Evaluating new species for aquaculture: A genomic dissection of growth in the New Zealand silver trevally (Pseudocaranx georgianus)

Abstract Aquaculture is the fastest‐growing food production sector worldwide, yet industry has been slow to implement genomic techniques as routine tools. Applying genomics to new breeding programmes can provide important information about pedigree structure and genetic diversity; key parameters for a successful long‐term breeding programme. It can also provide insights on potential gains for commercially important, yet complex, quantitative traits such as growth rate. Here we investigated a population of 1100 captive‐bred F1 silver trevally (Pseudocaranx georgianus), a promising new species for New Zealand aquaculture. We used whole‐genome information, coupled with image‐based phenotypic data collected over two years, to build the pedigree of the population, assess its genetic diversity, describe growth patterns of ten growth traits and estimate their genetic parameters. Successful parentage assignment of 664 F1 individuals showed that the pedigree consisted of a complex mixture of full‐ and half‐sib individuals, with skewed reproductive success among parents, especially in females. Growth patterns showed seasonal fluctuations (average increase across all traits of 27.3% in summer and only 7% in winter) and strong inter‐family differences. Heritability values for growth traits ranged from 0.27 to 0.76. Genetic and phenotypic correlations between traits were high and positive, ranging from 0.57 to 0.94 and 0.50 to 1.00 respectively. The implications of these findings are threefold: first, the best on‐growing conditions are in warmer months, where highest growth peaks can be achieved; second, size‐ and family‐based selection can be used as early selection criterion if pedigree structure and inbreeding risks are closely monitored; third, selection for body length results in concomitant increases in height and weight, traits of paramount importance for aquaculture. It is concluded that there is substantial potential for genetic improvement of economically important traits, suggesting that silver trevally is a promising species for selective breeding for enhanced growth.

F 1 individuals showed that the pedigree consisted of a complex mixture of full-and half-sib individuals, with skewed reproductive success among parents, especially in females. Growth patterns showed seasonal fluctuations (average increase across all traits of 27.3% in summer and only 7% in winter) and strong inter-family differences.
Heritability values for growth traits ranged from 0.27 to 0.76. Genetic and phenotypic correlations between traits were high and positive, ranging from 0.57 to 0.94 and 0.50 to 1.00 respectively. The implications of these findings are threefold: first, the best on-growing conditions are in warmer months, where highest growth peaks can be achieved; second, size-and family-based selection can be used as early selection criterion if pedigree structure and inbreeding risks are closely monitored; third, selection for body length results in concomitant increases in height and weight, traits of paramount importance for aquaculture. It is concluded that there is substantial potential for genetic improvement of economically important traits, suggesting that silver trevally is a promising species for selective breeding for enhanced growth.

K E Y W O R D S
aquaculture, Carangidae, genetic diversity, growth traits, heritability

