and then used for the prediction of genes in the repeat-masked genome. All of the gene models were merged, and redundancy was removed by MAKER (Cantarel et al. 2007). To annotate protein-coding genes functionally, gene sequences were searched using DIAMOND (v 0.9.19) with an e-value threshold of 1e–5 against the NCBI nonredundant protein (nr), TrEMBL, KOG, PFAM, and the SwissProtdatabases. Functional ontology and pathway information from the Gene Ontology (GO) and the Kytoto Encyclopedia of Genes and Genomes (KEGG) databases (Ogata et al. 1999) were assigned to the genes using BLAST2GO (Conesa et al. 2005) and KAAS (v2.1) (Moriya et al. 2007). Noncoding RNAs, including rRNAs, snRNAs, miRNAs, and tRNAs, were annotated by using TRNASCAN-SEv1.3.1 (Lowe and Eddy 1997) and the RFAM database (release 13.0) (Griffiths-Jones et al. 2003) using INFERNAL (v1.1.2) (Nawrocki and Eddy 2013).

Phylogenetic analysis and divergence time estimation

To identify gene families among M. albus and other species, the protein sequences of other nine fish species were downloaded from NCBI. These species included four close relative species of Anabantaria [Anabas testudineus (GCF_900324465.2), Betta splendens (GCF_900634795.2), Mastacembelus armatus (GCF_900324485.2), and Channa argus (GCA_004786185.1)], and five other teleost species [Oryzias latipes (GCF_002234675.1), Seriola lalandi (GCF_002814215.1), Takifugu rubripes (GCF_901000725.2), Danio rerio (GCF_000002035.6), and Gadus morhua (GCF_902167405.1)] as the outgroup. After removing those sequences less than 30 amino acids, these protein sequences were aligned to each other with BLASTP (Altschul et al. 1990) programs with an e-value threshold of 1e–5. Then, OrthoMCL (1.4) (Li et al. 2003) were used to cluster gene families with a Markov inflation index of 1.5 and a maximum e value of 1e–5. All gene families were ascertained from at least two genomes.

Protein sequences of the obtained one-to-one orthologous genes obtained from OrthoMCL analysis were used for phylogenetic tree reconstruction. MUSCLE (Edgar 2004) was used to generate multiple sequence alignments for protein sequences in each single-copy family with default parameters. Then we used Gblocks (Talavera and Castresana 2007) to extract the well-aligned regions of each gene family alignment and converted protein alignments to the corresponding coding DNA sequence alignments using an in-house script. Then, the alignments of each family were concatenated to a super alignment matrix. The super alignment matrix was subjected to phylogenetic analysis using RAxML (v8.2.11) (Stamatakis 2014) with the GTRGAMMA model and 100 bootstrap replicates.

Bayesian molecular dating was adopted to estimate the neutral evolutionary rate and species divergence time with MCMCTREE from PAML (version 4.4 b) (Yang and Rannala 2006) with the options “correlated molecular clock” and “JC69” model. Three calibration times were obtained from TimeTree database (http://www.timetree.org/, last accessed June 26, 2020) (Hedges et al. 2015) and previous report (Benton and Donoghue 2007). The MCMC (Markov chain Monte Carlo) chain length was set to 20,000 generations using a burn-in of 2000 iterations.

Expansion and contraction of gene families

The orthologous genes and phylogenetic tree topology inferred from the OrthoMCL analysis were taken into CAFÉ v4.2 (De Bie et al. 2006), which used a random birth and death model, to estimate the size of each family at each ancestral node and obtain a family-wise p-value (based on a Monte–Carlo re-sampling procedure) to indicate whether a significant expansion or contraction occurred in each gene family across species.

Data availability

The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive (Wang et al. 2017) in National Genomics Data Center (National Genomics Data Center and Partners 2020), Beijing Institute of Genomics (China National Center for Bioinformation), Chinese Academy of Sciences, under accession number CRA003062 that are publicly accessible at https://bigd.big.ac.cn/gsa.

Supplementary material is available at figshare DOI: https://doi.org/10.6084/m9.figshare.13228784.

