Background & Summary

With an annual growth rate of 5.8%, aquaculture is one of the most promising sectors of food production worldwide. World aquaculture production was 82 million tons in 2018, of which 54.3 million tons were contributed by finfish aquaculture1. Marine aquaculture has the potential to meet the increasing global demand for animal protein-based foods. Cobia (R. canadum), the only extant species in the family Rachycentridae, is a marine warm-water species distributed worldwide, particularly in tropical and subtropical climates, except for the central and eastern Pacific. In recent decades, cobia emerged as one of the most promising species for mariculture due to certain attributes like rapid growth rate, good meat quality and high market value, with a global production of around 40,000 tons2,3,4,5. In India, cobia was first successfully bred in 2010 at the Mandapam Regional Centre of the Central Marine Fisheries Research Institute in Tamil Nadu6,7,8. However, cobia aquaculture has been hampered by the deficiency of nutritional information, which limits the productivity of industrial forms of aquaculture9.

In order to optimise the efficient culture system of a species, we need to address the fundamental knowledge gap related to aspects of culture such as reproductive biology, digestive physiology and nutritional genetics10. To fill such a knowledge gap, an integrative study using different techniques is needed. Next-generation sequencing (NGS) studies can holistically elucidate the structures and functions of genes, as well as the molecular mechanisms underlying biological processes such as growth, nutrition, metabolism, immune function, stress, adaptation, and differential gene expression in response to factors such as diet, stress and other environmental factors11,12,13. Data from such systems has aided in the production of several commercially important fishes such as Chinese seabass (Lateolabrax maculatus14); Atlantic salmon (Salmo salar15), and Rainbow trout (Oncorhychus mykiss16). Thus, this information can be used to develop nutritional markers for different developmental stages and optimised feeding protocols17. For example, the effects of selected nutrients on target genes can be studied to adjust diet composition to improve growth, condition, and survival of fish larvae10,18,19. The identification of potential genes involved in key pathways involved in carbohydrate, lipid, amino acid, nucleotide, cofactor and vitamin metabolism will aid in the formulation of species and stage–specific diets for commercially important mariculture species such as cobia. However, previous studies on transcriptome analysis in cobia used short-read based platforms and were limited to a few tissue types17,20,21,22. Discovering novel transcripts, supporting genome annotation and identifying alternative splices and gene fusions require full-length transcripts, and as such, genetic data on cobia remain insufficient, limiting the scope of such research.

The full-length protein coding transcriptome of a species (including CDS and 5′ - and 3′ - UTRs) and its collection of splice variants are a crucial resource for the accurate annotation of protein-coding transcripts and for understanding how structural variants affect nutritional status, health, and economically significant traits in livestock23,24. Although next-generation short-read based sequencing has numerous advantages—for instance low cost, quantifiability and high throughput—it is less effective for assembling full-length transcripts with short sequencing runs without a reference genome, which could lead to inappropriate annotations25,26. The scope of studying alternative splice variants and corrected annotations is limited by low-quality transcripts attained by Illumina sequencing27. The most advanced third-generation sequencing platform (TGS) can aid us obtain a long-read or full-length transcriptome without assembly to study the structure of mRNAs, allowing us to discover more genes, detect alternative splicing, polyadenylation as well as long non-coding RNAs (LncRNAs)28,29. The TGS platforms have recently emerged as new genomic research tools owing to the advent of high-throughput sequencing technology.

The present study aims to generate full-length transcriptome for the commercially important mariculture fish, R. canadum, by sequencing individuals from different life stages using a TGS platform. The information generated from this research could be used to complement the genome for discovering new genes, gaining knowledge on the physiological properties and structure of mRNAs as well as for identifying potential nutritional markers in cobia.

Methods

Sample collection, preservation and RNA preparation

The animal experimental methods in this study were performed according to the ARRIVE recommendations30. The live fish were treated in accordance with the UK legislation: Animals (Scientific Procedures) Act (1986) of the United Kingdom (https://www.legislation.gov.uk/ukpga/1986/14/contents) and EU Directive on animal studies,2010/63/EU (2019)31. The experimental protocols used to conduct this study were approved by the ICAR-CMFRI, Kochi, India (BT/AAQ/3/SP28267/2018).

