Whole genome sequencing and assembly

Various technologies have been used for the generation of the whole genome sequence of aquaculture species. However, the Illumina and PacBio platforms have contributed the most to the progress of aquaculture genome sequencing. Illumina sequencing generates accurate, but short reads at a relatively low cost, while PacBio sequencing generates longer, but less accurate reads at a higher cost. The proportion of sequences generated using these two platforms varies depending on the species and the status of the sequencing technology when the genome sequencing projects were initiated. A significant decrease in the cost of PacBio sequencing has generally led to increased use of this technology and the enhancement of contig/scaffold lengths in sequence assemblies.

Like the selection of sequencing technologies, various sequencing templates were used for the generation of the whole genome sequence assemblies in aquaculture species. These included mixtures of outbred individuals, single diploid males or females, individuals from inbred lines, and completely homozygous doubled haploids. The use of more homozygous templates greatly simplifies the computation of genome assemblies, which are complicated by the high levels of heterozygosity and sequence polymorphism characteristic of several aquaculture species [6, 7, 10]. The choice of sequencing templates has largely been dictated by the availability of the preferred homozygous templates. For instance, doubled haploids, produced through gynogenesis or androgenesis, are the preferred sequencing template for most teleost sequencing projects. However, the generation of doubled haploids is not generally feasible in many shellfish species, due to the unequal first cleavage that is sensitive to manipulation [11]. For some species, multiple individuals must be used because the DNA extracted from a single individual is not sufficient for the sequencing process.

While the mechanics of generating a large number of sequence reads is no longer difficult, calculation of a high quality of sequence assembly remains a challenging task. Four specific metrics are generally used to evaluate the quality of whole genome sequence assemblies including 1) Contiguity, as reflected in contig numbers and distribution of contig sizes; 2) Connectivity, as reflected in the number of scaffolds and distribution of scaffold sizes; 3) Completeness, as reflected in the total size of the genome assemblies and the percentage of coverage of the whole genome; and 4) Accuracy, as validated by at least one additional methodology such as genetic linkage mapping, physical mapping, or optical mapping. In addition, the integration of the whole genome sequences with genetic linkage maps is important for genetic studies.

For species under the NRSP-8 program, reference genome sequence assemblies for catfish, tilapia, Atlantic salmon, and rainbow trout are of good quality. For catfish, 50% of the genome sequence is included in only 31 of the largest scaffolds; 90, 95, and 98% of the genome is included in 185, 314, and 594 scaffolds, respectively. The catfish reference genome sequence was assessed to be nearly complete as 99.7% of re-sequencing reads were mapped to the reference genome sequence. In addition, the number of complete genes included in the reference genome sequence is larger than that of any of the sequenced diploid fish species, including zebrafish [5]. The catfish reference genome sequence assembly was validated by genetic mapping. The positions of 253,744 genetically mapped SNPs were fully concordant with those on the reference genome sequence with four exceptions [13]. The vast majority of the reference genome sequence (99.1%) has been anchored to chromosomes [13].

The reference genome assembly of Atlantic salmon is also of high quality [6]. The genome was sequenced with Sanger and Illumina technologies. It is complete as 2.97 Gb reference genome sequences were assembled, with the unassembled sequences being just repetitive elements. The largest 9447 scaffolds accounted for 2.24 Gb of the 2.97 Gb genome sequence. This is a remarkable achievement considering the very complex nature of the genome. The Atlantic salmon genome is largely tetraploid due to a recent genome duplication. It also has a high repeat content (58–60%); the dispersed Tc1 transposons represented 12.89% of the genome [6]. Similarly, the assembly of the rainbow trout genome is of good quality [7]. Since the publication of the genome paper, the reference genome sequence of rainbow trout has been further improved. The contig N50 has increased from 7.7 Kb to 13.9 Kb, and the scaffold N50 has increased from 380 Kb to 1.72 Mb. More importantly, over 82% of the genome sequence has been mapped to chromosomes (Palti, personal communication).