Results and discussion

Genome size estimation and assembly of the M. albus genome

To estimate the genome size and heterozygosity of M. albus, 89.78 gigabyte (Gb) Illumina clean reads were obtained by Illumina paired-end libraries (Table 1). The mean reads length for Illumina data was 150 bp. The paired-end reads were primarily used for genome properties estimation, to assess and improve the base-level quality of the assembly. Based on the total number of 68,732, 704,793 17-mers and a peak 17-mer depth of 83, the estimated genome size of M. albus was calculated to be 785 Mb, with 42.07% GC content (Supplementary Figure S1). In addition, the estimated heterozygosity rate and repeats were approximately 0.49% and 47.32%, respectively (Supplementary Figure S1).

Table 1.

Statistics of the DNA sequence data used for M. albus genome assembly

Library resource	Sequencing platform	Insert size	Raw data (Gb)	Clean data (Gb)	Sequence coverage (X)
Genome	Illumina NovaSeq	350 bp	91.02	89.78	115.9
genome	PacBio SEQUEL II	20 kb	136.89	136.89	174.49
Hi-C	Illumina NovaSeq	350 bp	114.21	107.93	145.49
Transcriptome	Illumina NovaSeq	350 bp	8.93	8.67	11.38

We used the PacBio SEQUEL II platform to generate long genomic reads for the reference genome construction. A final total of 136.89 Gb sequencing data (∼174 X) were obtained (Table 1). The mean length of read and the N50 were 16 kb and 22.7 kb, respectively (Supplementary Tables S1). Then the PacBio long reads were assembled with CANUpackage and polished by Arrow and Pilon. We obtained a final assembled genome of 799 megabase (Mb) with a contig N50 length of 2.48 Mb (Supplementary Table S2) and GC content of 41.4%. The genome contained 809 contigs, and the longest contig was 14.73 Mb (Table 2; Supplementary Table S2). The GC content and sequencing depth for the genomic contigs were consistent with the estimation from the k-mer-based analysis (Supplementary Figure S2).

Table 2.

Assembly statistics for the swamp eel

	Length				Number
	Contig (Mb)		Scaffold (Mb)		Contig		Scaffold
Content	new	GCF_001952655.1	new	GCF_001952655.1	new	GCF_001952655.1	new	GCF_001952655.1
Total	799.64	635	799.72	689.52	809	117,579	71	62,978
Max	14.73	0.16	88.66	11.7	–	–	–	–
Number ≥2 kb	–	–	–	–	805	44,314	63	2,360
N50	2.44	0.02	67.24	2.11	97	8438	97	87
N90	0.54	0.005	49.70	0.37	348	33115	348	379

The Hi-C reads were used to generate a chromosome-level assembly of the genome. Using Hi-C library sequencing and read-quality filtering, we obtained approximately 107 Gb of clean bases with Q20 of 97.73% and Q30 of 93.22% (Table 2; Supplementary Table S3). Using our mapping strategy, we found that more than 95% of the total reads properly mapped to the genome, and 149 million read pairs (39.7%) provided valid interaction information for chromosome assembly. As a result, 750 contigs were successfully clustered, ordered, and oriented in 12 chromosomes (Figure 2), which represented 99.26% and 88.57% of all of the contigs at the base and sequence number level, respectively. The contig and scaffold N50 length of the final assembly reached 2.44 and 67.24 Mb, respectively, which represent ∼110-fold and ∼31.87-fold improvement compared with previously reported ∼22.24 Kb and 2.11 Mb (Zhao et al. 2018), respectively (Table 2). There were still 59 unanchored contigs after the Hi-C-based chromosome construction with an N50 length of 88.2 kb, which was significantly smaller than that of the anchored contigs. Although a primary genome assembly of M. albus was available (Zhao et al. 2018), its scaffold N50 (2.11 Mb) and contig N50 (22.23 kb) were shorter. Therefore, in this study, we improved the genome work with a high-quality assembly.

Figure 2.

Swamp eel genome contig contact matrix using Hi-C data, color bar indicates contact density from red (high) to white (low).