The different life stages of the cobia are depicted in Fig. 1. R. canadum larval samples were collected from the Marine Fish Hatchery at the Mandapam Regional Centre of ICAR-Central Marine Fisheries Research Institute, India. The juvenile and adult samples were collected from the fish maintained in the high-density polyethylene sea cages (6 m diameter, 4 m depth, 113 m3) at Mandapam, Tamil Nadu, India (site 9 16′ 11.9748” N, 79 7′ 56.0856” E; Lat-Long = 9.269993, 79.132246). For larvae samples, a weight of 500–800 mg (contained around 300 larvae of 5 dph and 15 larvae of 29 dph) was collected in triplicates and then immediately stored in RNA protection reagent (RNAlater, Sigma-Aldrich) at −80 °C until RNA extraction. Also, tissue samples (muscle, kidney, spleen, liver, intestine, 500–800 mg each) from 3 individuals of juvenile and adult fish were collected and immediately stored in RNA protection reagent and maintained at −80 °C until RNA extraction.

Fig. 1
figure 1

Different life-stages of cobia (R. canadum); (a) Embryo development. (b) Newly hatched larvae. (c) 3 dph larvae showing mouth opening. (d) Sub-adult.

Full size image

Total RNA from each sample was extracted using the lithium chloride approach32 and purified using the NucleoSpin RNA clean-up kit (MACHEREY NAGEL) following the manufacturer’s protocol. After isolation, the RNA samples were analysed for quantity and integrity using the Qubit 4.0 fluorometer (ThemoFisher Scientific, USA) and the AGILENT Bio-analyser 2100 (Agilent, USA).

PacBio Sequel Iso-seq3 library preparation and single molecule real-time (SMRT) sequencing

The RNA samples were divided into three pools prior to library construction. Pool 1 comprised whole larvae (5 and 29 days after hatching), Pool 2 comprised tissue samples from juveniles (muscle, kidney, spleen, liver, intestine) while Pool 3 comprised tissue samples from adults (muscle, pyloric caeca, spleen, intestine, kidney, liver). Equal amounts of RNA from each tissue were pooled to construct cDNA library.

Three Iso-Seq sequencing libraries were generated following the PacBio’s Iso-Seq3 protocol. Briefly, 2 µg of purified polyA mRNA was reverse transcribed into cDNA using the NEBNext Single Cell/Low input cDNA synthesis, while the second strand was synthesized by template switching. The cDNA preparation was purified using Pronex beads (Promega) and the purified cDNA was PCR amplified and repurified using specific Pronex beads to obtain standard transcripts, and analysed in the Bioanalyzer (Agilent Technologies, USA). After size selection using the BluePippinTM size selection system, DNA damage repair and terminal repair were performed on the SMRTbell libraries, followed by overhand adapter ligation, and equimolar amounts of the barcoded cDNA were pooled. A quantity of 132 ng HiFi SMRTbell libraries was prepared with a final concentration of 13.2 ng of purified cDNA. After polymerase binding and primer annealing with PacBio sequencing primers on SMRT templates, the SMRTbell containing 60 pM OPLC-purified polymerase-bound SMRTbell complex was finally processed for sequencing on the PacBio Sequel II platform at Nucleome Informatics (P) Ltd., Hyderabad, India.

The output of PacBio Sequel II sequencing and error rectifications

Three multiple tissue libraries were sequenced on the PacBio Sequel II platform and a total of 99 GB data was generated, mean sub-read lengths including 2109.9 bp, 1988.2 bp, and 1996.2 for larval (Pool 1), juvenile (Pool 2), and adult (Pool 3) sample pools, respectively (Table 1).

Table 1 Details of RNA sample pooling and PacBio Iso-seq output statistics.
Full size table

Further analysis revealed 219,072, 178,893 and 222,754 full-length non-chimeric reads (FLNC) for sample from Pool 1, Pool 2 and Pool 3, respectively. FLNC from all samples were combined to increase transcript recovery and resulted in 35661 high-quality, non-redundant isoform sequence sets with a total of 94193725 nucleotide bases, while the mean length of transcripts was 3110 bp, and N50 value was 2984 bp (Table 1).

Sequence data analysis

The raw data generated with the PacBio Sequel platform was analysed and processed using the standard protocol in SMRT Link software, while subreads were obtained by removing the adapters from the sequences and sorting out the polymerase reads with fragment lengths less than 50 bp, having a quality of 0.90. Meanwhile, subreads with a length of less than 50 bp were discarded, and the remaining subreads represented clean data. Circular consensus sequences (CCS) with full passes of ≥ 1 and a quality of > 0.90 were retrieved from the clean data, and by determining the presence of sequencing primers and terminal polyA sequence, the CCS were categorised into full-length nonchimeric CCSs and non-full-length nonchimeric CCSs. The presence of 5′ adapter, 3′ adapter sequence and poly A tails in the sequences was used to determine full-length non-chimeric readings (FLNC). Isoseq 3 software was used to extract and polish consensus isoforms in FLNC. The criterion for achieving high-quality, full-length transcripts was >99% post-correction accuracy. The CD-HIT software33 was used to eliminate redundant sequences from high-quality, full-length transcripts, and the full-length transcriptome from this step was used as the final isoform set of non-redundant transcripts used for further analysis. TransDecoder v3.0.1 software (TransDecoder. https://transdecoder.github.io/) was used to envisage the open reading frames (ORFs) of the non-redundant transcript isoform set with the lowest CDS of 100 bp. Finally, transcriptome completeness was analysed using the Benchmarking Universal Single Copy Orthologs (BUSCO) analysis34 based on the Ortholog database v935.