| INTRODUC TI ON
Aquaculture is the fastest growing food production sector worldwide (FAO, 2018). However, compared to terrestrial breeding programmes, the industry has been slow to implement the use of genomic information to inform selective decisions. This is despite the proven utility of genomics to provide insights into important commercial traits, such as faster growth and disease resistance (Gjedrem et al., 2012).
One of the most important and well-studied traits in breeding programmes is growth rate (e.g. Ashton, Hilario et al., 2019;. This is because growth is easily quantified and often highly heritable and has a pronounced impact on commercial returns. In particular, improvements in growth rates reduce the overall time needed to raise individuals to market size, thereby speeding up the turnover of production stock and decreasing costs (Gjedrem, 2005;Ye et al., 2017). Usually, growth rate in animals is measured using either body weight or body length as a proxy, but as phenotyping methods improve, other traits such as width, girth and height are starting to be used to provide additional insights (Zenger et al., 2017). Image-based phenotyping methods provide rapid and efficient measurements for multiple traits simultaneously. Moreover, datasets collected provide researchers with the opportunity to measure additional traits a posteriori, using computer vision based data mining at a later stage (e.g. body shape, health and colour) (Zenger et al., 2019).
While growth is of significant importance to aquaculture breeding programmes, genomic-assisted selection for this trait is complicated by its typically polygenic basis (Wellenreuther & Hansson, 2016) and consequent requirement of large numbers of molecular markers. Genome-wide markers have been used to identify loci associated with traits of importance and to guide selective breeding to efficiently increase phenotypic gains, but the application of genomic approaches has long been limited to model species or species of high economic value. Recently, the decrease in DNA sequencing and genotyping costs has meant that large numbers (e.g. thousands) of genome-wide markers can be easily generated for almost any species, at a cost that can be afforded by research groups (Bernatchez et al., 2017). Reduced representations of the genome, such as in genotyping by sequencing (GBS) (Elshire et al., 2011), offer a quick and cost-effective way of genotyping a large number of individuals.
Whole genome sequencing (WGS) can be used to generate a reference genome, particularly when combined with long read sequencing techniques and additional techniques, such as Hi-C.
Genomic insights can inform a number of breeding decisions.
First, a potential barrier for new breeding programmes is the lack of prior knowledge on key parameters such as the relatedness among individuals and number of contributing parents (Falconer & Mackay, 1996). This information is particularly challenging to gather for aquatic species where the founding individuals might be caught from the wild and/or exhibit mass spawning reproductive behaviours. If not carefully monitored, lack of control can lead to inbreeding pressure and early loss of genetic diversity, which must be minimized for a successful long-term breeding programme (Wang et al., 2002). Second, a useful parameter to asses is the selection potential of a trait. Estimates of trait heritability provide insights into the relative contribution of genetic variance compared with environmental variance. Knowing if the heritability of a trait is high (~0.9) or low (~0.1) will indicate how well the trait will respond to selection . Broad-sense heritability is defined as the ratio of the total genetic variance on the total phenotypic variance and includes effects such as dominance and epistasis, which do not respond to selection. Narrow-sense heritability reflects the ratio of the additive genetic variance on the total phenotypic variance and represents the part that will respond to selection . Third, another application of genomic tools for predicting the response of a trait to selection is the estimation of genetic correlations between two or more traits (Lynch, 1999). Because the genes that contribute to traits can be highly genetically correlated and co-inherited, it is particularly important to know the magnitude and direction of these correlations when predicting breeding values. The improvement of multiple traits simultaneously may seem efficient but, because of unexpected correlations, it could result in the co-selection of undesired phenotypes (Falconer & Mackay, 1996).
In Aotearoa New Zealand, aquaculture production relies almost exclusively on the farming of the three species: Greenshell™ mussels (Perna canaliculus); Pacific oysters (Crassostrea gigas); and only one finfish species, chinook/king salmon (Oncorhynchus tshawytscha), a species introduced from North America (Camara & Symonds, 2014;Davies et al., 2019). The paucity of species means that there is a strong need to add resilience to the sector by diversifying the range of species farmed. One possible candidate for commercial aquaculture is silver trevally (Pseudocaranx georgianus, Cuvier 1833) (referred to as trevally hereafter), a shoaling pelagic species found throughout the coastal waters of southern Australia and around New Zealand (Gomon et al., 2008;Smith-Vaniz & Jelks, 2006). Indigenous Māori people have a strong cultural connection to trevally, where it is considered as taonga (i.e. has value or is treasured) and is referred to as araara. In many regions, trevally is a major component of recreational and commercial fisheries (MPI, 2021). Pseudocaranx is also a genus that appears to do well in farm-like conditions, as demonstrated by the fact that other species in this genus have already been successfully developed for large-scale commercial farming in Asia (FAO, 2018). Furthermore, this genus belongs to the carangid family, which contains other species with well-established breeding programmes around the world, for example, for the yellowtail kingfish (Seriola lalandi).
In this study, we applied genomic tools to a recently assembled population of silver trevally (Pseudocaranx georgianus, Cuvier 1833).
The goal of this study was to use genome-wide data coupled with extensive phenotypic data to provide the first quantitative assessment of growth traits in a population of 1100 captive-bred F 1 trevally.
Specifically, we used a mixture of WGS and GBS data to (1) reconstruct the molecular pedigree of the population, (2) calculate inbreeding values within each generation, (3) describe the growth patterns by adding image-based phenotyping data collected over a period of two years to and finally (4) estimate the heritability and genetic correlations of ten growth traits to determine their potential for enhanced growth performance through a selective breeding programme.

| Populations studied and holding conditions
The trevally population studied consisted of an F 0 wild broodstock (n = 22) and a captive reared F 1 population (n = 1100).