The published tilapia genome sequence [8] was already of good quality, but the recent use of PacBio long sequencing technology allowed a new high quality assembly (Matthew Conte, personal communication). The contig L50 length reached 3.09 Mb, and 50% of the genome is included in the largest 93 contigs. Importantly, over 86.9% of the reference genome sequence is anchored to chromosomes, enhancing the utility of the reference genome sequence for genetic analyses. The whole genome sequences of striped bass, white bass, yellow perch, and bluegill sunfish are at the stage of draft assemblies.

A published genome sequence exists for the Pacific oyster, but the assembly is highly fragmented [9]. Efforts are ongoing to improve the genome assembly and contiguity, completeness, and accuracy are significantly better now (Zhang, personal communication). Linkage analyses were conducted to validate the genome sequence assembly [14]. The whole genome assembly of eastern oysters is at the draft sequence stage. Several strategies were employed to address challenges encountered in the assembly of the Pacific oyster genome. A single, highly inbred individual, produced through multiple generations of inbreeding and one generation of meiotic gynogenesis (Guo, personal communication) was used as a template. PacBio sequencing was used to provide 50x genome coverage in addition to Illumina sequencing ([15]; Gomez-Chiarri, personal communication). Initial statistics suggest this assembly is of much higher quality than that of the Pacific oyster. The draft assembly (Table 2) is now being validated using high-density linkage maps generated by the Guo laboratory.

The Pacific shrimp genome has been sequenced and assembled, but is not yet published. As shown in Table 2, the genome assembly is of high quality, with a contig N50 of 57.1 kb. The whole genome is included in 6007 scaffolds. Importantly, 71.6% of the genome sequence is anchored to chromosomes through linkage mapping (Xiang, personal communications). Of all the aquaculture genomes, the shrimp genome is perhaps the hardest to deal with because of the difficulty in isolating high molecular weight DNA due to enhanced DNase activity, the large chromosome number, and high levels of heterozygosity and repetitive elements. Physical mapping has been hindered by the lack of BAC libraries with very large inserts. The only BAC library of shrimp, pECBAC1, has an average insert size of approximately 101 kb [16].

For XY heterogametic species, often only the homogametic gender was used as sequencing template, and so information on the sex chromosomes is lacking. For instance, the catfish genome sequence was produced using a doubled haploid female produced through gynogenesis, and therefore the Y sex chromosome was not sequenced. Similarly, the Atlantic salmon genome was produced by using DNA template from a single double-haploid female produced by mitotic androgenesis. Therefore, the Y chromosome is not included in the reference genome. The rainbow trout genome was sequenced using a YY doubled haploid. While it provided Y chromosome information, the X chromosome was not covered in the reference genome sequence. Furthermore, sex determination in some fish and shellfish is complicated by having multifactorial sex determining mechanisms, including genetic sex determination (GSD), environmental sex determination (ESD) and their interactions. With WZ heterogametic species like some of tilapia species, sequencing a single representative of each gender may not be sufficient if there is a polygenic sex determination.

Genomic variations, polymorphic markers, and genotyping platforms

Catalogues of genome variations and efficient genotyping platforms are essential to fully exploit whole genome sequences. One of the most useful by-products of whole genome sequencing is the development of thousands of DNA markers. In the first decade of aquaculture genome research, major effort was focused on developing polymorphic markers [17]. As whole genome sequencing projects were conducted, large numbers of polymorphic markers were identified. Whole genome sequencing with diploid sequencing templates allows identification of both microsatellites and SNPs. Analysis of SNPs between the two alleles of the sequenced individual also allow a rough assessment of the level of heterozygosity of the species.

