Loading field map
The central challenge of aquaculture genetics is deceptively simple: how do you improve the growth rate, disease resistance, flesh quality, or environmental tolerance of a species when you cannot see the genes that control these traits, and when each generation takes months or years to produce? Unlike crop plants or laboratory model organisms, most farmed aquatic species lack deep genomic resources, have long generation intervals relative to research funding cycles, and are raised in environments that make precise measurement of individual performance difficult. The history of the subfield is therefore a story of steadily increasing resolution—from the statistical inference of hidden genetic merit, through the direct detection of DNA markers, to the genome-wide prediction of breeding values—but it is not a simple tale of one method replacing another. Each framework introduced new tools and assumptions, yet older approaches have persisted, been absorbed, or found new niches alongside their successors.
The first systematic framework for aquaculture genetics was built on the quantitative genetics tradition developed for terrestrial livestock. Selective Breeding and Genetic Improvement treated the traits of interest—body weight, fillet yield, survival—as polygenic, meaning they are influenced by many genes of small effect plus environmental noise. Breeders could not see individual genes, so they inferred an animal's genetic merit from its own performance and that of its relatives. The key statistical tool was the best linear unbiased prediction (BLUP) model, which used pedigree records to partition observed variation into additive genetic and environmental components. This framework was implemented at pioneering institutions such as Auburn University in the United States and the Norwegian salmon breeding program, where multi-generational selection produced dramatic gains: growth rate in Atlantic salmon increased by over 10% per generation. The core assumption was that additive genetic variance—the portion of heritable variation that responds to selection—was sufficient to drive steady improvement. This approach worked well for traits with moderate to high heritability, but it was blind to the molecular architecture underlying the response. Breeders could select the best fish without knowing which genes were changing.
By the 1990s, advances in DNA technology offered a way to look inside the black box. Molecular Genetics and Marker-Assisted Selection (MAS) aimed to identify specific regions of the genome—quantitative trait loci (QTL)—that had large effects on economically important traits. If a breeder could find a DNA marker linked to a QTL for, say, resistance to a viral disease, they could select carriers of the favorable marker without waiting for the fish to be exposed to the pathogen. This framework promised to accelerate genetic gain, especially for traits that were difficult or expensive to measure directly. Early successes came in species with well-characterized major genes, such as the MHC-linked resistance to infectious hematopoietic necrosis in salmonids. However, the broader application of MAS revealed a fundamental mismatch between its assumptions and the genetic architecture of most aquaculture traits. The majority of QTL mapping studies found that individual markers explained only a small fraction of the phenotypic variance—often less than 5%. The polygenic reality that the Selective Breeding framework had always assumed turned out to be largely correct. MAS did not replace selective breeding; instead, it became a complementary tool for the minority of traits where major QTL exist. The two frameworks coexisted, with MAS adding marker information to the pedigree-based BLUP models rather than supplanting them.
The turning point came when high-density single nucleotide polymorphism (SNP) arrays and later genotyping-by-sequencing made it affordable to survey thousands of markers across the genome. Genomic Selection, first developed in dairy cattle and rapidly adopted in aquaculture, abandoned the search for individual QTL altogether. Instead, it uses all markers simultaneously in a statistical model to predict the breeding value of each animal. The key insight is that with enough markers, every QTL—even those of tiny effect—will be in linkage disequilibrium with at least one marker. The breeder no longer needs to know which markers are causal; the model learns the association between each marker and the trait from a training population of genotyped and phenotyped individuals, then applies those weights to predict the genetic merit of selection candidates that have only genotype data. This framework transformed the field because it captured both the additive genetic variance that Selective Breeding had relied on and the small-effect QTL that MAS had missed. Genomic selection does not discard the pedigree or the phenotypic records of the Selective Breeding era; it incorporates them into a unified prediction framework. The Norwegian salmon breeding program, for example, now routinely uses genomic prediction alongside traditional BLUP, achieving 20–30% higher accuracy for growth traits than pedigree-based methods alone.
Today, the three frameworks are not arranged in a simple chronological sequence but form a layered toolkit. Selective Breeding remains the default for species or programs where genotyping costs are prohibitive or where the breeding population is too small to build a reliable genomic prediction model. MAS continues to be used for traits with known major genes, such as sex determination or specific disease resistance alleles. Genomic Selection is the leading framework for well-resourced programs in salmon, tilapia, shrimp, and several marine finfish species, precisely because it makes the most complete use of available data.
Yet the coexistence is not entirely harmonious. A central debate concerns the role of non-additive genetic variance—dominance and epistasis. The Selective Breeding framework and the standard genomic prediction models both assume that additive effects are sufficient for prediction. However, empirical studies in aquaculture species have found that non-additive variance can be substantial, particularly for traits like survival and stress tolerance. Some researchers argue that genomic models should incorporate dominance effects or even whole-genome epistatic terms, while others counter that additive models are robust enough for practical selection and that adding complexity risks overfitting. This disagreement reflects a deeper tension: should the field prioritize prediction accuracy, or should it aim for a mechanistic understanding of genetic architecture?
A second frontier is the integration of new data types. Transcriptomics, proteomics, and metabolomics are beginning to generate molecular phenotypes that could supplement traditional trait measurements. Some groups are exploring whether gene expression levels or metabolite concentrations can serve as intermediate predictors of complex traits, effectively building a bridge between the genome and the phenotype. This multi-omics approach does not replace genomic selection but extends its logic: if more information improves prediction, then any measurable molecular feature is a candidate predictor.
Despite their differences, the three frameworks share a foundational commitment to quantitative genetics. All assume that genetic improvement is possible through selection on heritable variation, and all rely on statistical models that partition variance into genetic and environmental components. They agree that additive genetic variance is the primary driver of response to selection, and that accurate phenotyping is the rate-limiting step for any breeding program. Where they disagree is on the optimal resolution of genetic information. Selective Breeding is content with pedigree-based relationships; MAS insists on identifying causal or linked markers; Genomic Selection argues that genome-wide marker density is both necessary and sufficient. The debate over non-additive variance is, at root, a disagreement about whether the additive model is a convenient approximation or a genuine biological description. The field has not resolved this question, but the practical success of genomic selection has shifted the burden of proof onto those who argue that more complex models are worth the cost.
Aquaculture genetics has moved from inferring the invisible to reading the genome, but it has not left its earlier tools behind. The pedigree, the scale, the growth tank, and the spreadsheet of phenotypic records remain essential. The frameworks that defined each era now coexist in a layered practice where the choice of method depends on the species, the trait, the budget, and the question being asked. The trajectory is not a ladder of replacement but an expanding repertoire of ways to see and shape the genetic potential of farmed aquatic organisms.