| Tissue sampling, DNA extraction and library preparations
Thirteen surviving F 0 were tagged and fin-clipped in January 2017.
Fin clips were placed directly into chilled 96% ethanol, heated to 80°C for 5 min within 1 h of collection, and then stored at −20°C until needed. Total DNA was extracted as described by Ashton, Hilario et al. (2019),  with the following modifications: proteinase K digestion time was increased to 1.5 h and the 80°C inactivation step was omitted; the RNA removal was performed after the salting-out step; and the DNA was quantified by fluorescence using the Qubit High Sensitivity dsDNA kit (Thermo F 1 individuals were collected during the first phenotyping round (November 2017) and stored as described above. Total genomic DNA was extracted by SlipStream Automation using the same protocol as for the F 0 . Genotyping for the F 1 was carried out using a modified GBS approach (Elshire et al., 2011;Hilario, 2015). DNA integrity was checked by capillary electrophoresis (High Sensitivity genomic DNA kit), Fragment Analyzer (Advanced Analytical). One microgram of total genomic DNA was used for digestion with restriction enzymes.
A double digestion was performed with Pst I and Msp I by incubation at 37°C for 3 h, the adaptor ligation step omitted drying out the DNA/adaptor mixture. The barcoded adaptors were associated with the Pst I cut sites and designed by Deena Bioinformatics. Adaptors were annealed according to Ko et al. (2003). A high-fidelity enzyme was used for amplifications (AccuPrime Taq DNA polymerase High Fidelity, Life Technologies). Amplification, quality check and clean up were done separately before pooling samples. A total of 12 pools of 96 samples each were prepared and sent to AGRF for sequencing on a HiSeq 2500 platform (single-end, 100 bp reads).

| Genotyping data quality checking and processing
Sequencing data quality for both F 0 and F 1 generations were checked using FastQC v0.11.7 (Andrews, 2010). As the F 0 and F 1 sequence data were generated using different sequencing technologies, different filtering parameters were used.

| F 1 filtering/pre-processing
The F 1 samples were de-multiplexed from the 12 sequencing libraries using the process_radtags module available in the STACKs v2.1 pipeline (Catchen et al., 2013), and the reads were trimmed using Fastq-mcf in ea-utils v1.1.2-806 (minimum sequence length = 50, quality threshold causing base removal = 33) (Aronesty, 2013). Like for the parents, read groups were also added and the bam files were sorted, indexed and aligned to the reference using BWA-mem. The variant calling was also done using samtools and BCFtools. Using VCFtools, indels and individuals with over 50% missing data were removed, the maximum depth was set at 8000 and SNPs with over 20% missing data were filtered.

| Recombining datasets
After indexing, a list of common SNPs between F 0 and F 1 was obtained using BCFtools isec and the two sets were merged using VCFtools vcf-merge. Finally, the resulting F 0 and F 1 dataset was filtered to keep only SNPs in common using VCFtools −positions.

| Pedigree reconstruction and F 0 sex prediction
Sequoia v2.0.7 (Huisman, 2017) was used in the R statistical environment (version 3.2.3) (R Core Team, 2013) to iteratively reconstruct a maximum-likelihood pedigree. To prepare the common SNP dataset for pedigree reconstruction, PLINK v1.9 (Purcell et al., 2007) was first used to test for and discard loci in linkage disequilibrium (LD) with the --indep function, evaluating 50 SNP windows, five SNPs at a time, with a variance inflation factor (VIF) cut-off = 1.5 and a minor allele frequency (MAF) of 0.4 was set for the population. The initial parentage assignment was accomplished with this genotype file and a life history file, using the parameters MaxSibiter = 0, Err = 0.05, MaxMismatch = 10, MaxSibshipSize = 900 and Tassign = 0.5. This allowed to scan the pedigree for obvious errors, as well as for duplicates that were accidentally retained. To construct the full pedigree, the parameter data frame (= Specs) was then altered to use the initial parentage assignment as prior information and MaxSibIter = 3, MaxSibshipSize = 900, Err = 0.2, Tfilter = −2, Tassign = 0.5. All other parameters were kept as default. We assessed the accuracy of the reconstructed pedigree and the ability of the SNP data set to correctly identify familial relationships by checking for Mendelian errors using PLINK.
Broodstock sex prediction based on the molecular pedigree was confirmed a posteriori by collecting a gonadal biopsy from each of the 13 F 0 individuals in December 2018. In brief, broodstock in the advanced stages of reproductive development were mass-sedated in tank (25 ppm Aqui-S; Aqui-S New Zealand Ltd). A gonadal biopsy was taken by inserting a glass cannula (Natelson tube, 3 mm outside diameter) connected to a plastic tubing into the gonopore of the fish and applying gentle aspiration by syringe. A portion of biopsy sample was placed in Ringer's solution (180 mM NaCl; 4 mM KCl; 1.5 mM CaCl 2 ; 1.2 mM MgSO 4 ; 3 mM NaH 2 PO 4 ; 12.5 mM NaHCO 3 -pH 7.5), and from this, a wet mount slide was prepared and examined under a compound microscope for the presence of oocytes or sperm.