We evaluated the accuracy of the initial assembly by mapping the 606,791,208 Illumina reads for genome survey to the assembly using BWA. As a result, 99.57% of the Illumina reads were successfully mapped to the assembled genome with a coverage of 99.84% (Supplementary Table S4). We also obtained 0.001% homozygous single nucleotide polymorphisms (SNP), suggesting the high accuracy of this assembly.

The comparisons between the two assemblies obtained 120,662 aligned blocks, which account for 94.85% of the previous assembly (GCF_001952655.1), but only account for 83.96% of our new assembly, and the dot plot for the whole-genome alignment shows good collinearity between the two assemblies though some gaps and inversions are also found (Supplementary Figure S3). This result indicates great improvements in terms of contiguity and gaps in our newly obtained genome assembly. To evaluate the completeness of the assembly, the assembled swap-eel genome was assessed by using BUSCO (Simao et al. 2015) with the actinopterygii_odb9 database (4584 core genes). We found that 95.99% of core genes were identified in full-length in this swamp eel genome assembly and 49 (1.07%) core genes were captured as fragments (Table 3). These results suggest that less than 2.95% of the core Ray-finned fish genes were missing in our swamp eel assembly. Furthermore, when assessed by using Merqury, an optimal k = 19 was calculated and the spectra-cn plot indicates that our assembly is a complete haplotype-resolved assembly (Supplementary Figure S4), and the k-mer completeness of our assembly is 92.6% and QV scores is 41.35 (Supplementary Table S5), indicating our assembled genome with high sequencing coverage and assembly accuracy. All these results suggested that these sequences represent a significant improvement in contiguity in contrast to previously published swamp eel genome (Zhao et al. 2018) and the genome assembly was complete and robust.

Table 3.

BUSCO results for analysis of genome completeness for M. albus

Type	Number of genes	%
Complete BUSCOs (C) Complete and single-copy BUSCOs (S) Complete and duplicated BUSCOs (D)	4,400	95.99
	4,247	92.65
	153	3.34
Fragmented BUSCOs (F)	49	1.07
Missing BUSCOs (M)	135	2.95
Total BUSCO groups searched	4,584	100.00

Genome annotation

Using de novo searching and homolog prediction, we found that ∼46.76% of genome contents were repetitive elements (Supplementary Table S6), which was comparable with our estimation from the k-mer-based method. The identified repetitive elements in this assembly are higher than that previous report identified by Illumina sequencing (Zhao et al. 2018) and those of Betta spenden (15%) and Channa argus (18.9%). With regard to transposable elements (TEs), DNA transposons (20.98%), long interspersed nuclear elements (LINEs, 17.39%) and long terminal repeats (LTRs, 9.64%) were the top three categories of repetitive elements in the M. albus genome (Table 4). Thus, altogether, 46.23% of the genome was predicted to be repeated.

Table 4.

Repetitive element annotations in the swamp eel

	Repbase TEs		TE proteins		De novo		Combined TEs
Type	Length (Bp)	% in genome	Length (Bp)	% in genome	Length (Bp)	% in genome	Length (Bp)	% in genome
DNA	64154403	8.02	22201254	2.78	140659060	17.59	167727458	20.98
LINE	65337162	8.17	48523109	6.07	111587057	13.95	139038129	17.39
SINE	6049424	0.76	0	0	5559625	0.7	11190306	1.4
LTR	21457697	2.68	18986592	2.37	68782334	8.6	77048713	9.64
Satellite	1279944	0.16	0	0	623725	0.08	1897697	0.24
Other	4012	0	0	0	0	0	4012	0
Unknown	1177982	0.15	870	0	22004421	2.75	23151650	2.9
Total	155589278	19.46	89653198	11.21	315767592	39.49	369647460	46.23

In the current assembly, a total of 22,373 protein-coding genes with an average coding DNA Sequence (CDS) of 1707 bp and 10.4 exons per gene were identified based on de novo, homology- and transcriptome sequencing-based methods (Supplementary Table S7; File S1). By comparing the distribution of genes, coding sequences (CDS), exon and intron lengths of M. albus and closely related species, we found that the distributions in the M. albus genome were comparable with those of other teleosts (Supplementary Figure S5). For clarity, the distributions of GC density, gene density and repeat density across the 12 chromosomes of M. albus genome were further illustrated in Figure 3. To understand gene function, all of the predicted genes were mapped against several public databases, and 21,769 genes were functionally annotated in at least one of the databases, and up to 97.3% of M. albus genes were functionally annotated (Table 5). Finally, 670 miRNAs, 296 rRNAs, 503 snRNAs, and 16,279 tRNAs were also identified (Supplementary Table S8).