Linkage mapping and physical mapping

Ultimately, genomic information must be translated into genetic terms to facilitate genetic enhancement in aquaculture. Genetic linkage maps derived from genetic analysis of recombination during meiosis are important for the assembly of chromosome-scale sequence scaffolds. Mapping of sequence-tagged genetic markers derived from the reference genome allows sequence contigs to be arranged in an order that corresponds to the linkage group or chromosome. In addition, linkage mapping is a good method for validating reference genome assemblies.

The major issue for linkage maps of aquaculture species is resolution. While the number of markers on the high density SNP arrays is large, map resolution has been limited by the size of the mapping populations. In most cases, the number of samples used for genetic mapping was not very large, leading to a high level of marker stacking. The exceptions are the Atlantic salmon and rainbow trout where over 2000 and 5000 individuals, respectively, were used for linkage analysis, leading to a very high resolution of the linkage map [6, 26]. While the high fecundity of fish and shellfish species makes it possible to generate large mapping families, the major limitation for high resolution linkage mapping is funding, as genotyping costs are directly proportional to the sample sizes in linkage analysis.

Transcriptome resources

Proper annotation of the genome sequences presents a challenge that can be at least partially overcome with transcriptome information. Specifically, gene models and gene structures need to be supported by experimental data; exon-intron borders need to be defined; alternatively spliced and differentially polyadenylated transcripts need to be identified and their translated proteins verified; and expression and function of the genes need to be studied. In addition to protein-coding genes, non-coding RNAs need to be identified and mechanisms of their target interactions need to be understood.

Large RNA-Seq datasets have been generated by various institutions for important aquaculture species in the United States (https://www.ncbi.nlm.nih.gov/sra) to characterize differentially expressed genes in response to disease or stress in catfish [37–40], disease in salmon [41, 42] and to identify markers associated with growth, heat stress, and disease and tissue specificity in rainbow trout [43–46]. In striped bass, RNA-Seq studies focused on reproduction traits and egg quality [47–49], while in tilapia, they were conducted to identify genes responsive to alkalinity stress [50], salinity adaptation [51], and adaptation to low or high fat diets [52]. In yellow perch and bluegill, RNA sequencing of neo-males (perch), neo-females (bluegill), regular males and regular females is being conducted to investigate epigenomic modification of SSD and sex determination in fish (Wang, personal communication). RNA-Seq studies have also been conducted to characterize the Pacific oyster response to environmental stress (e.g., temperature, salinity, air exposure and heavy metals) [9, 53–55] and Ostreid herpesvirus [52]. In eastern oysters, RNA-Seq studies identified genes associated with osmoregulation [12], characterized the transcriptomic response to a bacterial pathogen [56], and revealed extensive expansion of gene families associated with innate immunity [15, 57]. In shrimp, genes associated with early development [58] and resistance to Taura syndrome virus (TSV) [59] have been identified via RNA-Seq analysis, and improved shrimp transcriptome were reported [60]. When coupled with genetic analysis such as bulk segregant analysis (e.g., [38, 43]), transcriptome analyses using RNA-Seq will enable the identification of candidate genes for important aquaculture traits.

Transcriptome resources also empower proteomics analysis [48, 61, 62]. Proteomics offers great promise for advancing our understanding of the functions of genes that underlie important production traits, however these methods rely on existing homologous protein-coding sequence databases, which remain incomplete for many non-model organisms, including important aquaculture species. Tandem mass spectrometry approaches in proteomics use these databases to identify protein fragments by mass spectrometry and thus require amino acid (or protein-coding nucleic acid) sequence information, optimally from the research organism under investigation. Thousands of different proteins have already been identified and measured with tandem mass spectrometry approaches to answer important questions about reproduction in striped bass and the closely related white perch, which serves as a research model [61–64]. A similar proteomic approach identified important proteins related to muscle atrophy in rainbow trout [65].