| Inbreeding calculations
Using the same set of filtered SNPs used for the parentage assignment, inbreeding values for each individual found with both parents in the pedigree were calculated with a method-of-moments F coefficient (F H ) using PLINK. This statistic is equal to Nei's F IS statistic, but is calculated using a different formula: where O(Hom i ) is the observed number of homozygous loci for the ith individual, and E(Hom) is the Hardy-Weinberg expected mean number of homozygous genotypes across m loci (Kardos et al., 2015). The distribution of inbreeding values was then visualized using the ggplot2 library in R. A Welch two-sample t test was used to compare mean inbreeding values between F 0 and F 1 . To compare mean inbreeding values between families, a linear mixed model was fitted with family as a random effect to run an ANOVA. A post hoc Tukey correction for multiple comparisons was used to find which family values were significantly different from the others.

| Phenotyping, trait estimations and phenotypic correlations
In total, eight sets of external images were taken for each F 1 individual, corresponding to a measurement roughly every four months during a period of two years. Images were taken on a custom-built imaging rig using fitted Panasonic Lumix DMC-GH4 cameras. Images were analysed using PFR's Morphometric Sotfware™ (https://www. plant andfo od.co.nz/page/morph ometr ic-softw are-home/). The software extracts the outline of each individual fish from images, locates the XY coordinates of morphometric features on the outline (e.g. upper lip and narrowest cross section of the tail) and then uses those coordinates to make measurements. The measurements were converted from pixels to mm using the length of rulers also present in the images.

| Trait heritability and genetic correlations
Variance and covariance components were estimated using linear mixed animal models and restricted maximum likelihood methods with ASREML version 4.0 (Gilmour et al., 2015) in R. Narrow-sense heritability of each trait was estimated using a univariate analysis, modelled as follow: where is the population mean, a i is the breeding value and e i is a residual term (Galwey, 2014). The heritability models were run separately where X and Z are matrices and y, u and e are vectors (Thompson et al., 1995).

| Filtering allowed to increase quality of calls tenfold in offspring
A total of 1.23 billion DNA sequence reads were generated for the  (Table   S1). Quality was high across the full length of the reads for all F 1 plates, except towards the end of plate 5, where quality was slightly reduced ( Figure S1). Twenty-three F 1 sample extractions did not yield enough DNA for sequencing. A further 29 out of 1077 offspring libraries failed to be sequenced, probably due to low quality DNA.
The initial variant calling yielded 20.8 and 2.1 million markers for the F 0 and F 1 , respectively ( Figure S2A-D). After removing indels, 17.8 million sites were kept for the F 0 and 1.8 million for the F 1 . Fortyseven offspring were subsequently removed based on missing data ( Figure S2E,F for the parents and the offspring, respectively ( Figure S2G,H). The last filter for maximum depth in the parental call retained 16.9 million SNPs. These filtering steps resulted in a total of 171,923 SNPs shared between the F 0 and F 1 generations. Filtering enhanced largely the F 1 dataset, going from 68% missing data to 0.06%. The parental dataset went from 0.01% to 0% (Table S1).

| Reconstruction of the pedigree showed skewed parental contributions
A subset of 1525 SNPs was used for the parentage assignment.
Pedigree reconstruction allowed to determine sex for all F 0 individuals, which was further confirmed by biopsy. Visualization of the pedigree showed that out of 13 sequenced F 0 , 10 individuals participated in the spawning (three females and seven males), generating 21 families (Figure 2a). Both parents were identified for 63% (664) of the individuals in the F 1 population. The remaining F 1 had either one (31.4%) or both parents (3.6%) not genotyped. Although the mating ratios were equal among all females (1:7) and males (1:3), the contributions were skewed among both sexes, particularly so in females, with one of them contributing to up to 60.2% of the total F 1 population ( Figure 2b).