Figure 3.

Genome landscape of M. albus. From outer to inner circles: (A) 19 chromosomes; (B) GC density; (C) gene density; and (D) repeat density. Chr: chromosome. All statistics are based on 50 kb nonoverlapping windows.

Table 5.

Statistics of functional annotation of protein-coding genes

Database	Number	%
InterPro	19,893	88.92
GO	15,058	67.3
KEGG_ALL	21,371	95.52
KEGG_KO	13,605	60.81
Swissprot	18,961	84.75
TrEMBL	21,492	96.06
TF	3,601	16.1
Pfam	19,252	86.05
NR	21,684	96.92
KOG	17,519	78.3
At least one database	21,769	97.3
Total	22,373	–

Note that “at least one database” here refers to genes with at least one hit in multiple database.

Gene families and divergence time estimation

The consensus proteome set of the nine teleost species and the swamp eel were composed of a final data set of 239,948 protein sequences. A total of 19,705 gene families were predicted, including 4777 were one-to-one single-copy (Supplementary Table S9). M. albus contained 15,171 gene families. We studied the ortholog profiles of the five closely related species of Anabantaria (Anabas testudineus, Betta splendens, Mastacembelus armatus, Channa argus, and M. albus), a total of 12,353 (62.7%) gene families were shared by all five species (Figure 4) and 285 (∼0.1%) were M. albus specific ones.

Figure 4.

Venn diagram showing the distribution of shared gene families and their distribution between M. albus, Anabas testudineus, Betta splendens, Mastacembelus armatus, and Channa argus. The intersections between species indicate the numbers of shared gene families, whereas unique family numbers are shown in species-specific areas. The center represents the number of families shared by all species simultaneously. Analysis of families showed 12,535 families shared by all species simultaneously. Total number of gene families for each species is given in parenthesis.

In the phylogenetic tree obtained using 4777 single-copy ortholog gene families from the 10 teleost species, M. albus was recovered to be clustered with M. armatus, and then grouped with other teleosts, which was consistent with the fish species taxonomy (Hughes et al. 2018). Monopterus albus diverged from the common ancestor with M. armatus around 32.7–58.6 million years ago (MYA) (Figure 5).

Figure 5.

Phylogenetic analysis of M. albus and closely related teleost fish species. The species divergence time was shown at the branches of the phylogenetic tree, and the confidence intervals are given in parentheses. Except the figure of M. albus, all of the other species figures are retrieved from Wikimedia Commons. The species figures of Danio rerio (Mintern 1878), Gadus morhua, Seriola lalandi (Culloch 1911), Mastacembelus armatus (Ford 1878), Channa argus, and Anabas testudineus (France Day 1878) are distributed in public domain, and the species figures of Oryzias latipes (by Minami Kawasaki) (Kawasaki 2016), Takifugu rubripes [by DataBase Center for Life Science (DBCLS) ((DBCLS) DCFLS 2012)], and Betta splendens [by Pharaoh Hound (Hound 2007)] are distributed under Creative Commons Attribution 4.0 International license and Creative Commons Attribution 2.5 Generic license respectively.