Non-coding transcripts, regulation of genome expression, and epigenomics

Despite their importance in regulating gene expression, non-coding transcripts are much less understood than protein-coding transcripts in aquaculture species. Limited work has been conducted in this relatively new area of research. Among aquaculture species, most of the work on non-coding RNAs was conducted in rainbow trout. A few studies were devoted to identification of microRNAs and long non-coding RNAs [66–71]. In a number of cases, microRNAs were found to be associated with performance traits. For instance, a large number of microRNAs were differentially expressed between sexually mature and immature fish; in association with egg quality and muscle growth and quality [72–74]. In addition, differential expression of long non-coding RNAs studied in three genetic lines of rainbow trout identified important long-coding RNAs in response to infection with Flavobacterium psychrophilum [75].

In Atlantic salmon, several studies were conducted to characterize the microRNA repertoire. In one study, Bekaert et al. [76] identified 888 microRNA genes. In another study, Andreassen et al. [77] identified a total 180 distinct mature microRNAs, and found that many microRNAs were conserved across species, and a few microRNAs were expressed in a tissue-specific fashion. In another study, Kure et al. [78] found that 18 microRNAs were differentially expressed upon exposure to acidic aluminium-rich water.

Research on non-coding RNAs in catfish, striped bass, tilapia, oysters, and shrimp is limited. For instance, residue microRNA profiling was reported in catfish [79–81], tilapia [82], oysters [83, 84], and shrimp [85, 86]. However, now with the high quality reference genome sequences, it is expected that large numbers of projects will be conducted with aquaculture species in this area. This aligns very well with the FAANG (Functional Annotation of Animal Genomes) Project. As the importance and detailed operational protocols are well discussed in the white paper published in Genome Biology [87], we will not repeat them here, but this will be an important area for future research with aquaculture species as well, especially those with a well assembled reference genome sequence.

Genome scale analysis of epigenetic regulation have been conducted with oysters [88–93], Atlantic salmon [94], rainbow trout [95, 96], and tilapia [97], yellow perch, bluegill (Wang, personal communication) and additional projects are being initiated in several other major aquaculture species. Apparently, this is an area of active research, and functional annotation of non-protein coding genome elements is an important area. Again, this aligns well with those objectives of the FAANG Project [87].

Performance traits, phenotypic variations, and QTL analysis

The practical purpose of aquaculture genomics and genetics studies is to reveal the genetic basis of performance and production traits, and to use such information for genetic enhancement programs. Domestication of most aquaculture species is still in the early stages, occurring over the last few decades, compared to other food animals and crops which have been domesticated over hundreds or even thousands of years. Because of this short history of domestication, aquaculture species still segregate considerable genetic variation among strains, lines, families and individuals.

Many aquaculture phenotypes are complex and quantitative in nature. Therefore, a major goal of aquaculture genetics research is to leverage genome information to predict complex phenotypes. In aquaculture species, QTL mapping and GWAS analysis are well-established procedures for correlating genetic and phenotypic variation; however additional work is required to identify specific genetic variants responsible for phenotypic variations. The identification of the causal SNPs or the genes underlining the performance traits is not only important for aquaculture applications, but also important for understanding the molecular mechanisms of phenotypic expression.

Similarly, QTL have been identified for growth and reproductive traits, upper thermal tolerance, osmoregulation capacity, stress responses, and disease resistance in rainbow trout. Significant efforts have been devoted to the analysis of resistance to bacterial cold water disease (BCWD) [107–109]. QTL analysis and genome selection for BCWD resistance are facilitating significant genetic improvement for this trait in rainbow trout [3, 4].

Although QTL have been identified for disease resistance, viability and growth-related traits in eastern and Pacific oysters [110–113], low marker density limits QTL resolution. Candidate gene-based studies have led to the identification of variation in a serine protease inhibitor associated with Perkinsus marinus-resistance in the eastern oyster [114]. QTL analysis in tilapia, striped bass, and shrimp are at the early stages, but with the efficient genotyping systems, rapid progress is expected.