| Unbalanced inbreeding levels detected within families
The inbreeding F statistic (F H ) was calculated for 13 F 0 and 664 inbreeding values ranged from −0.02 to −0.11 except for families having female 8 as mother, which had higher average F H values, ranging from 0.06 to 0.14 ( Figure 3b). The mean inbreeding values for fam11_5, 12_2, 12_4 and 12_5 were significantly lower (ρ < 0.01) than those of other families (Table S2). Families with female 8 as a mother had a sample size too low (n = 1 to n = 5) to conclude if the means were statistically significantly different from those of the other groups.

| Growth pattern in trevally is seasonally influenced and is under family effects
For 2 years, phenotyping of the F 1 trevally cohort occurred roughly every 4 months, from the age of 2.1 to 4.1 years old. Ten growth traits that could either be directly extracted from the images (H25, High variation was observed in all traits ( Figure S3). Between the first and the last measurements, standard deviation less than doubled for all traits but increased fourfold for EW and ΔEW (Table   S3). Coefficients of variation (CV) of height, length and weight traits F I G U R E 3 Inbreeding scores for trevally individuals based on (a) generation, and (b) family. Visualized are the 1st, 2nd (median) and 3rd quartile and whiskers extending 1.5 times the interquartile range from the median (95% confidence interval). The coefficient of inbreeding used is Nei's F IS and significant differences between groups are shown in Table S2. No significant differences (p-value > 0.01) were found between the F 0 and F 1 generations. Significant differences (p-value < 0.01) were found between different families F I G U R E 4 Influence of seasonal temperature on the growth rate of trevally over two years for peduncle length (PL), height at 25% of PL (H25), height at 50% of PL (H50), height at 75% of PL

| Genetic correlations, phenotypic correlation and trait heritability
The estimates of heritability, variances, covariances and phenotypic correlations between traits are reported in Table S4.
Based on Pearson's correlation coefficients, strong phenotypic correlations were observed between all height, length and weight traits throughout the experiment (0.85 ± 0.02 to 1.00 ± 0.00) and moderate to strong correlations were found for the net growth traits (and 0.50 ± 0.00 to 0.98 ± 0.01) ( Figure 5).
Strong positive genetic correlations were found between all measured height, length and weight traits throughout the experiment (close to unity, 0.94 ± 0.00 to 1.00 ± 0.00) and between all net gain traits (0.84 ± 0.07 to 1.00 ± 0.03). Although calculations of weights were made using length measurements, such high genetic correlations indicate that the alleles associated have a pleiotropic effect and influence multiple traits simultaneously, Correlation between measurements of height, length, weight and net gain traits were genetically moderate to strong (0.59 ± 0.10 to 0.99 ± 0.00).

| DISCUSS ION
Here, we applied, for the first time, a genomics-informed approach to study a captive trevally population in New Zealand. Data from different genotyping methods were pooled to reconstruct a twogeneration pedigree of 13 F 0 and 1100 F 1 and to investigate the inbreeding levels in each generation and within families. Phenotyping data collected over 2 years were added to assess the quantitative genetic architecture of 10 growth traits. The results of this study will support breeding efforts in trevally, as well as inform more generally genomic work on other teleost species.
Marker-based pedigree reconstruction enabled to determine the sex of the broodstock and showed that most individuals con- The initial level of genetic variation entering a breeding programme is set by the amount of heterozygosity of the founding population. It is crucial to monitor its subsequent loss in the next generations, as this will help to avoid inbreeding depression and a diversity bottleneck for a breeding programme. In our study, the wild F 0 parents represented the baseline for inbreeding statistics. No statistical differences were observed between the two generations, which were both slightly outbred (−0.08 and −0.06 on average for F 0 and F 1 , respectively). Although the average inbreeding rates did not vary between generations, some slight differences could be observed between families in the F 1 generation.
Values were similar to the ones found in other wild-caught marine species such as orange clown fish (Amphiprion percula) (0.018) (Salles et al., 2016), Australasian snapper (0.055) (Ashton, Hilario et al., 2019; (Pauly, 1980). Similar results have been found in other species such as in chinook/king salmon where the optimal growing temperature is 19.0°C (Perry et al., 2015), or in yellowtail kingfish, where optimal rearing temperatures were found to be around 26.5°C (Abbink et al., 2012), and Australasian snapper, which show increased growth rates at 21.0°C compared to 13.0°C . However, too high temperatures can also have a negative effect on growth rates once it exceeds a tolerance threshold, as shown in varied species of coral reef fishes (Munday et al., 2008).
Genomic-based pedigree allowed the estimation of narrowsense heritability for measured growth traits, which were consistent over time and higher than those reported in other studies; for example, heritability estimates for growth traits (weight or length) ranged from 0.26 (Whatmore et al., 2013) (Thorland et al., 2020). These estimates are population specific and can be inflated if traits are also influenced by non-additive genetic effects (such as epistasis and dominance) Visscher et al., 2008;. It is also to be noted that estimates are often biased upwards due to difficulties to separate environmental and non-additive genetic effects common to full-sibs from additive genetic effects when full-sib families are reared in separate tanks until tagging (Kause et al., 2005).