Functional annotation of full-length transcriptome

Full-length transcripts were annotated by BLASTx and BLASTp searches against NCBInr (http://www.ncbi.nlm.nih.gov), RefSeq36, UniProtKB, KOG (http://www.expasy.ch/sprot, version: 2019-8-14) and Pfam (v26.0) databases with an E-value cut off of 1e-537. We found one best match among each transcript and a known sequence in the database based on bit score. Metascape38 and EggNOG39 analyses were performed for Gene Ontology (GO) annotation, and to classify the function of the transcript based on cellular components, molecular functions and biological process features. To obtain the overall biological function of R. canadum transcriptome, the full-length transcripts were mapped into canonical reference pathways in KEGG using KEGG KASS40, while the TransDecoder v3.0.1 software was employed to find functional protein domains and to predict the ORFs of the non-redundant transcripts.

Annotation of the transcriptome with several databases (NCBI nr, RefSeq and UniprotKB) revealed in a functional assignment for 19081 transcripts (53.51%). Most sequence similarities were against the NCBI nr. (34783 transcripts, 97.54%), followed by the UniprotKB database (33321 transcripts, 93.44%), the Pfam database (32888, 92.22%) and the RefSeq database (19081 transcripts, 53.5%) (Table 2). In the NCBI nr annotation, 11948 (34.35%) of the homologous sequence was aligned to Seriola dumerili, followed by Seriola lalandi dorsalis (5895, 16.95%), Echeneis naucrates (4697, 13.50%), and Lates calcarifer (4063, 11.68%).

Table 2 Annotation statistics.
Full size table

The KOG-annotated transcripts were grouped into 26 KOG classifications, with the highest number of transcripts in the function unknown category (S) (9038, 24.75%), signal transduction mechanism (7026, 19.24%) followed by posttranslational modification, protein turnover, chaperones (3196, 8.75%), transcription (2765, 7.57%) and intracellular trafficking, secretion, and vesicular transport (1822, 4.99%) (Fig. 2). For KEGG annotation, transcripts were mainly grouped into 398 signalling pathways in 48 level 2 pathways, among which, the signal transduction pathway (T) had the highest number of transcripts (2091), followed by infectious diseases- viral (1360) and immune system (1150) (Fig. 3).

Fig. 2
figure 2

KOG function classification of transcript of R. canadum. x- axis represent different KOG categories (denoted by legends on right), and the y- axis, the number of the transcripts.

Full size image
Fig. 3
figure 3

Identified KEGG pathways of transcript isoforms of R. canadum. The x-axis signifies the number of genes, and the y-axis, different KEGG pathways.

Full size image

After GO annotation, a total of 20975 (58.82%) transcripts were allocated to multiple GO terms, among which 6829 transcripts (19.15%) were allotted to biological process, 3057 transcripts (8.57%) to molecular function and 1533 transcripts (4.30%) to cellular component (Fig. 4). Of all transcripts, 35526 (99.62%) were successfully annotated in at least one database and 7736 (21.69%) were annotated in all databases (Table 2).

Fig. 4
figure 4

Gene function classification of R. canadum based on Gene ontology (GO) annotation. The x-axis denotes the number of genes, and the y-axis- different GO categories.

Full size image

Genes related to nutrition

Several important functional genes involved in vertebrate nutrition and their isoforms have been identified from functionally annotated transcripts, which can be used in future nutrigenomic studies on cobia. Genes involved in the following biological processes were selected as marker genes: amino acid metabolism, digestive system, lipid metabolism, carbohydrate metabolism, endocrine system and metabolism of other amino acids (Table 3). The KEGG classification of nutritionally important genes is shown in Fig. 5. Of the 129 identified genes of amino acid metabolism, 32 were involved in cysteine and methionine metabolism, 39 in lysine degradation, 28 in glutathione metabolism and 30 in tryptophan metabolism pathways. Under the carbohydrate metabolism pathway, 33 genes were involved in glycolysis/gluconeogenesis, 17 in starch and sucrose metabolism, and 10 in ascorbate and aldarate metabolism. A total of 159 genes involved in the digestive system have been identified, with the following distribution of genes: gastric acid secretion (28), protein digestion and absorption (53), carbohydrate digestion and absorption (15), fat digestion and absorption (21), vitamin digestion and absorption (17) and mineral absorption (25). Among the 168 genes involved in the endocrine system, 56 genes were involved in growth hormone synthesis, secretion and action, 65 in insulin signalling and 47 in glucagon signalling. A total of 98 genes were identified for pathways in lipid metabolism and are distributed as follows: fatty acid biosynthesis (9), fatty acid elongation (16), fatty acid degradation (25), arachidonic acid metabolism (19), linoleic acid metabolism (6), alpha-linolenic acid metabolism (7) and unsaturated fatty acid biosynthesis (16). Isoforms of lipoprotein lipase and insulin- like growth factor genes showing the isoform diversity in full-length transcript data of R. canadum is given in Fig. 6a.