Aside from lack of genetic and genomic resources (e.g., inbred lines/families, sequenced genomes, efficient genotyping platforms) in some aquaculture species, several additional challenges face aquaculture researchers. First, unlike many livestock species where phenotypic and genotypic data can be collected on a large proportion of the cultured animals, phenotypic and genotypic data collection on the entire population of an aquaculture species is impossible. It is therefore essential that aquaculture geneticists understand QTL in all strains used in the industry, because a QTL present in one population may not be present in another. Second, fish and shellfish are outbred species with extremely large numbers of founders. Their high fecundities make QTL analysis within families extremely efficient, but whether the identified QTL are conserved across families, strains, and populations are unknown.

Genome-based technologies and regulatory framework

A number of technologies, including polyploidization, gynogenesis, androgenesis, sex reversal, gamete cryopreservation, and gene transfer, are still very useful for aquaculture breeding programs. There are opportunities for enhancing these technologies by using genomic information. At the same time, genomic research has generated new technologies that can be used for genetic enhancement of aquaculture species, including marker-assisted selection (MAS), genome selection (GS), and genome editing.

Marker-assisted selection has been successfully used in aquaculture. The best example of MAS in an aquaculture species is selection for disease resistance in Japanese flounder. A microsatellite locus, Poli9-8TUF, was mapped near the major QTL for resistance to lymphocystis disease. Additional analysis indicated that the disease resistance was controlled by a single gene, and that the resistance allele was dominant. Based on the marker linkage information, Fuji et al. [115] developed a new population of Japanese flounder using MAS with the marker Poli9-8TUF. They selected a female homozygous for the favourable allele (B-favourable) and a male with a higher growth rate and good body shape, but without the resistant allele as parents. All the progeny are heterozygotes with the resistance allele and entirely resistant to lymphocystis disease, while the control group without B-favourable alleles showed incidences of 4.5 and 6.3% of mortality due to lymphocystis disease. These results clearly demonstrate that MAS is an efficient strategy for breeding [116].

Another good example of MAS is the selection of IPN resistance in Atlantic salmon. One major QTL was mapped to linkage group 21, which accounts for 29% and 83% of the phenotypic and genetic variances, respectively. Three microsatellite markers were tightly linked to the QTL, and these markers have been used for the selection of IPN resistance [99]. Recently, the gene responsible for IPN resistance was identified as a cadherin expressed in the epithelium where the protein binds to IPNV virions [101]. Marker-assisted selection allowed production of IPN-resistant salmon, leading to a 75% reduction in the number of IPN outbreaks in the salmon farming industry [101].

Sex identification using sex markers is a special case of MAS. Sex markers have been developed and used in quite a few aquaculture species, including common carp [117], tilapia [118], catfish [119], zhikong scallop [120], half-smooth tongue sole [121], white shrimp [122], kuruma prawn [123], yellowtail [124] and rainbow trout [125]. These sex-linked markers have been useful for the identification of sex without phenotypic data.

Recent advances in genome analysis including the availability of a large number of polymorphic markers, highly efficient genotyping platforms such as SNP arrays, and the application of next generation sequencing technologies, allowed mapping of dense markers across the entire genome, which in turn enables an estimation of the genetic merit of every chromosome fragment contributing variation in a population with phenotypic observations. Not only can the merit of every chromosomal segment be estimated, but also all the traits of interest can be estimated simultaneously. Whole genome selection is based on estimating the value of every chromosomal fragment contributing variation in a population with phenotypic observations (Training), and then the results of training are used to predict the merit of new animals (Testing) that are not included in the training dataset.

Genome selection was first proposed by Meuwissen et al. [126]. Since then it has gained tremendous attention in the animal genetics community. Compared with MAS, genomic selection uses the estimated effect of many loci across the entire genome at once, not just the small number of linked loci as done with MAS. Although genome selection has been successfully used in dairy cow and beef cattle and other livestock species [127], its use in aquaculture species has been limited to just a few species [128, 129]. In rainbow trout, genome selection was carried out for the selection of bacterial cold water disease [3]. In Atlantic salmon, genome selection was used to predict breeding values for resistance to sea lice [130]. Although demonstrated to be effective, genome selection has not been commercially applied in aquaculture species primarily due to financial limitations.