| Future directions and management implications
This study represents the first in-depth genetic investigation of reproductive success and growth rates for a captive trevally population in New Zealand. Genome-wide marker sets combined with reference genomes will be useful for a wide range of future applications. Investigating the basic genetic structure of a founding population yields fundamental insights into the biology of a species and is of primary importance when establishing a long-term breeding programme. Furthermore, understanding underlying genetic mechanisms of growth can support informed decisions about how to selectively breed species to fast track gains, while at the same time ensuring the long-term viability of the breeding programme as a whole (e.g. avoiding inbreeding).
In trevally, like in other teleost, seasonal effects have a significant influence on the realized growth rate. This finding has important implications for the selection of aquaculture locations and monitoring of optimal rearing temperatures; holding fish in a warmer environment could increase the growth rate and reduce the time taken to reach harvesting size. Our study showed that families exhibiting larger measurements initially, remained the largest individuals in the population throughout the duration of the experiment. Early removal of small individuals can thus be a useful hatchery-management method to maximize the realized growth potential overall. However, if there is no accompanying genetic management plan, it can result in a genetic bottleneck as often only a few families are selected, which increases the risk of inbreeding depression over several generations (Kincaid, 1983). The implementation of genomic information will enable selection decisions to be made on earlier measurements (e.g. March 2018 measure) while confidently preserving genetic diversity in the programme. Early rearing effects are common in land-based facilities, where populations are kept in separate or small tanks, and this can influence the estimates of heritability of commercially important traits. We found that the inclusion of the net growth for each trait can improve the estimate of heritability, particularly if environmental differences had an impact on early performance. Focusing selection based on the net gain traits could help reduce the impact of environmental effect on the heritability estimates that would otherwise be difficult to separate out.
The results of this research indicate that trevally have skewed parental contributions, which may be a consequence of uncontrolled tank-based spawning or subsequent family-specific mortality. A cohort mating strategy could be used to increase the number of parents contributing to even out the mating differences. Tank-based spawning of broodstock selected based on genotype data could help select less closely related individuals in the next generation. This approach, also called walk-back selection, can be mixed with additional steps that reduce skewed contribution, such as holding the broodstock in multiple tanks and standardizing the volumes of fertilized eggs from each tank. However, these techniques introduce environmental variation, which would need to be accounted for in downstream analyses. If walk-back selection were to be implemented, further research would also be needed to determine its effectiveness in controlling inbreeding over multiple generations.
The high heritability estimates found in this study highlight that there is potential for making strong genetic improvements via selective breeding. Phenotypic and genotypic correlations between traits were all positive and moderate to strong. A breeding objective for increased fish length is thus expected to produce a positive response on the other traits measured in this study, provided that they have the same allometric relationship later in life. Future work on this species could focus on a single trait such as length, as it can be easily measured using high-throughput phenotyping methods and has been shown to be highly genetically correlated with other commercially relevant traits. Indeed, the speed of phenotyping will become increasingly more important as the range of locations for image-based phenotyping grows wider. Using correlated traits like length and body shape from images to move into high frequency underwater environmentally linked measurements will enable more complete Genotype × Environment × Phenotype studies rather than single point Genotype × Phenotype studies in the future. Taken together, these results suggest that trevally is a suitable future candidate for enhanced growth.

ACK N OWLED G EM ENTS
We would like to acknowledge all the PFR staff who have been involved with breeding and husbandry for the trevally populations; in particular Warren Fantham who manages the fish breeding, and Therese Wells who manages the post-juvenile husbandry. We would also like to thank Matthew Wylie and Dafni Anastasiadi for their constructive comments on the manuscript. This project was funded through the MBIE Endeavour Programme "Accelerated breeding for enhanced seafood production" (#C11X1603).