Table 3 Genes related to nutrition.
Full size table
Fig. 5
figure 5

KEGG classification of nutritionally important genes.

Full size image
Fig. 6
figure 6

(a) The isoforms of lipoprotein lipase and insulin like growth factor_1genes in R. canadum showing isoform diversity. (b) LncRNA prediction. (c) Distribution of microsatellites (SSRs). (d) ORF prediction.

Full size image

Long non-coding RNAs (LncRNAs) prediction

LncRNAs were predicted using three methods including PLEK41, Coding Potential Calculator (CPC)42 and Pfam structural domain analysis. The common non-coding hits/intersection of the three results were then filtered and considered as LncRNA.

We obtained 4321, 1347 and 937 candidate LncRNAs determined using PLEK, CPC, and Pfam, respectively, and among these 497 (5.97%) were identified in all analyses (Fig. 6b). The length of the LncRNA transcripts ranged from 200 bp to 8198 bp, with a mean length of 1918 bp. The LncRNA results are given in Table 4.

Table 4 LncRNA prediction results.
Full size table

Detection of Simple sequence repeats (SSRs)

The MISA software (http://pgrc.ipk-gatersleben.de/misa/misa.html) was used to predict the simple repeat markers in the non-redundant reference transcriptome of R. canadum, and the minimum repetition time for core-repeat motifs was fixed as follows: 10 for mononucleotides, six for di-nucleotides and five for tri-nucleotides, tetra-nucleotides, penta-nucleotides and hexa-nucleotides. Furthermore, the SSRs were categorized into perfect and complicated (compound or discontinuous) SSRs based on the structural organisation of the repeat motifs.

A total of 35661 transcripts with a total length of 94193725 bp were used for SSR prediction and it was observed that 10449 sequences contained more than one SSR marker. The number of SSRs found in compound formation was 7901, most of which were mononucleotide repeats (25133, 59.35%), dinucleotide repeats (9824, 23.20%), tri-nucleotide repeats (6183, 14.6%), tetra-nucleotide repeats (914, 2.16%), hexa-nucleotide repeats (157, 0.37%) and penta-nucleotide repeats (135, 0.32%). The results of SSR prediction are given in Table 5 and represented in Fig. 6c.

Table 5 Number and unit size of SSR identified in the transcriptome.
Full size table

ORF prediction

In total, 38243 coding sequences were predicted from 35661 transcripts using TransDecoder, with an average length of 448 bp, and there were 2075 transcripts with a length >1000 bp. The coding sequence lengths of ORFs is presented in Fig. 6d.

Data Records

The raw full-length data (Table 1) were deposited in the NCBI Sequence Read Archive (SRA)43 under accession numbers SRR1937012544, SRR1937012445 and SRR1937012346, while the respective BioSamples accession numbers are SAMN28614395, SAMN28614396 and SAMN28614397. Data regarding the identified nutritionally important genes was deposited at the figshare platform47. The file contains multiple spreadsheets with the annotated list of genes involved in the metabolism of carbohydrate, protein, lipid, vitamin, mineral, digestive function and bone development in spreadsheets 1 to 7 respectively.

Technical Validation

The BUSCO analysis results showed that among the 255 conserved eukaryotic orthologous genes, 85.5% complete genes (218 genes) were found in the R. canadum transcriptome with an additional 0.78% (2 genes) as fragmented BUSCOs (Fig. 7). Of these, 41.17% were complete single-copy BUSCOs and 5.09% were complete duplicate BUSCOs. A total of 71.94% (2413 genes) of the 3354 orthologues searched for in vertebrates were found in full, with another 1.72% (58 genes) as partial sequences. Of the 3640 orthologues in the eukaryote, 69.72% (2538 genes) were found in full and a further 1.29% (47 genes) as a partial sequence.

Fig. 7
figure 7

BUSCO analysis results.

Full size image