Supervised machine learning is similar in concept to whole genome selection using Training and Testing datasets, and includes Support Vector Machines (SVMs) and Artificial Neural Networks (ANNs). These are systems that can be trained to recognize certain data input patterns and then can be used to predict outcomes or classify data. Machine learning has been used to classify transcriptome and proteome data by pattern recognition (expression “fingerprinting”) in an analytical bioinformatics approach [49]. Expression patterns of genes and proteins can be modelled to identify the most important ones contributing to a trait or response. Machine learning ANNs have been used to analyze tens of thousands of expressed genes in microarray and RNA-Seq studies to show that the collective changes in the expression of 233 ovary genes (less than 2% of the genes measured) explained over 90% of the variation in striped bass embryo survival [47, 49]. These trained ANNs also predict, with a correct classification rate over 80%, which female striped bass will produce fertile or infertile eggs based on gene expression profiles of ovary tissues sampled prior to ovulation. Additionally, SVMs have been used to model the striped bass ovary proteome (355 proteins) and this system can predict the specific ovary growth stage with 83% accuracy based on quantitative tandem mass spectrometry data [61]. A portion of the plasma proteome (94 proteins) also has been similarly modelled to accurately predict gender of white perch [131]. Therefore, machine learning additionally poses a potential use as a diagnostic tool, for example in identifying those females that will produce poor quality eggs, or determining reproductive state or gender. Future applications of machine learning could include modelling genomic markers, such as SNPs, to identify those most important to a particular trait and then to predict the future performance of an individual based on the presence or absence of those SNP markers.

Genome editing refers to the ability to make specific changes at targeted genomic sites [132]. With the initial zinc finger nuclease (ZFN) technology developed in 1996, genome editing technologies have evolved and become more and more efficient, with the development of TALEN (transcription activator-like effector nucleases) and CRISPR/Cas9 (clustered regulatory interspaced short palindromic repeats). These new genome editing technologies overcome the disadvantages of ZFN technology and they have become very efficient for the modification of genomes. CRISPR/Cas9 has been demonstrated to be very efficient in zebrafish [133, 134], tilapia [135] and catfish (Liu, unpublished data). Mutation rates of 70–100% can be achieved with very low levels of mosacism in channel catfish [136].

Genome editing technologies can be used to introduce an immediate improvement in a phenotype in a single generation; hence, these technologies hold great promise for improving aquaculture. However, genetically modified organisms (GMO) have encountered low public acceptance, especially with aquaculture species. As demonstrated with the lengthy approval process of AquAdvantage transgenic Atlantic salmon, decades of time and millions of dollars were spent in coping with the regulatory issues ([137]; Hackett, 2016, personal communications). It could be argued that genome editing technologies differ from traditional gene transfer technologies because no foreign DNA is introduced. The scientific community must be proactive of research in the area of regulatory issues and public perception. Escaping the GMO label is possible with genome editing. For example, recently, the USDA decided that a CRISPR-modified mushroom can be cultured and sold without passing through the agency’s regulatory process [138].

Leveraging Investments in Genomics through Integration with Germplasm Repositories