Gene family expansion and contraction analysis

We analyzed gene family expansion and contraction in the M. albus lineage. There were 769 and 2414 gene families that expanded and contracted, respectively, after speciation from M. armatus (Figure 6A). Using the Gene Ontology (GO) and KEGG databases, we observed that 1514 genes from expanded gene families were significantly enriched in 479 and 179 (q < 0.05) GO terms (Supplementary Table S10) and KEGG pathways (Supplementary Table S11), respectively. The expanded gene families were mainly found on immune system pathways, especially on Intestinal immune network for IgA production (q = 1.96E–33); Hematopoietic cell lineage (q = 7.05E–24); Th1 and Th2 cell differentiation (q = 3.90E–20); Th17 cell differentiation (q = 1.92E–16) on KEGG pathways; Transport and catabolism, including Phagosome (q = 1.39E–32), Autophagy-animal (q =2.66e–21); Sensory system, including Olfactory transduction (q  = 2.77E–29); signal transduction pathways, including NF-kappa B signaling pathway (q  = 1.19E–18), Notch signaling pathway (q = 1.84E–12), Calcium signaling pathway (q = 1.82E–9), and Rap1 signaling pathway (q = 6.44E–8); Metabolism of cofactors and vitamins (q = 1.82E–9). In addition, a total of 195 genes in the contracted families enriched in 412 GO terms and 49 KEGG pathways (P < 0.05) were enriched, respectively (Supplementary Tables S12 and S13). These enriched terms were mainly involved in Cellular community-eukaryotes, including Tight junction (q = 1.59E–7), Gap junction (q = 9.13E–3); Carbohydrate metabolism, including Ascorbate and aldarate metabolism (q = 1.59E–7), Pentose and glucuronate interconversions (q = 1.59E–7). These biological processes may be related to the special characteristics of the swamp eel.

Figure 6.

(A) Dynamic evolution and distribution of gene families among 10 teleost species, including Monopterus albus, Anabas testudineus, Betta splendens, Mastacembelus armatus, and Channa argus, Takifugu rubripes, Oryzias latipes, Gadus morhua, and Danio rerio. The green and red numbers represent the expanded and extracted gene families, respectively. MRCA: most recent common ancestor. (B) The distribution of single-copy, multiple-copy, unique, other orthologs, and unclustered genes in 10 teleost species.

Conclusions

Due to the unique protogynous hermaphroditism of M. albus (Liu 1944) and high economic value in fisheries in China and Southeast Asia (Matsumoto et al. 2010), highly accurate, chromosome-level reference genome would provide a sound support for subsequent investigations of its sexual reversion, genome-scale selective breeding, and future investigations of population and conservation genetics.

In the present study, combing Illumina, PacBio long-read sequencing, and Hi-C technologies, we reported a high-quality chromosome-level genome assembly for the localstrain (deep yellow and big spots) of swamp eel. The contig and scaffold N50 reached 2.44 and 67.24 Mb, respectively, suggesting this genome assembly improved substantially in the primary assembly contiguity in contrast to previous assembly (Zhao et al. 2018). More than 95.99% of the complete Actinopterygii BUSCO genes (n = 4,400) were identified in the genome and 99.57% of the high-quality Illumina reads can be mapped onto the de novo genome assembly, suggesting the quality of our de novo assembled genome was high for both completeness and base-level accuracy. Besides, good k-mer completeness (92.6%) and high QV value (41.35) calculated with Merqury suggesting high coverage and accuracy in our assembly. More repetitive sequences (46.23%) were identified than those in the previous report (∼28%)(Zhao et al. 2018), which might be attributed to the higher contiguity and accuracy of our assembly. Such high percent of repetitive elements identified in this species is also higher than many of the sequenced teleost genomes (Reichwald et al. 2009), but lower than that of the zebrafish (54.3%) (Howe et al. 2013) and Atlantic salmon (58%) (Lien et al. 2016). Besides, a total of 22,373 protein-coding genes were annotated based on this assembly.

The phylogenetic analysis of related species showed that swamp eel was diverged ∼49.9 MYA from the common ancestor of M. armatus. Expanded gene families were significantly enriched in several important biological pathways, mainly in immune system, sensory system, and Transport and catabolism.

In summary, we have completed a chromosome genome assembly of M. albus. Given the economic importance of swamp eel for aquaculture production and the increasing research interests for its unique sex reversal, we expect this well-assembled genomic data could offer a valuable resource for future genomic-scale selective breeding programs and provide valuable information for improving economically relevant culture traits, such as disease resistance.