Rapid development and adaptation of genomic tools among various aquatic species is a double-edged sword in terms of how such tools may create genetic diversity and thereby limit industry options in the future. Evidence of such contractions have been demonstrated with livestock and in particular the Holstein cow and how gene banks can facilitate the alleviation of genetic bottlenecks [139, 140]. While genomic research continues to rapidly proceed among various aquaculture species, there are some major technological gaps preventing the aquaculture sector from securing and utilizing improved genetic resources. As shown in other life forms such as livestock species, there is a critical need to understand and acquire genetically diverse samples from all major aquatic species, cryopreserve those samples, and to present them in publically available databases for viewing of information about the sampled populations. Such information would include phenotypes, management system descriptors, environmental conditions, locality data, and comprehensive genomic information. The Animal-GRIN information system operated by USDA/ARS is designed in this manner and is publically accessible via the internet. Acquiring and integrating this wide range of data not only serves to make germplasm and tissue samples more useful in the present, it will also allow researchers to perform studies not foreseen today and to respond to future challenges such as disease outbreaks or losses of critical genetic diversity in cultured lines. Viewing of genetic resources (via germplasm) and its associated detailed information as a public resource serves to speed innovation, as well as to leverage the considerable investments being made in genomic research. In essence this affords us new and more cost-effective approaches to produce, maintain, and distribute genetic improvement across the breadth of cultured aquatic species [141].

To respond to these needs, there is a requirement to collect and cryogenically store gametes and tissues from a wide range of species that can be used by industry members and public researchers alike. Coupled with these samples should be the ability to store genomic information from publicly funded research, as well as from industry. Such an information system would link samples with genomic, phenotypic, locality (GIS-based), and environmental descriptors and make this information publically available through a user interface via the internet. Aquatic species researchers could use this resource for varied experimental purposes (e.g., of crossing spring and fall spawning populations) and for corrective mating. As such, the collection and curation of germplasm or tissue samples has value, just as does the determination of genomic information. It is the purposeful integration of these genetic and informational resources that provides a synergistic leveraging or expansion of value and potential utility. Indeed, the value of information or germplasm samples is directly magnified by their coupling or association in a comprehensive repository system.

Future research priorities

Economically important aquaculture species are a diverse group of organisms and research priorities vary depending on the unique biology of each species. Although fishes are the most diverse vertebrate group, aquacultured teleosts are similar enough phylogenetically and biologically that they can follow a similar research program. Invertebrates are not as uniform, and may each have special properties that require different approaches.

The research tasks needed to develop a program of genetic enhancement in aquaculture species can be divided into two phases (Fig. 1). The first phase, development of species-specific genomic resources, is the one that has been pursued for the major aquaculture species over the past 20 years. It includes the development of genetic and physical maps, annotated genome sequences, and platforms for high-throughput genotyping. Some species (e.g., catfish, tilapia, rainbow trout, and salmon) may be nearing the completion of Phase I. Other species are just beginning this phase, hopefully benefitting from the experience of other species, and taking shortcuts available with new technologies. From the perspective of genetics and breeding, generation of complete sets of heritabilities and genetic correlations is also needed.

Within Phase I, we can make a distinction between development of resources and application of resources to commercial aquaculture. For instance, genetic maps enable QTL/MAS, and genome sequences enable genomic selection. Application of MAS/genomic selection is generally beyond the scope and funding of the academic laboratories that have participated in the development of the genomic tools that enable it, although not beyond the scope and mission of government laboratories and academic laboratories that take on genetic stock enhancement for smaller or regional aquaculture species.

In Phase II, the genomic resources developed in Phase I can be used to develop a functional understanding of animal systems. As an example, a number of laboratories are working to develop an understanding of the gene regulatory network underlying sex determination in fishes. The research program involves not only RNA-Seq to characterize patterns of gene expression in the developing gonad, but also CRISPR modifications to test linkages in the gene regulatory network model. It should be possible to develop an understanding of the gene network underlying sex determination that will be broadly applicable among aquaculture species. Similar research programs are underway to understand the genetic basis of other important traits including growth, disease resistance, etc. In each case, the goal is to develop an understanding of animal systems that can be easily transferred to related species.

When this more detailed understanding of animal systems is complete, it will become possible to make specific genetic modifications (e.g., using CRISPR) to improve animals for commercial production. The safety and effectiveness of such modifications are important topics for research, but commercial application of these technologies will require stable business models and well established regulatory frameworks that ensure the safe application of these technologies to the species being targeted, the public, and the environment.